Pileus Git - ~andy/linux/blob - fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
  99                                            u64 len, u64 orig_start,
 100                                            u64 block_start, u64 block_len,
 101                                            u64 orig_block_len, int type);
 102
 103 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 104                                      struct inode *inode,  struct inode *dir,
 105                                      const struct qstr *qstr)
 106 {
 107         int err;
 108
 109         err = btrfs_init_acl(trans, inode, dir);
 110         if (!err)
 111                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 112         return err;
 113 }
 114
 115 /*
 116  * this does all the hard work for inserting an inline extent into
 117  * the btree.  The caller should have done a btrfs_drop_extents so that
 118  * no overlapping inline items exist in the btree
 119  */
 120 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 121                                 struct btrfs_root *root, struct inode *inode,
 122                                 u64 start, size_t size, size_t compressed_size,
 123                                 int compress_type,
 124                                 struct page **compressed_pages)
 125 {
 126         struct btrfs_key key;
 127         struct btrfs_path *path;
 128         struct extent_buffer *leaf;
 129         struct page *page = NULL;
 130         char *kaddr;
 131         unsigned long ptr;
 132         struct btrfs_file_extent_item *ei;
 133         int err = 0;
 134         int ret;
 135         size_t cur_size = size;
 136         size_t datasize;
 137         unsigned long offset;
 138
 139         if (compressed_size && compressed_pages)
 140                 cur_size = compressed_size;
 141
 142         path = btrfs_alloc_path();
 143         if (!path)
 144                 return -ENOMEM;
 145
 146         path->leave_spinning = 1;
 147
 148         key.objectid = btrfs_ino(inode);
 149         key.offset = start;
 150         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 151         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 152
 153         inode_add_bytes(inode, size);
 154         ret = btrfs_insert_empty_item(trans, root, path, &key,
 155                                       datasize);
 156         if (ret) {
 157                 err = ret;
 158                 goto fail;
 159         }
 160         leaf = path->nodes[0];
 161         ei = btrfs_item_ptr(leaf, path->slots[0],
 162                             struct btrfs_file_extent_item);
 163         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 164         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 165         btrfs_set_file_extent_encryption(leaf, ei, 0);
 166         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 167         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 168         ptr = btrfs_file_extent_inline_start(ei);
 169
 170         if (compress_type != BTRFS_COMPRESS_NONE) {
 171                 struct page *cpage;
 172                 int i = 0;
 173                 while (compressed_size > 0) {
 174                         cpage = compressed_pages[i];
 175                         cur_size = min_t(unsigned long, compressed_size,
 176                                        PAGE_CACHE_SIZE);
 177
 178                         kaddr = kmap_atomic(cpage);
 179                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 180                         kunmap_atomic(kaddr);
 181
 182                         i++;
 183                         ptr += cur_size;
 184                         compressed_size -= cur_size;
 185                 }
 186                 btrfs_set_file_extent_compression(leaf, ei,
 187                                                   compress_type);
 188         } else {
 189                 page = find_get_page(inode->i_mapping,
 190                                      start >> PAGE_CACHE_SHIFT);
 191                 btrfs_set_file_extent_compression(leaf, ei, 0);
 192                 kaddr = kmap_atomic(page);
 193                 offset = start & (PAGE_CACHE_SIZE - 1);
 194                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 195                 kunmap_atomic(kaddr);
 196                 page_cache_release(page);
 197         }
 198         btrfs_mark_buffer_dirty(leaf);
 199         btrfs_free_path(path);
 200
 201         /*
 202          * we're an inline extent, so nobody can
 203          * extend the file past i_size without locking
 204          * a page we already have locked.
 205          *
 206          * We must do any isize and inode updates
 207          * before we unlock the pages.  Otherwise we
 208          * could end up racing with unlink.
 209          */
 210         BTRFS_I(inode)->disk_i_size = inode->i_size;
 211         ret = btrfs_update_inode(trans, root, inode);
 212
 213         return ret;
 214 fail:
 215         btrfs_free_path(path);
 216         return err;
 217 }
 218
 219
 220 /*
 221  * conditionally insert an inline extent into the file.  This
 222  * does the checks required to make sure the data is small enough
 223  * to fit as an inline extent.
 224  */
 225 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 226                                  struct btrfs_root *root,
 227                                  struct inode *inode, u64 start, u64 end,
 228                                  size_t compressed_size, int compress_type,
 229                                  struct page **compressed_pages)
 230 {
 231         u64 isize = i_size_read(inode);
 232         u64 actual_end = min(end + 1, isize);
 233         u64 inline_len = actual_end - start;
 234         u64 aligned_end = (end + root->sectorsize - 1) &
 235                         ~((u64)root->sectorsize - 1);
 236         u64 data_len = inline_len;
 237         int ret;
 238
 239         if (compressed_size)
 240                 data_len = compressed_size;
 241
 242         if (start > 0 ||
 243             actual_end >= PAGE_CACHE_SIZE ||
 244             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 245             (!compressed_size &&
 246             (actual_end & (root->sectorsize - 1)) == 0) ||
 247             end + 1 < isize ||
 248             data_len > root->fs_info->max_inline) {
 249                 return 1;
 250         }
 251
 252         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 253         if (ret)
 254                 return ret;
 255
 256         if (isize > actual_end)
 257                 inline_len = min_t(u64, isize, actual_end);
 258         ret = insert_inline_extent(trans, root, inode, start,
 259                                    inline_len, compressed_size,
 260                                    compress_type, compressed_pages);
 261         if (ret && ret != -ENOSPC) {
 262                 btrfs_abort_transaction(trans, root, ret);
 263                 return ret;
 264         } else if (ret == -ENOSPC) {
 265                 return 1;
 266         }
 267
 268         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 269         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 270         return 0;
 271 }
 272
 273 struct async_extent {
 274         u64 start;
 275         u64 ram_size;
 276         u64 compressed_size;
 277         struct page **pages;
 278         unsigned long nr_pages;
 279         int compress_type;
 280         struct list_head list;
 281 };
 282
 283 struct async_cow {
 284         struct inode *inode;
 285         struct btrfs_root *root;
 286         struct page *locked_page;
 287         u64 start;
 288         u64 end;
 289         struct list_head extents;
 290         struct btrfs_work work;
 291 };
 292
 293 static noinline int add_async_extent(struct async_cow *cow,
 294                                      u64 start, u64 ram_size,
 295                                      u64 compressed_size,
 296                                      struct page **pages,
 297                                      unsigned long nr_pages,
 298                                      int compress_type)
 299 {
 300         struct async_extent *async_extent;
 301
 302         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 303         BUG_ON(!async_extent); /* -ENOMEM */
 304         async_extent->start = start;
 305         async_extent->ram_size = ram_size;
 306         async_extent->compressed_size = compressed_size;
 307         async_extent->pages = pages;
 308         async_extent->nr_pages = nr_pages;
 309         async_extent->compress_type = compress_type;
 310         list_add_tail(&async_extent->list, &cow->extents);
 311         return 0;
 312 }
 313
 314 /*
 315  * we create compressed extents in two phases.  The first
 316  * phase compresses a range of pages that have already been
 317  * locked (both pages and state bits are locked).
 318  *
 319  * This is done inside an ordered work queue, and the compression
 320  * is spread across many cpus.  The actual IO submission is step
 321  * two, and the ordered work queue takes care of making sure that
 322  * happens in the same order things were put onto the queue by
 323  * writepages and friends.
 324  *
 325  * If this code finds it can't get good compression, it puts an
 326  * entry onto the work queue to write the uncompressed bytes.  This
 327  * makes sure that both compressed inodes and uncompressed inodes
 328  * are written in the same order that the flusher thread sent them
 329  * down.
 330  */
 331 static noinline int compress_file_range(struct inode *inode,
 332                                         struct page *locked_page,
 333                                         u64 start, u64 end,
 334                                         struct async_cow *async_cow,
 335                                         int *num_added)
 336 {
 337         struct btrfs_root *root = BTRFS_I(inode)->root;
 338         struct btrfs_trans_handle *trans;
 339         u64 num_bytes;
 340         u64 blocksize = root->sectorsize;
 341         u64 actual_end;
 342         u64 isize = i_size_read(inode);
 343         int ret = 0;
 344         struct page **pages = NULL;
 345         unsigned long nr_pages;
 346         unsigned long nr_pages_ret = 0;
 347         unsigned long total_compressed = 0;
 348         unsigned long total_in = 0;
 349         unsigned long max_compressed = 128 * 1024;
 350         unsigned long max_uncompressed = 128 * 1024;
 351         int i;
 352         int will_compress;
 353         int compress_type = root->fs_info->compress_type;
 354
 355         /* if this is a small write inside eof, kick off a defrag */
 356         if ((end - start + 1) < 16 * 1024 &&
 357             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 358                 btrfs_add_inode_defrag(NULL, inode);
 359
 360         actual_end = min_t(u64, isize, end + 1);
 361 again:
 362         will_compress = 0;
 363         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 364         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 365
 366         /*
 367          * we don't want to send crud past the end of i_size through
 368          * compression, that's just a waste of CPU time.  So, if the
 369          * end of the file is before the start of our current
 370          * requested range of bytes, we bail out to the uncompressed
 371          * cleanup code that can deal with all of this.
 372          *
 373          * It isn't really the fastest way to fix things, but this is a
 374          * very uncommon corner.
 375          */
 376         if (actual_end <= start)
 377                 goto cleanup_and_bail_uncompressed;
 378
 379         total_compressed = actual_end - start;
 380
 381         /* we want to make sure that amount of ram required to uncompress
 382          * an extent is reasonable, so we limit the total size in ram
 383          * of a compressed extent to 128k.  This is a crucial number
 384          * because it also controls how easily we can spread reads across
 385          * cpus for decompression.
 386          *
 387          * We also want to make sure the amount of IO required to do
 388          * a random read is reasonably small, so we limit the size of
 389          * a compressed extent to 128k.
 390          */
 391         total_compressed = min(total_compressed, max_uncompressed);
 392         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 393         num_bytes = max(blocksize,  num_bytes);
 394         total_in = 0;
 395         ret = 0;
 396
 397         /*
 398          * we do compression for mount -o compress and when the
 399          * inode has not been flagged as nocompress.  This flag can
 400          * change at any time if we discover bad compression ratios.
 401          */
 402         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 403             (btrfs_test_opt(root, COMPRESS) ||
 404              (BTRFS_I(inode)->force_compress) ||
 405              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 406                 WARN_ON(pages);
 407                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 408                 if (!pages) {
 409                         /* just bail out to the uncompressed code */
 410                         goto cont;
 411                 }
 412
 413                 if (BTRFS_I(inode)->force_compress)
 414                         compress_type = BTRFS_I(inode)->force_compress;
 415
 416                 ret = btrfs_compress_pages(compress_type,
 417                                            inode->i_mapping, start,
 418                                            total_compressed, pages,
 419                                            nr_pages, &nr_pages_ret,
 420                                            &total_in,
 421                                            &total_compressed,
 422                                            max_compressed);
 423
 424                 if (!ret) {
 425                         unsigned long offset = total_compressed &
 426                                 (PAGE_CACHE_SIZE - 1);
 427                         struct page *page = pages[nr_pages_ret - 1];
 428                         char *kaddr;
 429
 430                         /* zero the tail end of the last page, we might be
 431                          * sending it down to disk
 432                          */
 433                         if (offset) {
 434                                 kaddr = kmap_atomic(page);
 435                                 memset(kaddr + offset, 0,
 436                                        PAGE_CACHE_SIZE - offset);
 437                                 kunmap_atomic(kaddr);
 438                         }
 439                         will_compress = 1;
 440                 }
 441         }
 442 cont:
 443         if (start == 0) {
 444                 trans = btrfs_join_transaction(root);
 445                 if (IS_ERR(trans)) {
 446                         ret = PTR_ERR(trans);
 447                         trans = NULL;
 448                         goto cleanup_and_out;
 449                 }
 450                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 451
 452                 /* lets try to make an inline extent */
 453                 if (ret || total_in < (actual_end - start)) {
 454                         /* we didn't compress the entire range, try
 455                          * to make an uncompressed inline extent.
 456                          */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end, 0, 0, NULL);
 459                 } else {
 460                         /* try making a compressed inline extent */
 461                         ret = cow_file_range_inline(trans, root, inode,
 462                                                     start, end,
 463                                                     total_compressed,
 464                                                     compress_type, pages);
 465                 }
 466                 if (ret <= 0) {
 467                         /*
 468                          * inline extent creation worked or returned error,
 469                          * we don't need to create any more async work items.
 470                          * Unlock and free up our temp pages.
 471                          */
 472                         extent_clear_unlock_delalloc(inode,
 473                              &BTRFS_I(inode)->io_tree,
 474                              start, end, NULL,
 475                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 476                              EXTENT_CLEAR_DELALLOC |
 477                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 478
 479                         btrfs_end_transaction(trans, root);
 480                         goto free_pages_out;
 481                 }
 482                 btrfs_end_transaction(trans, root);
 483         }
 484
 485         if (will_compress) {
 486                 /*
 487                  * we aren't doing an inline extent round the compressed size
 488                  * up to a block size boundary so the allocator does sane
 489                  * things
 490                  */
 491                 total_compressed = (total_compressed + blocksize - 1) &
 492                         ~(blocksize - 1);
 493
 494                 /*
 495                  * one last check to make sure the compression is really a
 496                  * win, compare the page count read with the blocks on disk
 497                  */
 498                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 499                         ~(PAGE_CACHE_SIZE - 1);
 500                 if (total_compressed >= total_in) {
 501                         will_compress = 0;
 502                 } else {
 503                         num_bytes = total_in;
 504                 }
 505         }
 506         if (!will_compress && pages) {
 507                 /*
 508                  * the compression code ran but failed to make things smaller,
 509                  * free any pages it allocated and our page pointer array
 510                  */
 511                 for (i = 0; i < nr_pages_ret; i++) {
 512                         WARN_ON(pages[i]->mapping);
 513                         page_cache_release(pages[i]);
 514                 }
 515                 kfree(pages);
 516                 pages = NULL;
 517                 total_compressed = 0;
 518                 nr_pages_ret = 0;
 519
 520                 /* flag the file so we don't compress in the future */
 521                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 522                     !(BTRFS_I(inode)->force_compress)) {
 523                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 524                 }
 525         }
 526         if (will_compress) {
 527                 *num_added += 1;
 528
 529                 /* the async work queues will take care of doing actual
 530                  * allocation on disk for these compressed pages,
 531                  * and will submit them to the elevator.
 532                  */
 533                 add_async_extent(async_cow, start, num_bytes,
 534                                  total_compressed, pages, nr_pages_ret,
 535                                  compress_type);
 536
 537                 if (start + num_bytes < end) {
 538                         start += num_bytes;
 539                         pages = NULL;
 540                         cond_resched();
 541                         goto again;
 542                 }
 543         } else {
 544 cleanup_and_bail_uncompressed:
 545                 /*
 546                  * No compression, but we still need to write the pages in
 547                  * the file we've been given so far.  redirty the locked
 548                  * page if it corresponds to our extent and set things up
 549                  * for the async work queue to run cow_file_range to do
 550                  * the normal delalloc dance
 551                  */
 552                 if (page_offset(locked_page) >= start &&
 553                     page_offset(locked_page) <= end) {
 554                         __set_page_dirty_nobuffers(locked_page);
 555                         /* unlocked later on in the async handlers */
 556                 }
 557                 add_async_extent(async_cow, start, end - start + 1,
 558                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 559                 *num_added += 1;
 560         }
 561
 562 out:
 563         return ret;
 564
 565 free_pages_out:
 566         for (i = 0; i < nr_pages_ret; i++) {
 567                 WARN_ON(pages[i]->mapping);
 568                 page_cache_release(pages[i]);
 569         }
 570         kfree(pages);
 571
 572         goto out;
 573
 574 cleanup_and_out:
 575         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 576                                      start, end, NULL,
 577                                      EXTENT_CLEAR_UNLOCK_PAGE |
 578                                      EXTENT_CLEAR_DIRTY |
 579                                      EXTENT_CLEAR_DELALLOC |
 580                                      EXTENT_SET_WRITEBACK |
 581                                      EXTENT_END_WRITEBACK);
 582         if (!trans || IS_ERR(trans))
 583                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 584         else
 585                 btrfs_abort_transaction(trans, root, ret);
 586         goto free_pages_out;
 587 }
 588
 589 /*
 590  * phase two of compressed writeback.  This is the ordered portion
 591  * of the code, which only gets called in the order the work was
 592  * queued.  We walk all the async extents created by compress_file_range
 593  * and send them down to the disk.
 594  */
 595 static noinline int submit_compressed_extents(struct inode *inode,
 596                                               struct async_cow *async_cow)
 597 {
 598         struct async_extent *async_extent;
 599         u64 alloc_hint = 0;
 600         struct btrfs_trans_handle *trans;
 601         struct btrfs_key ins;
 602         struct extent_map *em;
 603         struct btrfs_root *root = BTRFS_I(inode)->root;
 604         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 605         struct extent_io_tree *io_tree;
 606         int ret = 0;
 607
 608         if (list_empty(&async_cow->extents))
 609                 return 0;
 610
 611
 612         while (!list_empty(&async_cow->extents)) {
 613                 async_extent = list_entry(async_cow->extents.next,
 614                                           struct async_extent, list);
 615                 list_del(&async_extent->list);
 616
 617                 io_tree = &BTRFS_I(inode)->io_tree;
 618
 619 retry:
 620                 /* did the compression code fall back to uncompressed IO? */
 621                 if (!async_extent->pages) {
 622                         int page_started = 0;
 623                         unsigned long nr_written = 0;
 624
 625                         lock_extent(io_tree, async_extent->start,
 626                                          async_extent->start +
 627                                          async_extent->ram_size - 1);
 628
 629                         /* allocate blocks */
 630                         ret = cow_file_range(inode, async_cow->locked_page,
 631                                              async_extent->start,
 632                                              async_extent->start +
 633                                              async_extent->ram_size - 1,
 634                                              &page_started, &nr_written, 0);
 635
 636                         /* JDM XXX */
 637
 638                         /*
 639                          * if page_started, cow_file_range inserted an
 640                          * inline extent and took care of all the unlocking
 641                          * and IO for us.  Otherwise, we need to submit
 642                          * all those pages down to the drive.
 643                          */
 644                         if (!page_started && !ret)
 645                                 extent_write_locked_range(io_tree,
 646                                                   inode, async_extent->start,
 647                                                   async_extent->start +
 648                                                   async_extent->ram_size - 1,
 649                                                   btrfs_get_extent,
 650                                                   WB_SYNC_ALL);
 651                         kfree(async_extent);
 652                         cond_resched();
 653                         continue;
 654                 }
 655
 656                 lock_extent(io_tree, async_extent->start,
 657                             async_extent->start + async_extent->ram_size - 1);
 658
 659                 trans = btrfs_join_transaction(root);
 660                 if (IS_ERR(trans)) {
 661                         ret = PTR_ERR(trans);
 662                 } else {
 663                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 664                         ret = btrfs_reserve_extent(trans, root,
 665                                            async_extent->compressed_size,
 666                                            async_extent->compressed_size,
 667                                            0, alloc_hint, &ins, 1);
 668                         if (ret && ret != -ENOSPC)
 669                                 btrfs_abort_transaction(trans, root, ret);
 670                         btrfs_end_transaction(trans, root);
 671                 }
 672
 673                 if (ret) {
 674                         int i;
 675                         for (i = 0; i < async_extent->nr_pages; i++) {
 676                                 WARN_ON(async_extent->pages[i]->mapping);
 677                                 page_cache_release(async_extent->pages[i]);
 678                         }
 679                         kfree(async_extent->pages);
 680                         async_extent->nr_pages = 0;
 681                         async_extent->pages = NULL;
 682                         unlock_extent(io_tree, async_extent->start,
 683                                       async_extent->start +
 684                                       async_extent->ram_size - 1);
 685                         if (ret == -ENOSPC)
 686                                 goto retry;
 687                         goto out_free; /* JDM: Requeue? */
 688                 }
 689
 690                 /*
 691                  * here we're doing allocation and writeback of the
 692                  * compressed pages
 693                  */
 694                 btrfs_drop_extent_cache(inode, async_extent->start,
 695                                         async_extent->start +
 696                                         async_extent->ram_size - 1, 0);
 697
 698                 em = alloc_extent_map();
 699                 BUG_ON(!em); /* -ENOMEM */
 700                 em->start = async_extent->start;
 701                 em->len = async_extent->ram_size;
 702                 em->orig_start = em->start;
 703
 704                 em->block_start = ins.objectid;
 705                 em->block_len = ins.offset;
 706                 em->orig_block_len = ins.offset;
 707                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 708                 em->compress_type = async_extent->compress_type;
 709                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 710                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 711                 em->generation = -1;
 712
 713                 while (1) {
 714                         write_lock(&em_tree->lock);
 715                         ret = add_extent_mapping(em_tree, em);
 716                         if (!ret)
 717                                 list_move(&em->list,
 718                                           &em_tree->modified_extents);
 719                         write_unlock(&em_tree->lock);
 720                         if (ret != -EEXIST) {
 721                                 free_extent_map(em);
 722                                 break;
 723                         }
 724                         btrfs_drop_extent_cache(inode, async_extent->start,
 725                                                 async_extent->start +
 726                                                 async_extent->ram_size - 1, 0);
 727                 }
 728
 729                 ret = btrfs_add_ordered_extent_compress(inode,
 730                                                 async_extent->start,
 731                                                 ins.objectid,
 732                                                 async_extent->ram_size,
 733                                                 ins.offset,
 734                                                 BTRFS_ORDERED_COMPRESSED,
 735                                                 async_extent->compress_type);
 736                 BUG_ON(ret); /* -ENOMEM */
 737
 738                 /*
 739                  * clear dirty, set writeback and unlock the pages.
 740                  */
 741                 extent_clear_unlock_delalloc(inode,
 742                                 &BTRFS_I(inode)->io_tree,
 743                                 async_extent->start,
 744                                 async_extent->start +
 745                                 async_extent->ram_size - 1,
 746                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 747                                 EXTENT_CLEAR_UNLOCK |
 748                                 EXTENT_CLEAR_DELALLOC |
 749                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 750
 751                 ret = btrfs_submit_compressed_write(inode,
 752                                     async_extent->start,
 753                                     async_extent->ram_size,
 754                                     ins.objectid,
 755                                     ins.offset, async_extent->pages,
 756                                     async_extent->nr_pages);
 757
 758                 BUG_ON(ret); /* -ENOMEM */
 759                 alloc_hint = ins.objectid + ins.offset;
 760                 kfree(async_extent);
 761                 cond_resched();
 762         }
 763         ret = 0;
 764 out:
 765         return ret;
 766 out_free:
 767         kfree(async_extent);
 768         goto out;
 769 }
 770
 771 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 772                                       u64 num_bytes)
 773 {
 774         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 775         struct extent_map *em;
 776         u64 alloc_hint = 0;
 777
 778         read_lock(&em_tree->lock);
 779         em = search_extent_mapping(em_tree, start, num_bytes);
 780         if (em) {
 781                 /*
 782                  * if block start isn't an actual block number then find the
 783                  * first block in this inode and use that as a hint.  If that
 784                  * block is also bogus then just don't worry about it.
 785                  */
 786                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 787                         free_extent_map(em);
 788                         em = search_extent_mapping(em_tree, 0, 0);
 789                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 790                                 alloc_hint = em->block_start;
 791                         if (em)
 792                                 free_extent_map(em);
 793                 } else {
 794                         alloc_hint = em->block_start;
 795                         free_extent_map(em);
 796                 }
 797         }
 798         read_unlock(&em_tree->lock);
 799
 800         return alloc_hint;
 801 }
 802
 803 /*
 804  * when extent_io.c finds a delayed allocation range in the file,
 805  * the call backs end up in this code.  The basic idea is to
 806  * allocate extents on disk for the range, and create ordered data structs
 807  * in ram to track those extents.
 808  *
 809  * locked_page is the page that writepage had locked already.  We use
 810  * it to make sure we don't do extra locks or unlocks.
 811  *
 812  * *page_started is set to one if we unlock locked_page and do everything
 813  * required to start IO on it.  It may be clean and already done with
 814  * IO when we return.
 815  */
 816 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 817                                      struct inode *inode,
 818                                      struct btrfs_root *root,
 819                                      struct page *locked_page,
 820                                      u64 start, u64 end, int *page_started,
 821                                      unsigned long *nr_written,
 822                                      int unlock)
 823 {
 824         u64 alloc_hint = 0;
 825         u64 num_bytes;
 826         unsigned long ram_size;
 827         u64 disk_num_bytes;
 828         u64 cur_alloc_size;
 829         u64 blocksize = root->sectorsize;
 830         struct btrfs_key ins;
 831         struct extent_map *em;
 832         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 833         int ret = 0;
 834
 835         BUG_ON(btrfs_is_free_space_inode(inode));
 836
 837         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 838         num_bytes = max(blocksize,  num_bytes);
 839         disk_num_bytes = num_bytes;
 840
 841         /* if this is a small write inside eof, kick off defrag */
 842         if (num_bytes < 64 * 1024 &&
 843             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 844                 btrfs_add_inode_defrag(trans, inode);
 845
 846         if (start == 0) {
 847                 /* lets try to make an inline extent */
 848                 ret = cow_file_range_inline(trans, root, inode,
 849                                             start, end, 0, 0, NULL);
 850                 if (ret == 0) {
 851                         extent_clear_unlock_delalloc(inode,
 852                                      &BTRFS_I(inode)->io_tree,
 853                                      start, end, NULL,
 854                                      EXTENT_CLEAR_UNLOCK_PAGE |
 855                                      EXTENT_CLEAR_UNLOCK |
 856                                      EXTENT_CLEAR_DELALLOC |
 857                                      EXTENT_CLEAR_DIRTY |
 858                                      EXTENT_SET_WRITEBACK |
 859                                      EXTENT_END_WRITEBACK);
 860
 861                         *nr_written = *nr_written +
 862                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 863                         *page_started = 1;
 864                         goto out;
 865                 } else if (ret < 0) {
 866                         btrfs_abort_transaction(trans, root, ret);
 867                         goto out_unlock;
 868                 }
 869         }
 870
 871         BUG_ON(disk_num_bytes >
 872                btrfs_super_total_bytes(root->fs_info->super_copy));
 873
 874         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 875         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 876
 877         while (disk_num_bytes > 0) {
 878                 unsigned long op;
 879
 880                 cur_alloc_size = disk_num_bytes;
 881                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 882                                            root->sectorsize, 0, alloc_hint,
 883                                            &ins, 1);
 884                 if (ret < 0) {
 885                         btrfs_abort_transaction(trans, root, ret);
 886                         goto out_unlock;
 887                 }
 888
 889                 em = alloc_extent_map();
 890                 BUG_ON(!em); /* -ENOMEM */
 891                 em->start = start;
 892                 em->orig_start = em->start;
 893                 ram_size = ins.offset;
 894                 em->len = ins.offset;
 895
 896                 em->block_start = ins.objectid;
 897                 em->block_len = ins.offset;
 898                 em->orig_block_len = ins.offset;
 899                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 900                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 901                 em->generation = -1;
 902
 903                 while (1) {
 904                         write_lock(&em_tree->lock);
 905                         ret = add_extent_mapping(em_tree, em);
 906                         if (!ret)
 907                                 list_move(&em->list,
 908                                           &em_tree->modified_extents);
 909                         write_unlock(&em_tree->lock);
 910                         if (ret != -EEXIST) {
 911                                 free_extent_map(em);
 912                                 break;
 913                         }
 914                         btrfs_drop_extent_cache(inode, start,
 915                                                 start + ram_size - 1, 0);
 916                 }
 917
 918                 cur_alloc_size = ins.offset;
 919                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 920                                                ram_size, cur_alloc_size, 0);
 921                 BUG_ON(ret); /* -ENOMEM */
 922
 923                 if (root->root_key.objectid ==
 924                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 925                         ret = btrfs_reloc_clone_csums(inode, start,
 926                                                       cur_alloc_size);
 927                         if (ret) {
 928                                 btrfs_abort_transaction(trans, root, ret);
 929                                 goto out_unlock;
 930                         }
 931                 }
 932
 933                 if (disk_num_bytes < cur_alloc_size)
 934                         break;
 935
 936                 /* we're not doing compressed IO, don't unlock the first
 937                  * page (which the caller expects to stay locked), don't
 938                  * clear any dirty bits and don't set any writeback bits
 939                  *
 940                  * Do set the Private2 bit so we know this page was properly
 941                  * setup for writepage
 942                  */
 943                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 944                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 945                         EXTENT_SET_PRIVATE2;
 946
 947                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 948                                              start, start + ram_size - 1,
 949                                              locked_page, op);
 950                 disk_num_bytes -= cur_alloc_size;
 951                 num_bytes -= cur_alloc_size;
 952                 alloc_hint = ins.objectid + ins.offset;
 953                 start += cur_alloc_size;
 954         }
 955 out:
 956         return ret;
 957
 958 out_unlock:
 959         extent_clear_unlock_delalloc(inode,
 960                      &BTRFS_I(inode)->io_tree,
 961                      start, end, locked_page,
 962                      EXTENT_CLEAR_UNLOCK_PAGE |
 963                      EXTENT_CLEAR_UNLOCK |
 964                      EXTENT_CLEAR_DELALLOC |
 965                      EXTENT_CLEAR_DIRTY |
 966                      EXTENT_SET_WRITEBACK |
 967                      EXTENT_END_WRITEBACK);
 968
 969         goto out;
 970 }
 971
 972 static noinline int cow_file_range(struct inode *inode,
 973                                    struct page *locked_page,
 974                                    u64 start, u64 end, int *page_started,
 975                                    unsigned long *nr_written,
 976                                    int unlock)
 977 {
 978         struct btrfs_trans_handle *trans;
 979         struct btrfs_root *root = BTRFS_I(inode)->root;
 980         int ret;
 981
 982         trans = btrfs_join_transaction(root);
 983         if (IS_ERR(trans)) {
 984                 extent_clear_unlock_delalloc(inode,
 985                              &BTRFS_I(inode)->io_tree,
 986                              start, end, locked_page,
 987                              EXTENT_CLEAR_UNLOCK_PAGE |
 988                              EXTENT_CLEAR_UNLOCK |
 989                              EXTENT_CLEAR_DELALLOC |
 990                              EXTENT_CLEAR_DIRTY |
 991                              EXTENT_SET_WRITEBACK |
 992                              EXTENT_END_WRITEBACK);
 993                 return PTR_ERR(trans);
 994         }
 995         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 996
 997         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
 998                                page_started, nr_written, unlock);
 999
1000         btrfs_end_transaction(trans, root);
1001
1002         return ret;
1003 }
1004
1005 /*
1006  * work queue call back to started compression on a file and pages
1007  */
1008 static noinline void async_cow_start(struct btrfs_work *work)
1009 {
1010         struct async_cow *async_cow;
1011         int num_added = 0;
1012         async_cow = container_of(work, struct async_cow, work);
1013
1014         compress_file_range(async_cow->inode, async_cow->locked_page,
1015                             async_cow->start, async_cow->end, async_cow,
1016                             &num_added);
1017         if (num_added == 0) {
1018                 btrfs_add_delayed_iput(async_cow->inode);
1019                 async_cow->inode = NULL;
1020         }
1021 }
1022
1023 /*
1024  * work queue call back to submit previously compressed pages
1025  */
1026 static noinline void async_cow_submit(struct btrfs_work *work)
1027 {
1028         struct async_cow *async_cow;
1029         struct btrfs_root *root;
1030         unsigned long nr_pages;
1031
1032         async_cow = container_of(work, struct async_cow, work);
1033
1034         root = async_cow->root;
1035         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1036                 PAGE_CACHE_SHIFT;
1037
1038         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1039             5 * 1024 * 1024 &&
1040             waitqueue_active(&root->fs_info->async_submit_wait))
1041                 wake_up(&root->fs_info->async_submit_wait);
1042
1043         if (async_cow->inode)
1044                 submit_compressed_extents(async_cow->inode, async_cow);
1045 }
1046
1047 static noinline void async_cow_free(struct btrfs_work *work)
1048 {
1049         struct async_cow *async_cow;
1050         async_cow = container_of(work, struct async_cow, work);
1051         if (async_cow->inode)
1052                 btrfs_add_delayed_iput(async_cow->inode);
1053         kfree(async_cow);
1054 }
1055
1056 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1057                                 u64 start, u64 end, int *page_started,
1058                                 unsigned long *nr_written)
1059 {
1060         struct async_cow *async_cow;
1061         struct btrfs_root *root = BTRFS_I(inode)->root;
1062         unsigned long nr_pages;
1063         u64 cur_end;
1064         int limit = 10 * 1024 * 1024;
1065
1066         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1067                          1, 0, NULL, GFP_NOFS);
1068         while (start < end) {
1069                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1070                 BUG_ON(!async_cow); /* -ENOMEM */
1071                 async_cow->inode = igrab(inode);
1072                 async_cow->root = root;
1073                 async_cow->locked_page = locked_page;
1074                 async_cow->start = start;
1075
1076                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1077                         cur_end = end;
1078                 else
1079                         cur_end = min(end, start + 512 * 1024 - 1);
1080
1081                 async_cow->end = cur_end;
1082                 INIT_LIST_HEAD(&async_cow->extents);
1083
1084                 async_cow->work.func = async_cow_start;
1085                 async_cow->work.ordered_func = async_cow_submit;
1086                 async_cow->work.ordered_free = async_cow_free;
1087                 async_cow->work.flags = 0;
1088
1089                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1090                         PAGE_CACHE_SHIFT;
1091                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1092
1093                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1094                                    &async_cow->work);
1095
1096                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1097                         wait_event(root->fs_info->async_submit_wait,
1098                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1099                             limit));
1100                 }
1101
1102                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1103                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1104                         wait_event(root->fs_info->async_submit_wait,
1105                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1106                            0));
1107                 }
1108
1109                 *nr_written += nr_pages;
1110                 start = cur_end + 1;
1111         }
1112         *page_started = 1;
1113         return 0;
1114 }
1115
1116 static noinline int csum_exist_in_range(struct btrfs_root *root,
1117                                         u64 bytenr, u64 num_bytes)
1118 {
1119         int ret;
1120         struct btrfs_ordered_sum *sums;
1121         LIST_HEAD(list);
1122
1123         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1124                                        bytenr + num_bytes - 1, &list, 0);
1125         if (ret == 0 && list_empty(&list))
1126                 return 0;
1127
1128         while (!list_empty(&list)) {
1129                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1130                 list_del(&sums->list);
1131                 kfree(sums);
1132         }
1133         return 1;
1134 }
1135
1136 /*
1137  * when nowcow writeback call back.  This checks for snapshots or COW copies
1138  * of the extents that exist in the file, and COWs the file as required.
