Pileus Git - ~andy/linux/blob - fs/nilfs2/inode.c

   1 /*
   2  * inode.c - NILFS inode operations.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/buffer_head.h>
  25 #include <linux/gfp.h>
  26 #include <linux/mpage.h>
  27 #include <linux/writeback.h>
  28 #include <linux/uio.h>
  29 #include "nilfs.h"
  30 #include "btnode.h"
  31 #include "segment.h"
  32 #include "page.h"
  33 #include "mdt.h"
  34 #include "cpfile.h"
  35 #include "ifile.h"
  36
  37 /**
  38  * struct nilfs_iget_args - arguments used during comparison between inodes
  39  * @ino: inode number
  40  * @cno: checkpoint number
  41  * @root: pointer on NILFS root object (mounted checkpoint)
  42  * @for_gc: inode for GC flag
  43  */
  44 struct nilfs_iget_args {
  45         u64 ino;
  46         __u64 cno;
  47         struct nilfs_root *root;
  48         int for_gc;
  49 };
  50
  51 void nilfs_inode_add_blocks(struct inode *inode, int n)
  52 {
  53         struct nilfs_root *root = NILFS_I(inode)->i_root;
  54
  55         inode_add_bytes(inode, (1 << inode->i_blkbits) * n);
  56         if (root)
  57                 atomic_add(n, &root->blocks_count);
  58 }
  59
  60 void nilfs_inode_sub_blocks(struct inode *inode, int n)
  61 {
  62         struct nilfs_root *root = NILFS_I(inode)->i_root;
  63
  64         inode_sub_bytes(inode, (1 << inode->i_blkbits) * n);
  65         if (root)
  66                 atomic_sub(n, &root->blocks_count);
  67 }
  68
  69 /**
  70  * nilfs_get_block() - get a file block on the filesystem (callback function)
  71  * @inode - inode struct of the target file
  72  * @blkoff - file block number
  73  * @bh_result - buffer head to be mapped on
  74  * @create - indicate whether allocating the block or not when it has not
  75  *      been allocated yet.
  76  *
  77  * This function does not issue actual read request of the specified data
  78  * block. It is done by VFS.
  79  */
  80 int nilfs_get_block(struct inode *inode, sector_t blkoff,
  81                     struct buffer_head *bh_result, int create)
  82 {
  83         struct nilfs_inode_info *ii = NILFS_I(inode);
  84         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
  85         __u64 blknum = 0;
  86         int err = 0, ret;
  87         unsigned maxblocks = bh_result->b_size >> inode->i_blkbits;
  88
  89         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  90         ret = nilfs_bmap_lookup_contig(ii->i_bmap, blkoff, &blknum, maxblocks);
  91         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
  92         if (ret >= 0) { /* found */
  93                 map_bh(bh_result, inode->i_sb, blknum);
  94                 if (ret > 0)
  95                         bh_result->b_size = (ret << inode->i_blkbits);
  96                 goto out;
  97         }
  98         /* data block was not found */
  99         if (ret == -ENOENT && create) {
 100                 struct nilfs_transaction_info ti;
 101
 102                 bh_result->b_blocknr = 0;
 103                 err = nilfs_transaction_begin(inode->i_sb, &ti, 1);
 104                 if (unlikely(err))
 105                         goto out;
 106                 err = nilfs_bmap_insert(ii->i_bmap, (unsigned long)blkoff,
 107                                         (unsigned long)bh_result);
 108                 if (unlikely(err != 0)) {
 109                         if (err == -EEXIST) {
 110                                 /*
 111                                  * The get_block() function could be called
 112                                  * from multiple callers for an inode.
 113                                  * However, the page having this block must
 114                                  * be locked in this case.
 115                                  */
 116                                 printk(KERN_WARNING
 117                                        "nilfs_get_block: a race condition "
 118                                        "while inserting a data block. "
 119                                        "(inode number=%lu, file block "
 120                                        "offset=%llu)\n",
 121                                        inode->i_ino,
 122                                        (unsigned long long)blkoff);
 123                                 err = 0;
 124                         }
 125                         nilfs_transaction_abort(inode->i_sb);
 126                         goto out;
 127                 }
 128                 nilfs_mark_inode_dirty(inode);
 129                 nilfs_transaction_commit(inode->i_sb); /* never fails */
 130                 /* Error handling should be detailed */
 131                 set_buffer_new(bh_result);
 132                 set_buffer_delay(bh_result);
 133                 map_bh(bh_result, inode->i_sb, 0); /* dbn must be changed
 134                                                       to proper value */
 135         } else if (ret == -ENOENT) {
 136                 /* not found is not error (e.g. hole); must return without
 137                    the mapped state flag. */
 138                 ;
 139         } else {
 140                 err = ret;
 141         }
 142
 143  out:
 144         return err;
 145 }
 146
 147 /**
 148  * nilfs_readpage() - implement readpage() method of nilfs_aops {}
 149  * address_space_operations.
