Pileus Git - ~andy/linux/blob - fs/ext4/move_extent.c

   1 /*
   2  * Copyright (c) 2008,2009 NEC Software Tohoku, Ltd.
   3  * Written by Takashi Sato <t-sato@yk.jp.nec.com>
   4  *            Akira Fujita <a-fujita@rs.jp.nec.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of version 2.1 of the GNU Lesser General Public License
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  */
  15
  16 #include <linux/fs.h>
  17 #include <linux/quotaops.h>
  18 #include <linux/slab.h>
  19 #include "ext4_jbd2.h"
  20 #include "ext4.h"
  21
  22 /**
  23  * get_ext_path - Find an extent path for designated logical block number.
  24  *
  25  * @inode:      an inode which is searched
  26  * @lblock:     logical block number to find an extent path
  27  * @path:       pointer to an extent path pointer (for output)
  28  *
  29  * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value
  30  * on failure.
  31  */
  32 static inline int
  33 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
  34                 struct ext4_ext_path **path)
  35 {
  36         int ret = 0;
  37
  38         *path = ext4_ext_find_extent(inode, lblock, *path);
  39         if (IS_ERR(*path)) {
  40                 ret = PTR_ERR(*path);
  41                 *path = NULL;
  42         } else if ((*path)[ext_depth(inode)].p_ext == NULL)
  43                 ret = -ENODATA;
  44
  45         return ret;
  46 }
  47
  48 /**
  49  * copy_extent_status - Copy the extent's initialization status
  50  *
  51  * @src:        an extent for getting initialize status
  52  * @dest:       an extent to be set the status
  53  */
  54 static void
  55 copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
  56 {
  57         if (ext4_ext_is_uninitialized(src))
  58                 ext4_ext_mark_uninitialized(dest);
  59         else
  60                 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
  61 }
  62
  63 /**
  64  * mext_next_extent - Search for the next extent and set it to "extent"
  65  *
  66  * @inode:      inode which is searched
  67  * @path:       this will obtain data for the next extent
  68  * @extent:     pointer to the next extent we have just gotten
  69  *
  70  * Search the next extent in the array of ext4_ext_path structure (@path)
  71  * and set it to ext4_extent structure (@extent). In addition, the member of
  72  * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
  73  * ext4_ext_path structure refers to the last extent, or a negative error
  74  * value on failure.
  75  */
  76 static int
  77 mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
  78                       struct ext4_extent **extent)
  79 {
  80         struct ext4_extent_header *eh;
  81         int ppos, leaf_ppos = path->p_depth;
  82
  83         ppos = leaf_ppos;
  84         if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
  85                 /* leaf block */
  86                 *extent = ++path[ppos].p_ext;
  87                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
  88                 return 0;
  89         }
  90
  91         while (--ppos >= 0) {
  92                 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
  93                     path[ppos].p_idx) {
  94                         int cur_ppos = ppos;
  95
  96                         /* index block */
  97                         path[ppos].p_idx++;
  98                         path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
  99                         if (path[ppos+1].p_bh)
 100                                 brelse(path[ppos+1].p_bh);
 101                         path[ppos+1].p_bh =
 102                                 sb_bread(inode->i_sb, path[ppos].p_block);
 103                         if (!path[ppos+1].p_bh)
 104                                 return -EIO;
 105                         path[ppos+1].p_hdr =
 106                                 ext_block_hdr(path[ppos+1].p_bh);
 107
 108                         /* Halfway index block */
 109                         while (++cur_ppos < leaf_ppos) {
 110                                 path[cur_ppos].p_idx =
 111                                         EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
 112                                 path[cur_ppos].p_block =
 113                                         ext4_idx_pblock(path[cur_ppos].p_idx);
 114                                 if (path[cur_ppos+1].p_bh)
 115                                         brelse(path[cur_ppos+1].p_bh);
 116                                 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
 117                                         path[cur_ppos].p_block);
 118                                 if (!path[cur_ppos+1].p_bh)
 119                                         return -EIO;
 120                                 path[cur_ppos+1].p_hdr =
 121                                         ext_block_hdr(path[cur_ppos+1].p_bh);
 122                         }
 123
 124                         path[leaf_ppos].p_ext = *extent = NULL;
 125
 126                         eh = path[leaf_ppos].p_hdr;
 127                         if (le16_to_cpu(eh->eh_entries) == 0)
 128                                 /* empty leaf is found */
 129                                 return -ENODATA;
 130
 131                         /* leaf block */
 132                         path[leaf_ppos].p_ext = *extent =
 133                                 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
 134                         path[leaf_ppos].p_block =
 135                                         ext4_ext_pblock(path[leaf_ppos].p_ext);
 136                         return 0;
 137                 }
 138         }
 139         /* We found the last extent */
 140         return 1;
 141 }
 142
 143 /**
 144  * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
 145  *
 146  * Acquire write lock of i_data_sem of the two inodes
 147  */
 148 static void
 149 double_down_write_data_sem(struct inode *first, struct inode *second)
 150 {
 151         if (first < second) {
 152                 down_write(&EXT4_I(first)->i_data_sem);
 153                 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
 154         } else {
 155                 down_write(&EXT4_I(second)->i_data_sem);
 156                 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING);
 157
 158         }
 159 }
 160
 161 /**
 162  * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
 163  *
 164  * @orig_inode:         original inode structure to be released its lock first
 165  * @donor_inode:        donor inode structure to be released its lock second
 166  * Release write lock of i_data_sem of two inodes (orig and donor).