1139  *
1140  * If no cow copies or snapshots exist, we write directly to the existing
1141  * blocks on disk
1142  */
1143 static noinline int run_delalloc_nocow(struct inode *inode,
1144                                        struct page *locked_page,
1145                               u64 start, u64 end, int *page_started, int force,
1146                               unsigned long *nr_written)
1147 {
1148         struct btrfs_root *root = BTRFS_I(inode)->root;
1149         struct btrfs_trans_handle *trans;
1150         struct extent_buffer *leaf;
1151         struct btrfs_path *path;
1152         struct btrfs_file_extent_item *fi;
1153         struct btrfs_key found_key;
1154         u64 cow_start;
1155         u64 cur_offset;
1156         u64 extent_end;
1157         u64 extent_offset;
1158         u64 disk_bytenr;
1159         u64 num_bytes;
1160         u64 disk_num_bytes;
1161         int extent_type;
1162         int ret, err;
1163         int type;
1164         int nocow;
1165         int check_prev = 1;
1166         bool nolock;
1167         u64 ino = btrfs_ino(inode);
1168
1169         path = btrfs_alloc_path();
1170         if (!path) {
1171                 extent_clear_unlock_delalloc(inode,
1172                              &BTRFS_I(inode)->io_tree,
1173                              start, end, locked_page,
1174                              EXTENT_CLEAR_UNLOCK_PAGE |
1175                              EXTENT_CLEAR_UNLOCK |
1176                              EXTENT_CLEAR_DELALLOC |
1177                              EXTENT_CLEAR_DIRTY |
1178                              EXTENT_SET_WRITEBACK |
1179                              EXTENT_END_WRITEBACK);
1180                 return -ENOMEM;
1181         }
1182
1183         nolock = btrfs_is_free_space_inode(inode);
1184
1185         if (nolock)
1186                 trans = btrfs_join_transaction_nolock(root);
1187         else
1188                 trans = btrfs_join_transaction(root);
1189
1190         if (IS_ERR(trans)) {
1191                 extent_clear_unlock_delalloc(inode,
1192                              &BTRFS_I(inode)->io_tree,
1193                              start, end, locked_page,
1194                              EXTENT_CLEAR_UNLOCK_PAGE |
1195                              EXTENT_CLEAR_UNLOCK |
1196                              EXTENT_CLEAR_DELALLOC |
1197                              EXTENT_CLEAR_DIRTY |
1198                              EXTENT_SET_WRITEBACK |
1199                              EXTENT_END_WRITEBACK);
1200                 btrfs_free_path(path);
1201                 return PTR_ERR(trans);
1202         }
1203
1204         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1205
1206         cow_start = (u64)-1;
1207         cur_offset = start;
1208         while (1) {
1209                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1210                                                cur_offset, 0);
1211                 if (ret < 0) {
1212                         btrfs_abort_transaction(trans, root, ret);
1213                         goto error;
1214                 }
1215                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1216                         leaf = path->nodes[0];
1217                         btrfs_item_key_to_cpu(leaf, &found_key,
1218                                               path->slots[0] - 1);
1219                         if (found_key.objectid == ino &&
1220                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1221                                 path->slots[0]--;
1222                 }
1223                 check_prev = 0;
1224 next_slot:
1225                 leaf = path->nodes[0];
1226                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1227                         ret = btrfs_next_leaf(root, path);
1228                         if (ret < 0) {
1229                                 btrfs_abort_transaction(trans, root, ret);
1230                                 goto error;
1231                         }
1232                         if (ret > 0)
1233                                 break;
1234                         leaf = path->nodes[0];
1235                 }
1236
1237                 nocow = 0;
1238                 disk_bytenr = 0;
1239                 num_bytes = 0;
1240                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1241
1242                 if (found_key.objectid > ino ||
1243                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1244                     found_key.offset > end)
1245                         break;
1246
1247                 if (found_key.offset > cur_offset) {
1248                         extent_end = found_key.offset;
1249                         extent_type = 0;
1250                         goto out_check;
1251                 }
1252
1253                 fi = btrfs_item_ptr(leaf, path->slots[0],
1254                                     struct btrfs_file_extent_item);
1255                 extent_type = btrfs_file_extent_type(leaf, fi);
1256
1257                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1258                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1259                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1260                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1261                         extent_end = found_key.offset +
1262                                 btrfs_file_extent_num_bytes(leaf, fi);
1263                         disk_num_bytes =
1264                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1265                         if (extent_end <= start) {
1266                                 path->slots[0]++;
1267                                 goto next_slot;
1268                         }
1269                         if (disk_bytenr == 0)
1270                                 goto out_check;
1271                         if (btrfs_file_extent_compression(leaf, fi) ||
1272                             btrfs_file_extent_encryption(leaf, fi) ||
1273                             btrfs_file_extent_other_encoding(leaf, fi))
1274                                 goto out_check;
1275                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1276                                 goto out_check;
1277                         if (btrfs_extent_readonly(root, disk_bytenr))
1278                                 goto out_check;
1279                         if (btrfs_cross_ref_exist(trans, root, ino,
1280                                                   found_key.offset -
1281                                                   extent_offset, disk_bytenr))
1282                                 goto out_check;
1283                         disk_bytenr += extent_offset;
1284                         disk_bytenr += cur_offset - found_key.offset;
1285                         num_bytes = min(end + 1, extent_end) - cur_offset;
1286                         /*
1287                          * force cow if csum exists in the range.
1288                          * this ensure that csum for a given extent are
1289                          * either valid or do not exist.
1290                          */
1291                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1292                                 goto out_check;
1293                         nocow = 1;
1294                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1295                         extent_end = found_key.offset +
1296                                 btrfs_file_extent_inline_len(leaf, fi);
1297                         extent_end = ALIGN(extent_end, root->sectorsize);
1298                 } else {
1299                         BUG_ON(1);
1300                 }
1301 out_check:
1302                 if (extent_end <= start) {
1303                         path->slots[0]++;
1304                         goto next_slot;
1305                 }
1306                 if (!nocow) {
1307                         if (cow_start == (u64)-1)
1308                                 cow_start = cur_offset;
1309                         cur_offset = extent_end;
1310                         if (cur_offset > end)
1311                                 break;
1312                         path->slots[0]++;
1313                         goto next_slot;
1314                 }
1315
1316                 btrfs_release_path(path);
1317                 if (cow_start != (u64)-1) {
1318                         ret = __cow_file_range(trans, inode, root, locked_page,
1319                                                cow_start, found_key.offset - 1,
1320                                                page_started, nr_written, 1);
1321                         if (ret) {
1322                                 btrfs_abort_transaction(trans, root, ret);
1323                                 goto error;
1324                         }
1325                         cow_start = (u64)-1;
1326                 }
1327
1328                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1329                         struct extent_map *em;
1330                         struct extent_map_tree *em_tree;
1331                         em_tree = &BTRFS_I(inode)->extent_tree;
1332                         em = alloc_extent_map();
1333                         BUG_ON(!em); /* -ENOMEM */
1334                         em->start = cur_offset;
1335                         em->orig_start = found_key.offset - extent_offset;
1336                         em->len = num_bytes;
1337                         em->block_len = num_bytes;
1338                         em->block_start = disk_bytenr;
1339                         em->orig_block_len = disk_num_bytes;
1340                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1341                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1342                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1343                         em->generation = -1;
1344                         while (1) {
1345                                 write_lock(&em_tree->lock);
1346                                 ret = add_extent_mapping(em_tree, em);
1347                                 if (!ret)
1348                                         list_move(&em->list,
1349                                                   &em_tree->modified_extents);
1350                                 write_unlock(&em_tree->lock);
1351                                 if (ret != -EEXIST) {
1352                                         free_extent_map(em);
1353                                         break;
1354                                 }
1355                                 btrfs_drop_extent_cache(inode, em->start,
1356                                                 em->start + em->len - 1, 0);
1357                         }
1358                         type = BTRFS_ORDERED_PREALLOC;
1359                 } else {
1360                         type = BTRFS_ORDERED_NOCOW;
1361                 }
1362
1363                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1364                                                num_bytes, num_bytes, type);
1365                 BUG_ON(ret); /* -ENOMEM */
1366
1367                 if (root->root_key.objectid ==
1368                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1369                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1370                                                       num_bytes);
1371                         if (ret) {
1372                                 btrfs_abort_transaction(trans, root, ret);
1373                                 goto error;
1374                         }
1375                 }
1376
1377                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1378                                 cur_offset, cur_offset + num_bytes - 1,
1379                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1380                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1381                                 EXTENT_SET_PRIVATE2);
1382                 cur_offset = extent_end;
1383                 if (cur_offset > end)
1384                         break;
1385         }
1386         btrfs_release_path(path);
1387
1388         if (cur_offset <= end && cow_start == (u64)-1) {
1389                 cow_start = cur_offset;
1390                 cur_offset = end;
1391         }
1392
1393         if (cow_start != (u64)-1) {
1394                 ret = __cow_file_range(trans, inode, root, locked_page,
1395                                        cow_start, end,
1396                                        page_started, nr_written, 1);
1397                 if (ret) {
1398                         btrfs_abort_transaction(trans, root, ret);
1399                         goto error;
1400                 }
1401         }
1402
1403 error:
1404         err = btrfs_end_transaction(trans, root);
1405         if (!ret)
1406                 ret = err;
1407
1408         if (ret && cur_offset < end)
1409                 extent_clear_unlock_delalloc(inode,
1410                              &BTRFS_I(inode)->io_tree,
1411                              cur_offset, end, locked_page,
1412                              EXTENT_CLEAR_UNLOCK_PAGE |
1413                              EXTENT_CLEAR_UNLOCK |
1414                              EXTENT_CLEAR_DELALLOC |
1415                              EXTENT_CLEAR_DIRTY |
1416                              EXTENT_SET_WRITEBACK |
1417                              EXTENT_END_WRITEBACK);
1418
1419         btrfs_free_path(path);
1420         return ret;
1421 }
1422
1423 /*
1424  * extent_io.c call back to do delayed allocation processing
1425  */
1426 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1427                               u64 start, u64 end, int *page_started,
1428                               unsigned long *nr_written)
1429 {
1430         int ret;
1431         struct btrfs_root *root = BTRFS_I(inode)->root;
1432
1433         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1434                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1435                                          page_started, 1, nr_written);
1436         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1437                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1438                                          page_started, 0, nr_written);
1439         } else if (!btrfs_test_opt(root, COMPRESS) &&
1440                    !(BTRFS_I(inode)->force_compress) &&
1441                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1442                 ret = cow_file_range(inode, locked_page, start, end,
1443                                       page_started, nr_written, 1);
1444         } else {
1445                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1446                         &BTRFS_I(inode)->runtime_flags);
1447                 ret = cow_file_range_async(inode, locked_page, start, end,
1448                                            page_started, nr_written);
1449         }
1450         return ret;
1451 }
1452
1453 static void btrfs_split_extent_hook(struct inode *inode,
1454                                     struct extent_state *orig, u64 split)
1455 {
1456         /* not delalloc, ignore it */
1457         if (!(orig->state & EXTENT_DELALLOC))
1458                 return;
1459
1460         spin_lock(&BTRFS_I(inode)->lock);
1461         BTRFS_I(inode)->outstanding_extents++;
1462         spin_unlock(&BTRFS_I(inode)->lock);
1463 }
1464
1465 /*
1466  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1467  * extents so we can keep track of new extents that are just merged onto old
1468  * extents, such as when we are doing sequential writes, so we can properly
1469  * account for the metadata space we'll need.
1470  */
1471 static void btrfs_merge_extent_hook(struct inode *inode,
1472                                     struct extent_state *new,
1473                                     struct extent_state *other)
1474 {
1475         /* not delalloc, ignore it */
1476         if (!(other->state & EXTENT_DELALLOC))
1477                 return;
1478
1479         spin_lock(&BTRFS_I(inode)->lock);
1480         BTRFS_I(inode)->outstanding_extents--;
1481         spin_unlock(&BTRFS_I(inode)->lock);
1482 }
1483
1484 /*
1485  * extent_io.c set_bit_hook, used to track delayed allocation
1486  * bytes in this file, and to maintain the list of inodes that
1487  * have pending delalloc work to be done.
1488  */
1489 static void btrfs_set_bit_hook(struct inode *inode,
1490                                struct extent_state *state, int *bits)
1491 {
1492
1493         /*
1494          * set_bit and clear bit hooks normally require _irqsave/restore
1495          * but in this case, we are only testing for the DELALLOC
1496          * bit, which is only set or cleared with irqs on
1497          */
1498         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1499                 struct btrfs_root *root = BTRFS_I(inode)->root;
1500                 u64 len = state->end + 1 - state->start;
1501                 bool do_list = !btrfs_is_free_space_inode(inode);
1502
1503                 if (*bits & EXTENT_FIRST_DELALLOC) {
1504                         *bits &= ~EXTENT_FIRST_DELALLOC;
1505                 } else {
1506                         spin_lock(&BTRFS_I(inode)->lock);
1507                         BTRFS_I(inode)->outstanding_extents++;
1508                         spin_unlock(&BTRFS_I(inode)->lock);
1509                 }
1510
1511                 spin_lock(&root->fs_info->delalloc_lock);
1512                 BTRFS_I(inode)->delalloc_bytes += len;
1513                 root->fs_info->delalloc_bytes += len;
1514                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1515                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1516                                       &root->fs_info->delalloc_inodes);
1517                 }
1518                 spin_unlock(&root->fs_info->delalloc_lock);
1519         }
1520 }
1521
1522 /*
1523  * extent_io.c clear_bit_hook, see set_bit_hook for why
1524  */
1525 static void btrfs_clear_bit_hook(struct inode *inode,
1526                                  struct extent_state *state, int *bits)
1527 {
1528         /*
1529          * set_bit and clear bit hooks normally require _irqsave/restore
1530          * but in this case, we are only testing for the DELALLOC
1531          * bit, which is only set or cleared with irqs on
1532          */
1533         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1534                 struct btrfs_root *root = BTRFS_I(inode)->root;
1535                 u64 len = state->end + 1 - state->start;
1536                 bool do_list = !btrfs_is_free_space_inode(inode);
1537
1538                 if (*bits & EXTENT_FIRST_DELALLOC) {
1539                         *bits &= ~EXTENT_FIRST_DELALLOC;
1540                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541                         spin_lock(&BTRFS_I(inode)->lock);
1542                         BTRFS_I(inode)->outstanding_extents--;
1543                         spin_unlock(&BTRFS_I(inode)->lock);
1544                 }
1545
1546                 if (*bits & EXTENT_DO_ACCOUNTING)
1547                         btrfs_delalloc_release_metadata(inode, len);
1548
1549                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1550                     && do_list)
1551                         btrfs_free_reserved_data_space(inode, len);
1552
1553                 spin_lock(&root->fs_info->delalloc_lock);
1554                 root->fs_info->delalloc_bytes -= len;
1555                 BTRFS_I(inode)->delalloc_bytes -= len;
1556
1557                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1558                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1559                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1560                 }
1561                 spin_unlock(&root->fs_info->delalloc_lock);
1562         }
1563 }
1564
1565 /*
1566  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1567  * we don't create bios that span stripes or chunks
1568  */
1569 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1570                          size_t size, struct bio *bio,
1571                          unsigned long bio_flags)
1572 {
1573         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1574         u64 logical = (u64)bio->bi_sector << 9;
1575         u64 length = 0;
1576         u64 map_length;
1577         int ret;
1578
1579         if (bio_flags & EXTENT_BIO_COMPRESSED)
1580                 return 0;
1581
1582         length = bio->bi_size;
1583         map_length = length;
1584         ret = btrfs_map_block(root->fs_info, READ, logical,
1585                               &map_length, NULL, 0);
1586         /* Will always return 0 with map_multi == NULL */
1587         BUG_ON(ret < 0);
1588         if (map_length < length + size)
1589                 return 1;
1590         return 0;
1591 }
1592
1593 /*
1594  * in order to insert checksums into the metadata in large chunks,
1595  * we wait until bio submission time.   All the pages in the bio are
1596  * checksummed and sums are attached onto the ordered extent record.
1597  *
1598  * At IO completion time the cums attached on the ordered extent record
1599  * are inserted into the btree
1600  */
1601 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1602                                     struct bio *bio, int mirror_num,
1603                                     unsigned long bio_flags,
1604                                     u64 bio_offset)
1605 {
1606         struct btrfs_root *root = BTRFS_I(inode)->root;
1607         int ret = 0;
1608
1609         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1610         BUG_ON(ret); /* -ENOMEM */
1611         return 0;
1612 }
1613
1614 /*
1615  * in order to insert checksums into the metadata in large chunks,
1616  * we wait until bio submission time.   All the pages in the bio are
1617  * checksummed and sums are attached onto the ordered extent record.
1618  *
1619  * At IO completion time the cums attached on the ordered extent record
1620  * are inserted into the btree
1621  */
1622 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1623                           int mirror_num, unsigned long bio_flags,
1624                           u64 bio_offset)
1625 {
1626         struct btrfs_root *root = BTRFS_I(inode)->root;
1627         int ret;
1628
1629         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1630         if (ret)
1631                 bio_endio(bio, ret);
1632         return ret;
1633 }
1634
1635 /*
1636  * extent_io.c submission hook. This does the right thing for csum calculation
1637  * on write, or reading the csums from the tree before a read
1638  */
1639 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1640                           int mirror_num, unsigned long bio_flags,
1641                           u64 bio_offset)
1642 {
1643         struct btrfs_root *root = BTRFS_I(inode)->root;
1644         int ret = 0;
1645         int skip_sum;
1646         int metadata = 0;
1647         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1648
1649         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1650
1651         if (btrfs_is_free_space_inode(inode))
1652                 metadata = 2;
1653
1654         if (!(rw & REQ_WRITE)) {
1655                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1656                 if (ret)
1657                         goto out;
1658
1659                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1660                         ret = btrfs_submit_compressed_read(inode, bio,
1661                                                            mirror_num,
1662                                                            bio_flags);
1663                         goto out;
1664                 } else if (!skip_sum) {
1665                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1666                         if (ret)
1667                                 goto out;
1668                 }
1669                 goto mapit;
1670         } else if (async && !skip_sum) {
1671                 /* csum items have already been cloned */
1672                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1673                         goto mapit;
1674                 /* we're doing a write, do the async checksumming */
1675                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1676                                    inode, rw, bio, mirror_num,
1677                                    bio_flags, bio_offset,
1678                                    __btrfs_submit_bio_start,
1679                                    __btrfs_submit_bio_done);
1680                 goto out;
1681         } else if (!skip_sum) {
1682                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1683                 if (ret)
1684                         goto out;
1685         }
1686
1687 mapit:
1688         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1689
1690 out:
1691         if (ret < 0)
1692                 bio_endio(bio, ret);
1693         return ret;
1694 }
1695
1696 /*
1697  * given a list of ordered sums record them in the inode.  This happens
1698  * at IO completion time based on sums calculated at bio submission time.
1699  */
1700 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1701                              struct inode *inode, u64 file_offset,
1702                              struct list_head *list)
1703 {
1704         struct btrfs_ordered_sum *sum;
1705
1706         list_for_each_entry(sum, list, list) {
1707                 btrfs_csum_file_blocks(trans,
1708                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1709         }
1710         return 0;
1711 }
1712
1713 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1714                               struct extent_state **cached_state)
1715 {
1716         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1717         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1718                                    cached_state, GFP_NOFS);
1719 }
1720
1721 /* see btrfs_writepage_start_hook for details on why this is required */
1722 struct btrfs_writepage_fixup {
1723         struct page *page;
1724         struct btrfs_work work;
1725 };
1726
1727 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1728 {
1729         struct btrfs_writepage_fixup *fixup;
1730         struct btrfs_ordered_extent *ordered;
1731         struct extent_state *cached_state = NULL;
1732         struct page *page;
1733         struct inode *inode;
1734         u64 page_start;
1735         u64 page_end;
1736         int ret;
1737
1738         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1739         page = fixup->page;
1740 again:
1741         lock_page(page);
1742         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1743                 ClearPageChecked(page);
1744                 goto out_page;
1745         }
1746
1747         inode = page->mapping->host;
1748         page_start = page_offset(page);
1749         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1750
1751         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1752                          &cached_state);
1753
1754         /* already ordered? We're done */
1755         if (PagePrivate2(page))
1756                 goto out;
1757
1758         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1759         if (ordered) {
1760                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1761                                      page_end, &cached_state, GFP_NOFS);
1762                 unlock_page(page);
1763                 btrfs_start_ordered_extent(inode, ordered, 1);
1764                 btrfs_put_ordered_extent(ordered);
1765                 goto again;
1766         }
1767
1768         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1769         if (ret) {
1770                 mapping_set_error(page->mapping, ret);
1771                 end_extent_writepage(page, ret, page_start, page_end);
1772                 ClearPageChecked(page);
1773                 goto out;
1774          }
1775
1776         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1777         ClearPageChecked(page);
1778         set_page_dirty(page);
1779 out:
1780         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1781                              &cached_state, GFP_NOFS);
1782 out_page:
1783         unlock_page(page);
1784         page_cache_release(page);
1785         kfree(fixup);
1786 }
1787
1788 /*
1789  * There are a few paths in the higher layers of the kernel that directly
1790  * set the page dirty bit without asking the filesystem if it is a
1791  * good idea.  This causes problems because we want to make sure COW
1792  * properly happens and the data=ordered rules are followed.
1793  *
1794  * In our case any range that doesn't have the ORDERED bit set
1795  * hasn't been properly setup for IO.  We kick off an async process
1796  * to fix it up.  The async helper will wait for ordered extents, set
1797  * the delalloc bit and make it safe to write the page.
1798  */
1799 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1800 {
1801         struct inode *inode = page->mapping->host;
1802         struct btrfs_writepage_fixup *fixup;
1803         struct btrfs_root *root = BTRFS_I(inode)->root;
1804
1805         /* this page is properly in the ordered list */
1806         if (TestClearPagePrivate2(page))
1807                 return 0;
1808
1809         if (PageChecked(page))
1810                 return -EAGAIN;
1811
1812         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1813         if (!fixup)
1814                 return -EAGAIN;
1815
1816         SetPageChecked(page);
1817         page_cache_get(page);
1818         fixup->work.func = btrfs_writepage_fixup_worker;
1819         fixup->page = page;
1820         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1821         return -EBUSY;
1822 }
1823
1824 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1825                                        struct inode *inode, u64 file_pos,
1826                                        u64 disk_bytenr, u64 disk_num_bytes,
1827                                        u64 num_bytes, u64 ram_bytes,
1828                                        u8 compression, u8 encryption,
1829                                        u16 other_encoding, int extent_type)
1830 {
1831         struct btrfs_root *root = BTRFS_I(inode)->root;
1832         struct btrfs_file_extent_item *fi;
1833         struct btrfs_path *path;
1834         struct extent_buffer *leaf;
1835         struct btrfs_key ins;
1836         int ret;
1837
1838         path = btrfs_alloc_path();
1839         if (!path)
1840                 return -ENOMEM;
1841
1842         path->leave_spinning = 1;
1843
1844         /*
1845          * we may be replacing one extent in the tree with another.
1846          * The new extent is pinned in the extent map, and we don't want
1847          * to drop it from the cache until it is completely in the btree.
1848          *
1849          * So, tell btrfs_drop_extents to leave this extent in the cache.
1850          * the caller is expected to unpin it and allow it to be merged
1851          * with the others.
1852          */
1853         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1854                                  file_pos + num_bytes, 0);
1855         if (ret)
1856                 goto out;
1857
1858         ins.objectid = btrfs_ino(inode);
1859         ins.offset = file_pos;
1860         ins.type = BTRFS_EXTENT_DATA_KEY;
1861         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1862         if (ret)
1863                 goto out;
1864         leaf = path->nodes[0];
1865         fi = btrfs_item_ptr(leaf, path->slots[0],
1866                             struct btrfs_file_extent_item);
1867         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1868         btrfs_set_file_extent_type(leaf, fi, extent_type);
1869         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1870         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1871         btrfs_set_file_extent_offset(leaf, fi, 0);
1872         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1873         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1874         btrfs_set_file_extent_compression(leaf, fi, compression);
1875         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1876         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1877
1878         btrfs_mark_buffer_dirty(leaf);
1879         btrfs_release_path(path);
1880
1881         inode_add_bytes(inode, num_bytes);
1882
1883         ins.objectid = disk_bytenr;
1884         ins.offset = disk_num_bytes;
1885         ins.type = BTRFS_EXTENT_ITEM_KEY;
1886         ret = btrfs_alloc_reserved_file_extent(trans, root,
1887                                         root->root_key.objectid,
1888                                         btrfs_ino(inode), file_pos, &ins);
1889 out:
1890         btrfs_free_path(path);
1891
1892         return ret;
1893 }
1894
1895 /*
1896  * helper function for btrfs_finish_ordered_io, this
1897  * just reads in some of the csum leaves to prime them into ram
1898  * before we start the transaction.  It limits the amount of btree
1899  * reads required while inside the transaction.
1900  */
1901 /* as ordered data IO finishes, this gets called so we can finish
1902  * an ordered extent if the range of bytes in the file it covers are
1903  * fully written.