 150  * @file - file struct of the file to be read
 151  * @page - the page to be read
 152  */
 153 static int nilfs_readpage(struct file *file, struct page *page)
 154 {
 155         return mpage_readpage(page, nilfs_get_block);
 156 }
 157
 158 /**
 159  * nilfs_readpages() - implement readpages() method of nilfs_aops {}
 160  * address_space_operations.
 161  * @file - file struct of the file to be read
 162  * @mapping - address_space struct used for reading multiple pages
 163  * @pages - the pages to be read
 164  * @nr_pages - number of pages to be read
 165  */
 166 static int nilfs_readpages(struct file *file, struct address_space *mapping,
 167                            struct list_head *pages, unsigned nr_pages)
 168 {
 169         return mpage_readpages(mapping, pages, nr_pages, nilfs_get_block);
 170 }
 171
 172 static int nilfs_writepages(struct address_space *mapping,
 173                             struct writeback_control *wbc)
 174 {
 175         struct inode *inode = mapping->host;
 176         int err = 0;
 177
 178         if (wbc->sync_mode == WB_SYNC_ALL)
 179                 err = nilfs_construct_dsync_segment(inode->i_sb, inode,
 180                                                     wbc->range_start,
 181                                                     wbc->range_end);
 182         return err;
 183 }
 184
 185 static int nilfs_writepage(struct page *page, struct writeback_control *wbc)
 186 {
 187         struct inode *inode = page->mapping->host;
 188         int err;
 189
 190         redirty_page_for_writepage(wbc, page);
 191         unlock_page(page);
 192
 193         if (wbc->sync_mode == WB_SYNC_ALL) {
 194                 err = nilfs_construct_segment(inode->i_sb);
 195                 if (unlikely(err))
 196                         return err;
 197         } else if (wbc->for_reclaim)
 198                 nilfs_flush_segment(inode->i_sb, inode->i_ino);
 199
 200         return 0;
 201 }
 202
 203 static int nilfs_set_page_dirty(struct page *page)
 204 {
 205         int ret = __set_page_dirty_buffers(page);
 206
 207         if (ret) {
 208                 struct inode *inode = page->mapping->host;
 209                 unsigned nr_dirty = 1 << (PAGE_SHIFT - inode->i_blkbits);
 210
 211                 nilfs_set_file_dirty(inode, nr_dirty);
 212         }
 213         return ret;
 214 }
 215
 216 void nilfs_write_failed(struct address_space *mapping, loff_t to)
 217 {
 218         struct inode *inode = mapping->host;
 219
 220         if (to > inode->i_size) {
 221                 truncate_pagecache(inode, to, inode->i_size);
 222                 nilfs_truncate(inode);
 223         }
 224 }
 225
 226 static int nilfs_write_begin(struct file *file, struct address_space *mapping,
 227                              loff_t pos, unsigned len, unsigned flags,
 228                              struct page **pagep, void **fsdata)
 229
 230 {
 231         struct inode *inode = mapping->host;
 232         int err = nilfs_transaction_begin(inode->i_sb, NULL, 1);
 233
 234         if (unlikely(err))
 235                 return err;
 236
 237         err = block_write_begin(mapping, pos, len, flags, pagep,
 238                                 nilfs_get_block);
 239         if (unlikely(err)) {
 240                 nilfs_write_failed(mapping, pos + len);
 241                 nilfs_transaction_abort(inode->i_sb);
 242         }
 243         return err;
 244 }
 245
 246 static int nilfs_write_end(struct file *file, struct address_space *mapping,
 247                            loff_t pos, unsigned len, unsigned copied,
 248                            struct page *page, void *fsdata)
 249 {
 250         struct inode *inode = mapping->host;
 251         unsigned start = pos & (PAGE_CACHE_SIZE - 1);
 252         unsigned nr_dirty;
 253         int err;
 254
 255         nr_dirty = nilfs_page_count_clean_buffers(page, start,
 256                                                   start + copied);
 257         copied = generic_write_end(file, mapping, pos, len, copied, page,
 258                                    fsdata);
 259         nilfs_set_file_dirty(inode, nr_dirty);
 260         err = nilfs_transaction_commit(inode->i_sb);
 261         return err ? : copied;
 262 }
 263
 264 static ssize_t
 265 nilfs_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov,
 266                 loff_t offset, unsigned long nr_segs)
 267 {
 268         struct file *file = iocb->ki_filp;
 269         struct address_space *mapping = file->f_mapping;
 270         struct inode *inode = file->f_mapping->host;
 271         ssize_t size;
 272
 273         if (rw == WRITE)
 274                 return 0;
 275
 276         /* Needs synchronization with the cleaner */
 277         size = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs,
 278                                   nilfs_get_block);
 279
 280         /*
 281          * In case of error extending write may have instantiated a few
 282          * blocks outside i_size. Trim these off again.
 283          */
 284         if (unlikely((rw & WRITE) && size < 0)) {
 285                 loff_t isize = i_size_read(inode);
 286                 loff_t end = offset + iov_length(iov, nr_segs);
 287
 288                 if (end > isize)
 289                         nilfs_write_failed(mapping, end);
 290         }
 291
 292         return size;
 293 }
 294
 295 const struct address_space_operations nilfs_aops = {
 296         .writepage              = nilfs_writepage,
 297         .readpage               = nilfs_readpage,
 298         .writepages             = nilfs_writepages,
 299         .set_page_dirty         = nilfs_set_page_dirty,
 300         .readpages              = nilfs_readpages,
 301         .write_begin            = nilfs_write_begin,
 302         .write_end              = nilfs_write_end,
 303         /* .releasepage         = nilfs_releasepage, */
 304         .invalidatepage         = block_invalidatepage,
 305         .direct_IO              = nilfs_direct_IO,
 306         .is_partially_uptodate  = block_is_partially_uptodate,
 307 };
 308
 309 struct inode *nilfs_new_inode(struct inode *dir, umode_t mode)
 310 {
 311         struct super_block *sb = dir->i_sb;
 312         struct the_nilfs *nilfs = sb->s_fs_info;
 313         struct inode *inode;
 314         struct nilfs_inode_info *ii;
 315         struct nilfs_root *root;
 316         int err = -ENOMEM;
 317         ino_t ino;
 318
 319         inode = new_inode(sb);
 320         if (unlikely(!inode))
 321                 goto failed;
 322
 323         mapping_set_gfp_mask(inode->i_mapping,
 324                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 325
 326         root = NILFS_I(dir)->i_root;
 327         ii = NILFS_I(inode);
 328         ii->i_state = 1 << NILFS_I_NEW;
 329         ii->i_root = root;
 330
 331         err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
 332         if (unlikely(err))
 333                 goto failed_ifile_create_inode;
 334         /* reference count of i_bh inherits from nilfs_mdt_read_block() */
 335
 336         atomic_inc(&root->inodes_count);
 337         inode_init_owner(inode, dir, mode);
 338         inode->i_ino = ino;
 339         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 340
 341         if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
 342                 err = nilfs_bmap_read(ii->i_bmap, NULL);
 343                 if (err < 0)
 344                         goto failed_bmap;
 345
 346                 set_bit(NILFS_I_BMAP, &ii->i_state);
 347                 /* No lock is needed; iget() ensures it. */
 348         }
 349
 350         ii->i_flags = nilfs_mask_flags(
 351                 mode, NILFS_I(dir)->i_flags & NILFS_FL_INHERITED);
 352
 353         /* ii->i_file_acl = 0; */
 354         /* ii->i_dir_acl = 0; */
 355         ii->i_dir_start_lookup = 0;
 356         nilfs_set_inode_flags(inode);
 357         spin_lock(&nilfs->ns_next_gen_lock);
 358         inode->i_generation = nilfs->ns_next_generation++;
 359         spin_unlock(&nilfs->ns_next_gen_lock);
 360         insert_inode_hash(inode);
 361
 362         err = nilfs_init_acl(inode, dir);
 363         if (unlikely(err))
 364                 goto failed_acl; /* never occur. When supporting
 365                                     nilfs_init_acl(), proper cancellation of
 366                                     above jobs should be considered */
 367
 368         return inode;
 369
 370  failed_acl:
 371  failed_bmap:
 372         clear_nlink(inode);
 373         iput(inode);  /* raw_inode will be deleted through
 374                          generic_delete_inode() */
 375         goto failed;
 376
 377  failed_ifile_create_inode:
 378         make_bad_inode(inode);
 379         iput(inode);  /* if i_nlink == 1, generic_forget_inode() will be
 380                          called */
 381  failed:
 382         return ERR_PTR(err);
 383 }
 384
 385 void nilfs_set_inode_flags(struct inode *inode)
 386 {
 387         unsigned int flags = NILFS_I(inode)->i_flags;
 388
 389         inode->i_flags &= ~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME |
 390                             S_DIRSYNC);
 391         if (flags & FS_SYNC_FL)
 392                 inode->i_flags |= S_SYNC;
 393         if (flags & FS_APPEND_FL)
 394                 inode->i_flags |= S_APPEND;
 395         if (flags & FS_IMMUTABLE_FL)
 396                 inode->i_flags |= S_IMMUTABLE;
 397         if (flags & FS_NOATIME_FL)
 398                 inode->i_flags |= S_NOATIME;
 399         if (flags & FS_DIRSYNC_FL)
 400                 inode->i_flags |= S_DIRSYNC;
 401         mapping_set_gfp_mask(inode->i_mapping,
 402                              mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
 403 }
 404
 405 int nilfs_read_inode_common(struct inode *inode,
 406                             struct nilfs_inode *raw_inode)
 407 {
 408         struct nilfs_inode_info *ii = NILFS_I(inode);
 409         int err;
 410
 411         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 412         i_uid_write(inode, le32_to_cpu(raw_inode->i_uid));
 413         i_gid_write(inode, le32_to_cpu(raw_inode->i_gid));
 414         set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
 415         inode->i_size = le64_to_cpu(raw_inode->i_size);
 416         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 417         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 418         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 419         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 420         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 421         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 422         if (inode->i_nlink == 0 && inode->i_mode == 0)
 423                 return -EINVAL; /* this inode is deleted */
 424
 425         inode->i_blocks = le64_to_cpu(raw_inode->i_blocks);
 426         ii->i_flags = le32_to_cpu(raw_inode->i_flags);
 427 #if 0
 428         ii->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
 429         ii->i_dir_acl = S_ISREG(inode->i_mode) ?