1904  */
1905 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1906 {
1907         struct inode *inode = ordered_extent->inode;
1908         struct btrfs_root *root = BTRFS_I(inode)->root;
1909         struct btrfs_trans_handle *trans = NULL;
1910         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1911         struct extent_state *cached_state = NULL;
1912         int compress_type = 0;
1913         int ret;
1914         bool nolock;
1915
1916         nolock = btrfs_is_free_space_inode(inode);
1917
1918         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1919                 ret = -EIO;
1920                 goto out;
1921         }
1922
1923         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1924                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1925                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1926                 if (nolock)
1927                         trans = btrfs_join_transaction_nolock(root);
1928                 else
1929                         trans = btrfs_join_transaction(root);
1930                 if (IS_ERR(trans)) {
1931                         ret = PTR_ERR(trans);
1932                         trans = NULL;
1933                         goto out;
1934                 }
1935                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1936                 ret = btrfs_update_inode_fallback(trans, root, inode);
1937                 if (ret) /* -ENOMEM or corruption */
1938                         btrfs_abort_transaction(trans, root, ret);
1939                 goto out;
1940         }
1941
1942         lock_extent_bits(io_tree, ordered_extent->file_offset,
1943                          ordered_extent->file_offset + ordered_extent->len - 1,
1944                          0, &cached_state);
1945
1946         if (nolock)
1947                 trans = btrfs_join_transaction_nolock(root);
1948         else
1949                 trans = btrfs_join_transaction(root);
1950         if (IS_ERR(trans)) {
1951                 ret = PTR_ERR(trans);
1952                 trans = NULL;
1953                 goto out_unlock;
1954         }
1955         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1956
1957         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1958                 compress_type = ordered_extent->compress_type;
1959         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1960                 BUG_ON(compress_type);
1961                 ret = btrfs_mark_extent_written(trans, inode,
1962                                                 ordered_extent->file_offset,
1963                                                 ordered_extent->file_offset +
1964                                                 ordered_extent->len);
1965         } else {
1966                 BUG_ON(root == root->fs_info->tree_root);
1967                 ret = insert_reserved_file_extent(trans, inode,
1968                                                 ordered_extent->file_offset,
1969                                                 ordered_extent->start,
1970                                                 ordered_extent->disk_len,
1971                                                 ordered_extent->len,
1972                                                 ordered_extent->len,
1973                                                 compress_type, 0, 0,
1974                                                 BTRFS_FILE_EXTENT_REG);
1975         }
1976         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1977                            ordered_extent->file_offset, ordered_extent->len,
1978                            trans->transid);
1979         if (ret < 0) {
1980                 btrfs_abort_transaction(trans, root, ret);
1981                 goto out_unlock;
1982         }
1983
1984         add_pending_csums(trans, inode, ordered_extent->file_offset,
1985                           &ordered_extent->list);
1986
1987         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1988         ret = btrfs_update_inode_fallback(trans, root, inode);
1989         if (ret) { /* -ENOMEM or corruption */
1990                 btrfs_abort_transaction(trans, root, ret);
1991                 goto out_unlock;
1992         }
1993         ret = 0;
1994 out_unlock:
1995         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1996                              ordered_extent->file_offset +
1997                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1998 out:
1999         if (root != root->fs_info->tree_root)
2000                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2001         if (trans)
2002                 btrfs_end_transaction(trans, root);
2003
2004         if (ret)
2005                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2006                                       ordered_extent->file_offset +
2007                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2008
2009         /*
2010          * This needs to be done to make sure anybody waiting knows we are done
2011          * updating everything for this ordered extent.
2012          */
2013         btrfs_remove_ordered_extent(inode, ordered_extent);
2014
2015         /* once for us */
2016         btrfs_put_ordered_extent(ordered_extent);
2017         /* once for the tree */
2018         btrfs_put_ordered_extent(ordered_extent);
2019
2020         return ret;
2021 }
2022
2023 static void finish_ordered_fn(struct btrfs_work *work)
2024 {
2025         struct btrfs_ordered_extent *ordered_extent;
2026         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2027         btrfs_finish_ordered_io(ordered_extent);
2028 }
2029
2030 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2031                                 struct extent_state *state, int uptodate)
2032 {
2033         struct inode *inode = page->mapping->host;
2034         struct btrfs_root *root = BTRFS_I(inode)->root;
2035         struct btrfs_ordered_extent *ordered_extent = NULL;
2036         struct btrfs_workers *workers;
2037
2038         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2039
2040         ClearPagePrivate2(page);
2041         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2042                                             end - start + 1, uptodate))
2043                 return 0;
2044
2045         ordered_extent->work.func = finish_ordered_fn;
2046         ordered_extent->work.flags = 0;
2047
2048         if (btrfs_is_free_space_inode(inode))
2049                 workers = &root->fs_info->endio_freespace_worker;
2050         else
2051                 workers = &root->fs_info->endio_write_workers;
2052         btrfs_queue_worker(workers, &ordered_extent->work);
2053
2054         return 0;
2055 }
2056
2057 /*
2058  * when reads are done, we need to check csums to verify the data is correct
2059  * if there's a match, we allow the bio to finish.  If not, the code in
2060  * extent_io.c will try to find good copies for us.
2061  */
2062 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2063                                struct extent_state *state, int mirror)
2064 {
2065         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2066         struct inode *inode = page->mapping->host;
2067         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2068         char *kaddr;
2069         u64 private = ~(u32)0;
2070         int ret;
2071         struct btrfs_root *root = BTRFS_I(inode)->root;
2072         u32 csum = ~(u32)0;
2073
2074         if (PageChecked(page)) {
2075                 ClearPageChecked(page);
2076                 goto good;
2077         }
2078
2079         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2080                 goto good;
2081
2082         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2083             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2084                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2085                                   GFP_NOFS);
2086                 return 0;
2087         }
2088
2089         if (state && state->start == start) {
2090                 private = state->private;
2091                 ret = 0;
2092         } else {
2093                 ret = get_state_private(io_tree, start, &private);
2094         }
2095         kaddr = kmap_atomic(page);
2096         if (ret)
2097                 goto zeroit;
2098
2099         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2100         btrfs_csum_final(csum, (char *)&csum);
2101         if (csum != private)
2102                 goto zeroit;
2103
2104         kunmap_atomic(kaddr);
2105 good:
2106         return 0;
2107
2108 zeroit:
2109         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2110                        "private %llu\n",
2111                        (unsigned long long)btrfs_ino(page->mapping->host),
2112                        (unsigned long long)start, csum,
2113                        (unsigned long long)private);
2114         memset(kaddr + offset, 1, end - start + 1);
2115         flush_dcache_page(page);
2116         kunmap_atomic(kaddr);
2117         if (private == 0)
2118                 return 0;
2119         return -EIO;
2120 }
2121
2122 struct delayed_iput {
2123         struct list_head list;
2124         struct inode *inode;
2125 };
2126
2127 /* JDM: If this is fs-wide, why can't we add a pointer to
2128  * btrfs_inode instead and avoid the allocation? */
2129 void btrfs_add_delayed_iput(struct inode *inode)
2130 {
2131         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2132         struct delayed_iput *delayed;
2133
2134         if (atomic_add_unless(&inode->i_count, -1, 1))
2135                 return;
2136
2137         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2138         delayed->inode = inode;
2139
2140         spin_lock(&fs_info->delayed_iput_lock);
2141         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2142         spin_unlock(&fs_info->delayed_iput_lock);
2143 }
2144
2145 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2146 {
2147         LIST_HEAD(list);
2148         struct btrfs_fs_info *fs_info = root->fs_info;
2149         struct delayed_iput *delayed;
2150         int empty;
2151
2152         spin_lock(&fs_info->delayed_iput_lock);
2153         empty = list_empty(&fs_info->delayed_iputs);
2154         spin_unlock(&fs_info->delayed_iput_lock);
2155         if (empty)
2156                 return;
2157
2158         spin_lock(&fs_info->delayed_iput_lock);
2159         list_splice_init(&fs_info->delayed_iputs, &list);
2160         spin_unlock(&fs_info->delayed_iput_lock);
2161
2162         while (!list_empty(&list)) {
2163                 delayed = list_entry(list.next, struct delayed_iput, list);
2164                 list_del(&delayed->list);
2165                 iput(delayed->inode);
2166                 kfree(delayed);
2167         }
2168 }
2169
2170 enum btrfs_orphan_cleanup_state {
2171         ORPHAN_CLEANUP_STARTED  = 1,
2172         ORPHAN_CLEANUP_DONE     = 2,
2173 };
2174
2175 /*
2176  * This is called in transaction commit time. If there are no orphan
2177  * files in the subvolume, it removes orphan item and frees block_rsv
2178  * structure.
2179  */
2180 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2181                               struct btrfs_root *root)
2182 {
2183         struct btrfs_block_rsv *block_rsv;
2184         int ret;
2185
2186         if (atomic_read(&root->orphan_inodes) ||
2187             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2188                 return;
2189
2190         spin_lock(&root->orphan_lock);
2191         if (atomic_read(&root->orphan_inodes)) {
2192                 spin_unlock(&root->orphan_lock);
2193                 return;
2194         }
2195
2196         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2197                 spin_unlock(&root->orphan_lock);
2198                 return;
2199         }
2200
2201         block_rsv = root->orphan_block_rsv;
2202         root->orphan_block_rsv = NULL;
2203         spin_unlock(&root->orphan_lock);
2204
2205         if (root->orphan_item_inserted &&
2206             btrfs_root_refs(&root->root_item) > 0) {
2207                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2208                                             root->root_key.objectid);
2209                 BUG_ON(ret);
2210                 root->orphan_item_inserted = 0;
2211         }
2212
2213         if (block_rsv) {
2214                 WARN_ON(block_rsv->size > 0);
2215                 btrfs_free_block_rsv(root, block_rsv);
2216         }
2217 }
2218
2219 /*
2220  * This creates an orphan entry for the given inode in case something goes
2221  * wrong in the middle of an unlink/truncate.
2222  *
2223  * NOTE: caller of this function should reserve 5 units of metadata for
2224  *       this function.
2225  */
2226 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2227 {
2228         struct btrfs_root *root = BTRFS_I(inode)->root;
2229         struct btrfs_block_rsv *block_rsv = NULL;
2230         int reserve = 0;
2231         int insert = 0;
2232         int ret;
2233
2234         if (!root->orphan_block_rsv) {
2235                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2236                 if (!block_rsv)
2237                         return -ENOMEM;
2238         }
2239
2240         spin_lock(&root->orphan_lock);
2241         if (!root->orphan_block_rsv) {
2242                 root->orphan_block_rsv = block_rsv;
2243         } else if (block_rsv) {
2244                 btrfs_free_block_rsv(root, block_rsv);
2245                 block_rsv = NULL;
2246         }
2247
2248         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2249                               &BTRFS_I(inode)->runtime_flags)) {
2250 #if 0
2251                 /*
2252                  * For proper ENOSPC handling, we should do orphan
2253                  * cleanup when mounting. But this introduces backward
2254                  * compatibility issue.
2255                  */
2256                 if (!xchg(&root->orphan_item_inserted, 1))
2257                         insert = 2;
2258                 else
2259                         insert = 1;
2260 #endif
2261                 insert = 1;
2262                 atomic_inc(&root->orphan_inodes);
2263         }
2264
2265         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2266                               &BTRFS_I(inode)->runtime_flags))
2267                 reserve = 1;
2268         spin_unlock(&root->orphan_lock);
2269
2270         /* grab metadata reservation from transaction handle */
2271         if (reserve) {
2272                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2273                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2274         }
2275
2276         /* insert an orphan item to track this unlinked/truncated file */
2277         if (insert >= 1) {
2278                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2279                 if (ret && ret != -EEXIST) {
2280                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2281                                   &BTRFS_I(inode)->runtime_flags);
2282                         btrfs_abort_transaction(trans, root, ret);
2283                         return ret;
2284                 }
2285                 ret = 0;
2286         }
2287
2288         /* insert an orphan item to track subvolume contains orphan files */
2289         if (insert >= 2) {
2290                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2291                                                root->root_key.objectid);
2292                 if (ret && ret != -EEXIST) {
2293                         btrfs_abort_transaction(trans, root, ret);
2294                         return ret;
2295                 }
2296         }
2297         return 0;
2298 }
2299
2300 /*
2301  * We have done the truncate/delete so we can go ahead and remove the orphan
2302  * item for this particular inode.
2303  */
2304 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2305 {
2306         struct btrfs_root *root = BTRFS_I(inode)->root;
2307         int delete_item = 0;
2308         int release_rsv = 0;
2309         int ret = 0;
2310
2311         spin_lock(&root->orphan_lock);
2312         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2313                                &BTRFS_I(inode)->runtime_flags))
2314                 delete_item = 1;
2315
2316         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2317                                &BTRFS_I(inode)->runtime_flags))
2318                 release_rsv = 1;
2319         spin_unlock(&root->orphan_lock);
2320
2321         if (trans && delete_item) {
2322                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2323                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2324         }
2325
2326         if (release_rsv) {
2327                 btrfs_orphan_release_metadata(inode);
2328                 atomic_dec(&root->orphan_inodes);
2329         }
2330
2331         return 0;
2332 }
2333
2334 /*
2335  * this cleans up any orphans that may be left on the list from the last use
2336  * of this root.
2337  */
2338 int btrfs_orphan_cleanup(struct btrfs_root *root)
2339 {
2340         struct btrfs_path *path;
2341         struct extent_buffer *leaf;
2342         struct btrfs_key key, found_key;
2343         struct btrfs_trans_handle *trans;
2344         struct inode *inode;
2345         u64 last_objectid = 0;
2346         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2347
2348         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2349                 return 0;
2350
2351         path = btrfs_alloc_path();
2352         if (!path) {
2353                 ret = -ENOMEM;
2354                 goto out;
2355         }
2356         path->reada = -1;
2357
2358         key.objectid = BTRFS_ORPHAN_OBJECTID;
2359         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2360         key.offset = (u64)-1;
2361
2362         while (1) {
2363                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2364                 if (ret < 0)
2365                         goto out;
2366
2367                 /*
2368                  * if ret == 0 means we found what we were searching for, which
2369                  * is weird, but possible, so only screw with path if we didn't
2370                  * find the key and see if we have stuff that matches
2371                  */
2372                 if (ret > 0) {
2373                         ret = 0;
2374                         if (path->slots[0] == 0)
2375                                 break;
2376                         path->slots[0]--;
2377                 }
2378
2379                 /* pull out the item */
2380                 leaf = path->nodes[0];
2381                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2382
2383                 /* make sure the item matches what we want */
2384                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2385                         break;
2386                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2387                         break;
2388
2389                 /* release the path since we're done with it */
2390                 btrfs_release_path(path);
2391
2392                 /*
2393                  * this is where we are basically btrfs_lookup, without the
2394                  * crossing root thing.  we store the inode number in the
2395                  * offset of the orphan item.
2396                  */
2397
2398                 if (found_key.offset == last_objectid) {
2399                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2400                                "stopping orphan cleanup\n");
2401                         ret = -EINVAL;
2402                         goto out;
2403                 }
2404
2405                 last_objectid = found_key.offset;
2406
2407                 found_key.objectid = found_key.offset;
2408                 found_key.type = BTRFS_INODE_ITEM_KEY;
2409                 found_key.offset = 0;
2410                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2411                 ret = PTR_RET(inode);
2412                 if (ret && ret != -ESTALE)
2413                         goto out;
2414
2415                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2416                         struct btrfs_root *dead_root;
2417                         struct btrfs_fs_info *fs_info = root->fs_info;
2418                         int is_dead_root = 0;
2419
2420                         /*
2421                          * this is an orphan in the tree root. Currently these
2422                          * could come from 2 sources:
2423                          *  a) a snapshot deletion in progress
2424                          *  b) a free space cache inode
2425                          * We need to distinguish those two, as the snapshot
2426                          * orphan must not get deleted.
2427                          * find_dead_roots already ran before us, so if this
2428                          * is a snapshot deletion, we should find the root
2429                          * in the dead_roots list
2430                          */
2431                         spin_lock(&fs_info->trans_lock);
2432                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2433                                             root_list) {
2434                                 if (dead_root->root_key.objectid ==
2435                                     found_key.objectid) {
2436                                         is_dead_root = 1;
2437                                         break;
2438                                 }
2439                         }
2440                         spin_unlock(&fs_info->trans_lock);
2441                         if (is_dead_root) {
2442                                 /* prevent this orphan from being found again */
2443                                 key.offset = found_key.objectid - 1;
2444                                 continue;
2445                         }
2446                 }
2447                 /*
2448                  * Inode is already gone but the orphan item is still there,
2449                  * kill the orphan item.
2450                  */
2451                 if (ret == -ESTALE) {
2452                         trans = btrfs_start_transaction(root, 1);
2453                         if (IS_ERR(trans)) {
2454                                 ret = PTR_ERR(trans);
2455                                 goto out;
2456                         }
2457                         printk(KERN_ERR "auto deleting %Lu\n",
2458                                found_key.objectid);
2459                         ret = btrfs_del_orphan_item(trans, root,
2460                                                     found_key.objectid);
2461                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2462                         btrfs_end_transaction(trans, root);
2463                         continue;
2464                 }
2465
2466                 /*
2467                  * add this inode to the orphan list so btrfs_orphan_del does
2468                  * the proper thing when we hit it
2469                  */
2470                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2471                         &BTRFS_I(inode)->runtime_flags);
2472
2473                 /* if we have links, this was a truncate, lets do that */
2474                 if (inode->i_nlink) {
2475                         if (!S_ISREG(inode->i_mode)) {
2476                                 WARN_ON(1);
2477                                 iput(inode);
2478                                 continue;
2479                         }
2480                         nr_truncate++;
2481                         ret = btrfs_truncate(inode);
2482                 } else {
2483                         nr_unlink++;
2484                 }
2485
2486                 /* this will do delete_inode and everything for us */
2487                 iput(inode);
2488                 if (ret)
2489                         goto out;
2490         }
2491         /* release the path since we're done with it */
2492         btrfs_release_path(path);
2493
2494         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2495
2496         if (root->orphan_block_rsv)
2497                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2498                                         (u64)-1);
2499
2500         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2501                 trans = btrfs_join_transaction(root);
2502                 if (!IS_ERR(trans))
2503                         btrfs_end_transaction(trans, root);
2504         }
2505
2506         if (nr_unlink)
2507                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2508         if (nr_truncate)
2509                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2510
2511 out:
2512         if (ret)
2513                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2514         btrfs_free_path(path);
2515         return ret;
2516 }
2517
2518 /*
2519  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2520  * don't find any xattrs, we know there can't be any acls.
2521  *
2522  * slot is the slot the inode is in, objectid is the objectid of the inode
2523  */
2524 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2525                                           int slot, u64 objectid)
2526 {
2527         u32 nritems = btrfs_header_nritems(leaf);
2528         struct btrfs_key found_key;
2529         int scanned = 0;
2530
2531         slot++;
2532         while (slot < nritems) {
2533                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2534
2535                 /* we found a different objectid, there must not be acls */
2536                 if (found_key.objectid != objectid)
2537                         return 0;
2538
2539                 /* we found an xattr, assume we've got an acl */
2540                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2541                         return 1;
2542
2543                 /*
2544                  * we found a key greater than an xattr key, there can't
2545                  * be any acls later on
2546                  */
2547                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2548                         return 0;
2549
2550                 slot++;
2551                 scanned++;
2552
2553                 /*
2554                  * it goes inode, inode backrefs, xattrs, extents,
2555                  * so if there are a ton of hard links to an inode there can
2556                  * be a lot of backrefs.  Don't waste time searching too hard,
2557                  * this is just an optimization
2558                  */
2559                 if (scanned >= 8)
2560                         break;
2561         }
2562         /* we hit the end of the leaf before we found an xattr or
2563          * something larger than an xattr.  We have to assume the inode
2564          * has acls
2565          */
2566         return 1;
2567 }
2568
2569 /*
2570  * read an inode from the btree into the in-memory inode
2571  */
2572 static void btrfs_read_locked_inode(struct inode *inode)
2573 {
2574         struct btrfs_path *path;
2575         struct extent_buffer *leaf;
2576         struct btrfs_inode_item *inode_item;
2577         struct btrfs_timespec *tspec;
2578         struct btrfs_root *root = BTRFS_I(inode)->root;
2579         struct btrfs_key location;
2580         int maybe_acls;
2581         u32 rdev;
2582         int ret;
2583         bool filled = false;
2584
2585         ret = btrfs_fill_inode(inode, &rdev);
2586         if (!ret)
2587                 filled = true;
2588
2589         path = btrfs_alloc_path();
2590         if (!path)
2591                 goto make_bad;
2592
2593         path->leave_spinning = 1;
2594         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2595
2596         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2597         if (ret)
2598                 goto make_bad;
2599
2600         leaf = path->nodes[0];
2601
2602         if (filled)
2603                 goto cache_acl;
2604
2605         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2606                                     struct btrfs_inode_item);
2607         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2608         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2609         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2610         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2611         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2612
2613         tspec = btrfs_inode_atime(inode_item);
2614         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2615         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2616
2617         tspec = btrfs_inode_mtime(inode_item);
2618         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2619         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2620
2621         tspec = btrfs_inode_ctime(inode_item);
2622         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2623         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2624
2625         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2626         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2627         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2628
2629         /*
2630          * If we were modified in the current generation and evicted from memory
2631          * and then re-read we need to do a full sync since we don't have any
2632          * idea about which extents were modified before we were evicted from
2633          * cache.
2634          */
2635         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2636                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2637                         &BTRFS_I(inode)->runtime_flags);
2638
2639         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2640         inode->i_generation = BTRFS_I(inode)->generation;
2641         inode->i_rdev = 0;
2642         rdev = btrfs_inode_rdev(leaf, inode_item);
2643
2644         BTRFS_I(inode)->index_cnt = (u64)-1;
2645         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2646 cache_acl:
2647         /*
2648          * try to precache a NULL acl entry for files that don't have
2649          * any xattrs or acls
2650          */
2651         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2652                                            btrfs_ino(inode));
2653         if (!maybe_acls)
2654                 cache_no_acl(inode);
2655
2656         btrfs_free_path(path);
2657
2658         switch (inode->i_mode & S_IFMT) {
2659         case S_IFREG:
2660                 inode->i_mapping->a_ops = &btrfs_aops;
2661                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2662                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2663                 inode->i_fop = &btrfs_file_operations;
2664                 inode->i_op = &btrfs_file_inode_operations;
2665                 break;
2666         case S_IFDIR:
2667                 inode->i_fop = &btrfs_dir_file_operations;
2668                 if (root == root->fs_info->tree_root)
2669                         inode->i_op = &btrfs_dir_ro_inode_operations;
2670                 else
2671                         inode->i_op = &btrfs_dir_inode_operations;
2672                 break;
2673         case S_IFLNK:
2674                 inode->i_op = &btrfs_symlink_inode_operations;
2675                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2676                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2677                 break;
2678         default:
2679                 inode->i_op = &btrfs_special_inode_operations;
2680                 init_special_inode(inode, inode->i_mode, rdev);
2681                 break;
2682         }
2683
2684         btrfs_update_iflags(inode);
2685         return;
2686
2687 make_bad:
2688         btrfs_free_path(path);
2689         make_bad_inode(inode);
2690 }
2691
2692 /*
2693  * given a leaf and an inode, copy the inode fields into the leaf
2694  */
2695 static void fill_inode_item(struct btrfs_trans_handle *trans,
2696                             struct extent_buffer *leaf,
2697                             struct btrfs_inode_item *item,
2698                             struct inode *inode)
2699 {
2700         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2701         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2702         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2703         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2704         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2705
2706         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2707                                inode->i_atime.tv_sec);
2708         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2709                                 inode->i_atime.tv_nsec);
2710
2711         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2712                                inode->i_mtime.tv_sec);
2713         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2714                                 inode->i_mtime.tv_nsec);
2715
2716         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2717                                inode->i_ctime.tv_sec);
2718         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2719                                 inode->i_ctime.tv_nsec);
2720
2721         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2722         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2723         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2724         btrfs_set_inode_transid(leaf, item, trans->transid);
2725         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2726         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2727         btrfs_set_inode_block_group(leaf, item, 0);
2728 }
2729
2730 /*
2731  * copy everything in the in-memory inode into the btree.
2732  */
2733 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2734                                 struct btrfs_root *root, struct inode *inode)
2735 {
2736         struct btrfs_inode_item *inode_item;
2737         struct btrfs_path *path;
2738         struct extent_buffer *leaf;
2739         int ret;
2740
2741         path = btrfs_alloc_path();
2742         if (!path)
2743                 return -ENOMEM;
2744
2745         path->leave_spinning = 1;
2746         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2747                                  1);
2748         if (ret) {
2749                 if (ret > 0)
2750                         ret = -ENOENT;
2751                 goto failed;
2752         }
2753
2754         btrfs_unlock_up_safe(path, 1);
2755         leaf = path->nodes[0];
2756         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2757                                     struct btrfs_inode_item);
2758
2759         fill_inode_item(trans, leaf, inode_item, inode);
2760         btrfs_mark_buffer_dirty(leaf);
2761         btrfs_set_inode_last_trans(trans, inode);
2762         ret = 0;
2763 failed:
2764         btrfs_free_path(path);
2765         return ret;
2766 }
2767
2768 /*
2769  * copy everything in the in-memory inode into the btree.
2770  */
2771 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2772                                 struct btrfs_root *root, struct inode *inode)
2773 {
2774         int ret;
2775
2776         /*
2777          * If the inode is a free space inode, we can deadlock during commit
2778          * if we put it into the delayed code.
2779          *
2780          * The data relocation inode should also be directly updated
2781          * without delay
2782          */
2783         if (!btrfs_is_free_space_inode(inode)
2784             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2785                 btrfs_update_root_times(trans, root);
2786
2787                 ret = btrfs_delayed_update_inode(trans, root, inode);
2788                 if (!ret)
2789                         btrfs_set_inode_last_trans(trans, inode);
2790                 return ret;
2791         }
2792
2793         return btrfs_update_inode_item(trans, root, inode);
2794 }
2795
2796 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2797                                          struct btrfs_root *root,
2798                                          struct inode *inode)
2799 {
2800         int ret;
2801
2802         ret = btrfs_update_inode(trans, root, inode);
2803         if (ret == -ENOSPC)
2804                 return btrfs_update_inode_item(trans, root, inode);
2805         return ret;
2806 }
2807
2808 /*
2809  * unlink helper that gets used here in inode.c and in the tree logging
2810  * recovery code.  It remove a link in a directory with a given name, and
2811  * also drops the back refs in the inode to the directory
2812  */
2813 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2814                                 struct btrfs_root *root,
2815                                 struct inode *dir, struct inode *inode,
2816                                 const char *name, int name_len)
2817 {
2818         struct btrfs_path *path;
2819         int ret = 0;
2820         struct extent_buffer *leaf;
2821         struct btrfs_dir_item *di;
2822         struct btrfs_key key;
2823         u64 index;
2824         u64 ino = btrfs_ino(inode);
2825         u64 dir_ino = btrfs_ino(dir);
2826
2827         path = btrfs_alloc_path();
2828         if (!path) {
2829                 ret = -ENOMEM;
2830                 goto out;
2831         }
2832
2833         path->leave_spinning = 1;
2834         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2835                                     name, name_len, -1);
2836         if (IS_ERR(di)) {
2837                 ret = PTR_ERR(di);
2838                 goto err;
2839         }
2840         if (!di) {
2841                 ret = -ENOENT;
2842                 goto err;
2843         }
2844         leaf = path->nodes[0];
2845         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2846         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2847         if (ret)
2848                 goto err;
2849         btrfs_release_path(path);
2850
2851         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2852                                   dir_ino, &index);
2853         if (ret) {
2854                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2855                        "inode %llu parent %llu\n", name_len, name,
2856                        (unsigned long long)ino, (unsigned long long)dir_ino);
2857                 btrfs_abort_transaction(trans, root, ret);
2858                 goto err;
2859         }
2860
2861         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2862         if (ret) {
2863                 btrfs_abort_transaction(trans, root, ret);
2864                 goto err;
2865         }
2866
2867         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2868                                          inode, dir_ino);
2869         if (ret != 0 && ret != -ENOENT) {
2870                 btrfs_abort_transaction(trans, root, ret);
2871                 goto err;
2872         }
2873
2874         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2875                                            dir, index);
2876         if (ret == -ENOENT)
2877                 ret = 0;
2878 err:
2879         btrfs_free_path(path);
2880         if (ret)
2881                 goto out;
2882
2883         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2884         inode_inc_iversion(inode);
2885         inode_inc_iversion(dir);
2886         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2887         ret = btrfs_update_inode(trans, root, dir);
2888 out:
2889         return ret;
2890 }
2891
2892 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2893                        struct btrfs_root *root,
2894                        struct inode *dir, struct inode *inode,
2895                        const char *name, int name_len)
2896 {
2897         int ret;
2898         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2899         if (!ret) {
2900                 btrfs_drop_nlink(inode);
2901                 ret = btrfs_update_inode(trans, root, inode);
2902         }
2903         return ret;
2904 }
2905
2906
2907 /* helper to check if there is any shared block in the path */
2908 static int check_path_shared(struct btrfs_root *root,
2909                              struct btrfs_path *path)
2910 {
2911         struct extent_buffer *eb;
2912         int level;
2913         u64 refs = 1;
2914
2915         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2916                 int ret;
2917
2918                 if (!path->nodes[level])
2919                         break;
2920                 eb = path->nodes[level];
2921                 if (!btrfs_block_can_be_shared(root, eb))
2922                         continue;
2923                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2924                                                &refs, NULL);
2925                 if (refs > 1)
2926                         return 1;
2927         }
2928         return 0;
2929 }
2930
2931 /*
2932  * helper to start transaction for unlink and rmdir.
2933  *
2934  * unlink and rmdir are special in btrfs, they do not always free space.
2935  * so in enospc case, we should make sure they will free space before
2936  * allowing them to use the global metadata reservation.
2937  */
2938 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2939                                                        struct dentry *dentry)
2940 {
2941         struct btrfs_trans_handle *trans;
2942         struct btrfs_root *root = BTRFS_I(dir)->root;
2943         struct btrfs_path *path;
2944         struct btrfs_dir_item *di;
2945         struct inode *inode = dentry->d_inode;
2946         u64 index;
2947         int check_link = 1;
2948         int err = -ENOSPC;
2949         int ret;
2950         u64 ino = btrfs_ino(inode);
2951         u64 dir_ino = btrfs_ino(dir);
2952
2953         /*
2954          * 1 for the possible orphan item
2955          * 1 for the dir item
2956          * 1 for the dir index
2957          * 1 for the inode ref
2958          * 1 for the inode ref in the tree log
2959          * 2 for the dir entries in the log
2960          * 1 for the inode
2961          */
2962         trans = btrfs_start_transaction(root, 8);
2963         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2964                 return trans;
2965
2966         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2967                 return ERR_PTR(-ENOSPC);
2968
2969         /* check if there is someone else holds reference */
2970         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2971                 return ERR_PTR(-ENOSPC);
2972
2973         if (atomic_read(&inode->i_count) > 2)
2974                 return ERR_PTR(-ENOSPC);
2975
2976         if (xchg(&root->fs_info->enospc_unlink, 1))
2977                 return ERR_PTR(-ENOSPC);
2978
2979         path = btrfs_alloc_path();
2980         if (!path) {
2981                 root->fs_info->enospc_unlink = 0;
2982                 return ERR_PTR(-ENOMEM);
2983         }
2984
2985         /* 1 for the orphan item */
2986         trans = btrfs_start_transaction(root, 1);
2987         if (IS_ERR(trans)) {
2988                 btrfs_free_path(path);
2989                 root->fs_info->enospc_unlink = 0;
2990                 return trans;
2991         }
2992
2993         path->skip_locking = 1;
2994         path->search_commit_root = 1;
2995
2996         ret = btrfs_lookup_inode(trans, root, path,
2997                                 &BTRFS_I(dir)->location, 0);
2998         if (ret < 0) {
2999                 err = ret;
3000                 goto out;
3001         }
3002         if (ret == 0) {
3003                 if (check_path_shared(root, path))
3004                         goto out;
3005         } else {
3006                 check_link = 0;
3007         }
3008         btrfs_release_path(path);
3009
3010         ret = btrfs_lookup_inode(trans, root, path,
3011                                 &BTRFS_I(inode)->location, 0);
3012         if (ret < 0) {
3013                 err = ret;
3014                 goto out;
3015         }
3016         if (ret == 0) {
3017                 if (check_path_shared(root, path))
3018                         goto out;
3019         } else {
3020                 check_link = 0;
3021         }
3022         btrfs_release_path(path);
3023
3024         if (ret == 0 && S_ISREG(inode->i_mode)) {
3025                 ret = btrfs_lookup_file_extent(trans, root, path,
3026                                                ino, (u64)-1, 0);
3027                 if (ret < 0) {
3028                         err = ret;
3029                         goto out;
3030                 }
3031                 BUG_ON(ret == 0); /* Corruption */
3032                 if (check_path_shared(root, path))
3033                         goto out;
3034                 btrfs_release_path(path);
3035         }
3036
3037         if (!check_link) {
3038                 err = 0;
3039                 goto out;
3040         }
3041
3042         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3043                                 dentry->d_name.name, dentry->d_name.len, 0);
3044         if (IS_ERR(di)) {
3045                 err = PTR_ERR(di);
3046                 goto out;
3047         }
3048         if (di) {
3049                 if (check_path_shared(root, path))
3050                         goto out;
3051         } else {
3052                 err = 0;
3053                 goto out;
3054         }
3055         btrfs_release_path(path);
3056
3057         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3058                                         dentry->d_name.len, ino, dir_ino, 0,
3059                                         &index);
3060         if (ret) {
3061                 err = ret;
3062                 goto out;
3063         }
3064
3065         if (check_path_shared(root, path))
3066                 goto out;
3067
3068         btrfs_release_path(path);
3069
3070         /*
3071          * This is a commit root search, if we can lookup inode item and other
3072          * relative items in the commit root, it means the transaction of
3073          * dir/file creation has been committed, and the dir index item that we
3074          * delay to insert has also been inserted into the commit root. So
3075          * we needn't worry about the delayed insertion of the dir index item
3076          * here.