 430                 0 : le32_to_cpu(raw_inode->i_dir_acl);
 431 #endif
 432         ii->i_dir_start_lookup = 0;
 433         inode->i_generation = le32_to_cpu(raw_inode->i_generation);
 434
 435         if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
 436             S_ISLNK(inode->i_mode)) {
 437                 err = nilfs_bmap_read(ii->i_bmap, raw_inode);
 438                 if (err < 0)
 439                         return err;
 440                 set_bit(NILFS_I_BMAP, &ii->i_state);
 441                 /* No lock is needed; iget() ensures it. */
 442         }
 443         return 0;
 444 }
 445
 446 static int __nilfs_read_inode(struct super_block *sb,
 447                               struct nilfs_root *root, unsigned long ino,
 448                               struct inode *inode)
 449 {
 450         struct the_nilfs *nilfs = sb->s_fs_info;
 451         struct buffer_head *bh;
 452         struct nilfs_inode *raw_inode;
 453         int err;
 454
 455         down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 456         err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
 457         if (unlikely(err))
 458                 goto bad_inode;
 459
 460         raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
 461
 462         err = nilfs_read_inode_common(inode, raw_inode);
 463         if (err)
 464                 goto failed_unmap;
 465
 466         if (S_ISREG(inode->i_mode)) {
 467                 inode->i_op = &nilfs_file_inode_operations;
 468                 inode->i_fop = &nilfs_file_operations;
 469                 inode->i_mapping->a_ops = &nilfs_aops;
 470         } else if (S_ISDIR(inode->i_mode)) {
 471                 inode->i_op = &nilfs_dir_inode_operations;
 472                 inode->i_fop = &nilfs_dir_operations;
 473                 inode->i_mapping->a_ops = &nilfs_aops;
 474         } else if (S_ISLNK(inode->i_mode)) {
 475                 inode->i_op = &nilfs_symlink_inode_operations;
 476                 inode->i_mapping->a_ops = &nilfs_aops;
 477         } else {
 478                 inode->i_op = &nilfs_special_inode_operations;
 479                 init_special_inode(
 480                         inode, inode->i_mode,
 481                         huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
 482         }
 483         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 484         brelse(bh);
 485         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 486         nilfs_set_inode_flags(inode);
 487         return 0;
 488
 489  failed_unmap:
 490         nilfs_ifile_unmap_inode(root->ifile, ino, bh);
 491         brelse(bh);
 492
 493  bad_inode:
 494         up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
 495         return err;
 496 }
 497
 498 static int nilfs_iget_test(struct inode *inode, void *opaque)
 499 {
 500         struct nilfs_iget_args *args = opaque;
 501         struct nilfs_inode_info *ii;
 502
 503         if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
 504                 return 0;
 505
 506         ii = NILFS_I(inode);
 507         if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
 508                 return !args->for_gc;
 509
 510         return args->for_gc && args->cno == ii->i_cno;
 511 }
 512
 513 static int nilfs_iget_set(struct inode *inode, void *opaque)
 514 {
 515         struct nilfs_iget_args *args = opaque;
 516
 517         inode->i_ino = args->ino;
 518         if (args->for_gc) {
 519                 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
 520                 NILFS_I(inode)->i_cno = args->cno;
 521                 NILFS_I(inode)->i_root = NULL;
 522         } else {
 523                 if (args->root && args->ino == NILFS_ROOT_INO)
 524                         nilfs_get_root(args->root);
 525                 NILFS_I(inode)->i_root = args->root;
 526         }
 527         return 0;
 528 }
 529
 530 struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
 531                             unsigned long ino)
 532 {
 533         struct nilfs_iget_args args = {
 534                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 535         };
 536
 537         return ilookup5(sb, ino, nilfs_iget_test, &args);
 538 }
 539
 540 struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
 541                                 unsigned long ino)
 542 {
 543         struct nilfs_iget_args args = {
 544                 .ino = ino, .root = root, .cno = 0, .for_gc = 0
 545         };
 546
 547         return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 548 }
 549
 550 struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
 551                          unsigned long ino)
 552 {
 553         struct inode *inode;
 554         int err;
 555
 556         inode = nilfs_iget_locked(sb, root, ino);
 557         if (unlikely(!inode))
 558                 return ERR_PTR(-ENOMEM);
 559         if (!(inode->i_state & I_NEW))
 560                 return inode;
 561
 562         err = __nilfs_read_inode(sb, root, ino, inode);
 563         if (unlikely(err)) {
 564                 iget_failed(inode);
 565                 return ERR_PTR(err);
 566         }
 567         unlock_new_inode(inode);
 568         return inode;
 569 }
 570
 571 struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
 572                                 __u64 cno)
 573 {
 574         struct nilfs_iget_args args = {
 575                 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
 576         };
 577         struct inode *inode;
 578         int err;
 579
 580         inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