3077          */
3078         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3079                                 dentry->d_name.name, dentry->d_name.len, 0);
3080         if (IS_ERR(di)) {
3081                 err = PTR_ERR(di);
3082                 goto out;
3083         }
3084         BUG_ON(ret == -ENOENT);
3085         if (check_path_shared(root, path))
3086                 goto out;
3087
3088         err = 0;
3089 out:
3090         btrfs_free_path(path);
3091         /* Migrate the orphan reservation over */
3092         if (!err)
3093                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3094                                 &root->fs_info->global_block_rsv,
3095                                 trans->bytes_reserved);
3096
3097         if (err) {
3098                 btrfs_end_transaction(trans, root);
3099                 root->fs_info->enospc_unlink = 0;
3100                 return ERR_PTR(err);
3101         }
3102
3103         trans->block_rsv = &root->fs_info->global_block_rsv;
3104         return trans;
3105 }
3106
3107 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3108                                struct btrfs_root *root)
3109 {
3110         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3111                 btrfs_block_rsv_release(root, trans->block_rsv,
3112                                         trans->bytes_reserved);
3113                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3114                 BUG_ON(!root->fs_info->enospc_unlink);
3115                 root->fs_info->enospc_unlink = 0;
3116         }
3117         btrfs_end_transaction(trans, root);
3118 }
3119
3120 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3121 {
3122         struct btrfs_root *root = BTRFS_I(dir)->root;
3123         struct btrfs_trans_handle *trans;
3124         struct inode *inode = dentry->d_inode;
3125         int ret;
3126
3127         trans = __unlink_start_trans(dir, dentry);
3128         if (IS_ERR(trans))
3129                 return PTR_ERR(trans);
3130
3131         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3132
3133         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3134                                  dentry->d_name.name, dentry->d_name.len);
3135         if (ret)
3136                 goto out;
3137
3138         if (inode->i_nlink == 0) {
3139                 ret = btrfs_orphan_add(trans, inode);
3140                 if (ret)
3141                         goto out;
3142         }
3143
3144 out:
3145         __unlink_end_trans(trans, root);
3146         btrfs_btree_balance_dirty(root);
3147         return ret;
3148 }
3149
3150 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3151                         struct btrfs_root *root,
3152                         struct inode *dir, u64 objectid,
3153                         const char *name, int name_len)
3154 {
3155         struct btrfs_path *path;
3156         struct extent_buffer *leaf;
3157         struct btrfs_dir_item *di;
3158         struct btrfs_key key;
3159         u64 index;
3160         int ret;
3161         u64 dir_ino = btrfs_ino(dir);
3162
3163         path = btrfs_alloc_path();
3164         if (!path)
3165                 return -ENOMEM;
3166
3167         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3168                                    name, name_len, -1);
3169         if (IS_ERR_OR_NULL(di)) {
3170                 if (!di)
3171                         ret = -ENOENT;
3172                 else
3173                         ret = PTR_ERR(di);
3174                 goto out;
3175         }
3176
3177         leaf = path->nodes[0];
3178         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3179         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3180         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3181         if (ret) {
3182                 btrfs_abort_transaction(trans, root, ret);
3183                 goto out;
3184         }
3185         btrfs_release_path(path);
3186
3187         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3188                                  objectid, root->root_key.objectid,
3189                                  dir_ino, &index, name, name_len);
3190         if (ret < 0) {
3191                 if (ret != -ENOENT) {
3192                         btrfs_abort_transaction(trans, root, ret);
3193                         goto out;
3194                 }
3195                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3196                                                  name, name_len);
3197                 if (IS_ERR_OR_NULL(di)) {
3198                         if (!di)
3199                                 ret = -ENOENT;
3200                         else
3201                                 ret = PTR_ERR(di);
3202                         btrfs_abort_transaction(trans, root, ret);
3203                         goto out;
3204                 }
3205
3206                 leaf = path->nodes[0];
3207                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3208                 btrfs_release_path(path);
3209                 index = key.offset;
3210         }
3211         btrfs_release_path(path);
3212
3213         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3214         if (ret) {
3215                 btrfs_abort_transaction(trans, root, ret);
3216                 goto out;
3217         }
3218
3219         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3220         inode_inc_iversion(dir);
3221         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3222         ret = btrfs_update_inode_fallback(trans, root, dir);
3223         if (ret)
3224                 btrfs_abort_transaction(trans, root, ret);
3225 out:
3226         btrfs_free_path(path);
3227         return ret;
3228 }
3229
3230 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3231 {
3232         struct inode *inode = dentry->d_inode;
3233         int err = 0;
3234         struct btrfs_root *root = BTRFS_I(dir)->root;
3235         struct btrfs_trans_handle *trans;
3236
3237         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3238                 return -ENOTEMPTY;
3239         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3240                 return -EPERM;
3241
3242         trans = __unlink_start_trans(dir, dentry);
3243         if (IS_ERR(trans))
3244                 return PTR_ERR(trans);
3245
3246         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3247                 err = btrfs_unlink_subvol(trans, root, dir,
3248                                           BTRFS_I(inode)->location.objectid,
3249                                           dentry->d_name.name,
3250                                           dentry->d_name.len);
3251                 goto out;
3252         }
3253
3254         err = btrfs_orphan_add(trans, inode);
3255         if (err)
3256                 goto out;
3257
3258         /* now the directory is empty */
3259         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3260                                  dentry->d_name.name, dentry->d_name.len);
3261         if (!err)
3262                 btrfs_i_size_write(inode, 0);
3263 out:
3264         __unlink_end_trans(trans, root);
3265         btrfs_btree_balance_dirty(root);
3266
3267         return err;
3268 }
3269
3270 /*
3271  * this can truncate away extent items, csum items and directory items.
3272  * It starts at a high offset and removes keys until it can't find
3273  * any higher than new_size
3274  *
3275  * csum items that cross the new i_size are truncated to the new size
3276  * as well.
3277  *
3278  * min_type is the minimum key type to truncate down to.  If set to 0, this
3279  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3280  */
3281 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3282                                struct btrfs_root *root,
3283                                struct inode *inode,
3284                                u64 new_size, u32 min_type)
3285 {
3286         struct btrfs_path *path;
3287         struct extent_buffer *leaf;
3288         struct btrfs_file_extent_item *fi;
3289         struct btrfs_key key;
3290         struct btrfs_key found_key;
3291         u64 extent_start = 0;
3292         u64 extent_num_bytes = 0;
3293         u64 extent_offset = 0;
3294         u64 item_end = 0;
3295         u64 mask = root->sectorsize - 1;
3296         u32 found_type = (u8)-1;
3297         int found_extent;
3298         int del_item;
3299         int pending_del_nr = 0;
3300         int pending_del_slot = 0;
3301         int extent_type = -1;
3302         int ret;
3303         int err = 0;
3304         u64 ino = btrfs_ino(inode);
3305
3306         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3307
3308         path = btrfs_alloc_path();
3309         if (!path)
3310                 return -ENOMEM;
3311         path->reada = -1;
3312
3313         /*
3314          * We want to drop from the next block forward in case this new size is
3315          * not block aligned since we will be keeping the last block of the
3316          * extent just the way it is.
3317          */
3318         if (root->ref_cows || root == root->fs_info->tree_root)
3319                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3320
3321         /*
3322          * This function is also used to drop the items in the log tree before
3323          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3324          * it is used to drop the loged items. So we shouldn't kill the delayed
3325          * items.
3326          */
3327         if (min_type == 0 && root == BTRFS_I(inode)->root)
3328                 btrfs_kill_delayed_inode_items(inode);
3329
3330         key.objectid = ino;
3331         key.offset = (u64)-1;
3332         key.type = (u8)-1;
3333
3334 search_again:
3335         path->leave_spinning = 1;
3336         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3337         if (ret < 0) {
3338                 err = ret;
3339                 goto out;
3340         }
3341
3342         if (ret > 0) {
3343                 /* there are no items in the tree for us to truncate, we're
3344                  * done
3345                  */
3346                 if (path->slots[0] == 0)
3347                         goto out;
3348                 path->slots[0]--;
3349         }
3350
3351         while (1) {
3352                 fi = NULL;
3353                 leaf = path->nodes[0];
3354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3355                 found_type = btrfs_key_type(&found_key);
3356
3357                 if (found_key.objectid != ino)
3358                         break;
3359
3360                 if (found_type < min_type)
3361                         break;
3362
3363                 item_end = found_key.offset;
3364                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3365                         fi = btrfs_item_ptr(leaf, path->slots[0],
3366                                             struct btrfs_file_extent_item);
3367                         extent_type = btrfs_file_extent_type(leaf, fi);
3368                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3369                                 item_end +=
3370                                     btrfs_file_extent_num_bytes(leaf, fi);
3371                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3372                                 item_end += btrfs_file_extent_inline_len(leaf,
3373                                                                          fi);
3374                         }
3375                         item_end--;
3376                 }
3377                 if (found_type > min_type) {
3378                         del_item = 1;
3379                 } else {
3380                         if (item_end < new_size)
3381                                 break;
3382                         if (found_key.offset >= new_size)
3383                                 del_item = 1;
3384                         else
3385                                 del_item = 0;
3386                 }
3387                 found_extent = 0;
3388                 /* FIXME, shrink the extent if the ref count is only 1 */
3389                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3390                         goto delete;
3391
3392                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3393                         u64 num_dec;
3394                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3395                         if (!del_item) {
3396                                 u64 orig_num_bytes =
3397                                         btrfs_file_extent_num_bytes(leaf, fi);
3398                                 extent_num_bytes = new_size -
3399                                         found_key.offset + root->sectorsize - 1;
3400                                 extent_num_bytes = extent_num_bytes &
3401                                         ~((u64)root->sectorsize - 1);
3402                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3403                                                          extent_num_bytes);
3404                                 num_dec = (orig_num_bytes -
3405                                            extent_num_bytes);
3406                                 if (root->ref_cows && extent_start != 0)
3407                                         inode_sub_bytes(inode, num_dec);
3408                                 btrfs_mark_buffer_dirty(leaf);
3409                         } else {
3410                                 extent_num_bytes =
3411                                         btrfs_file_extent_disk_num_bytes(leaf,
3412                                                                          fi);
3413                                 extent_offset = found_key.offset -
3414                                         btrfs_file_extent_offset(leaf, fi);
3415
3416                                 /* FIXME blocksize != 4096 */
3417                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3418                                 if (extent_start != 0) {
3419                                         found_extent = 1;
3420                                         if (root->ref_cows)
3421                                                 inode_sub_bytes(inode, num_dec);
3422                                 }
3423                         }
3424                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3425                         /*
3426                          * we can't truncate inline items that have had
3427                          * special encodings
3428                          */
3429                         if (!del_item &&
3430                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3431                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3432                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3433                                 u32 size = new_size - found_key.offset;
3434
3435                                 if (root->ref_cows) {
3436                                         inode_sub_bytes(inode, item_end + 1 -
3437                                                         new_size);
3438                                 }
3439                                 size =
3440                                     btrfs_file_extent_calc_inline_size(size);
3441                                 btrfs_truncate_item(trans, root, path,
3442                                                     size, 1);
3443                         } else if (root->ref_cows) {
3444                                 inode_sub_bytes(inode, item_end + 1 -
3445                                                 found_key.offset);
3446                         }
3447                 }
3448 delete:
3449                 if (del_item) {
3450                         if (!pending_del_nr) {
3451                                 /* no pending yet, add ourselves */
3452                                 pending_del_slot = path->slots[0];
3453                                 pending_del_nr = 1;
3454                         } else if (pending_del_nr &&
3455                                    path->slots[0] + 1 == pending_del_slot) {
3456                                 /* hop on the pending chunk */
3457                                 pending_del_nr++;
3458                                 pending_del_slot = path->slots[0];
3459                         } else {
3460                                 BUG();
3461                         }
3462                 } else {
3463                         break;
3464                 }
3465                 if (found_extent && (root->ref_cows ||
3466                                      root == root->fs_info->tree_root)) {
3467                         btrfs_set_path_blocking(path);
3468                         ret = btrfs_free_extent(trans, root, extent_start,
3469                                                 extent_num_bytes, 0,
3470                                                 btrfs_header_owner(leaf),
3471                                                 ino, extent_offset, 0);
3472                         BUG_ON(ret);
3473                 }
3474
3475                 if (found_type == BTRFS_INODE_ITEM_KEY)
3476                         break;
3477
3478                 if (path->slots[0] == 0 ||
3479                     path->slots[0] != pending_del_slot) {
3480                         if (pending_del_nr) {
3481                                 ret = btrfs_del_items(trans, root, path,
3482                                                 pending_del_slot,
3483                                                 pending_del_nr);
3484                                 if (ret) {
3485                                         btrfs_abort_transaction(trans,
3486                                                                 root, ret);
3487                                         goto error;
3488                                 }
3489                                 pending_del_nr = 0;
3490                         }
3491                         btrfs_release_path(path);
3492                         goto search_again;
3493                 } else {
3494                         path->slots[0]--;
3495                 }
3496         }
3497 out:
3498         if (pending_del_nr) {
3499                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3500                                       pending_del_nr);
3501                 if (ret)
3502                         btrfs_abort_transaction(trans, root, ret);
3503         }
3504 error:
3505         btrfs_free_path(path);
3506         return err;
3507 }
3508
3509 /*
3510  * btrfs_truncate_page - read, zero a chunk and write a page
3511  * @inode - inode that we're zeroing
3512  * @from - the offset to start zeroing
3513  * @len - the length to zero, 0 to zero the entire range respective to the
3514  *      offset
3515  * @front - zero up to the offset instead of from the offset on
3516  *
3517  * This will find the page for the "from" offset and cow the page and zero the
3518  * part we want to zero.  This is used with truncate and hole punching.
3519  */
3520 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3521                         int front)
3522 {
3523         struct address_space *mapping = inode->i_mapping;
3524         struct btrfs_root *root = BTRFS_I(inode)->root;
3525         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3526         struct btrfs_ordered_extent *ordered;
3527         struct extent_state *cached_state = NULL;
3528         char *kaddr;
3529         u32 blocksize = root->sectorsize;
3530         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3531         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3532         struct page *page;
3533         gfp_t mask = btrfs_alloc_write_mask(mapping);
3534         int ret = 0;
3535         u64 page_start;
3536         u64 page_end;
3537
3538         if ((offset & (blocksize - 1)) == 0 &&
3539             (!len || ((len & (blocksize - 1)) == 0)))
3540                 goto out;
3541         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3542         if (ret)
3543                 goto out;
3544
3545 again:
3546         page = find_or_create_page(mapping, index, mask);
3547         if (!page) {
3548                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3549                 ret = -ENOMEM;
3550                 goto out;
3551         }
3552
3553         page_start = page_offset(page);
3554         page_end = page_start + PAGE_CACHE_SIZE - 1;
3555
3556         if (!PageUptodate(page)) {
3557                 ret = btrfs_readpage(NULL, page);
3558                 lock_page(page);
3559                 if (page->mapping != mapping) {
3560                         unlock_page(page);
3561                         page_cache_release(page);
3562                         goto again;
3563                 }
3564                 if (!PageUptodate(page)) {
3565                         ret = -EIO;
3566                         goto out_unlock;
3567                 }
3568         }
3569         wait_on_page_writeback(page);
3570
3571         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3572         set_page_extent_mapped(page);
3573
3574         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3575         if (ordered) {
3576                 unlock_extent_cached(io_tree, page_start, page_end,
3577                                      &cached_state, GFP_NOFS);
3578                 unlock_page(page);
3579                 page_cache_release(page);
3580                 btrfs_start_ordered_extent(inode, ordered, 1);
3581                 btrfs_put_ordered_extent(ordered);
3582                 goto again;
3583         }
3584
3585         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3586                           EXTENT_DIRTY | EXTENT_DELALLOC |
3587                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3588                           0, 0, &cached_state, GFP_NOFS);
3589
3590         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3591                                         &cached_state);
3592         if (ret) {
3593                 unlock_extent_cached(io_tree, page_start, page_end,
3594                                      &cached_state, GFP_NOFS);
3595                 goto out_unlock;
3596         }
3597
3598         if (offset != PAGE_CACHE_SIZE) {
3599                 if (!len)
3600                         len = PAGE_CACHE_SIZE - offset;
3601                 kaddr = kmap(page);
3602                 if (front)
3603                         memset(kaddr, 0, offset);
3604                 else
3605                         memset(kaddr + offset, 0, len);
3606                 flush_dcache_page(page);
3607                 kunmap(page);
3608         }
3609         ClearPageChecked(page);
3610         set_page_dirty(page);
3611         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3612                              GFP_NOFS);
3613
3614 out_unlock:
3615         if (ret)
3616                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3617         unlock_page(page);
3618         page_cache_release(page);
3619 out:
3620         return ret;
3621 }
3622
3623 /*
3624  * This function puts in dummy file extents for the area we're creating a hole
3625  * for.  So if we are truncating this file to a larger size we need to insert
3626  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3627  * the range between oldsize and size
3628  */
3629 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3630 {
3631         struct btrfs_trans_handle *trans;
3632         struct btrfs_root *root = BTRFS_I(inode)->root;
3633         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3634         struct extent_map *em = NULL;
3635         struct extent_state *cached_state = NULL;
3636         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3637         u64 mask = root->sectorsize - 1;
3638         u64 hole_start = (oldsize + mask) & ~mask;
3639         u64 block_end = (size + mask) & ~mask;
3640         u64 last_byte;
3641         u64 cur_offset;
3642         u64 hole_size;
3643         int err = 0;
3644
3645         if (size <= hole_start)
3646                 return 0;
3647
3648         while (1) {
3649                 struct btrfs_ordered_extent *ordered;
3650                 btrfs_wait_ordered_range(inode, hole_start,
3651                                          block_end - hole_start);
3652                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3653                                  &cached_state);
3654                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3655                 if (!ordered)
3656                         break;
3657                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3658                                      &cached_state, GFP_NOFS);
3659                 btrfs_put_ordered_extent(ordered);
3660         }
3661
3662         cur_offset = hole_start;
3663         while (1) {
3664                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3665                                 block_end - cur_offset, 0);
3666                 if (IS_ERR(em)) {
3667                         err = PTR_ERR(em);
3668                         break;
3669                 }
3670                 last_byte = min(extent_map_end(em), block_end);
3671                 last_byte = (last_byte + mask) & ~mask;
3672                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3673                         struct extent_map *hole_em;
3674                         hole_size = last_byte - cur_offset;
3675
3676                         trans = btrfs_start_transaction(root, 3);
3677                         if (IS_ERR(trans)) {
3678                                 err = PTR_ERR(trans);
3679                                 break;
3680                         }
3681
3682                         err = btrfs_drop_extents(trans, root, inode,
3683                                                  cur_offset,
3684                                                  cur_offset + hole_size, 1);
3685                         if (err) {
3686                                 btrfs_abort_transaction(trans, root, err);
3687                                 btrfs_end_transaction(trans, root);
3688                                 break;
3689                         }
3690
3691                         err = btrfs_insert_file_extent(trans, root,
3692                                         btrfs_ino(inode), cur_offset, 0,
3693                                         0, hole_size, 0, hole_size,
3694                                         0, 0, 0);
3695                         if (err) {
3696                                 btrfs_abort_transaction(trans, root, err);
3697                                 btrfs_end_transaction(trans, root);
3698                                 break;
3699                         }
3700
3701                         btrfs_drop_extent_cache(inode, cur_offset,
3702                                                 cur_offset + hole_size - 1, 0);
3703                         hole_em = alloc_extent_map();
3704                         if (!hole_em) {
3705                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3706                                         &BTRFS_I(inode)->runtime_flags);
3707                                 goto next;
3708                         }
3709                         hole_em->start = cur_offset;
3710                         hole_em->len = hole_size;
3711                         hole_em->orig_start = cur_offset;
3712
3713                         hole_em->block_start = EXTENT_MAP_HOLE;
3714                         hole_em->block_len = 0;
3715                         hole_em->orig_block_len = 0;
3716                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3717                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3718                         hole_em->generation = trans->transid;
3719
3720                         while (1) {
3721                                 write_lock(&em_tree->lock);
3722                                 err = add_extent_mapping(em_tree, hole_em);
3723                                 if (!err)
3724                                         list_move(&hole_em->list,
3725                                                   &em_tree->modified_extents);
3726                                 write_unlock(&em_tree->lock);
3727                                 if (err != -EEXIST)
3728                                         break;
3729                                 btrfs_drop_extent_cache(inode, cur_offset,
3730                                                         cur_offset +
3731                                                         hole_size - 1, 0);
3732                         }
3733                         free_extent_map(hole_em);
3734 next:
3735                         btrfs_update_inode(trans, root, inode);
3736                         btrfs_end_transaction(trans, root);
3737                 }
3738                 free_extent_map(em);
3739                 em = NULL;
3740                 cur_offset = last_byte;
3741                 if (cur_offset >= block_end)
3742                         break;
3743         }
3744
3745         free_extent_map(em);
3746         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3747                              GFP_NOFS);
3748         return err;
3749 }
3750
3751 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3752 {
3753         struct btrfs_root *root = BTRFS_I(inode)->root;
3754         struct btrfs_trans_handle *trans;
3755         loff_t oldsize = i_size_read(inode);
3756         int ret;
3757
3758         if (newsize == oldsize)
3759                 return 0;
3760
3761         if (newsize > oldsize) {
3762                 truncate_pagecache(inode, oldsize, newsize);
3763                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3764                 if (ret)
3765                         return ret;
3766
3767                 trans = btrfs_start_transaction(root, 1);
3768                 if (IS_ERR(trans))
3769                         return PTR_ERR(trans);
3770
3771                 i_size_write(inode, newsize);
3772                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3773                 ret = btrfs_update_inode(trans, root, inode);
3774                 btrfs_end_transaction(trans, root);
3775         } else {
3776
3777                 /*
3778                  * We're truncating a file that used to have good data down to
3779                  * zero. Make sure it gets into the ordered flush list so that
3780                  * any new writes get down to disk quickly.
3781                  */
3782                 if (newsize == 0)
3783                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3784                                 &BTRFS_I(inode)->runtime_flags);
3785
3786                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3787                 truncate_setsize(inode, newsize);
3788                 ret = btrfs_truncate(inode);
3789         }
3790
3791         return ret;
3792 }
3793
3794 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3795 {
3796         struct inode *inode = dentry->d_inode;
3797         struct btrfs_root *root = BTRFS_I(inode)->root;
3798         int err;
3799
3800         if (btrfs_root_readonly(root))
3801                 return -EROFS;
3802
3803         err = inode_change_ok(inode, attr);
3804         if (err)
3805                 return err;
3806
3807         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3808                 err = btrfs_setsize(inode, attr->ia_size);
3809                 if (err)
3810                         return err;
3811         }
3812
3813         if (attr->ia_valid) {
3814                 setattr_copy(inode, attr);
3815                 inode_inc_iversion(inode);
3816                 err = btrfs_dirty_inode(inode);
3817
3818                 if (!err && attr->ia_valid & ATTR_MODE)
3819                         err = btrfs_acl_chmod(inode);
3820         }
3821
3822         return err;
3823 }
3824
3825 void btrfs_evict_inode(struct inode *inode)
3826 {
3827         struct btrfs_trans_handle *trans;
3828         struct btrfs_root *root = BTRFS_I(inode)->root;
3829         struct btrfs_block_rsv *rsv, *global_rsv;
3830         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3831         int ret;
3832
3833         trace_btrfs_inode_evict(inode);
3834
3835         truncate_inode_pages(&inode->i_data, 0);
3836         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3837                                btrfs_is_free_space_inode(inode)))
3838                 goto no_delete;
3839
3840         if (is_bad_inode(inode)) {
3841                 btrfs_orphan_del(NULL, inode);
3842                 goto no_delete;
3843         }
3844         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3845         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3846
3847         if (root->fs_info->log_root_recovering) {
3848                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3849                                  &BTRFS_I(inode)->runtime_flags));
3850                 goto no_delete;
3851         }
3852
3853         if (inode->i_nlink > 0) {
3854                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3855                 goto no_delete;
3856         }
3857
3858         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3859         if (!rsv) {
3860                 btrfs_orphan_del(NULL, inode);
3861                 goto no_delete;
3862         }
3863         rsv->size = min_size;
3864         rsv->failfast = 1;
3865         global_rsv = &root->fs_info->global_block_rsv;
3866
3867         btrfs_i_size_write(inode, 0);
3868
3869         /*
3870          * This is a bit simpler than btrfs_truncate since we've already
3871          * reserved our space for our orphan item in the unlink, so we just
3872          * need to reserve some slack space in case we add bytes and update
3873          * inode item when doing the truncate.
3874          */
3875         while (1) {
3876                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3877                                              BTRFS_RESERVE_FLUSH_LIMIT);
3878
3879                 /*
3880                  * Try and steal from the global reserve since we will
3881                  * likely not use this space anyway, we want to try as
3882                  * hard as possible to get this to work.
3883                  */
3884                 if (ret)
3885                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3886
3887                 if (ret) {
3888                         printk(KERN_WARNING "Could not get space for a "
3889                                "delete, will truncate on mount %d\n", ret);
3890                         btrfs_orphan_del(NULL, inode);
3891                         btrfs_free_block_rsv(root, rsv);
3892                         goto no_delete;
3893                 }
3894
3895                 trans = btrfs_start_transaction_lflush(root, 1);
3896                 if (IS_ERR(trans)) {
3897                         btrfs_orphan_del(NULL, inode);
3898                         btrfs_free_block_rsv(root, rsv);
3899                         goto no_delete;
3900                 }
3901
3902                 trans->block_rsv = rsv;
3903
3904                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3905                 if (ret != -ENOSPC)
3906                         break;
3907
3908                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3909                 ret = btrfs_update_inode(trans, root, inode);
3910                 BUG_ON(ret);
3911
3912                 btrfs_end_transaction(trans, root);
3913                 trans = NULL;
3914                 btrfs_btree_balance_dirty(root);
3915         }
3916
3917         btrfs_free_block_rsv(root, rsv);
3918
3919         if (ret == 0) {
3920                 trans->block_rsv = root->orphan_block_rsv;
3921                 ret = btrfs_orphan_del(trans, inode);
3922                 BUG_ON(ret);
3923         }
3924
3925         trans->block_rsv = &root->fs_info->trans_block_rsv;
3926         if (!(root == root->fs_info->tree_root ||
3927               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3928                 btrfs_return_ino(root, btrfs_ino(inode));
3929
3930         btrfs_end_transaction(trans, root);
3931         btrfs_btree_balance_dirty(root);
3932 no_delete:
3933         clear_inode(inode);
3934         return;
3935 }
3936
3937 /*
3938  * this returns the key found in the dir entry in the location pointer.
3939  * If no dir entries were found, location->objectid is 0.
3940  */
3941 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3942                                struct btrfs_key *location)
3943 {
3944         const char *name = dentry->d_name.name;
3945         int namelen = dentry->d_name.len;
3946         struct btrfs_dir_item *di;
3947         struct btrfs_path *path;
3948         struct btrfs_root *root = BTRFS_I(dir)->root;
3949         int ret = 0;
3950
3951         path = btrfs_alloc_path();
3952         if (!path)
3953                 return -ENOMEM;
3954
3955         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3956                                     namelen, 0);
3957         if (IS_ERR(di))
3958                 ret = PTR_ERR(di);
3959
3960         if (IS_ERR_OR_NULL(di))
3961                 goto out_err;
3962
3963         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3964 out:
3965         btrfs_free_path(path);
3966         return ret;
3967 out_err:
3968         location->objectid = 0;
3969         goto out;
3970 }
3971
3972 /*
3973  * when we hit a tree root in a directory, the btrfs part of the inode
3974  * needs to be changed to reflect the root directory of the tree root.  This
3975  * is kind of like crossing a mount point.