 581         if (unlikely(!inode))
 582                 return ERR_PTR(-ENOMEM);
 583         if (!(inode->i_state & I_NEW))
 584                 return inode;
 585
 586         err = nilfs_init_gcinode(inode);
 587         if (unlikely(err)) {
 588                 iget_failed(inode);
 589                 return ERR_PTR(err);
 590         }
 591         unlock_new_inode(inode);
 592         return inode;
 593 }
 594
 595 void nilfs_write_inode_common(struct inode *inode,
 596                               struct nilfs_inode *raw_inode, int has_bmap)
 597 {
 598         struct nilfs_inode_info *ii = NILFS_I(inode);
 599
 600         raw_inode->i_mode = cpu_to_le16(inode->i_mode);
 601         raw_inode->i_uid = cpu_to_le32(i_uid_read(inode));
 602         raw_inode->i_gid = cpu_to_le32(i_gid_read(inode));
 603         raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
 604         raw_inode->i_size = cpu_to_le64(inode->i_size);
 605         raw_inode->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 606         raw_inode->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 607         raw_inode->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 608         raw_inode->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 609         raw_inode->i_blocks = cpu_to_le64(inode->i_blocks);
 610
 611         raw_inode->i_flags = cpu_to_le32(ii->i_flags);
 612         raw_inode->i_generation = cpu_to_le32(inode->i_generation);
 613
 614         if (NILFS_ROOT_METADATA_FILE(inode->i_ino)) {
 615                 struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 616
 617                 /* zero-fill unused portion in the case of super root block */
 618                 raw_inode->i_xattr = 0;
 619                 raw_inode->i_pad = 0;
 620                 memset((void *)raw_inode + sizeof(*raw_inode), 0,
 621                        nilfs->ns_inode_size - sizeof(*raw_inode));
 622         }
 623
 624         if (has_bmap)
 625                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 626         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 627                 raw_inode->i_device_code =
 628                         cpu_to_le64(huge_encode_dev(inode->i_rdev));
 629         /* When extending inode, nilfs->ns_inode_size should be checked
 630            for substitutions of appended fields */
 631 }
 632
 633 void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
 634 {
 635         ino_t ino = inode->i_ino;
 636         struct nilfs_inode_info *ii = NILFS_I(inode);
 637         struct inode *ifile = ii->i_root->ifile;
 638         struct nilfs_inode *raw_inode;
 639
 640         raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
 641
 642         if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
 643                 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
 644         set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 645
 646         nilfs_write_inode_common(inode, raw_inode, 0);
 647                 /* XXX: call with has_bmap = 0 is a workaround to avoid
 648                    deadlock of bmap. This delays update of i_bmap to just
 649                    before writing */
 650         nilfs_ifile_unmap_inode(ifile, ino, ibh);
 651 }
 652
 653 #define NILFS_MAX_TRUNCATE_BLOCKS       16384  /* 64MB for 4KB block */
 654
 655 static void nilfs_truncate_bmap(struct nilfs_inode_info *ii,
 656                                 unsigned long from)
 657 {
 658         unsigned long b;
 659         int ret;
 660
 661         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 662                 return;
 663 repeat:
 664         ret = nilfs_bmap_last_key(ii->i_bmap, &b);
 665         if (ret == -ENOENT)
 666                 return;
 667         else if (ret < 0)
 668                 goto failed;
 669
 670         if (b < from)
 671                 return;
 672
 673         b -= min_t(unsigned long, NILFS_MAX_TRUNCATE_BLOCKS, b - from);
 674         ret = nilfs_bmap_truncate(ii->i_bmap, b);
 675         nilfs_relax_pressure_in_lock(ii->vfs_inode.i_sb);
 676         if (!ret || (ret == -ENOMEM &&
 677                      nilfs_bmap_truncate(ii->i_bmap, b) == 0))
 678                 goto repeat;
 679
 680 failed:
 681         nilfs_warning(ii->vfs_inode.i_sb, __func__,
 682                       "failed to truncate bmap (ino=%lu, err=%d)",
 683                       ii->vfs_inode.i_ino, ret);
 684 }
 685
 686 void nilfs_truncate(struct inode *inode)
 687 {
 688         unsigned long blkoff;
 689         unsigned int blocksize;
 690         struct nilfs_transaction_info ti;
 691         struct super_block *sb = inode->i_sb;
 692         struct nilfs_inode_info *ii = NILFS_I(inode);
 693
 694         if (!test_bit(NILFS_I_BMAP, &ii->i_state))
 695                 return;
 696         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
 697                 return;
 698
 699         blocksize = sb->s_blocksize;
 700         blkoff = (inode->i_size + blocksize - 1) >> sb->s_blocksize_bits;
 701         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 702
 703         block_truncate_page(inode->i_mapping, inode->i_size, nilfs_get_block);
 704
 705         nilfs_truncate_bmap(ii, blkoff);
 706
 707         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 708         if (IS_SYNC(inode))
 709                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 710
 711         nilfs_mark_inode_dirty(inode);
 712         nilfs_set_file_dirty(inode, 0);
 713         nilfs_transaction_commit(sb);
 714         /* May construct a logical segment and may fail in sync mode.