3976  */
3977 static int fixup_tree_root_location(struct btrfs_root *root,
3978                                     struct inode *dir,
3979                                     struct dentry *dentry,
3980                                     struct btrfs_key *location,
3981                                     struct btrfs_root **sub_root)
3982 {
3983         struct btrfs_path *path;
3984         struct btrfs_root *new_root;
3985         struct btrfs_root_ref *ref;
3986         struct extent_buffer *leaf;
3987         int ret;
3988         int err = 0;
3989
3990         path = btrfs_alloc_path();
3991         if (!path) {
3992                 err = -ENOMEM;
3993                 goto out;
3994         }
3995
3996         err = -ENOENT;
3997         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3998                                   BTRFS_I(dir)->root->root_key.objectid,
3999                                   location->objectid);
4000         if (ret) {
4001                 if (ret < 0)
4002                         err = ret;
4003                 goto out;
4004         }
4005
4006         leaf = path->nodes[0];
4007         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4008         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4009             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4010                 goto out;
4011
4012         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4013                                    (unsigned long)(ref + 1),
4014                                    dentry->d_name.len);
4015         if (ret)
4016                 goto out;
4017
4018         btrfs_release_path(path);
4019
4020         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4021         if (IS_ERR(new_root)) {
4022                 err = PTR_ERR(new_root);
4023                 goto out;
4024         }
4025
4026         if (btrfs_root_refs(&new_root->root_item) == 0) {
4027                 err = -ENOENT;
4028                 goto out;
4029         }
4030
4031         *sub_root = new_root;
4032         location->objectid = btrfs_root_dirid(&new_root->root_item);
4033         location->type = BTRFS_INODE_ITEM_KEY;
4034         location->offset = 0;
4035         err = 0;
4036 out:
4037         btrfs_free_path(path);
4038         return err;
4039 }
4040
4041 static void inode_tree_add(struct inode *inode)
4042 {
4043         struct btrfs_root *root = BTRFS_I(inode)->root;
4044         struct btrfs_inode *entry;
4045         struct rb_node **p;
4046         struct rb_node *parent;
4047         u64 ino = btrfs_ino(inode);
4048 again:
4049         p = &root->inode_tree.rb_node;
4050         parent = NULL;
4051
4052         if (inode_unhashed(inode))
4053                 return;
4054
4055         spin_lock(&root->inode_lock);
4056         while (*p) {
4057                 parent = *p;
4058                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4059
4060                 if (ino < btrfs_ino(&entry->vfs_inode))
4061                         p = &parent->rb_left;
4062                 else if (ino > btrfs_ino(&entry->vfs_inode))
4063                         p = &parent->rb_right;
4064                 else {
4065                         WARN_ON(!(entry->vfs_inode.i_state &
4066                                   (I_WILL_FREE | I_FREEING)));
4067                         rb_erase(parent, &root->inode_tree);
4068                         RB_CLEAR_NODE(parent);
4069                         spin_unlock(&root->inode_lock);
4070                         goto again;
4071                 }
4072         }
4073         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4074         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4075         spin_unlock(&root->inode_lock);
4076 }
4077
4078 static void inode_tree_del(struct inode *inode)
4079 {
4080         struct btrfs_root *root = BTRFS_I(inode)->root;
4081         int empty = 0;
4082
4083         spin_lock(&root->inode_lock);
4084         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4085                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4086                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4087                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4088         }
4089         spin_unlock(&root->inode_lock);
4090
4091         /*
4092          * Free space cache has inodes in the tree root, but the tree root has a
4093          * root_refs of 0, so this could end up dropping the tree root as a
4094          * snapshot, so we need the extra !root->fs_info->tree_root check to
4095          * make sure we don't drop it.
4096          */
4097         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4098             root != root->fs_info->tree_root) {
4099                 synchronize_srcu(&root->fs_info->subvol_srcu);
4100                 spin_lock(&root->inode_lock);
4101                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4102                 spin_unlock(&root->inode_lock);
4103                 if (empty)
4104                         btrfs_add_dead_root(root);
4105         }
4106 }
4107
4108 void btrfs_invalidate_inodes(struct btrfs_root *root)
4109 {
4110         struct rb_node *node;
4111         struct rb_node *prev;
4112         struct btrfs_inode *entry;
4113         struct inode *inode;
4114         u64 objectid = 0;
4115
4116         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4117
4118         spin_lock(&root->inode_lock);
4119 again:
4120         node = root->inode_tree.rb_node;
4121         prev = NULL;
4122         while (node) {
4123                 prev = node;
4124                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4125
4126                 if (objectid < btrfs_ino(&entry->vfs_inode))
4127                         node = node->rb_left;
4128                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4129                         node = node->rb_right;
4130                 else
4131                         break;
4132         }
4133         if (!node) {
4134                 while (prev) {
4135                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4136                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4137                                 node = prev;
4138                                 break;
4139                         }
4140                         prev = rb_next(prev);
4141                 }
4142         }
4143         while (node) {
4144                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4145                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4146                 inode = igrab(&entry->vfs_inode);
4147                 if (inode) {
4148                         spin_unlock(&root->inode_lock);
4149                         if (atomic_read(&inode->i_count) > 1)
4150                                 d_prune_aliases(inode);
4151                         /*
4152                          * btrfs_drop_inode will have it removed from
4153                          * the inode cache when its usage count
4154                          * hits zero.
4155                          */
4156                         iput(inode);
4157                         cond_resched();
4158                         spin_lock(&root->inode_lock);
4159                         goto again;
4160                 }
4161
4162                 if (cond_resched_lock(&root->inode_lock))
4163                         goto again;
4164
4165                 node = rb_next(node);
4166         }
4167         spin_unlock(&root->inode_lock);
4168 }
4169
4170 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4171 {
4172         struct btrfs_iget_args *args = p;
4173         inode->i_ino = args->ino;
4174         BTRFS_I(inode)->root = args->root;
4175         return 0;
4176 }
4177
4178 static int btrfs_find_actor(struct inode *inode, void *opaque)
4179 {
4180         struct btrfs_iget_args *args = opaque;
4181         return args->ino == btrfs_ino(inode) &&
4182                 args->root == BTRFS_I(inode)->root;
4183 }
4184
4185 static struct inode *btrfs_iget_locked(struct super_block *s,
4186                                        u64 objectid,
4187                                        struct btrfs_root *root)
4188 {
4189         struct inode *inode;
4190         struct btrfs_iget_args args;
4191         args.ino = objectid;
4192         args.root = root;
4193
4194         inode = iget5_locked(s, objectid, btrfs_find_actor,
4195                              btrfs_init_locked_inode,
4196                              (void *)&args);
4197         return inode;
4198 }
4199
4200 /* Get an inode object given its location and corresponding root.
4201  * Returns in *is_new if the inode was read from disk
4202  */
4203 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4204                          struct btrfs_root *root, int *new)
4205 {
4206         struct inode *inode;
4207
4208         inode = btrfs_iget_locked(s, location->objectid, root);
4209         if (!inode)
4210                 return ERR_PTR(-ENOMEM);
4211
4212         if (inode->i_state & I_NEW) {
4213                 BTRFS_I(inode)->root = root;
4214                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4215                 btrfs_read_locked_inode(inode);
4216                 if (!is_bad_inode(inode)) {
4217                         inode_tree_add(inode);
4218                         unlock_new_inode(inode);
4219                         if (new)
4220                                 *new = 1;
4221                 } else {
4222                         unlock_new_inode(inode);
4223                         iput(inode);
4224                         inode = ERR_PTR(-ESTALE);
4225                 }
4226         }
4227
4228         return inode;
4229 }
4230
4231 static struct inode *new_simple_dir(struct super_block *s,
4232                                     struct btrfs_key *key,
4233                                     struct btrfs_root *root)
4234 {
4235         struct inode *inode = new_inode(s);
4236
4237         if (!inode)
4238                 return ERR_PTR(-ENOMEM);
4239
4240         BTRFS_I(inode)->root = root;
4241         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4242         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4243
4244         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4245         inode->i_op = &btrfs_dir_ro_inode_operations;
4246         inode->i_fop = &simple_dir_operations;
4247         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4248         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4249
4250         return inode;
4251 }
4252
4253 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4254 {
4255         struct inode *inode;
4256         struct btrfs_root *root = BTRFS_I(dir)->root;
4257         struct btrfs_root *sub_root = root;
4258         struct btrfs_key location;
4259         int index;
4260         int ret = 0;
4261
4262         if (dentry->d_name.len > BTRFS_NAME_LEN)
4263                 return ERR_PTR(-ENAMETOOLONG);
4264
4265         if (unlikely(d_need_lookup(dentry))) {
4266                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4267                 kfree(dentry->d_fsdata);
4268                 dentry->d_fsdata = NULL;
4269                 /* This thing is hashed, drop it for now */
4270                 d_drop(dentry);
4271         } else {
4272                 ret = btrfs_inode_by_name(dir, dentry, &location);
4273         }
4274
4275         if (ret < 0)
4276                 return ERR_PTR(ret);
4277
4278         if (location.objectid == 0)
4279                 return NULL;
4280
4281         if (location.type == BTRFS_INODE_ITEM_KEY) {
4282                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4283                 return inode;
4284         }
4285
4286         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4287
4288         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4289         ret = fixup_tree_root_location(root, dir, dentry,
4290                                        &location, &sub_root);
4291         if (ret < 0) {
4292                 if (ret != -ENOENT)
4293                         inode = ERR_PTR(ret);
4294                 else
4295                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4296         } else {
4297                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4298         }
4299         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4300
4301         if (!IS_ERR(inode) && root != sub_root) {
4302                 down_read(&root->fs_info->cleanup_work_sem);
4303                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4304                         ret = btrfs_orphan_cleanup(sub_root);
4305                 up_read(&root->fs_info->cleanup_work_sem);
4306                 if (ret)
4307                         inode = ERR_PTR(ret);
4308         }
4309
4310         return inode;
4311 }
4312
4313 static int btrfs_dentry_delete(const struct dentry *dentry)
4314 {
4315         struct btrfs_root *root;
4316         struct inode *inode = dentry->d_inode;
4317
4318         if (!inode && !IS_ROOT(dentry))
4319                 inode = dentry->d_parent->d_inode;
4320
4321         if (inode) {
4322                 root = BTRFS_I(inode)->root;
4323                 if (btrfs_root_refs(&root->root_item) == 0)
4324                         return 1;
4325
4326                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4327                         return 1;
4328         }
4329         return 0;
4330 }
4331
4332 static void btrfs_dentry_release(struct dentry *dentry)
4333 {
4334         if (dentry->d_fsdata)
4335                 kfree(dentry->d_fsdata);
4336 }
4337
4338 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4339                                    unsigned int flags)
4340 {
4341         struct dentry *ret;
4342
4343         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4344         if (unlikely(d_need_lookup(dentry))) {
4345                 spin_lock(&dentry->d_lock);
4346                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4347                 spin_unlock(&dentry->d_lock);
4348         }
4349         return ret;
4350 }
4351
4352 unsigned char btrfs_filetype_table[] = {
4353         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4354 };
4355
4356 static int btrfs_real_readdir(struct file *filp, void *dirent,
4357                               filldir_t filldir)
4358 {
4359         struct inode *inode = filp->f_dentry->d_inode;
4360         struct btrfs_root *root = BTRFS_I(inode)->root;
4361         struct btrfs_item *item;
4362         struct btrfs_dir_item *di;
4363         struct btrfs_key key;
4364         struct btrfs_key found_key;
4365         struct btrfs_path *path;
4366         struct list_head ins_list;
4367         struct list_head del_list;
4368         int ret;
4369         struct extent_buffer *leaf;
4370         int slot;
4371         unsigned char d_type;
4372         int over = 0;
4373         u32 di_cur;
4374         u32 di_total;
4375         u32 di_len;
4376         int key_type = BTRFS_DIR_INDEX_KEY;
4377         char tmp_name[32];
4378         char *name_ptr;
4379         int name_len;
4380         int is_curr = 0;        /* filp->f_pos points to the current index? */
4381
4382         /* FIXME, use a real flag for deciding about the key type */
4383         if (root->fs_info->tree_root == root)
4384                 key_type = BTRFS_DIR_ITEM_KEY;
4385
4386         /* special case for "." */
4387         if (filp->f_pos == 0) {
4388                 over = filldir(dirent, ".", 1,
4389                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4390                 if (over)
4391                         return 0;
4392                 filp->f_pos = 1;
4393         }
4394         /* special case for .., just use the back ref */
4395         if (filp->f_pos == 1) {
4396                 u64 pino = parent_ino(filp->f_path.dentry);
4397                 over = filldir(dirent, "..", 2,
4398                                filp->f_pos, pino, DT_DIR);
4399                 if (over)
4400                         return 0;
4401                 filp->f_pos = 2;
4402         }
4403         path = btrfs_alloc_path();
4404         if (!path)
4405                 return -ENOMEM;
4406
4407         path->reada = 1;
4408
4409         if (key_type == BTRFS_DIR_INDEX_KEY) {
4410                 INIT_LIST_HEAD(&ins_list);
4411                 INIT_LIST_HEAD(&del_list);
4412                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4413         }
4414
4415         btrfs_set_key_type(&key, key_type);
4416         key.offset = filp->f_pos;
4417         key.objectid = btrfs_ino(inode);
4418
4419         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4420         if (ret < 0)
4421                 goto err;
4422
4423         while (1) {
4424                 leaf = path->nodes[0];
4425                 slot = path->slots[0];
4426                 if (slot >= btrfs_header_nritems(leaf)) {
4427                         ret = btrfs_next_leaf(root, path);
4428                         if (ret < 0)
4429                                 goto err;
4430                         else if (ret > 0)
4431                                 break;
4432                         continue;
4433                 }
4434
4435                 item = btrfs_item_nr(leaf, slot);
4436                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4437
4438                 if (found_key.objectid != key.objectid)
4439                         break;
4440                 if (btrfs_key_type(&found_key) != key_type)
4441                         break;
4442                 if (found_key.offset < filp->f_pos)
4443                         goto next;
4444                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4445                     btrfs_should_delete_dir_index(&del_list,
4446                                                   found_key.offset))
4447                         goto next;
4448
4449                 filp->f_pos = found_key.offset;
4450                 is_curr = 1;
4451
4452                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4453                 di_cur = 0;
4454                 di_total = btrfs_item_size(leaf, item);
4455
4456                 while (di_cur < di_total) {
4457                         struct btrfs_key location;
4458
4459                         if (verify_dir_item(root, leaf, di))
4460                                 break;
4461
4462                         name_len = btrfs_dir_name_len(leaf, di);
4463                         if (name_len <= sizeof(tmp_name)) {
4464                                 name_ptr = tmp_name;
4465                         } else {
4466                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4467                                 if (!name_ptr) {
4468                                         ret = -ENOMEM;
4469                                         goto err;
4470                                 }
4471                         }
4472                         read_extent_buffer(leaf, name_ptr,
4473                                            (unsigned long)(di + 1), name_len);
4474
4475                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4476                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4477
4478
4479                         /* is this a reference to our own snapshot? If so
4480                          * skip it.
4481                          *
4482                          * In contrast to old kernels, we insert the snapshot's
4483                          * dir item and dir index after it has been created, so
4484                          * we won't find a reference to our own snapshot. We
4485                          * still keep the following code for backward
4486                          * compatibility.
4487                          */
4488                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4489                             location.objectid == root->root_key.objectid) {
4490                                 over = 0;
4491                                 goto skip;
4492                         }
4493                         over = filldir(dirent, name_ptr, name_len,
4494                                        found_key.offset, location.objectid,
4495                                        d_type);
4496
4497 skip:
4498                         if (name_ptr != tmp_name)
4499                                 kfree(name_ptr);
4500
4501                         if (over)
4502                                 goto nopos;
4503                         di_len = btrfs_dir_name_len(leaf, di) +
4504                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4505                         di_cur += di_len;
4506                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4507                 }
4508 next:
4509                 path->slots[0]++;
4510         }
4511
4512         if (key_type == BTRFS_DIR_INDEX_KEY) {
4513                 if (is_curr)
4514                         filp->f_pos++;
4515                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4516                                                       &ins_list);
4517                 if (ret)
4518                         goto nopos;
4519         }
4520
4521         /* Reached end of directory/root. Bump pos past the last item. */
4522         if (key_type == BTRFS_DIR_INDEX_KEY)
4523                 /*
4524                  * 32-bit glibc will use getdents64, but then strtol -
4525                  * so the last number we can serve is this.
4526                  */
4527                 filp->f_pos = 0x7fffffff;
4528         else
4529                 filp->f_pos++;
4530 nopos:
4531         ret = 0;
4532 err:
4533         if (key_type == BTRFS_DIR_INDEX_KEY)
4534                 btrfs_put_delayed_items(&ins_list, &del_list);
4535         btrfs_free_path(path);
4536         return ret;
4537 }
4538
4539 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4540 {
4541         struct btrfs_root *root = BTRFS_I(inode)->root;
4542         struct btrfs_trans_handle *trans;
4543         int ret = 0;
4544         bool nolock = false;
4545
4546         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4547                 return 0;
4548
4549         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4550                 nolock = true;
4551
4552         if (wbc->sync_mode == WB_SYNC_ALL) {
4553                 if (nolock)
4554                         trans = btrfs_join_transaction_nolock(root);
4555                 else
4556                         trans = btrfs_join_transaction(root);
4557                 if (IS_ERR(trans))
4558                         return PTR_ERR(trans);
4559                 ret = btrfs_commit_transaction(trans, root);
4560         }
4561         return ret;
4562 }
4563
4564 /*
4565  * This is somewhat expensive, updating the tree every time the
4566  * inode changes.  But, it is most likely to find the inode in cache.
4567  * FIXME, needs more benchmarking...there are no reasons other than performance
4568  * to keep or drop this code.
4569  */
4570 int btrfs_dirty_inode(struct inode *inode)
4571 {
4572         struct btrfs_root *root = BTRFS_I(inode)->root;
4573         struct btrfs_trans_handle *trans;
4574         int ret;
4575
4576         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4577                 return 0;
4578
4579         trans = btrfs_join_transaction(root);
4580         if (IS_ERR(trans))
4581                 return PTR_ERR(trans);
4582
4583         ret = btrfs_update_inode(trans, root, inode);
4584         if (ret && ret == -ENOSPC) {
4585                 /* whoops, lets try again with the full transaction */
4586                 btrfs_end_transaction(trans, root);
4587                 trans = btrfs_start_transaction(root, 1);
4588                 if (IS_ERR(trans))
4589                         return PTR_ERR(trans);
4590
4591                 ret = btrfs_update_inode(trans, root, inode);
4592         }
4593         btrfs_end_transaction(trans, root);
4594         if (BTRFS_I(inode)->delayed_node)
4595                 btrfs_balance_delayed_items(root);
4596
4597         return ret;
4598 }
4599
4600 /*
4601  * This is a copy of file_update_time.  We need this so we can return error on
4602  * ENOSPC for updating the inode in the case of file write and mmap writes.
4603  */
4604 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4605                              int flags)
4606 {
4607         struct btrfs_root *root = BTRFS_I(inode)->root;
4608
4609         if (btrfs_root_readonly(root))
4610                 return -EROFS;
4611
4612         if (flags & S_VERSION)
4613                 inode_inc_iversion(inode);
4614         if (flags & S_CTIME)
4615                 inode->i_ctime = *now;
4616         if (flags & S_MTIME)
4617                 inode->i_mtime = *now;
4618         if (flags & S_ATIME)
4619                 inode->i_atime = *now;
4620         return btrfs_dirty_inode(inode);
4621 }
4622
4623 /*
4624  * find the highest existing sequence number in a directory
4625  * and then set the in-memory index_cnt variable to reflect
4626  * free sequence numbers
4627  */
4628 static int btrfs_set_inode_index_count(struct inode *inode)
4629 {
4630         struct btrfs_root *root = BTRFS_I(inode)->root;
4631         struct btrfs_key key, found_key;
4632         struct btrfs_path *path;
4633         struct extent_buffer *leaf;
4634         int ret;
4635
4636         key.objectid = btrfs_ino(inode);
4637         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4638         key.offset = (u64)-1;
4639
4640         path = btrfs_alloc_path();
4641         if (!path)
4642                 return -ENOMEM;
4643
4644         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4645         if (ret < 0)
4646                 goto out;
4647         /* FIXME: we should be able to handle this */
4648         if (ret == 0)
4649                 goto out;
4650         ret = 0;
4651
4652         /*
4653          * MAGIC NUMBER EXPLANATION:
4654          * since we search a directory based on f_pos we have to start at 2
4655          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4656          * else has to start at 2
4657          */
4658         if (path->slots[0] == 0) {
4659                 BTRFS_I(inode)->index_cnt = 2;
4660                 goto out;
4661         }
4662
4663         path->slots[0]--;
4664
4665         leaf = path->nodes[0];
4666         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4667
4668         if (found_key.objectid != btrfs_ino(inode) ||
4669             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4670                 BTRFS_I(inode)->index_cnt = 2;
4671                 goto out;
4672         }
4673
4674         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4675 out:
4676         btrfs_free_path(path);
4677         return ret;
4678 }
4679
4680 /*
4681  * helper to find a free sequence number in a given directory.  This current
4682  * code is very simple, later versions will do smarter things in the btree
4683  */
4684 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4685 {
4686         int ret = 0;
4687
4688         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4689                 ret = btrfs_inode_delayed_dir_index_count(dir);
4690                 if (ret) {
4691                         ret = btrfs_set_inode_index_count(dir);
4692                         if (ret)
4693                                 return ret;
4694                 }
4695         }
4696
4697         *index = BTRFS_I(dir)->index_cnt;
4698         BTRFS_I(dir)->index_cnt++;
4699
4700         return ret;
4701 }
4702
4703 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4704                                      struct btrfs_root *root,
4705                                      struct inode *dir,
4706                                      const char *name, int name_len,
4707                                      u64 ref_objectid, u64 objectid,
4708                                      umode_t mode, u64 *index)
4709 {
4710         struct inode *inode;
4711         struct btrfs_inode_item *inode_item;
4712         struct btrfs_key *location;
4713         struct btrfs_path *path;
4714         struct btrfs_inode_ref *ref;
4715         struct btrfs_key key[2];
4716         u32 sizes[2];
4717         unsigned long ptr;
4718         int ret;
4719         int owner;
4720
4721         path = btrfs_alloc_path();
4722         if (!path)
4723                 return ERR_PTR(-ENOMEM);
4724
4725         inode = new_inode(root->fs_info->sb);
4726         if (!inode) {
4727                 btrfs_free_path(path);
4728                 return ERR_PTR(-ENOMEM);
4729         }
4730
4731         /*
4732          * we have to initialize this early, so we can reclaim the inode
4733          * number if we fail afterwards in this function.
4734          */
4735         inode->i_ino = objectid;
4736
4737         if (dir) {
4738                 trace_btrfs_inode_request(dir);
4739
4740                 ret = btrfs_set_inode_index(dir, index);
4741                 if (ret) {
4742                         btrfs_free_path(path);
4743                         iput(inode);
4744                         return ERR_PTR(ret);
4745                 }
4746         }
4747         /*
4748          * index_cnt is ignored for everything but a dir,
4749          * btrfs_get_inode_index_count has an explanation for the magic
4750          * number
4751          */
4752         BTRFS_I(inode)->index_cnt = 2;
4753         BTRFS_I(inode)->root = root;
4754         BTRFS_I(inode)->generation = trans->transid;
4755         inode->i_generation = BTRFS_I(inode)->generation;
4756
4757         /*
4758          * We could have gotten an inode number from somebody who was fsynced
4759          * and then removed in this same transaction, so let's just set full
4760          * sync since it will be a full sync anyway and this will blow away the
4761          * old info in the log.
4762          */
4763         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4764
4765         if (S_ISDIR(mode))
4766                 owner = 0;
4767         else
4768                 owner = 1;
4769
4770         key[0].objectid = objectid;
4771         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4772         key[0].offset = 0;
4773
4774         /*
4775          * Start new inodes with an inode_ref. This is slightly more
4776          * efficient for small numbers of hard links since they will
4777          * be packed into one item. Extended refs will kick in if we
4778          * add more hard links than can fit in the ref item.
4779          */
4780         key[1].objectid = objectid;
4781         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4782         key[1].offset = ref_objectid;
4783
4784         sizes[0] = sizeof(struct btrfs_inode_item);
4785         sizes[1] = name_len + sizeof(*ref);
4786
4787         path->leave_spinning = 1;
4788         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4789         if (ret != 0)
4790                 goto fail;
4791
4792         inode_init_owner(inode, dir, mode);
4793         inode_set_bytes(inode, 0);
4794         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4795         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4796                                   struct btrfs_inode_item);
4797         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4798                              sizeof(*inode_item));
4799         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4800
4801         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4802                              struct btrfs_inode_ref);
4803         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4804         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4805         ptr = (unsigned long)(ref + 1);
4806         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4807
4808         btrfs_mark_buffer_dirty(path->nodes[0]);
4809         btrfs_free_path(path);
4810
4811         location = &BTRFS_I(inode)->location;
4812         location->objectid = objectid;
4813         location->offset = 0;
4814         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4815
4816         btrfs_inherit_iflags(inode, dir);
4817
4818         if (S_ISREG(mode)) {
4819                 if (btrfs_test_opt(root, NODATASUM))
4820                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4821                 if (btrfs_test_opt(root, NODATACOW))
4822                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4823         }
4824
4825         insert_inode_hash(inode);
4826         inode_tree_add(inode);
4827
4828         trace_btrfs_inode_new(inode);
4829         btrfs_set_inode_last_trans(trans, inode);
4830
4831         btrfs_update_root_times(trans, root);
4832
4833         return inode;
4834 fail:
4835         if (dir)
4836                 BTRFS_I(dir)->index_cnt--;
4837         btrfs_free_path(path);
4838         iput(inode);
4839         return ERR_PTR(ret);
4840 }
4841
4842 static inline u8 btrfs_inode_type(struct inode *inode)
4843 {
4844         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4845 }
4846
4847 /*
4848  * utility function to add 'inode' into 'parent_inode' with
4849  * a give name and a given sequence number.
4850  * if 'add_backref' is true, also insert a backref from the
4851  * inode to the parent directory.
4852  */
4853 int btrfs_add_link(struct btrfs_trans_handle *trans,
4854                    struct inode *parent_inode, struct inode *inode,
4855                    const char *name, int name_len, int add_backref, u64 index)
4856 {
4857         int ret = 0;
4858         struct btrfs_key key;
4859         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4860         u64 ino = btrfs_ino(inode);
4861         u64 parent_ino = btrfs_ino(parent_inode);
4862
4863         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4864                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4865         } else {
4866                 key.objectid = ino;
4867                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4868                 key.offset = 0;
4869         }
4870
4871         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4872                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4873                                          key.objectid, root->root_key.objectid,
4874                                          parent_ino, index, name, name_len);
4875         } else if (add_backref) {
4876                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4877                                              parent_ino, index);
4878         }
4879
4880         /* Nothing to clean up yet */
4881         if (ret)
4882                 return ret;
4883
4884         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4885                                     parent_inode, &key,
4886                                     btrfs_inode_type(inode), index);
4887         if (ret == -EEXIST || ret == -EOVERFLOW)
4888                 goto fail_dir_item;
4889         else if (ret) {
4890                 btrfs_abort_transaction(trans, root, ret);
4891                 return ret;
4892         }
4893
4894         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4895                            name_len * 2);
4896         inode_inc_iversion(parent_inode);
4897         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4898         ret = btrfs_update_inode(trans, root, parent_inode);
4899         if (ret)
4900                 btrfs_abort_transaction(trans, root, ret);
4901         return ret;
4902
4903 fail_dir_item:
4904         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4905                 u64 local_index;
4906                 int err;
4907                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4908                                  key.objectid, root->root_key.objectid,
4909                                  parent_ino, &local_index, name, name_len);
4910
4911         } else if (add_backref) {
4912                 u64 local_index;
4913                 int err;
4914
4915                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4916                                           ino, parent_ino, &local_index);
4917         }
4918         return ret;
4919 }
4920
4921 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4922                             struct inode *dir, struct dentry *dentry,
4923                             struct inode *inode, int backref, u64 index)
4924 {
4925         int err = btrfs_add_link(trans, dir, inode,
4926                                  dentry->d_name.name, dentry->d_name.len,
4927                                  backref, index);
4928         if (err > 0)
4929                 err = -EEXIST;
4930         return err;
4931 }
4932
4933 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4934                         umode_t mode, dev_t rdev)
4935 {
4936         struct btrfs_trans_handle *trans;
4937         struct btrfs_root *root = BTRFS_I(dir)->root;
4938         struct inode *inode = NULL;
4939         int err;
4940         int drop_inode = 0;
4941         u64 objectid;
4942         u64 index = 0;
4943
4944         if (!new_valid_dev(rdev))
4945                 return -EINVAL;
4946
4947         /*
4948          * 2 for inode item and ref
4949          * 2 for dir items
4950          * 1 for xattr if selinux is on
4951          */
4952         trans = btrfs_start_transaction(root, 5);
4953         if (IS_ERR(trans))
4954                 return PTR_ERR(trans);
4955
4956         err = btrfs_find_free_ino(root, &objectid);
4957         if (err)
4958                 goto out_unlock;
4959
4960         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4961                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4962                                 mode, &index);
4963         if (IS_ERR(inode)) {
4964                 err = PTR_ERR(inode);
4965                 goto out_unlock;
4966         }
4967
4968         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4969         if (err) {
4970                 drop_inode = 1;
4971                 goto out_unlock;
4972         }
4973
4974         err = btrfs_update_inode(trans, root, inode);
4975         if (err) {
4976                 drop_inode = 1;
4977                 goto out_unlock;
4978         }
4979
4980         /*
4981         * If the active LSM wants to access the inode during
4982         * d_instantiate it needs these. Smack checks to see
4983         * if the filesystem supports xattrs by looking at the
4984         * ops vector.
4985         */
4986
4987         inode->i_op = &btrfs_special_inode_operations;
4988         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4989         if (err)
4990                 drop_inode = 1;
4991         else {
4992                 init_special_inode(inode, inode->i_mode, rdev);
4993                 btrfs_update_inode(trans, root, inode);
4994                 d_instantiate(dentry, inode);
4995         }
4996 out_unlock:
4997         btrfs_end_transaction(trans, root);
4998         btrfs_btree_balance_dirty(root);
4999         if (drop_inode) {
5000                 inode_dec_link_count(inode);
5001                 iput(inode);
5002         }
5003         return err;
5004 }
5005
5006 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5007                         umode_t mode, bool excl)
5008 {
5009         struct btrfs_trans_handle *trans;
5010         struct btrfs_root *root = BTRFS_I(dir)->root;
5011         struct inode *inode = NULL;
5012         int drop_inode_on_err = 0;
5013         int err;
5014         u64 objectid;
5015         u64 index = 0;
5016
5017         /*
5018          * 2 for inode item and ref
5019          * 2 for dir items
5020          * 1 for xattr if selinux is on
5021          */
5022         trans = btrfs_start_transaction(root, 5);
5023         if (IS_ERR(trans))
5024                 return PTR_ERR(trans);
5025
5026         err = btrfs_find_free_ino(root, &objectid);
5027         if (err)
5028                 goto out_unlock;
5029
5030         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5031                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5032                                 mode, &index);
5033         if (IS_ERR(inode)) {
5034                 err = PTR_ERR(inode);
5035                 goto out_unlock;
5036         }
5037         drop_inode_on_err = 1;
5038
5039         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5040         if (err)
5041                 goto out_unlock;
5042
5043         err = btrfs_update_inode(trans, root, inode);
5044         if (err)
5045                 goto out_unlock;
5046
5047         /*
5048         * If the active LSM wants to access the inode during
5049         * d_instantiate it needs these. Smack checks to see
5050         * if the filesystem supports xattrs by looking at the
5051         * ops vector.
5052         */
5053         inode->i_fop = &btrfs_file_operations;
5054         inode->i_op = &btrfs_file_inode_operations;
5055
5056         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5057         if (err)
5058                 goto out_unlock;
5059
5060         inode->i_mapping->a_ops = &btrfs_aops;
5061         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5062         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5063         d_instantiate(dentry, inode);
5064
5065 out_unlock:
5066         btrfs_end_transaction(trans, root);
5067         if (err && drop_inode_on_err) {
5068                 inode_dec_link_count(inode);
5069                 iput(inode);
5070         }
5071         btrfs_btree_balance_dirty(root);
5072         return err;
5073 }
5074
5075 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5076                       struct dentry *dentry)
5077 {
5078         struct btrfs_trans_handle *trans;
5079         struct btrfs_root *root = BTRFS_I(dir)->root;
5080         struct inode *inode = old_dentry->d_inode;
5081         u64 index;
5082         int err;
5083         int drop_inode = 0;
5084
5085         /* do not allow sys_link's with other subvols of the same device */
5086         if (root->objectid != BTRFS_I(inode)->root->objectid)
5087                 return -EXDEV;
5088
5089         if (inode->i_nlink >= BTRFS_LINK_MAX)
5090                 return -EMLINK;
5091
5092         err = btrfs_set_inode_index(dir, &index);
5093         if (err)
5094                 goto fail;
5095
5096         /*
5097          * 2 items for inode and inode ref
5098          * 2 items for dir items
5099          * 1 item for parent inode
5100          */
5101         trans = btrfs_start_transaction(root, 5);
5102         if (IS_ERR(trans)) {
5103                 err = PTR_ERR(trans);
5104                 goto fail;
5105         }
5106
5107         btrfs_inc_nlink(inode);
5108         inode_inc_iversion(inode);
5109         inode->i_ctime = CURRENT_TIME;
5110         ihold(inode);
5111         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5112
5113         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5114
5115         if (err) {
5116                 drop_inode = 1;
5117         } else {
5118                 struct dentry *parent = dentry->d_parent;
5119                 err = btrfs_update_inode(trans, root, inode);
5120                 if (err)
5121                         goto fail;
5122                 d_instantiate(dentry, inode);
5123                 btrfs_log_new_name(trans, inode, NULL, parent);
5124         }
5125
5126         btrfs_end_transaction(trans, root);
5127 fail:
5128         if (drop_inode) {
5129                 inode_dec_link_count(inode);
5130                 iput(inode);
5131         }
5132         btrfs_btree_balance_dirty(root);
5133         return err;
5134 }
5135
5136 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5137 {
5138         struct inode *inode = NULL;
5139         struct btrfs_trans_handle *trans;
5140         struct btrfs_root *root = BTRFS_I(dir)->root;
5141         int err = 0;
5142         int drop_on_err = 0;
5143         u64 objectid = 0;
5144         u64 index = 0;
5145
5146         /*
5147          * 2 items for inode and ref
5148          * 2 items for dir items
5149          * 1 for xattr if selinux is on
5150          */
5151         trans = btrfs_start_transaction(root, 5);
5152         if (IS_ERR(trans))
5153                 return PTR_ERR(trans);
5154
5155         err = btrfs_find_free_ino(root, &objectid);
5156         if (err)
5157                 goto out_fail;
5158
5159         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5160                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5161                                 S_IFDIR | mode, &index);
5162         if (IS_ERR(inode)) {
5163                 err = PTR_ERR(inode);
5164                 goto out_fail;
5165         }
5166
5167         drop_on_err = 1;
5168
5169         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5170         if (err)
5171                 goto out_fail;
5172
5173         inode->i_op = &btrfs_dir_inode_operations;
5174         inode->i_fop = &btrfs_dir_file_operations;
5175
5176         btrfs_i_size_write(inode, 0);
5177         err = btrfs_update_inode(trans, root, inode);
5178         if (err)
5179                 goto out_fail;
5180
5181         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5182                              dentry->d_name.len, 0, index);
5183         if (err)
5184                 goto out_fail;
5185
5186         d_instantiate(dentry, inode);
5187         drop_on_err = 0;
5188
5189 out_fail:
5190         btrfs_end_transaction(trans, root);
5191         if (drop_on_err)
5192                 iput(inode);
5193         btrfs_btree_balance_dirty(root);
5194         return err;
5195 }
5196
5197 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5198  * and an extent that you want to insert, deal with overlap and insert
5199  * the new extent into the tree.
5200  */
5201 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5202                                 struct extent_map *existing,
5203                                 struct extent_map *em,
5204                                 u64 map_start, u64 map_len)
5205 {
5206         u64 start_diff;
5207
5208         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5209         start_diff = map_start - em->start;
5210         em->start = map_start;
5211         em->len = map_len;
5212         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5213             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5214                 em->block_start += start_diff;
5215                 em->block_len -= start_diff;
5216         }
5217         return add_extent_mapping(em_tree, em);
5218 }
5219
5220 static noinline int uncompress_inline(struct btrfs_path *path,
5221                                       struct inode *inode, struct page *page,
5222                                       size_t pg_offset, u64 extent_offset,
5223                                       struct btrfs_file_extent_item *item)
5224 {
5225         int ret;
5226         struct extent_buffer *leaf = path->nodes[0];
5227         char *tmp;
5228         size_t max_size;
5229         unsigned long inline_size;
5230         unsigned long ptr;
5231         int compress_type;
5232
5233         WARN_ON(pg_offset != 0);
5234         compress_type = btrfs_file_extent_compression(leaf, item);
5235         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5236         inline_size = btrfs_file_extent_inline_item_len(leaf,
5237                                         btrfs_item_nr(leaf, path->slots[0]));
5238         tmp = kmalloc(inline_size, GFP_NOFS);
5239         if (!tmp)
5240                 return -ENOMEM;
5241         ptr = btrfs_file_extent_inline_start(item);
5242
5243         read_extent_buffer(leaf, tmp, ptr, inline_size);
5244
5245         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5246         ret = btrfs_decompress(compress_type, tmp, page,
5247                                extent_offset, inline_size, max_size);
5248         if (ret) {
5249                 char *kaddr = kmap_atomic(page);
5250                 unsigned long copy_size = min_t(u64,
5251                                   PAGE_CACHE_SIZE - pg_offset,
5252                                   max_size - extent_offset);
5253                 memset(kaddr + pg_offset, 0, copy_size);
5254                 kunmap_atomic(kaddr);
5255         }
5256         kfree(tmp);
5257         return 0;
5258 }
5259
5260 /*
5261  * a bit scary, this does extent mapping from logical file offset to the disk.
5262  * the ugly parts come from merging extents from the disk with the in-ram
5263  * representation.  This gets more complex because of the data=ordered code,
5264  * where the in-ram extents might be locked pending data=ordered completion.
5265  *
5266  * This also copies inline extents directly into the page.
5267  */
5268
5269 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5270                                     size_t pg_offset, u64 start, u64 len,
5271                                     int create)
5272 {
5273         int ret;
5274         int err = 0;
5275         u64 bytenr;
5276         u64 extent_start = 0;
5277         u64 extent_end = 0;
5278         u64 objectid = btrfs_ino(inode);
5279         u32 found_type;
5280         struct btrfs_path *path = NULL;
5281         struct btrfs_root *root = BTRFS_I(inode)->root;
5282         struct btrfs_file_extent_item *item;
5283         struct extent_buffer *leaf;
5284         struct btrfs_key found_key;
5285         struct extent_map *em = NULL;
5286         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5287         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5288         struct btrfs_trans_handle *trans = NULL;
5289         int compress_type;
5290
5291 again:
5292         read_lock(&em_tree->lock);
5293         em = lookup_extent_mapping(em_tree, start, len);
5294         if (em)
5295                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5296         read_unlock(&em_tree->lock);
5297
5298         if (em) {
5299                 if (em->start > start || em->start + em->len <= start)
5300                         free_extent_map(em);
5301                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5302                         free_extent_map(em);
5303                 else
5304                         goto out;
5305         }
5306         em = alloc_extent_map();
5307         if (!em) {
5308                 err = -ENOMEM;
5309                 goto out;
5310         }
5311         em->bdev = root->fs_info->fs_devices->latest_bdev;
5312         em->start = EXTENT_MAP_HOLE;
5313         em->orig_start = EXTENT_MAP_HOLE;
5314         em->len = (u64)-1;
5315         em->block_len = (u64)-1;
5316
5317         if (!path) {
5318                 path = btrfs_alloc_path();
5319                 if (!path) {
5320                         err = -ENOMEM;
5321                         goto out;
5322                 }
5323                 /*
5324                  * Chances are we'll be called again, so go ahead and do
5325                  * readahead
5326                  */
5327                 path->reada = 1;
5328         }
5329
5330         ret = btrfs_lookup_file_extent(trans, root, path,
5331                                        objectid, start, trans != NULL);
5332         if (ret < 0) {
5333                 err = ret;
5334                 goto out;
5335         }
5336
5337         if (ret != 0) {
5338                 if (path->slots[0] == 0)
5339                         goto not_found;
5340                 path->slots[0]--;
5341         }
5342
5343         leaf = path->nodes[0];
5344         item = btrfs_item_ptr(leaf, path->slots[0],
5345                               struct btrfs_file_extent_item);
5346         /* are we inside the extent that was found? */
5347         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5348         found_type = btrfs_key_type(&found_key);
5349         if (found_key.objectid != objectid ||
5350             found_type != BTRFS_EXTENT_DATA_KEY) {
5351                 goto not_found;
5352         }
5353
5354         found_type = btrfs_file_extent_type(leaf, item);
5355         extent_start = found_key.offset;
5356         compress_type = btrfs_file_extent_compression(leaf, item);
5357         if (found_type == BTRFS_FILE_EXTENT_REG ||
5358             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5359                 extent_end = extent_start +
5360                        btrfs_file_extent_num_bytes(leaf, item);
5361         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5362                 size_t size;
5363                 size = btrfs_file_extent_inline_len(leaf, item);
5364                 extent_end = (extent_start + size + root->sectorsize - 1) &
5365                         ~((u64)root->sectorsize - 1);
5366         }
5367
5368         if (start >= extent_end) {
5369                 path->slots[0]++;
5370                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5371                         ret = btrfs_next_leaf(root, path);
5372                         if (ret < 0) {
5373                                 err = ret;
5374                                 goto out;
5375                         }
5376                         if (ret > 0)
5377                                 goto not_found;
5378                         leaf = path->nodes[0];
5379                 }
5380                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5381                 if (found_key.objectid != objectid ||
5382                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5383                         goto not_found;
5384                 if (start + len <= found_key.offset)
5385                         goto not_found;
5386                 em->start = start;
5387                 em->orig_start = start;
5388                 em->len = found_key.offset - start;
5389                 goto not_found_em;
5390         }
5391
5392         if (found_type == BTRFS_FILE_EXTENT_REG ||
5393             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5394                 em->start = extent_start;
5395                 em->len = extent_end - extent_start;
5396                 em->orig_start = extent_start -
5397                                  btrfs_file_extent_offset(leaf, item);
5398                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5399                                                                       item);
5400                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5401                 if (bytenr == 0) {
5402                         em->block_start = EXTENT_MAP_HOLE;
5403                         goto insert;
5404                 }
5405                 if (compress_type != BTRFS_COMPRESS_NONE) {
5406                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5407                         em->compress_type = compress_type;
5408                         em->block_start = bytenr;
5409                         em->block_len = em->orig_block_len;
5410                 } else {
5411                         bytenr += btrfs_file_extent_offset(leaf, item);
5412                         em->block_start = bytenr;
5413                         em->block_len = em->len;
5414                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5415                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5416                 }
5417                 goto insert;
5418         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5419                 unsigned long ptr;
5420                 char *map;
5421                 size_t size;
5422                 size_t extent_offset;
5423                 size_t copy_size;
5424
5425                 em->block_start = EXTENT_MAP_INLINE;
5426                 if (!page || create) {
5427                         em->start = extent_start;
5428                         em->len = extent_end - extent_start;
5429                         goto out;
5430                 }
5431
5432                 size = btrfs_file_extent_inline_len(leaf, item);
5433                 extent_offset = page_offset(page) + pg_offset - extent_start;
5434                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5435                                 size - extent_offset);
5436                 em->start = extent_start + extent_offset;
5437                 em->len = (copy_size + root->sectorsize - 1) &
5438                         ~((u64)root->sectorsize - 1);
5439                 em->orig_block_len = em->len;
5440                 em->orig_start = em->start;
5441                 if (compress_type) {
5442                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5443                         em->compress_type = compress_type;
5444                 }
5445                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5446                 if (create == 0 && !PageUptodate(page)) {
5447                         if (btrfs_file_extent_compression(leaf, item) !=
5448                             BTRFS_COMPRESS_NONE) {
5449                                 ret = uncompress_inline(path, inode, page,
5450                                                         pg_offset,
5451                                                         extent_offset, item);
5452                                 BUG_ON(ret); /* -ENOMEM */
5453                         } else {
5454                                 map = kmap(page);
5455                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5456                                                    copy_size);
5457                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5458                                         memset(map + pg_offset + copy_size, 0,
5459                                                PAGE_CACHE_SIZE - pg_offset -
5460                                                copy_size);
5461                                 }
5462                                 kunmap(page);
5463                         }
5464                         flush_dcache_page(page);
5465                 } else if (create && PageUptodate(page)) {
5466                         BUG();
5467                         if (!trans) {
5468                                 kunmap(page);
5469                                 free_extent_map(em);
5470                                 em = NULL;
5471
5472                                 btrfs_release_path(path);
5473                                 trans = btrfs_join_transaction(root);
5474
5475                                 if (IS_ERR(trans))
5476                                         return ERR_CAST(trans);
5477                                 goto again;
5478                         }
5479                         map = kmap(page);
5480                         write_extent_buffer(leaf, map + pg_offset, ptr,
5481                                             copy_size);
5482                         kunmap(page);
5483                         btrfs_mark_buffer_dirty(leaf);
5484                 }
5485                 set_extent_uptodate(io_tree, em->start,
5486                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5487                 goto insert;
5488         } else {
5489                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5490         }
5491 not_found:
5492         em->start = start;
5493         em->orig_start = start;
5494         em->len = len;
5495 not_found_em:
5496         em->block_start = EXTENT_MAP_HOLE;
5497         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5498 insert:
5499         btrfs_release_path(path);
5500         if (em->start > start || extent_map_end(em) <= start) {
5501                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5502                        "[%llu %llu]\n", (unsigned long long)em->start,
5503                        (unsigned long long)em->len,
5504                        (unsigned long long)start,
5505                        (unsigned long long)len);
5506                 err = -EIO;
5507                 goto out;
5508         }
5509
5510         err = 0;
5511         write_lock(&em_tree->lock);
5512         ret = add_extent_mapping(em_tree, em);
5513         /* it is possible that someone inserted the extent into the tree
5514          * while we had the lock dropped.  It is also possible that
5515          * an overlapping map exists in the tree
5516          */
5517         if (ret == -EEXIST) {
5518                 struct extent_map *existing;
5519
5520                 ret = 0;
5521
5522                 existing = lookup_extent_mapping(em_tree, start, len);
5523                 if (existing && (existing->start > start ||
5524                     existing->start + existing->len <= start)) {
5525                         free_extent_map(existing);
5526                         existing = NULL;
5527                 }
5528                 if (!existing) {
5529                         existing = lookup_extent_mapping(em_tree, em->start,
5530                                                          em->len);
5531                         if (existing) {
5532                                 err = merge_extent_mapping(em_tree, existing,
5533                                                            em, start,
5534                                                            root->sectorsize);
5535                                 free_extent_map(existing);
5536                                 if (err) {
5537                                         free_extent_map(em);
5538                                         em = NULL;
5539                                 }
5540                         } else {
5541                                 err = -EIO;
5542                                 free_extent_map(em);
5543                                 em = NULL;
5544                         }
5545                 } else {
5546                         free_extent_map(em);
5547                         em = existing;
5548                         err = 0;
5549                 }
5550         }
5551         write_unlock(&em_tree->lock);
5552 out:
5553
5554         if (em)
5555                 trace_btrfs_get_extent(root, em);
5556
5557         if (path)
5558                 btrfs_free_path(path);
5559         if (trans) {
5560                 ret = btrfs_end_transaction(trans, root);
5561                 if (!err)
5562                         err = ret;
5563         }
5564         if (err) {
5565                 free_extent_map(em);
5566                 return ERR_PTR(err);
5567         }
5568         BUG_ON(!em); /* Error is always set */
5569         return em;
5570 }
5571
5572 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5573                                            size_t pg_offset, u64 start, u64 len,
5574                                            int create)
5575 {
5576         struct extent_map *em;
5577         struct extent_map *hole_em = NULL;
5578         u64 range_start = start;
5579         u64 end;
5580         u64 found;
5581         u64 found_end;
5582         int err = 0;
5583
5584         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5585         if (IS_ERR(em))
5586                 return em;
5587         if (em) {
5588                 /*
5589                  * if our em maps to a hole, there might
5590                  * actually be delalloc bytes behind it
5591                  */
5592                 if (em->block_start != EXTENT_MAP_HOLE)
5593                         return em;
5594                 else
5595                         hole_em = em;
5596         }
5597
5598         /* check to see if we've wrapped (len == -1 or similar) */
5599         end = start + len;
5600         if (end < start)
5601                 end = (u64)-1;
5602         else
5603                 end -= 1;
5604
5605         em = NULL;
5606
5607         /* ok, we didn't find anything, lets look for delalloc */
5608         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5609                                  end, len, EXTENT_DELALLOC, 1);
5610         found_end = range_start + found;
5611         if (found_end < range_start)
5612                 found_end = (u64)-1;
5613
5614         /*
5615          * we didn't find anything useful, return
5616          * the original results from get_extent()
5617          */
5618         if (range_start > end || found_end <= start) {
5619                 em = hole_em;
5620                 hole_em = NULL;
5621                 goto out;
5622         }
5623
5624         /* adjust the range_start to make sure it doesn't
5625          * go backwards from the start they passed in
5626          */
5627         range_start = max(start,range_start);
5628         found = found_end - range_start;
5629
5630         if (found > 0) {
5631                 u64 hole_start = start;
5632                 u64 hole_len = len;
5633
5634                 em = alloc_extent_map();
5635                 if (!em) {
5636                         err = -ENOMEM;
5637                         goto out;
5638                 }
5639                 /*
5640                  * when btrfs_get_extent can't find anything it
5641                  * returns one huge hole
5642                  *
5643                  * make sure what it found really fits our range, and
5644                  * adjust to make sure it is based on the start from
5645                  * the caller
5646                  */
5647                 if (hole_em) {
5648                         u64 calc_end = extent_map_end(hole_em);
5649
5650                         if (calc_end <= start || (hole_em->start > end)) {
5651                                 free_extent_map(hole_em);
5652                                 hole_em = NULL;
5653                         } else {
5654                                 hole_start = max(hole_em->start, start);
5655                                 hole_len = calc_end - hole_start;
5656                         }
5657                 }
5658                 em->bdev = NULL;
5659                 if (hole_em && range_start > hole_start) {
5660                         /* our hole starts before our delalloc, so we
5661                          * have to return just the parts of the hole
5662                          * that go until  the delalloc starts
5663                          */
5664                         em->len = min(hole_len,
5665                                       range_start - hole_start);
5666                         em->start = hole_start;
5667                         em->orig_start = hole_start;
5668                         /*
5669                          * don't adjust block start at all,
5670                          * it is fixed at EXTENT_MAP_HOLE
5671                          */
5672                         em->block_start = hole_em->block_start;
5673                         em->block_len = hole_len;
5674                 } else {
5675                         em->start = range_start;
5676                         em->len = found;
5677                         em->orig_start = range_start;
5678                         em->block_start = EXTENT_MAP_DELALLOC;
5679                         em->block_len = found;
5680                 }
5681         } else if (hole_em) {
5682                 return hole_em;
5683         }
5684 out:
5685
5686         free_extent_map(hole_em);
5687         if (err) {
5688                 free_extent_map(em);
5689                 return ERR_PTR(err);
5690         }
5691         return em;
5692 }
5693
5694 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5695                                                   u64 start, u64 len)
5696 {
5697         struct btrfs_root *root = BTRFS_I(inode)->root;
5698         struct btrfs_trans_handle *trans;
5699         struct extent_map *em;
5700         struct btrfs_key ins;
5701         u64 alloc_hint;
5702         int ret;
5703
5704         trans = btrfs_join_transaction(root);
5705         if (IS_ERR(trans))
5706                 return ERR_CAST(trans);
5707
5708         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5709
5710         alloc_hint = get_extent_allocation_hint(inode, start, len);
5711         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5712                                    alloc_hint, &ins, 1);
5713         if (ret) {
5714                 em = ERR_PTR(ret);
5715                 goto out;
5716         }
5717
5718         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5719                               ins.offset, ins.offset, 0);
5720         if (IS_ERR(em))
5721                 goto out;
5722
5723         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5724                                            ins.offset, ins.offset, 0);
5725         if (ret) {
5726                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5727                 em = ERR_PTR(ret);
5728         }
5729 out:
5730         btrfs_end_transaction(trans, root);
5731         return em;
5732 }
5733
5734 /*
5735  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5736  * block must be cow'd
5737  */
5738 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5739                                       struct inode *inode, u64 offset, u64 len)
5740 {
5741         struct btrfs_path *path;
5742         int ret;
5743         struct extent_buffer *leaf;
5744         struct btrfs_root *root = BTRFS_I(inode)->root;
5745         struct btrfs_file_extent_item *fi;
5746         struct btrfs_key key;
5747         u64 disk_bytenr;
5748         u64 backref_offset;
5749         u64 extent_end;
5750         u64 num_bytes;
5751         int slot;
5752         int found_type;
5753
5754         path = btrfs_alloc_path();
5755         if (!path)
5756                 return -ENOMEM;
5757
5758         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5759                                        offset, 0);
5760         if (ret < 0)
5761                 goto out;
5762
5763         slot = path->slots[0];
5764         if (ret == 1) {
5765                 if (slot == 0) {
5766                         /* can't find the item, must cow */
5767                         ret = 0;
5768                         goto out;
5769                 }
5770                 slot--;
5771         }
5772         ret = 0;
5773         leaf = path->nodes[0];
5774         btrfs_item_key_to_cpu(leaf, &key, slot);
5775         if (key.objectid != btrfs_ino(inode) ||
5776             key.type != BTRFS_EXTENT_DATA_KEY) {
5777                 /* not our file or wrong item type, must cow */
5778                 goto out;
5779         }
5780
5781         if (key.offset > offset) {
5782                 /* Wrong offset, must cow */
5783                 goto out;
5784         }
5785
5786         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5787         found_type = btrfs_file_extent_type(leaf, fi);
5788         if (found_type != BTRFS_FILE_EXTENT_REG &&
5789             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5790                 /* not a regular extent, must cow */
5791                 goto out;
5792         }
5793         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5794         backref_offset = btrfs_file_extent_offset(leaf, fi);
5795
5796         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5797         if (extent_end < offset + len) {
5798                 /* extent doesn't include our full range, must cow */
5799                 goto out;
5800         }
5801
5802         if (btrfs_extent_readonly(root, disk_bytenr))
5803                 goto out;
5804
5805         /*
5806          * look for other files referencing this extent, if we
5807          * find any we must cow
5808          */
5809         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5810                                   key.offset - backref_offset, disk_bytenr))
5811                 goto out;
5812
5813         /*
5814          * adjust disk_bytenr and num_bytes to cover just the bytes
5815          * in this extent we are about to write.  If there
5816          * are any csums in that range we have to cow in order
5817          * to keep the csums correct
5818          */
5819         disk_bytenr += backref_offset;
5820         disk_bytenr += offset - key.offset;
5821         num_bytes = min(offset + len, extent_end) - offset;
5822         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5823                                 goto out;
5824         /*
5825          * all of the above have passed, it is safe to overwrite this extent
5826          * without cow
5827          */
5828         ret = 1;
5829 out:
5830         btrfs_free_path(path);
5831         return ret;
5832 }
5833
5834 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5835                               struct extent_state **cached_state, int writing)
5836 {
5837         struct btrfs_ordered_extent *ordered;
5838         int ret = 0;
5839
5840         while (1) {
5841                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5842                                  0, cached_state);
5843                 /*
5844                  * We're concerned with the entire range that we're going to be
5845                  * doing DIO to, so we need to make sure theres no ordered
5846                  * extents in this range.
5847                  */
5848                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5849                                                      lockend - lockstart + 1);
5850
5851                 /*
5852                  * We need to make sure there are no buffered pages in this
5853                  * range either, we could have raced between the invalidate in
5854                  * generic_file_direct_write and locking the extent.  The
5855                  * invalidate needs to happen so that reads after a write do not
5856                  * get stale data.
5857                  */
5858                 if (!ordered && (!writing ||
5859                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5860                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5861                                     *cached_state)))
5862                         break;
5863
5864                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5865                                      cached_state, GFP_NOFS);
5866
5867                 if (ordered) {
5868                         btrfs_start_ordered_extent(inode, ordered, 1);
5869                         btrfs_put_ordered_extent(ordered);
5870                 } else {
5871                         /* Screw you mmap */
5872                         ret = filemap_write_and_wait_range(inode->i_mapping,
5873                                                            lockstart,
5874                                                            lockend);
5875                         if (ret)
5876                                 break;
5877
5878                         /*
5879                          * If we found a page that couldn't be invalidated just
5880                          * fall back to buffered.
5881                          */
5882                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5883                                         lockstart >> PAGE_CACHE_SHIFT,
5884                                         lockend >> PAGE_CACHE_SHIFT);
5885                         if (ret)
5886                                 break;
5887                 }
5888
5889                 cond_resched();
5890         }
5891
5892         return ret;
5893 }
5894
5895 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5896                                            u64 len, u64 orig_start,
5897                                            u64 block_start, u64 block_len,
5898                                            u64 orig_block_len, int type)
5899 {
5900         struct extent_map_tree *em_tree;
5901         struct extent_map *em;
5902         struct btrfs_root *root = BTRFS_I(inode)->root;
5903         int ret;
5904
5905         em_tree = &BTRFS_I(inode)->extent_tree;
5906         em = alloc_extent_map();
5907         if (!em)
5908                 return ERR_PTR(-ENOMEM);
5909
5910         em->start = start;
5911         em->orig_start = orig_start;
5912         em->len = len;
5913         em->block_len = block_len;
5914         em->block_start = block_start;
5915         em->bdev = root->fs_info->fs_devices->latest_bdev;
5916         em->orig_block_len = orig_block_len;
5917         em->generation = -1;
5918         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5919         if (type == BTRFS_ORDERED_PREALLOC)
5920                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
5921
5922         do {
5923                 btrfs_drop_extent_cache(inode, em->start,
5924                                 em->start + em->len - 1, 0);
5925                 write_lock(&em_tree->lock);
5926                 ret = add_extent_mapping(em_tree, em);
5927                 if (!ret)
5928                         list_move(&em->list,
5929                                   &em_tree->modified_extents);
5930                 write_unlock(&em_tree->lock);
5931         } while (ret == -EEXIST);
5932
5933         if (ret) {
5934                 free_extent_map(em);
5935                 return ERR_PTR(ret);
5936         }
5937
5938         return em;
5939 }
5940
5941
5942 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5943                                    struct buffer_head *bh_result, int create)
5944 {
5945         struct extent_map *em;
5946         struct btrfs_root *root = BTRFS_I(inode)->root;
5947         struct extent_state *cached_state = NULL;
5948         u64 start = iblock << inode->i_blkbits;
5949         u64 lockstart, lockend;
5950         u64 len = bh_result->b_size;
5951         struct btrfs_trans_handle *trans;
5952         int unlock_bits = EXTENT_LOCKED;
5953         int ret;
5954
5955         if (create) {
5956                 ret = btrfs_delalloc_reserve_space(inode, len);
5957                 if (ret)
5958                         return ret;
5959                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5960         } else {
5961                 len = min_t(u64, len, root->sectorsize);
5962         }
5963
5964         lockstart = start;
5965         lockend = start + len - 1;
5966
5967         /*
5968          * If this errors out it's because we couldn't invalidate pagecache for
5969          * this range and we need to fallback to buffered.
5970          */
5971         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5972                 return -ENOTBLK;
5973
5974         if (create) {
5975                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5976                                      lockend, EXTENT_DELALLOC, NULL,
5977                                      &cached_state, GFP_NOFS);
5978                 if (ret)
5979                         goto unlock_err;
5980         }
5981
5982         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5983         if (IS_ERR(em)) {
5984                 ret = PTR_ERR(em);
5985                 goto unlock_err;
5986         }
5987
5988         /*
5989          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5990          * io.  INLINE is special, and we could probably kludge it in here, but
5991          * it's still buffered so for safety lets just fall back to the generic
5992          * buffered path.
5993          *
5994          * For COMPRESSED we _have_ to read the entire extent in so we can
5995          * decompress it, so there will be buffering required no matter what we
5996          * do, so go ahead and fallback to buffered.
5997          *
5998          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5999          * to buffered IO.  Don't blame me, this is the price we pay for using
6000          * the generic code.
6001          */
6002         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6003             em->block_start == EXTENT_MAP_INLINE) {
6004                 free_extent_map(em);
6005                 ret = -ENOTBLK;
6006                 goto unlock_err;
6007         }
6008
6009         /* Just a good old fashioned hole, return */
6010         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6011                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6012                 free_extent_map(em);
6013                 ret = 0;
6014                 goto unlock_err;
6015         }
6016
6017         /*
6018          * We don't allocate a new extent in the following cases
6019          *
6020          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6021          * existing extent.
6022          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6023          * just use the extent.