 715            But truncate has no return value. */
 716 }
 717
 718 static void nilfs_clear_inode(struct inode *inode)
 719 {
 720         struct nilfs_inode_info *ii = NILFS_I(inode);
 721         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 722
 723         /*
 724          * Free resources allocated in nilfs_read_inode(), here.
 725          */
 726         BUG_ON(!list_empty(&ii->i_dirty));
 727         brelse(ii->i_bh);
 728         ii->i_bh = NULL;
 729
 730         if (mdi && mdi->mi_palloc_cache)
 731                 nilfs_palloc_destroy_cache(inode);
 732
 733         if (test_bit(NILFS_I_BMAP, &ii->i_state))
 734                 nilfs_bmap_clear(ii->i_bmap);
 735
 736         nilfs_btnode_cache_clear(&ii->i_btnode_cache);
 737
 738         if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
 739                 nilfs_put_root(ii->i_root);
 740 }
 741
 742 void nilfs_evict_inode(struct inode *inode)
 743 {
 744         struct nilfs_transaction_info ti;
 745         struct super_block *sb = inode->i_sb;
 746         struct nilfs_inode_info *ii = NILFS_I(inode);
 747         int ret;
 748
 749         if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
 750                 if (inode->i_data.nrpages)
 751                         truncate_inode_pages(&inode->i_data, 0);
 752                 clear_inode(inode);
 753                 nilfs_clear_inode(inode);
 754                 return;
 755         }
 756         nilfs_transaction_begin(sb, &ti, 0); /* never fails */
 757
 758         if (inode->i_data.nrpages)
 759                 truncate_inode_pages(&inode->i_data, 0);
 760
 761         /* TODO: some of the following operations may fail.  */
 762         nilfs_truncate_bmap(ii, 0);
 763         nilfs_mark_inode_dirty(inode);
 764         clear_inode(inode);
 765
 766         ret = nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
 767         if (!ret)
 768                 atomic_dec(&ii->i_root->inodes_count);
 769
 770         nilfs_clear_inode(inode);
 771
 772         if (IS_SYNC(inode))
 773                 nilfs_set_transaction_flag(NILFS_TI_SYNC);
 774         nilfs_transaction_commit(sb);
 775         /* May construct a logical segment and may fail in sync mode.
 776            But delete_inode has no return value. */
 777 }
 778
 779 int nilfs_setattr(struct dentry *dentry, struct iattr *iattr)
 780 {
 781         struct nilfs_transaction_info ti;
 782         struct inode *inode = dentry->d_inode;
 783         struct super_block *sb = inode->i_sb;
 784         int err;
 785
 786         err = inode_change_ok(inode, iattr);
 787         if (err)
 788                 return err;
 789
 790         err = nilfs_transaction_begin(sb, &ti, 0);
 791         if (unlikely(err))
 792                 return err;
 793
 794         if ((iattr->ia_valid & ATTR_SIZE) &&
 795             iattr->ia_size != i_size_read(inode)) {
 796                 inode_dio_wait(inode);
 797                 truncate_setsize(inode, iattr->ia_size);
 798                 nilfs_truncate(inode);
 799         }
 800
 801         setattr_copy(inode, iattr);
 802         mark_inode_dirty(inode);
 803
 804         if (iattr->ia_valid & ATTR_MODE) {
 805                 err = nilfs_acl_chmod(inode);
 806                 if (unlikely(err))
 807                         goto out_err;
 808         }
 809
 810         return nilfs_transaction_commit(sb);
 811
 812 out_err:
 813         nilfs_transaction_abort(sb);
 814         return err;
 815 }
 816
 817 int nilfs_permission(struct inode *inode, int mask)
 818 {
 819         struct nilfs_root *root = NILFS_I(inode)->i_root;
 820         if ((mask & MAY_WRITE) && root &&
 821             root->cno != NILFS_CPTREE_CURRENT_CNO)
 822                 return -EROFS; /* snapshot is not writable */
 823
 824         return generic_permission(inode, mask);
 825 }
 826
 827 int nilfs_load_inode_block(struct inode *inode, struct buffer_head **pbh)
 828 {
 829         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 830         struct nilfs_inode_info *ii = NILFS_I(inode);