6024          *
6025          */
6026         if (!create) {
6027                 len = min(len, em->len - (start - em->start));
6028                 lockstart = start + len;
6029                 goto unlock;
6030         }
6031
6032         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6033             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6034              em->block_start != EXTENT_MAP_HOLE)) {
6035                 int type;
6036                 int ret;
6037                 u64 block_start;
6038
6039                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6040                         type = BTRFS_ORDERED_PREALLOC;
6041                 else
6042                         type = BTRFS_ORDERED_NOCOW;
6043                 len = min(len, em->len - (start - em->start));
6044                 block_start = em->block_start + (start - em->start);
6045
6046                 /*
6047                  * we're not going to log anything, but we do need
6048                  * to make sure the current transaction stays open
6049                  * while we look for nocow cross refs
6050                  */
6051                 trans = btrfs_join_transaction(root);
6052                 if (IS_ERR(trans))
6053                         goto must_cow;
6054
6055                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6056                         u64 orig_start = em->orig_start;
6057                         u64 orig_block_len = em->orig_block_len;
6058
6059                         if (type == BTRFS_ORDERED_PREALLOC) {
6060                                 free_extent_map(em);
6061                                 em = create_pinned_em(inode, start, len,
6062                                                        orig_start,
6063                                                        block_start, len,
6064                                                        orig_block_len, type);
6065                                 if (IS_ERR(em)) {
6066                                         btrfs_end_transaction(trans, root);
6067                                         goto unlock_err;
6068                                 }
6069                         }
6070
6071                         ret = btrfs_add_ordered_extent_dio(inode, start,
6072                                            block_start, len, len, type);
6073                         btrfs_end_transaction(trans, root);
6074                         if (ret) {
6075                                 free_extent_map(em);
6076                                 goto unlock_err;
6077                         }
6078                         goto unlock;
6079                 }
6080                 btrfs_end_transaction(trans, root);
6081         }
6082 must_cow:
6083         /*
6084          * this will cow the extent, reset the len in case we changed
6085          * it above
6086          */
6087         len = bh_result->b_size;
6088         free_extent_map(em);
6089         em = btrfs_new_extent_direct(inode, start, len);
6090         if (IS_ERR(em)) {
6091                 ret = PTR_ERR(em);
6092                 goto unlock_err;
6093         }
6094         len = min(len, em->len - (start - em->start));
6095 unlock:
6096         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6097                 inode->i_blkbits;
6098         bh_result->b_size = len;
6099         bh_result->b_bdev = em->bdev;
6100         set_buffer_mapped(bh_result);
6101         if (create) {
6102                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6103                         set_buffer_new(bh_result);
6104
6105                 /*
6106                  * Need to update the i_size under the extent lock so buffered
6107                  * readers will get the updated i_size when we unlock.
6108                  */
6109                 if (start + len > i_size_read(inode))
6110                         i_size_write(inode, start + len);
6111         }
6112
6113         /*
6114          * In the case of write we need to clear and unlock the entire range,
6115          * in the case of read we need to unlock only the end area that we
6116          * aren't using if there is any left over space.
6117          */
6118         if (lockstart < lockend) {
6119                 if (create && len < lockend - lockstart) {
6120                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6121                                          lockstart + len - 1,
6122                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6123                                          &cached_state, GFP_NOFS);
6124                         /*
6125                          * Beside unlock, we also need to cleanup reserved space
6126                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6127                          */
6128                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6129                                          lockstart + len, lockend,
6130                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6131                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6132                 } else {
6133                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6134                                          lockend, unlock_bits, 1, 0,
6135                                          &cached_state, GFP_NOFS);
6136                 }
6137         } else {
6138                 free_extent_state(cached_state);
6139         }
6140
6141         free_extent_map(em);
6142
6143         return 0;
6144
6145 unlock_err:
6146         if (create)
6147                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6148
6149         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6150                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6151         return ret;
6152 }
6153
6154 struct btrfs_dio_private {
6155         struct inode *inode;
6156         u64 logical_offset;
6157         u64 disk_bytenr;
6158         u64 bytes;
6159         void *private;
6160
6161         /* number of bios pending for this dio */
6162         atomic_t pending_bios;
6163
6164         /* IO errors */
6165         int errors;
6166
6167         struct bio *orig_bio;
6168 };
6169
6170 static void btrfs_endio_direct_read(struct bio *bio, int err)
6171 {
6172         struct btrfs_dio_private *dip = bio->bi_private;
6173         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6174         struct bio_vec *bvec = bio->bi_io_vec;
6175         struct inode *inode = dip->inode;
6176         struct btrfs_root *root = BTRFS_I(inode)->root;
6177         u64 start;
6178
6179         start = dip->logical_offset;
6180         do {
6181                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6182                         struct page *page = bvec->bv_page;
6183                         char *kaddr;
6184                         u32 csum = ~(u32)0;
6185                         u64 private = ~(u32)0;
6186                         unsigned long flags;
6187
6188                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6189                                               start, &private))
6190                                 goto failed;
6191                         local_irq_save(flags);
6192                         kaddr = kmap_atomic(page);
6193                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6194                                                csum, bvec->bv_len);
6195                         btrfs_csum_final(csum, (char *)&csum);
6196                         kunmap_atomic(kaddr);
6197                         local_irq_restore(flags);
6198
6199                         flush_dcache_page(bvec->bv_page);
6200                         if (csum != private) {
6201 failed:
6202                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6203                                       " %llu csum %u private %u\n",
6204                                       (unsigned long long)btrfs_ino(inode),
6205                                       (unsigned long long)start,
6206                                       csum, (unsigned)private);
6207                                 err = -EIO;
6208                         }
6209                 }
6210
6211                 start += bvec->bv_len;
6212                 bvec++;
6213         } while (bvec <= bvec_end);
6214
6215         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6216                       dip->logical_offset + dip->bytes - 1);
6217         bio->bi_private = dip->private;
6218
6219         kfree(dip);
6220
6221         /* If we had a csum failure make sure to clear the uptodate flag */
6222         if (err)
6223                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6224         dio_end_io(bio, err);
6225 }
6226
6227 static void btrfs_endio_direct_write(struct bio *bio, int err)
6228 {
6229         struct btrfs_dio_private *dip = bio->bi_private;
6230         struct inode *inode = dip->inode;
6231         struct btrfs_root *root = BTRFS_I(inode)->root;
6232         struct btrfs_ordered_extent *ordered = NULL;
6233         u64 ordered_offset = dip->logical_offset;
6234         u64 ordered_bytes = dip->bytes;
6235         int ret;
6236
6237         if (err)
6238                 goto out_done;
6239 again:
6240         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6241                                                    &ordered_offset,
6242                                                    ordered_bytes, !err);
6243         if (!ret)
6244                 goto out_test;
6245
6246         ordered->work.func = finish_ordered_fn;
6247         ordered->work.flags = 0;
6248         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6249                            &ordered->work);
6250 out_test:
6251         /*
6252          * our bio might span multiple ordered extents.  If we haven't
6253          * completed the accounting for the whole dio, go back and try again
6254          */
6255         if (ordered_offset < dip->logical_offset + dip->bytes) {
6256                 ordered_bytes = dip->logical_offset + dip->bytes -
6257                         ordered_offset;
6258                 ordered = NULL;
6259                 goto again;
6260         }
6261 out_done:
6262         bio->bi_private = dip->private;
6263
6264         kfree(dip);
6265
6266         /* If we had an error make sure to clear the uptodate flag */
6267         if (err)
6268                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6269         dio_end_io(bio, err);
6270 }
6271
6272 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6273                                     struct bio *bio, int mirror_num,
6274                                     unsigned long bio_flags, u64 offset)
6275 {
6276         int ret;
6277         struct btrfs_root *root = BTRFS_I(inode)->root;
6278         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6279         BUG_ON(ret); /* -ENOMEM */
6280         return 0;
6281 }
6282
6283 static void btrfs_end_dio_bio(struct bio *bio, int err)
6284 {
6285         struct btrfs_dio_private *dip = bio->bi_private;
6286
6287         if (err) {
6288                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6289                       "sector %#Lx len %u err no %d\n",
6290                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6291                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6292                 dip->errors = 1;
6293
6294                 /*
6295                  * before atomic variable goto zero, we must make sure
6296                  * dip->errors is perceived to be set.
6297                  */
6298                 smp_mb__before_atomic_dec();
6299         }
6300
6301         /* if there are more bios still pending for this dio, just exit */
6302         if (!atomic_dec_and_test(&dip->pending_bios))
6303                 goto out;
6304
6305         if (dip->errors)
6306                 bio_io_error(dip->orig_bio);
6307         else {
6308                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6309                 bio_endio(dip->orig_bio, 0);
6310         }
6311 out:
6312         bio_put(bio);
6313 }
6314
6315 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6316                                        u64 first_sector, gfp_t gfp_flags)
6317 {
6318         int nr_vecs = bio_get_nr_vecs(bdev);
6319         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6320 }
6321
6322 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6323                                          int rw, u64 file_offset, int skip_sum,
6324                                          int async_submit)
6325 {
6326         int write = rw & REQ_WRITE;
6327         struct btrfs_root *root = BTRFS_I(inode)->root;
6328         int ret;
6329
6330         if (async_submit)
6331                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6332
6333         bio_get(bio);
6334
6335         if (!write) {
6336                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6337                 if (ret)
6338                         goto err;
6339         }
6340
6341         if (skip_sum)
6342                 goto map;
6343
6344         if (write && async_submit) {
6345                 ret = btrfs_wq_submit_bio(root->fs_info,
6346                                    inode, rw, bio, 0, 0,
6347                                    file_offset,
6348                                    __btrfs_submit_bio_start_direct_io,
6349                                    __btrfs_submit_bio_done);
6350                 goto err;
6351         } else if (write) {
6352                 /*
6353                  * If we aren't doing async submit, calculate the csum of the
6354                  * bio now.
6355                  */
6356                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6357                 if (ret)
6358                         goto err;
6359         } else if (!skip_sum) {
6360                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6361                 if (ret)
6362                         goto err;
6363         }
6364
6365 map:
6366         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6367 err:
6368         bio_put(bio);
6369         return ret;
6370 }
6371
6372 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6373                                     int skip_sum)
6374 {
6375         struct inode *inode = dip->inode;
6376         struct btrfs_root *root = BTRFS_I(inode)->root;
6377         struct bio *bio;
6378         struct bio *orig_bio = dip->orig_bio;
6379         struct bio_vec *bvec = orig_bio->bi_io_vec;
6380         u64 start_sector = orig_bio->bi_sector;
6381         u64 file_offset = dip->logical_offset;
6382         u64 submit_len = 0;
6383         u64 map_length;
6384         int nr_pages = 0;
6385         int ret = 0;
6386         int async_submit = 0;
6387
6388         map_length = orig_bio->bi_size;
6389         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6390                               &map_length, NULL, 0);
6391         if (ret) {
6392                 bio_put(orig_bio);
6393                 return -EIO;
6394         }
6395
6396         if (map_length >= orig_bio->bi_size) {
6397                 bio = orig_bio;
6398                 goto submit;
6399         }
6400
6401         async_submit = 1;
6402         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6403         if (!bio)
6404                 return -ENOMEM;
6405         bio->bi_private = dip;
6406         bio->bi_end_io = btrfs_end_dio_bio;
6407         atomic_inc(&dip->pending_bios);
6408
6409         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6410                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6411                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6412                                  bvec->bv_offset) < bvec->bv_len)) {
6413                         /*
6414                          * inc the count before we submit the bio so
6415                          * we know the end IO handler won't happen before
6416                          * we inc the count. Otherwise, the dip might get freed
6417                          * before we're done setting it up
6418                          */
6419                         atomic_inc(&dip->pending_bios);
6420                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6421                                                      file_offset, skip_sum,
6422                                                      async_submit);
6423                         if (ret) {
6424                                 bio_put(bio);
6425                                 atomic_dec(&dip->pending_bios);
6426                                 goto out_err;
6427                         }
6428
6429                         start_sector += submit_len >> 9;
6430                         file_offset += submit_len;
6431
6432                         submit_len = 0;
6433                         nr_pages = 0;
6434
6435                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6436                                                   start_sector, GFP_NOFS);
6437                         if (!bio)
6438                                 goto out_err;
6439                         bio->bi_private = dip;
6440                         bio->bi_end_io = btrfs_end_dio_bio;
6441
6442                         map_length = orig_bio->bi_size;
6443                         ret = btrfs_map_block(root->fs_info, READ,
6444                                               start_sector << 9,
6445                                               &map_length, NULL, 0);
6446                         if (ret) {
6447                                 bio_put(bio);
6448                                 goto out_err;
6449                         }
6450                 } else {
6451                         submit_len += bvec->bv_len;
6452                         nr_pages ++;
6453                         bvec++;
6454                 }
6455         }
6456
6457 submit:
6458         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6459                                      async_submit);
6460         if (!ret)
6461                 return 0;
6462
6463         bio_put(bio);
6464 out_err:
6465         dip->errors = 1;
6466         /*
6467          * before atomic variable goto zero, we must
6468          * make sure dip->errors is perceived to be set.
6469          */
6470         smp_mb__before_atomic_dec();
6471         if (atomic_dec_and_test(&dip->pending_bios))
6472                 bio_io_error(dip->orig_bio);
6473
6474         /* bio_end_io() will handle error, so we needn't return it */
6475         return 0;
6476 }
6477
6478 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6479                                 loff_t file_offset)
6480 {
6481         struct btrfs_root *root = BTRFS_I(inode)->root;
6482         struct btrfs_dio_private *dip;
6483         struct bio_vec *bvec = bio->bi_io_vec;
6484         int skip_sum;
6485         int write = rw & REQ_WRITE;
6486         int ret = 0;
6487
6488         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6489
6490         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6491         if (!dip) {
6492                 ret = -ENOMEM;
6493                 goto free_ordered;
6494         }
6495
6496         dip->private = bio->bi_private;
6497         dip->inode = inode;
6498         dip->logical_offset = file_offset;
6499
6500         dip->bytes = 0;
6501         do {
6502                 dip->bytes += bvec->bv_len;
6503                 bvec++;
6504         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6505
6506         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6507         bio->bi_private = dip;
6508         dip->errors = 0;
6509         dip->orig_bio = bio;
6510         atomic_set(&dip->pending_bios, 0);
6511
6512         if (write)
6513                 bio->bi_end_io = btrfs_endio_direct_write;
6514         else
6515                 bio->bi_end_io = btrfs_endio_direct_read;
6516
6517         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6518         if (!ret)
6519                 return;
6520 free_ordered:
6521         /*
6522          * If this is a write, we need to clean up the reserved space and kill
6523          * the ordered extent.
6524          */
6525         if (write) {
6526                 struct btrfs_ordered_extent *ordered;
6527                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6528                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6529                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6530                         btrfs_free_reserved_extent(root, ordered->start,
6531                                                    ordered->disk_len);
6532                 btrfs_put_ordered_extent(ordered);
6533                 btrfs_put_ordered_extent(ordered);
6534         }
6535         bio_endio(bio, ret);
6536 }
6537
6538 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6539                         const struct iovec *iov, loff_t offset,
6540                         unsigned long nr_segs)
6541 {
6542         int seg;
6543         int i;
6544         size_t size;
6545         unsigned long addr;
6546         unsigned blocksize_mask = root->sectorsize - 1;
6547         ssize_t retval = -EINVAL;
6548         loff_t end = offset;
6549
6550         if (offset & blocksize_mask)
6551                 goto out;
6552
6553         /* Check the memory alignment.  Blocks cannot straddle pages */
6554         for (seg = 0; seg < nr_segs; seg++) {
6555                 addr = (unsigned long)iov[seg].iov_base;
6556                 size = iov[seg].iov_len;
6557                 end += size;
6558                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6559                         goto out;
6560
6561                 /* If this is a write we don't need to check anymore */
6562                 if (rw & WRITE)
6563                         continue;
6564
6565                 /*
6566                  * Check to make sure we don't have duplicate iov_base's in this
6567                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6568                  * when reading back.
6569                  */
6570                 for (i = seg + 1; i < nr_segs; i++) {
6571                         if (iov[seg].iov_base == iov[i].iov_base)
6572                                 goto out;
6573                 }
6574         }
6575         retval = 0;
6576 out:
6577         return retval;
6578 }
6579
6580 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6581                         const struct iovec *iov, loff_t offset,
6582                         unsigned long nr_segs)
6583 {
6584         struct file *file = iocb->ki_filp;
6585         struct inode *inode = file->f_mapping->host;
6586
6587         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6588                             offset, nr_segs))
6589                 return 0;
6590
6591         return __blockdev_direct_IO(rw, iocb, inode,
6592                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6593                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6594                    btrfs_submit_direct, 0);
6595 }
6596
6597 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6598
6599 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6600                 __u64 start, __u64 len)
6601 {
6602         int     ret;
6603
6604         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6605         if (ret)
6606                 return ret;
6607
6608         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6609 }
6610
6611 int btrfs_readpage(struct file *file, struct page *page)
6612 {
6613         struct extent_io_tree *tree;
6614         tree = &BTRFS_I(page->mapping->host)->io_tree;
6615         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6616 }
6617
6618 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6619 {
6620         struct extent_io_tree *tree;
6621
6622
6623         if (current->flags & PF_MEMALLOC) {
6624                 redirty_page_for_writepage(wbc, page);
6625                 unlock_page(page);
6626                 return 0;
6627         }
6628         tree = &BTRFS_I(page->mapping->host)->io_tree;
6629         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6630 }
6631
6632 int btrfs_writepages(struct address_space *mapping,
6633                      struct writeback_control *wbc)
6634 {
6635         struct extent_io_tree *tree;
6636
6637         tree = &BTRFS_I(mapping->host)->io_tree;
6638         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6639 }
6640
6641 static int
6642 btrfs_readpages(struct file *file, struct address_space *mapping,
6643                 struct list_head *pages, unsigned nr_pages)
6644 {
6645         struct extent_io_tree *tree;
6646         tree = &BTRFS_I(mapping->host)->io_tree;
6647         return extent_readpages(tree, mapping, pages, nr_pages,
6648                                 btrfs_get_extent);
6649 }
6650 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6651 {
6652         struct extent_io_tree *tree;
6653         struct extent_map_tree *map;
6654         int ret;
6655
6656         tree = &BTRFS_I(page->mapping->host)->io_tree;
6657         map = &BTRFS_I(page->mapping->host)->extent_tree;
6658         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6659         if (ret == 1) {
6660                 ClearPagePrivate(page);
6661                 set_page_private(page, 0);
6662                 page_cache_release(page);
6663         }
6664         return ret;
6665 }
6666
6667 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6668 {
6669         if (PageWriteback(page) || PageDirty(page))
6670                 return 0;
6671         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6672 }
6673
6674 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6675 {
6676         struct inode *inode = page->mapping->host;
6677         struct extent_io_tree *tree;
6678         struct btrfs_ordered_extent *ordered;
6679         struct extent_state *cached_state = NULL;
6680         u64 page_start = page_offset(page);
6681         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6682
6683         /*
6684          * we have the page locked, so new writeback can't start,
6685          * and the dirty bit won't be cleared while we are here.
6686          *
6687          * Wait for IO on this page so that we can safely clear
6688          * the PagePrivate2 bit and do ordered accounting
6689          */
6690         wait_on_page_writeback(page);
6691
6692         tree = &BTRFS_I(inode)->io_tree;
6693         if (offset) {
6694                 btrfs_releasepage(page, GFP_NOFS);
6695                 return;
6696         }
6697         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6698         ordered = btrfs_lookup_ordered_extent(inode,
6699                                            page_offset(page));
6700         if (ordered) {
6701                 /*
6702                  * IO on this page will never be started, so we need
6703                  * to account for any ordered extents now
6704                  */
6705                 clear_extent_bit(tree, page_start, page_end,
6706                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6707                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6708                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6709                 /*
6710                  * whoever cleared the private bit is responsible
6711                  * for the finish_ordered_io
6712                  */
6713                 if (TestClearPagePrivate2(page) &&
6714                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6715                                                    PAGE_CACHE_SIZE, 1)) {
6716                         btrfs_finish_ordered_io(ordered);
6717                 }
6718                 btrfs_put_ordered_extent(ordered);
6719                 cached_state = NULL;
6720                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6721         }
6722         clear_extent_bit(tree, page_start, page_end,
6723                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6724                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6725                  &cached_state, GFP_NOFS);
6726         __btrfs_releasepage(page, GFP_NOFS);
6727
6728         ClearPageChecked(page);
6729         if (PagePrivate(page)) {
6730                 ClearPagePrivate(page);
6731                 set_page_private(page, 0);
6732                 page_cache_release(page);
6733         }
6734 }
6735
6736 /*
6737  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6738  * called from a page fault handler when a page is first dirtied. Hence we must
6739  * be careful to check for EOF conditions here. We set the page up correctly
6740  * for a written page which means we get ENOSPC checking when writing into
6741  * holes and correct delalloc and unwritten extent mapping on filesystems that
6742  * support these features.
6743  *
6744  * We are not allowed to take the i_mutex here so we have to play games to
6745  * protect against truncate races as the page could now be beyond EOF.  Because
6746  * vmtruncate() writes the inode size before removing pages, once we have the
6747  * page lock we can determine safely if the page is beyond EOF. If it is not
6748  * beyond EOF, then the page is guaranteed safe against truncation until we
6749  * unlock the page.
6750  */
6751 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6752 {
6753         struct page *page = vmf->page;
6754         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6755         struct btrfs_root *root = BTRFS_I(inode)->root;
6756         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6757         struct btrfs_ordered_extent *ordered;
6758         struct extent_state *cached_state = NULL;
6759         char *kaddr;
6760         unsigned long zero_start;
6761         loff_t size;
6762         int ret;
6763         int reserved = 0;
6764         u64 page_start;
6765         u64 page_end;
6766
6767         sb_start_pagefault(inode->i_sb);
6768         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6769         if (!ret) {
6770                 ret = file_update_time(vma->vm_file);
6771                 reserved = 1;
6772         }
6773         if (ret) {
6774                 if (ret == -ENOMEM)
6775                         ret = VM_FAULT_OOM;
6776                 else /* -ENOSPC, -EIO, etc */
6777                         ret = VM_FAULT_SIGBUS;
6778                 if (reserved)
6779                         goto out;
6780                 goto out_noreserve;
6781         }
6782
6783         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6784 again:
6785         lock_page(page);
6786         size = i_size_read(inode);
6787         page_start = page_offset(page);
6788         page_end = page_start + PAGE_CACHE_SIZE - 1;
6789
6790         if ((page->mapping != inode->i_mapping) ||
6791             (page_start >= size)) {
6792                 /* page got truncated out from underneath us */
6793                 goto out_unlock;
6794         }
6795         wait_on_page_writeback(page);
6796
6797         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6798         set_page_extent_mapped(page);
6799
6800         /*
6801          * we can't set the delalloc bits if there are pending ordered
6802          * extents.  Drop our locks and wait for them to finish
6803          */
6804         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6805         if (ordered) {
6806                 unlock_extent_cached(io_tree, page_start, page_end,
6807                                      &cached_state, GFP_NOFS);
6808                 unlock_page(page);
6809                 btrfs_start_ordered_extent(inode, ordered, 1);
6810                 btrfs_put_ordered_extent(ordered);
6811                 goto again;
6812         }
6813
6814         /*
6815          * XXX - page_mkwrite gets called every time the page is dirtied, even
6816          * if it was already dirty, so for space accounting reasons we need to
6817          * clear any delalloc bits for the range we are fixing to save.  There
6818          * is probably a better way to do this, but for now keep consistent with
6819          * prepare_pages in the normal write path.
6820          */
6821         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6822                           EXTENT_DIRTY | EXTENT_DELALLOC |
6823                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6824                           0, 0, &cached_state, GFP_NOFS);
6825
6826         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6827                                         &cached_state);
6828         if (ret) {
6829                 unlock_extent_cached(io_tree, page_start, page_end,
6830                                      &cached_state, GFP_NOFS);
6831                 ret = VM_FAULT_SIGBUS;
6832                 goto out_unlock;
6833         }
6834         ret = 0;
6835
6836         /* page is wholly or partially inside EOF */
6837         if (page_start + PAGE_CACHE_SIZE > size)
6838                 zero_start = size & ~PAGE_CACHE_MASK;
6839         else
6840                 zero_start = PAGE_CACHE_SIZE;
6841
6842         if (zero_start != PAGE_CACHE_SIZE) {
6843                 kaddr = kmap(page);
6844                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6845                 flush_dcache_page(page);
6846                 kunmap(page);
6847         }
6848         ClearPageChecked(page);
6849         set_page_dirty(page);
6850         SetPageUptodate(page);
6851
6852         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6853         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6854         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6855
6856         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6857
6858 out_unlock:
6859         if (!ret) {
6860                 sb_end_pagefault(inode->i_sb);
6861                 return VM_FAULT_LOCKED;
6862         }
6863         unlock_page(page);
6864 out:
6865         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6866 out_noreserve:
6867         sb_end_pagefault(inode->i_sb);
6868         return ret;
6869 }
6870
6871 static int btrfs_truncate(struct inode *inode)
6872 {
6873         struct btrfs_root *root = BTRFS_I(inode)->root;
6874         struct btrfs_block_rsv *rsv;
6875         int ret;
6876         int err = 0;
6877         struct btrfs_trans_handle *trans;
6878         u64 mask = root->sectorsize - 1;
6879         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6880
6881         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6882         if (ret)
6883                 return ret;
6884
6885         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6886         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6887
6888         /*
6889          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6890          * 3 things going on here
6891          *
6892          * 1) We need to reserve space for our orphan item and the space to
6893          * delete our orphan item.  Lord knows we don't want to have a dangling
6894          * orphan item because we didn't reserve space to remove it.
6895          *
6896          * 2) We need to reserve space to update our inode.
6897          *
6898          * 3) We need to have something to cache all the space that is going to
6899          * be free'd up by the truncate operation, but also have some slack
6900          * space reserved in case it uses space during the truncate (thank you
6901          * very much snapshotting).
6902          *
6903          * And we need these to all be seperate.  The fact is we can use alot of
6904          * space doing the truncate, and we have no earthly idea how much space
6905          * we will use, so we need the truncate reservation to be seperate so it
6906          * doesn't end up using space reserved for updating the inode or
6907          * removing the orphan item.  We also need to be able to stop the
6908          * transaction and start a new one, which means we need to be able to
6909          * update the inode several times, and we have no idea of knowing how
6910          * many times that will be, so we can't just reserve 1 item for the
6911          * entirety of the opration, so that has to be done seperately as well.
6912          * Then there is the orphan item, which does indeed need to be held on
6913          * to for the whole operation, and we need nobody to touch this reserved
6914          * space except the orphan code.
6915          *
6916          * So that leaves us with
6917          *
6918          * 1) root->orphan_block_rsv - for the orphan deletion.
6919          * 2) rsv - for the truncate reservation, which we will steal from the
6920          * transaction reservation.
6921          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6922          * updating the inode.
6923          */
6924         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6925         if (!rsv)
6926                 return -ENOMEM;
6927         rsv->size = min_size;
6928         rsv->failfast = 1;
6929
6930         /*
6931          * 1 for the truncate slack space
6932          * 1 for the orphan item we're going to add
6933          * 1 for the orphan item deletion
6934          * 1 for updating the inode.
6935          */
6936         trans = btrfs_start_transaction(root, 4);
6937         if (IS_ERR(trans)) {
6938                 err = PTR_ERR(trans);
6939                 goto out;
6940         }
6941
6942         /* Migrate the slack space for the truncate to our reserve */
6943         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6944                                       min_size);
6945         BUG_ON(ret);
6946
6947         ret = btrfs_orphan_add(trans, inode);
6948         if (ret) {
6949                 btrfs_end_transaction(trans, root);
6950                 goto out;
6951         }
6952
6953         /*
6954          * setattr is responsible for setting the ordered_data_close flag,
6955          * but that is only tested during the last file release.  That
6956          * could happen well after the next commit, leaving a great big
6957          * window where new writes may get lost if someone chooses to write
6958          * to this file after truncating to zero
6959          *
6960          * The inode doesn't have any dirty data here, and so if we commit
6961          * this is a noop.  If someone immediately starts writing to the inode
6962          * it is very likely we'll catch some of their writes in this
6963          * transaction, and the commit will find this file on the ordered
6964          * data list with good things to send down.
6965          *
6966          * This is a best effort solution, there is still a window where
6967          * using truncate to replace the contents of the file will
6968          * end up with a zero length file after a crash.
6969          */
6970         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6971                                            &BTRFS_I(inode)->runtime_flags))
6972                 btrfs_add_ordered_operation(trans, root, inode);
6973
6974         /*
6975          * So if we truncate and then write and fsync we normally would just
6976          * write the extents that changed, which is a problem if we need to
6977          * first truncate that entire inode.  So set this flag so we write out
6978          * all of the extents in the inode to the sync log so we're completely
6979          * safe.
6980          */
6981         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6982         trans->block_rsv = rsv;
6983
6984         while (1) {
6985                 ret = btrfs_truncate_inode_items(trans, root, inode,
6986                                                  inode->i_size,
6987                                                  BTRFS_EXTENT_DATA_KEY);
6988                 if (ret != -ENOSPC) {
6989                         err = ret;
6990                         break;
6991                 }
6992
6993                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6994                 ret = btrfs_update_inode(trans, root, inode);
6995                 if (ret) {
6996                         err = ret;
6997                         break;
6998                 }
6999
7000                 btrfs_end_transaction(trans, root);
7001                 btrfs_btree_balance_dirty(root);
7002
7003                 trans = btrfs_start_transaction(root, 2);
7004                 if (IS_ERR(trans)) {
7005                         ret = err = PTR_ERR(trans);
7006                         trans = NULL;
7007                         break;
7008                 }
7009
7010                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7011                                               rsv, min_size);
7012                 BUG_ON(ret);    /* shouldn't happen */
7013                 trans->block_rsv = rsv;
7014         }
7015
7016         if (ret == 0 && inode->i_nlink > 0) {
7017                 trans->block_rsv = root->orphan_block_rsv;
7018                 ret = btrfs_orphan_del(trans, inode);
7019                 if (ret)
7020                         err = ret;
7021         } else if (ret && inode->i_nlink > 0) {
7022                 /*
7023                  * Failed to do the truncate, remove us from the in memory
7024                  * orphan list.
7025                  */
7026                 ret = btrfs_orphan_del(NULL, inode);
7027         }
7028
7029         if (trans) {
7030                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7031                 ret = btrfs_update_inode(trans, root, inode);
7032                 if (ret && !err)
7033                         err = ret;
7034
7035                 ret = btrfs_end_transaction(trans, root);
7036                 btrfs_btree_balance_dirty(root);
7037         }
7038
7039 out:
7040         btrfs_free_block_rsv(root, rsv);
7041
7042         if (ret && !err)
7043                 err = ret;
7044
7045         return err;
7046 }
7047
7048 /*
7049  * create a new subvolume directory/inode (helper for the ioctl).
7050  */
7051 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7052                              struct btrfs_root *new_root, u64 new_dirid)
7053 {
7054         struct inode *inode;
7055         int err;
7056         u64 index = 0;
7057
7058         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7059                                 new_dirid, new_dirid,
7060                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7061                                 &index);
7062         if (IS_ERR(inode))
7063                 return PTR_ERR(inode);
7064         inode->i_op = &btrfs_dir_inode_operations;
7065         inode->i_fop = &btrfs_dir_file_operations;
7066
7067         set_nlink(inode, 1);
7068         btrfs_i_size_write(inode, 0);
7069
7070         err = btrfs_update_inode(trans, new_root, inode);
7071
7072         iput(inode);
7073         return err;
7074 }
7075
7076 struct inode *btrfs_alloc_inode(struct super_block *sb)
7077 {
7078         struct btrfs_inode *ei;
7079         struct inode *inode;
7080
7081         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7082         if (!ei)
7083                 return NULL;
7084
7085         ei->root = NULL;
7086         ei->generation = 0;
7087         ei->last_trans = 0;
7088         ei->last_sub_trans = 0;
7089         ei->logged_trans = 0;
7090         ei->delalloc_bytes = 0;
7091         ei->disk_i_size = 0;
7092         ei->flags = 0;
7093         ei->csum_bytes = 0;
7094         ei->index_cnt = (u64)-1;
7095         ei->last_unlink_trans = 0;
7096         ei->last_log_commit = 0;
7097
7098         spin_lock_init(&ei->lock);
7099         ei->outstanding_extents = 0;
7100         ei->reserved_extents = 0;
7101
7102         ei->runtime_flags = 0;
7103         ei->force_compress = BTRFS_COMPRESS_NONE;
7104
7105         ei->delayed_node = NULL;
7106
7107         inode = &ei->vfs_inode;
7108         extent_map_tree_init(&ei->extent_tree);
7109         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7110         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7111         ei->io_tree.track_uptodate = 1;
7112         ei->io_failure_tree.track_uptodate = 1;
7113         atomic_set(&ei->sync_writers, 0);
7114         mutex_init(&ei->log_mutex);
7115         mutex_init(&ei->delalloc_mutex);
7116         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7117         INIT_LIST_HEAD(&ei->delalloc_inodes);
7118         INIT_LIST_HEAD(&ei->ordered_operations);
7119         RB_CLEAR_NODE(&ei->rb_node);
7120
7121         return inode;
7122 }
7123
7124 static void btrfs_i_callback(struct rcu_head *head)
7125 {
7126         struct inode *inode = container_of(head, struct inode, i_rcu);
7127         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7128 }
7129
7130 void btrfs_destroy_inode(struct inode *inode)
7131 {
7132         struct btrfs_ordered_extent *ordered;
7133         struct btrfs_root *root = BTRFS_I(inode)->root;
7134
7135         WARN_ON(!hlist_empty(&inode->i_dentry));
7136         WARN_ON(inode->i_data.nrpages);
7137         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7138         WARN_ON(BTRFS_I(inode)->reserved_extents);
7139         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7140         WARN_ON(BTRFS_I(inode)->csum_bytes);
7141
7142         /*
7143          * This can happen where we create an inode, but somebody else also
7144          * created the same inode and we need to destroy the one we already
7145          * created.
7146          */
7147         if (!root)
7148                 goto free;
7149
7150         /*
7151          * Make sure we're properly removed from the ordered operation
7152          * lists.
7153          */
7154         smp_mb();
7155         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7156                 spin_lock(&root->fs_info->ordered_extent_lock);
7157                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7158                 spin_unlock(&root->fs_info->ordered_extent_lock);
7159         }
7160
7161         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7162                      &BTRFS_I(inode)->runtime_flags)) {
7163                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7164                        (unsigned long long)btrfs_ino(inode));
7165                 atomic_dec(&root->orphan_inodes);
7166         }
7167
7168         while (1) {
7169                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7170                 if (!ordered)
7171                         break;
7172                 else {
7173                         printk(KERN_ERR "btrfs found ordered "
7174                                "extent %llu %llu on inode cleanup\n",
7175                                (unsigned long long)ordered->file_offset,
7176                                (unsigned long long)ordered->len);
7177                         btrfs_remove_ordered_extent(inode, ordered);
7178                         btrfs_put_ordered_extent(ordered);
7179                         btrfs_put_ordered_extent(ordered);
7180                 }
7181         }
7182         inode_tree_del(inode);
7183         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7184 free:
7185         btrfs_remove_delayed_node(inode);
7186         call_rcu(&inode->i_rcu, btrfs_i_callback);
7187 }
7188
7189 int btrfs_drop_inode(struct inode *inode)
7190 {
7191         struct btrfs_root *root = BTRFS_I(inode)->root;
7192
7193         if (btrfs_root_refs(&root->root_item) == 0 &&
7194             !btrfs_is_free_space_inode(inode))
7195                 return 1;
7196         else
7197                 return generic_drop_inode(inode);
7198 }
7199
7200 static void init_once(void *foo)
7201 {
7202         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7203
7204         inode_init_once(&ei->vfs_inode);
7205 }
7206
7207 void btrfs_destroy_cachep(void)
7208 {
7209         /*
7210          * Make sure all delayed rcu free inodes are flushed before we
7211          * destroy cache.
7212          */
7213         rcu_barrier();
7214         if (btrfs_inode_cachep)
7215                 kmem_cache_destroy(btrfs_inode_cachep);
7216         if (btrfs_trans_handle_cachep)
7217                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7218         if (btrfs_transaction_cachep)
7219                 kmem_cache_destroy(btrfs_transaction_cachep);
7220         if (btrfs_path_cachep)
7221                 kmem_cache_destroy(btrfs_path_cachep);
7222         if (btrfs_free_space_cachep)
7223                 kmem_cache_destroy(btrfs_free_space_cachep);
7224         if (btrfs_delalloc_work_cachep)
7225                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7226 }
7227
7228 int btrfs_init_cachep(void)
7229 {
7230         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7231                         sizeof(struct btrfs_inode), 0,
7232                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7233         if (!btrfs_inode_cachep)
7234                 goto fail;
7235
7236         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7237                         sizeof(struct btrfs_trans_handle), 0,
7238                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7239         if (!btrfs_trans_handle_cachep)
7240                 goto fail;
7241
7242         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7243                         sizeof(struct btrfs_transaction), 0,
7244                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7245         if (!btrfs_transaction_cachep)
7246                 goto fail;
7247
7248         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7249                         sizeof(struct btrfs_path), 0,
7250                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7251         if (!btrfs_path_cachep)
7252                 goto fail;
7253
7254         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7255                         sizeof(struct btrfs_free_space), 0,
7256                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7257         if (!btrfs_free_space_cachep)
7258                 goto fail;
7259
7260         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7261                         sizeof(struct btrfs_delalloc_work), 0,
7262                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7263                         NULL);
7264         if (!btrfs_delalloc_work_cachep)
7265                 goto fail;
7266
7267         return 0;
7268 fail:
7269         btrfs_destroy_cachep();
7270         return -ENOMEM;
7271 }
7272
7273 static int btrfs_getattr(struct vfsmount *mnt,
7274                          struct dentry *dentry, struct kstat *stat)
7275 {
7276         struct inode *inode = dentry->d_inode;
7277         u32 blocksize = inode->i_sb->s_blocksize;
7278
7279         generic_fillattr(inode, stat);
7280         stat->dev = BTRFS_I(inode)->root->anon_dev;
7281         stat->blksize = PAGE_CACHE_SIZE;
7282         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7283                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7284         return 0;
7285 }
7286
7287 /*
7288  * If a file is moved, it will inherit the cow and compression flags of the new
7289  * directory.
7290  */
7291 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7292 {
7293         struct btrfs_inode *b_dir = BTRFS_I(dir);
7294         struct btrfs_inode *b_inode = BTRFS_I(inode);
7295
7296         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7297                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7298         else
7299                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7300
7301         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7302                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7303                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7304         } else {
7305                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7306                                     BTRFS_INODE_NOCOMPRESS);
7307         }
7308 }
7309
7310 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7311                            struct inode *new_dir, struct dentry *new_dentry)
7312 {
7313         struct btrfs_trans_handle *trans;
7314         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7315         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7316         struct inode *new_inode = new_dentry->d_inode;
7317         struct inode *old_inode = old_dentry->d_inode;
7318         struct timespec ctime = CURRENT_TIME;
7319         u64 index = 0;
7320         u64 root_objectid;
7321         int ret;
7322         u64 old_ino = btrfs_ino(old_inode);
7323
7324         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7325                 return -EPERM;
7326
7327         /* we only allow rename subvolume link between subvolumes */
7328         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7329                 return -EXDEV;
7330
7331         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7332             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7333                 return -ENOTEMPTY;
7334
7335         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7336             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7337                 return -ENOTEMPTY;
7338
7339
7340         /* check for collisions, even if the  name isn't there */
7341         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7342                              new_dentry->d_name.name,
7343                              new_dentry->d_name.len);
7344
7345         if (ret) {
7346                 if (ret == -EEXIST) {
7347                         /* we shouldn't get
7348                          * eexist without a new_inode */
7349                         if (!new_inode) {
7350                                 WARN_ON(1);
7351                                 return ret;
7352                         }
7353                 } else {
7354                         /* maybe -EOVERFLOW */
7355                         return ret;
7356                 }
7357         }
7358         ret = 0;
7359
7360         /*
7361          * we're using rename to replace one file with another.
7362          * and the replacement file is large.  Start IO on it now so
7363          * we don't add too much work to the end of the transaction
7364          */
7365         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7366             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7367                 filemap_flush(old_inode->i_mapping);
7368
7369         /* close the racy window with snapshot create/destroy ioctl */
7370         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7371                 down_read(&root->fs_info->subvol_sem);
7372         /*
7373          * We want to reserve the absolute worst case amount of items.  So if
7374          * both inodes are subvols and we need to unlink them then that would
7375          * require 4 item modifications, but if they are both normal inodes it
7376          * would require 5 item modifications, so we'll assume their normal
7377          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7378          * should cover the worst case number of items we'll modify.
7379          */
7380         trans = btrfs_start_transaction(root, 20);
7381         if (IS_ERR(trans)) {
7382                 ret = PTR_ERR(trans);
7383                 goto out_notrans;
7384         }
7385
7386         if (dest != root)
7387                 btrfs_record_root_in_trans(trans, dest);
7388
7389         ret = btrfs_set_inode_index(new_dir, &index);
7390         if (ret)
7391                 goto out_fail;
7392
7393         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7394                 /* force full log commit if subvolume involved. */
7395                 root->fs_info->last_trans_log_full_commit = trans->transid;
7396         } else {
7397                 ret = btrfs_insert_inode_ref(trans, dest,
7398                                              new_dentry->d_name.name,
7399                                              new_dentry->d_name.len,
7400                                              old_ino,
7401                                              btrfs_ino(new_dir), index);
7402                 if (ret)
7403                         goto out_fail;
7404                 /*
7405                  * this is an ugly little race, but the rename is required
7406                  * to make sure that if we crash, the inode is either at the
7407                  * old name or the new one.  pinning the log transaction lets
7408                  * us make sure we don't allow a log commit to come in after
7409                  * we unlink the name but before we add the new name back in.
7410                  */
7411                 btrfs_pin_log_trans(root);
7412         }
7413         /*
7414          * make sure the inode gets flushed if it is replacing
7415          * something.
7416          */
7417         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7418                 btrfs_add_ordered_operation(trans, root, old_inode);
7419
7420         inode_inc_iversion(old_dir);
7421         inode_inc_iversion(new_dir);
7422         inode_inc_iversion(old_inode);
7423         old_dir->i_ctime = old_dir->i_mtime = ctime;
7424         new_dir->i_ctime = new_dir->i_mtime = ctime;
7425         old_inode->i_ctime = ctime;
7426
7427         if (old_dentry->d_parent != new_dentry->d_parent)
7428                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7429
7430         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7431                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7432                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7433                                         old_dentry->d_name.name,
7434                                         old_dentry->d_name.len);
7435         } else {
7436                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7437                                         old_dentry->d_inode,
7438                                         old_dentry->d_name.name,
7439                                         old_dentry->d_name.len);
7440                 if (!ret)
7441                         ret = btrfs_update_inode(trans, root, old_inode);
7442         }
7443         if (ret) {
7444                 btrfs_abort_transaction(trans, root, ret);
7445                 goto out_fail;
7446         }
7447
7448         if (new_inode) {
7449                 inode_inc_iversion(new_inode);
7450                 new_inode->i_ctime = CURRENT_TIME;
7451                 if (unlikely(btrfs_ino(new_inode) ==
7452                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7453                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7454                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7455                                                 root_objectid,
7456                                                 new_dentry->d_name.name,
7457                                                 new_dentry->d_name.len);
7458                         BUG_ON(new_inode->i_nlink == 0);
7459                 } else {
7460                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7461                                                  new_dentry->d_inode,
7462                                                  new_dentry->d_name.name,
7463                                                  new_dentry->d_name.len);
7464                 }
7465                 if (!ret && new_inode->i_nlink == 0) {
7466                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7467                         BUG_ON(ret);
7468                 }
7469                 if (ret) {
7470                         btrfs_abort_transaction(trans, root, ret);
7471                         goto out_fail;
7472                 }
7473         }
7474
7475         fixup_inode_flags(new_dir, old_inode);
7476
7477         ret = btrfs_add_link(trans, new_dir, old_inode,
7478                              new_dentry->d_name.name,
7479                              new_dentry->d_name.len, 0, index);
7480         if (ret) {
7481                 btrfs_abort_transaction(trans, root, ret);
7482                 goto out_fail;
7483         }
7484
7485         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7486                 struct dentry *parent = new_dentry->d_parent;
7487                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7488                 btrfs_end_log_trans(root);
7489         }
7490 out_fail:
7491         btrfs_end_transaction(trans, root);
7492 out_notrans:
7493         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7494                 up_read(&root->fs_info->subvol_sem);
7495
7496         return ret;
7497 }
7498
7499 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7500 {
7501         struct btrfs_delalloc_work *delalloc_work;
7502
7503         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7504                                      work);
7505         if (delalloc_work->wait)
7506                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7507         else
7508                 filemap_flush(delalloc_work->inode->i_mapping);
7509
7510         if (delalloc_work->delay_iput)
7511                 btrfs_add_delayed_iput(delalloc_work->inode);
7512         else
7513                 iput(delalloc_work->inode);
7514         complete(&delalloc_work->completion);
7515 }
7516
7517 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7518                                                     int wait, int delay_iput)
7519 {
7520         struct btrfs_delalloc_work *work;
7521
7522         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7523         if (!work)
7524                 return NULL;
7525
7526         init_completion(&work->completion);
7527         INIT_LIST_HEAD(&work->list);
7528         work->inode = inode;
7529         work->wait = wait;
7530         work->delay_iput = delay_iput;
7531         work->work.func = btrfs_run_delalloc_work;
7532
7533         return work;
7534 }
7535
7536 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7537 {
7538         wait_for_completion(&work->completion);
7539         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7540 }
7541
7542 /*
7543  * some fairly slow code that needs optimization. This walks the list
7544  * of all the inodes with pending delalloc and forces them to disk.
7545  */
7546 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7547 {
7548         struct list_head *head = &root->fs_info->delalloc_inodes;
7549         struct btrfs_inode *binode;
7550         struct inode *inode;
7551         struct btrfs_delalloc_work *work, *next;
7552         struct list_head works;
7553         int ret = 0;
7554
7555         if (root->fs_info->sb->s_flags & MS_RDONLY)
7556                 return -EROFS;
7557
7558         INIT_LIST_HEAD(&works);
7559
7560         spin_lock(&root->fs_info->delalloc_lock);
7561         while (!list_empty(head)) {
7562                 binode = list_entry(head->next, struct btrfs_inode,
7563                                     delalloc_inodes);
7564                 inode = igrab(&binode->vfs_inode);
7565                 if (!inode)
7566                         list_del_init(&binode->delalloc_inodes);
7567                 spin_unlock(&root->fs_info->delalloc_lock);
7568                 if (inode) {
7569                         work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7570                         if (!work) {
7571                                 ret = -ENOMEM;
7572                                 goto out;
7573                         }
7574                         list_add_tail(&work->list, &works);
7575                         btrfs_queue_worker(&root->fs_info->flush_workers,
7576                                            &work->work);
7577                 }
7578                 cond_resched();
7579                 spin_lock(&root->fs_info->delalloc_lock);
7580         }
7581         spin_unlock(&root->fs_info->delalloc_lock);
7582
7583         /* the filemap_flush will queue IO into the worker threads, but
7584          * we have to make sure the IO is actually started and that
7585          * ordered extents get created before we return
7586          */
7587         atomic_inc(&root->fs_info->async_submit_draining);
7588         while (atomic_read(&root->fs_info->nr_async_submits) ||
7589               atomic_read(&root->fs_info->async_delalloc_pages)) {
7590                 wait_event(root->fs_info->async_submit_wait,
7591                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7592                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7593         }
7594         atomic_dec(&root->fs_info->async_submit_draining);
7595 out:
7596         list_for_each_entry_safe(work, next, &works, list) {
7597                 list_del_init(&work->list);
7598                 btrfs_wait_and_free_delalloc_work(work);
7599         }
7600         return ret;
7601 }
7602
7603 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7604                          const char *symname)
7605 {
7606         struct btrfs_trans_handle *trans;
7607         struct btrfs_root *root = BTRFS_I(dir)->root;
7608         struct btrfs_path *path;
7609         struct btrfs_key key;
7610         struct inode *inode = NULL;
7611         int err;
7612         int drop_inode = 0;
7613         u64 objectid;
7614         u64 index = 0 ;
7615         int name_len;
7616         int datasize;
7617         unsigned long ptr;
7618         struct btrfs_file_extent_item *ei;
7619         struct extent_buffer *leaf;
7620
7621         name_len = strlen(symname) + 1;
7622         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7623                 return -ENAMETOOLONG;
7624
7625         /*
7626          * 2 items for inode item and ref
7627          * 2 items for dir items
7628          * 1 item for xattr if selinux is on
7629          */
7630         trans = btrfs_start_transaction(root, 5);
7631         if (IS_ERR(trans))
7632                 return PTR_ERR(trans);
7633
7634         err = btrfs_find_free_ino(root, &objectid);
7635         if (err)
7636                 goto out_unlock;
7637
7638         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7639                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7640                                 S_IFLNK|S_IRWXUGO, &index);
7641         if (IS_ERR(inode)) {
7642                 err = PTR_ERR(inode);
7643                 goto out_unlock;
7644         }
7645
7646         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7647         if (err) {
7648                 drop_inode = 1;
7649                 goto out_unlock;
7650         }
7651
7652         /*
7653         * If the active LSM wants to access the inode during
7654         * d_instantiate it needs these. Smack checks to see
7655         * if the filesystem supports xattrs by looking at the
7656         * ops vector.
7657         */
7658         inode->i_fop = &btrfs_file_operations;
7659         inode->i_op = &btrfs_file_inode_operations;
7660
7661         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7662         if (err)
7663                 drop_inode = 1;
7664         else {
7665                 inode->i_mapping->a_ops = &btrfs_aops;
7666                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7667                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7668         }
7669         if (drop_inode)
7670                 goto out_unlock;
7671
7672         path = btrfs_alloc_path();
7673         if (!path) {
7674                 err = -ENOMEM;
7675                 drop_inode = 1;
7676                 goto out_unlock;
7677         }
7678         key.objectid = btrfs_ino(inode);
7679         key.offset = 0;
7680         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7681         datasize = btrfs_file_extent_calc_inline_size(name_len);
7682         err = btrfs_insert_empty_item(trans, root, path, &key,
7683                                       datasize);
7684         if (err) {
7685                 drop_inode = 1;
7686                 btrfs_free_path(path);
7687                 goto out_unlock;
7688         }
7689         leaf = path->nodes[0];
7690         ei = btrfs_item_ptr(leaf, path->slots[0],
7691                             struct btrfs_file_extent_item);
7692         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7693         btrfs_set_file_extent_type(leaf, ei,
7694                                    BTRFS_FILE_EXTENT_INLINE);
7695         btrfs_set_file_extent_encryption(leaf, ei, 0);
7696         btrfs_set_file_extent_compression(leaf, ei, 0);
7697         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7698         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7699
7700         ptr = btrfs_file_extent_inline_start(ei);
7701         write_extent_buffer(leaf, symname, ptr, name_len);
7702         btrfs_mark_buffer_dirty(leaf);
7703         btrfs_free_path(path);
7704
7705         inode->i_op = &btrfs_symlink_inode_operations;
7706         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7707         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7708         inode_set_bytes(inode, name_len);
7709         btrfs_i_size_write(inode, name_len - 1);
7710         err = btrfs_update_inode(trans, root, inode);
7711         if (err)
7712                 drop_inode = 1;
7713
7714 out_unlock:
7715         if (!err)
7716                 d_instantiate(dentry, inode);
7717         btrfs_end_transaction(trans, root);
7718         if (drop_inode) {
7719                 inode_dec_link_count(inode);
7720                 iput(inode);
7721         }
7722         btrfs_btree_balance_dirty(root);
7723         return err;
7724 }
7725
7726 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7727                                        u64 start, u64 num_bytes, u64 min_size,
7728                                        loff_t actual_len, u64 *alloc_hint,
7729                                        struct btrfs_trans_handle *trans)
7730 {
7731         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7732         struct extent_map *em;
7733         struct btrfs_root *root = BTRFS_I(inode)->root;
7734         struct btrfs_key ins;
7735         u64 cur_offset = start;
7736         u64 i_size;
7737         int ret = 0;
7738         bool own_trans = true;
7739
7740         if (trans)
7741                 own_trans = false;
7742         while (num_bytes > 0) {
7743                 if (own_trans) {
7744                         trans = btrfs_start_transaction(root, 3);
7745                         if (IS_ERR(trans)) {
7746                                 ret = PTR_ERR(trans);
7747                                 break;
7748                         }
7749                 }
7750
7751                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7752                                            0, *alloc_hint, &ins, 1);
7753                 if (ret) {
7754                         if (own_trans)
7755                                 btrfs_end_transaction(trans, root);
7756                         break;
7757                 }
7758
7759                 ret = insert_reserved_file_extent(trans, inode,
7760                                                   cur_offset, ins.objectid,
7761                                                   ins.offset, ins.offset,
7762                                                   ins.offset, 0, 0, 0,
7763                                                   BTRFS_FILE_EXTENT_PREALLOC);
7764                 if (ret) {
7765                         btrfs_abort_transaction(trans, root, ret);
7766                         if (own_trans)
7767                                 btrfs_end_transaction(trans, root);
7768                         break;
7769                 }
7770                 btrfs_drop_extent_cache(inode, cur_offset,
7771                                         cur_offset + ins.offset -1, 0);
7772
7773                 em = alloc_extent_map();
7774                 if (!em) {
7775                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7776                                 &BTRFS_I(inode)->runtime_flags);
7777                         goto next;
7778                 }
7779
7780                 em->start = cur_offset;
7781                 em->orig_start = cur_offset;
7782                 em->len = ins.offset;
7783                 em->block_start = ins.objectid;
7784                 em->block_len = ins.offset;
7785                 em->orig_block_len = ins.offset;
7786                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7787                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7788                 em->generation = trans->transid;
7789
7790                 while (1) {
7791                         write_lock(&em_tree->lock);
7792                         ret = add_extent_mapping(em_tree, em);
7793                         if (!ret)
7794                                 list_move(&em->list,
7795                                           &em_tree->modified_extents);
7796                         write_unlock(&em_tree->lock);
7797                         if (ret != -EEXIST)
7798                                 break;
7799                         btrfs_drop_extent_cache(inode, cur_offset,
7800                                                 cur_offset + ins.offset - 1,
7801                                                 0);
7802                 }
7803                 free_extent_map(em);
7804 next:
7805                 num_bytes -= ins.offset;
7806                 cur_offset += ins.offset;
7807                 *alloc_hint = ins.objectid + ins.offset;
7808
7809                 inode_inc_iversion(inode);
7810                 inode->i_ctime = CURRENT_TIME;
7811                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7812                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7813                     (actual_len > inode->i_size) &&
7814                     (cur_offset > inode->i_size)) {
7815                         if (cur_offset > actual_len)
7816                                 i_size = actual_len;
7817                         else
7818                                 i_size = cur_offset;
7819                         i_size_write(inode, i_size);
7820                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7821                 }
7822
7823                 ret = btrfs_update_inode(trans, root, inode);
7824
7825                 if (ret) {
7826                         btrfs_abort_transaction(trans, root, ret);
7827                         if (own_trans)
7828                                 btrfs_end_transaction(trans, root);
7829                         break;
7830                 }
7831
7832                 if (own_trans)
7833                         btrfs_end_transaction(trans, root);
7834         }
7835         return ret;
7836 }
7837
7838 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7839                               u64 start, u64 num_bytes, u64 min_size,
7840                               loff_t actual_len, u64 *alloc_hint)
7841 {
7842         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7843                                            min_size, actual_len, alloc_hint,
7844                                            NULL);
7845 }
7846
7847 int btrfs_prealloc_file_range_trans(struct inode *inode,
7848                                     struct btrfs_trans_handle *trans, int mode,
7849                                     u64 start, u64 num_bytes, u64 min_size,
7850                                     loff_t actual_len, u64 *alloc_hint)
7851 {
7852         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7853                                            min_size, actual_len, alloc_hint, trans);
7854 }
7855
7856 static int btrfs_set_page_dirty(struct page *page)
7857 {
7858         return __set_page_dirty_nobuffers(page);
7859 }
7860
7861 static int btrfs_permission(struct inode *inode, int mask)
7862 {
7863         struct btrfs_root *root = BTRFS_I(inode)->root;
7864         umode_t mode = inode->i_mode;
7865
7866         if (mask & MAY_WRITE &&
7867             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7868                 if (btrfs_root_readonly(root))
7869                         return -EROFS;
7870                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7871                         return -EACCES;
7872         }
7873         return generic_permission(inode, mask);
7874 }
7875
7876 static const struct inode_operations btrfs_dir_inode_operations = {
7877         .getattr        = btrfs_getattr,
7878         .lookup         = btrfs_lookup,
7879         .create         = btrfs_create,
7880         .unlink         = btrfs_unlink,
7881         .link           = btrfs_link,
7882         .mkdir          = btrfs_mkdir,
7883         .rmdir          = btrfs_rmdir,
7884         .rename         = btrfs_rename,
7885         .symlink        = btrfs_symlink,
7886         .setattr        = btrfs_setattr,
7887         .mknod          = btrfs_mknod,
7888         .setxattr       = btrfs_setxattr,
7889         .getxattr       = btrfs_getxattr,
7890         .listxattr      = btrfs_listxattr,
7891         .removexattr    = btrfs_removexattr,
7892         .permission     = btrfs_permission,
7893         .get_acl        = btrfs_get_acl,
7894 };
7895 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7896         .lookup         = btrfs_lookup,
7897         .permission     = btrfs_permission,
7898         .get_acl        = btrfs_get_acl,
7899 };
7900
7901 static const struct file_operations btrfs_dir_file_operations = {
7902         .llseek         = generic_file_llseek,
7903         .read           = generic_read_dir,
7904         .readdir        = btrfs_real_readdir,
7905         .unlocked_ioctl = btrfs_ioctl,
7906 #ifdef CONFIG_COMPAT
7907         .compat_ioctl   = btrfs_ioctl,
7908 #endif
7909         .release        = btrfs_release_file,
7910         .fsync          = btrfs_sync_file,
7911 };
7912
7913 static struct extent_io_ops btrfs_extent_io_ops = {
7914         .fill_delalloc = run_delalloc_range,
7915         .submit_bio_hook = btrfs_submit_bio_hook,
7916         .merge_bio_hook = btrfs_merge_bio_hook,
7917         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7918         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7919         .writepage_start_hook = btrfs_writepage_start_hook,
7920         .set_bit_hook = btrfs_set_bit_hook,
7921         .clear_bit_hook = btrfs_clear_bit_hook,
7922         .merge_extent_hook = btrfs_merge_extent_hook,
7923         .split_extent_hook = btrfs_split_extent_hook,
7924 };
7925
7926 /*
7927  * btrfs doesn't support the bmap operation because swapfiles
7928  * use bmap to make a mapping of extents in the file.  They assume
7929  * these extents won't change over the life of the file and they
7930  * use the bmap result to do IO directly to the drive.
7931  *
7932  * the btrfs bmap call would return logical addresses that aren't
7933  * suitable for IO and they also will change frequently as COW
7934  * operations happen.  So, swapfile + btrfs == corruption.
7935  *
7936  * For now we're avoiding this by dropping bmap.
7937  */
7938 static const struct address_space_operations btrfs_aops = {
7939         .readpage       = btrfs_readpage,
7940         .writepage      = btrfs_writepage,
7941         .writepages     = btrfs_writepages,
7942         .readpages      = btrfs_readpages,
7943         .direct_IO      = btrfs_direct_IO,
7944         .invalidatepage = btrfs_invalidatepage,
7945         .releasepage    = btrfs_releasepage,
7946         .set_page_dirty = btrfs_set_page_dirty,
7947         .error_remove_page = generic_error_remove_page,
7948 };
7949
7950 static const struct address_space_operations btrfs_symlink_aops = {
7951         .readpage       = btrfs_readpage,
7952         .writepage      = btrfs_writepage,
7953         .invalidatepage = btrfs_invalidatepage,
7954         .releasepage    = btrfs_releasepage,
7955 };
7956
7957 static const struct inode_operations btrfs_file_inode_operations = {
7958         .getattr        = btrfs_getattr,
7959         .setattr        = btrfs_setattr,
7960         .setxattr       = btrfs_setxattr,
7961         .getxattr       = btrfs_getxattr,
7962         .listxattr      = btrfs_listxattr,
7963         .removexattr    = btrfs_removexattr,
7964         .permission     = btrfs_permission,
7965         .fiemap         = btrfs_fiemap,
7966         .get_acl        = btrfs_get_acl,
7967         .update_time    = btrfs_update_time,
7968 };
7969 static const struct inode_operations btrfs_special_inode_operations = {
7970         .getattr        = btrfs_getattr,
7971         .setattr        = btrfs_setattr,
7972         .permission     = btrfs_permission,
7973         .setxattr       = btrfs_setxattr,
7974         .getxattr       = btrfs_getxattr,
7975         .listxattr      = btrfs_listxattr,
7976         .removexattr    = btrfs_removexattr,
7977         .get_acl        = btrfs_get_acl,
7978         .update_time    = btrfs_update_time,
7979 };
7980 static const struct inode_operations btrfs_symlink_inode_operations = {
7981         .readlink       = generic_readlink,
7982         .follow_link    = page_follow_link_light,
7983         .put_link       = page_put_link,
7984         .getattr        = btrfs_getattr,
7985         .setattr        = btrfs_setattr,
7986         .permission     = btrfs_permission,
7987         .setxattr       = btrfs_setxattr,
7988         .getxattr       = btrfs_getxattr,
7989         .listxattr      = btrfs_listxattr,
7990         .removexattr    = btrfs_removexattr,
7991         .get_acl        = btrfs_get_acl,
7992         .update_time    = btrfs_update_time,
7993 };
7994
7995 const struct dentry_operations btrfs_dentry_operations = {
7996         .d_delete       = btrfs_dentry_delete,
7997         .d_release      = btrfs_dentry_release,
7998 };