 831         int err;
 832
 833         spin_lock(&nilfs->ns_inode_lock);
 834         if (ii->i_bh == NULL) {
 835                 spin_unlock(&nilfs->ns_inode_lock);
 836                 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
 837                                                   inode->i_ino, pbh);
 838                 if (unlikely(err))
 839                         return err;
 840                 spin_lock(&nilfs->ns_inode_lock);
 841                 if (ii->i_bh == NULL)
 842                         ii->i_bh = *pbh;
 843                 else {
 844                         brelse(*pbh);
 845                         *pbh = ii->i_bh;
 846                 }
 847         } else
 848                 *pbh = ii->i_bh;
 849
 850         get_bh(*pbh);
 851         spin_unlock(&nilfs->ns_inode_lock);
 852         return 0;
 853 }
 854
 855 int nilfs_inode_dirty(struct inode *inode)
 856 {
 857         struct nilfs_inode_info *ii = NILFS_I(inode);
 858         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 859         int ret = 0;
 860
 861         if (!list_empty(&ii->i_dirty)) {
 862                 spin_lock(&nilfs->ns_inode_lock);
 863                 ret = test_bit(NILFS_I_DIRTY, &ii->i_state) ||
 864                         test_bit(NILFS_I_BUSY, &ii->i_state);
 865                 spin_unlock(&nilfs->ns_inode_lock);
 866         }
 867         return ret;
 868 }
 869
 870 int nilfs_set_file_dirty(struct inode *inode, unsigned nr_dirty)
 871 {
 872         struct nilfs_inode_info *ii = NILFS_I(inode);
 873         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 874
 875         atomic_add(nr_dirty, &nilfs->ns_ndirtyblks);
 876
 877         if (test_and_set_bit(NILFS_I_DIRTY, &ii->i_state))
 878                 return 0;
 879
 880         spin_lock(&nilfs->ns_inode_lock);
 881         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
 882             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
 883                 /* Because this routine may race with nilfs_dispose_list(),
 884                    we have to check NILFS_I_QUEUED here, too. */
 885                 if (list_empty(&ii->i_dirty) && igrab(inode) == NULL) {
 886                         /* This will happen when somebody is freeing
 887                            this inode. */
 888                         nilfs_warning(inode->i_sb, __func__,
 889                                       "cannot get inode (ino=%lu)\n",
 890                                       inode->i_ino);
 891                         spin_unlock(&nilfs->ns_inode_lock);
 892                         return -EINVAL; /* NILFS_I_DIRTY may remain for
 893                                            freeing inode */
 894                 }
 895                 list_move_tail(&ii->i_dirty, &nilfs->ns_dirty_files);
 896                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 897         }
 898         spin_unlock(&nilfs->ns_inode_lock);
 899         return 0;
 900 }
 901
 902 int nilfs_mark_inode_dirty(struct inode *inode)
 903 {
 904         struct buffer_head *ibh;
 905         int err;
 906
 907         err = nilfs_load_inode_block(inode, &ibh);
 908         if (unlikely(err)) {
 909                 nilfs_warning(inode->i_sb, __func__,
 910                               "failed to reget inode block.\n");
 911                 return err;
 912         }
 913         nilfs_update_inode(inode, ibh);
 914         mark_buffer_dirty(ibh);
 915         nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
 916         brelse(ibh);
 917         return 0;
 918 }
 919
 920 /**
 921  * nilfs_dirty_inode - reflect changes on given inode to an inode block.
 922  * @inode: inode of the file to be registered.
 923  *
 924  * nilfs_dirty_inode() loads a inode block containing the specified
 925  * @inode and copies data from a nilfs_inode to a corresponding inode
 926  * entry in the inode block. This operation is excluded from the segment
 927  * construction. This function can be called both as a single operation
 928  * and as a part of indivisible file operations.
 929  */
 930 void nilfs_dirty_inode(struct inode *inode, int flags)
 931 {
 932         struct nilfs_transaction_info ti;
 933         struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
 934
 935         if (is_bad_inode(inode)) {
 936                 nilfs_warning(inode->i_sb, __func__,
 937                               "tried to mark bad_inode dirty. ignored.\n");
 938                 dump_stack();
 939                 return;
 940         }
 941         if (mdi) {
 942                 nilfs_mdt_mark_dirty(inode);
 943                 return;
 944         }
 945         nilfs_transaction_begin(inode->i_sb, &ti, 0);
 946         nilfs_mark_inode_dirty(inode);
 947         nilfs_transaction_commit(inode->i_sb); /* never fails */
 948 }
 949
 950 int nilfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 951                  __u64 start, __u64 len)
 952 {
 953         struct the_nilfs *nilfs = inode->i_sb->s_fs_info;
 954         __u64 logical = 0, phys = 0, size = 0;
 955         __u32 flags = 0;
 956         loff_t isize;
 957         sector_t blkoff, end_blkoff;
 958         sector_t delalloc_blkoff;
 959         unsigned long delalloc_blklen;
 960         unsigned int blkbits = inode->i_blkbits;
 961         int ret, n;
 962
 963         ret = fiemap_check_flags(fieinfo, FIEMAP_FLAG_SYNC);
 964         if (ret)
 965                 return ret;
 966
 967         mutex_lock(&inode->i_mutex);
 968
 969         isize = i_size_read(inode);
 970
 971         blkoff = start >> blkbits;
 972         end_blkoff = (start + len - 1) >> blkbits;
 973
 974         delalloc_blklen = nilfs_find_uncommitted_extent(inode, blkoff,
 975                                                         &delalloc_blkoff);
 976
 977         do {
 978                 __u64 blkphy;
 979                 unsigned int maxblocks;
 980
 981                 if (delalloc_blklen && blkoff == delalloc_blkoff) {
 982                         if (size) {
 983                                 /* End of the current extent */
 984                                 ret = fiemap_fill_next_extent(
 985                                         fieinfo, logical, phys, size, flags);
 986                                 if (ret)
 987                                         break;
 988                         }
 989                         if (blkoff > end_blkoff)
 990                                 break;
 991
 992                         flags = FIEMAP_EXTENT_MERGED | FIEMAP_EXTENT_DELALLOC;
 993                         logical = blkoff << blkbits;
 994                         phys = 0;
 995                         size = delalloc_blklen << blkbits;
 996
 997                         blkoff = delalloc_blkoff + delalloc_blklen;
 998                         delalloc_blklen = nilfs_find_uncommitted_extent(
 999                                 inode, blkoff, &delalloc_blkoff);
1000                         continue;
1001                 }
1002
1003                 /*
1004                  * Limit the number of blocks that we look up so as
1005                  * not to get into the next delayed allocation extent.
1006                  */
1007                 maxblocks = INT_MAX;
1008                 if (delalloc_blklen)
1009                         maxblocks = min_t(sector_t, delalloc_blkoff - blkoff,
1010                                           maxblocks);
1011                 blkphy = 0;
1012
1013                 down_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1014                 n = nilfs_bmap_lookup_contig(
1015                         NILFS_I(inode)->i_bmap, blkoff, &blkphy, maxblocks);
1016                 up_read(&NILFS_MDT(nilfs->ns_dat)->mi_sem);
1017
1018                 if (n < 0) {
1019                         int past_eof;
1020
1021                         if (unlikely(n != -ENOENT))
1022                                 break; /* error */
1023
1024                         /* HOLE */
1025                         blkoff++;
1026                         past_eof = ((blkoff << blkbits) >= isize);
1027
1028                         if (size) {
1029                                 /* End of the current extent */
1030
1031                                 if (past_eof)
1032                                         flags |= FIEMAP_EXTENT_LAST;
1033
1034                                 ret = fiemap_fill_next_extent(
1035                                         fieinfo, logical, phys, size, flags);
1036                                 if (ret)
1037                                         break;
1038                                 size = 0;
1039                         }
1040                         if (blkoff > end_blkoff || past_eof)
1041                                 break;
1042                 } else {
1043                         if (size) {
1044                                 if (phys && blkphy << blkbits == phys + size) {
1045                                         /* The current extent goes on */
1046                                         size += n << blkbits;
1047                                 } else {
1048                                         /* Terminate the current extent */
1049                                         ret = fiemap_fill_next_extent(
1050                                                 fieinfo, logical, phys, size,
1051                                                 flags);
1052                                         if (ret || blkoff > end_blkoff)
1053                                                 break;
1054
1055                                         /* Start another extent */
1056                                         flags = FIEMAP_EXTENT_MERGED;
1057                                         logical = blkoff << blkbits;
1058                                         phys = blkphy << blkbits;
1059                                         size = n << blkbits;
1060                                 }
1061                         } else {
1062                                 /* Start a new extent */
1063                                 flags = FIEMAP_EXTENT_MERGED;
1064                                 logical = blkoff << blkbits;
1065                                 phys = blkphy << blkbits;
1066                                 size = n << blkbits;
1067                         }
1068                         blkoff += n;
1069                 }
1070                 cond_resched();
1071         } while (true);
1072
1073         /* If ret is 1 then we just hit the end of the extent array */
1074         if (ret == 1)
1075                 ret = 0;
1076
1077         mutex_unlock(&inode->i_mutex);
1078         return ret;
1079 }