 167  */
 168 static void
 169 double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
 170 {
 171         up_write(&EXT4_I(orig_inode)->i_data_sem);
 172         up_write(&EXT4_I(donor_inode)->i_data_sem);
 173 }
 174
 175 /**
 176  * mext_insert_across_blocks - Insert extents across leaf block
 177  *
 178  * @handle:             journal handle
 179  * @orig_inode:         original inode
 180  * @o_start:            first original extent to be changed
 181  * @o_end:              last original extent to be changed
 182  * @start_ext:          first new extent to be inserted
 183  * @new_ext:            middle of new extent to be inserted
 184  * @end_ext:            last new extent to be inserted
 185  *
 186  * Allocate a new leaf block and insert extents into it. Return 0 on success,
 187  * or a negative error value on failure.
 188  */
 189 static int
 190 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
 191                 struct ext4_extent *o_start, struct ext4_extent *o_end,
 192                 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
 193                 struct ext4_extent *end_ext)
 194 {
 195         struct ext4_ext_path *orig_path = NULL;
 196         ext4_lblk_t eblock = 0;
 197         int new_flag = 0;
 198         int end_flag = 0;
 199         int err = 0;
 200
 201         if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
 202                 if (o_start == o_end) {
 203
 204                         /*       start_ext   new_ext    end_ext
 205                          * donor |---------|-----------|--------|
 206                          * orig  |------------------------------|
 207                          */
 208                         end_flag = 1;
 209                 } else {
 210
 211                         /*       start_ext   new_ext   end_ext
 212                          * donor |---------|----------|---------|
 213                          * orig  |---------------|--------------|
 214                          */
 215                         o_end->ee_block = end_ext->ee_block;
 216                         o_end->ee_len = end_ext->ee_len;
 217                         ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
 218                 }
 219
 220                 o_start->ee_len = start_ext->ee_len;
 221                 eblock = le32_to_cpu(start_ext->ee_block);
 222                 new_flag = 1;
 223
 224         } else if (start_ext->ee_len && new_ext->ee_len &&
 225                    !end_ext->ee_len && o_start == o_end) {
 226
 227                 /*       start_ext      new_ext
 228                  * donor |--------------|---------------|
 229                  * orig  |------------------------------|
 230                  */
 231                 o_start->ee_len = start_ext->ee_len;
 232                 eblock = le32_to_cpu(start_ext->ee_block);
 233                 new_flag = 1;
 234
 235         } else if (!start_ext->ee_len && new_ext->ee_len &&
 236                    end_ext->ee_len && o_start == o_end) {
 237
 238                 /*        new_ext       end_ext
 239                  * donor |--------------|---------------|
 240                  * orig  |------------------------------|
 241                  */
 242                 o_end->ee_block = end_ext->ee_block;
 243                 o_end->ee_len = end_ext->ee_len;
 244                 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
 245
 246                 /*
 247                  * Set 0 to the extent block if new_ext was
 248                  * the first block.
 249                  */
 250                 if (new_ext->ee_block)
 251                         eblock = le32_to_cpu(new_ext->ee_block);
 252
 253                 new_flag = 1;
 254         } else {
 255                 ext4_debug("ext4 move extent: Unexpected insert case\n");
 256                 return -EIO;
 257         }
 258
 259         if (new_flag) {
 260                 err = get_ext_path(orig_inode, eblock, &orig_path);
 261                 if (err)
 262                         goto out;
 263
 264                 if (ext4_ext_insert_extent(handle, orig_inode,
 265                                         orig_path, new_ext, 0))
 266                         goto out;
 267         }
 268
 269         if (end_flag) {
 270                 err = get_ext_path(orig_inode,
 271                                 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
 272                 if (err)
 273                         goto out;
 274
 275                 if (ext4_ext_insert_extent(handle, orig_inode,
 276                                            orig_path, end_ext, 0))
 277                         goto out;
 278         }
 279 out:
 280         if (orig_path) {
 281                 ext4_ext_drop_refs(orig_path);
 282                 kfree(orig_path);
 283         }
 284
 285         return err;
 286
 287 }
 288
 289 /**
 290  * mext_insert_inside_block - Insert new extent to the extent block
 291  *
 292  * @o_start:            first original extent to be moved
 293  * @o_end:              last original extent to be moved
 294  * @start_ext:          first new extent to be inserted
 295  * @new_ext:            middle of new extent to be inserted
 296  * @end_ext:            last new extent to be inserted
 297  * @eh:                 extent header of target leaf block
 298  * @range_to_move:      used to decide how to insert extent
 299  *
 300  * Insert extents into the leaf block. The extent (@o_start) is overwritten
 301  * by inserted extents.
 302  */
 303 static void
 304 mext_insert_inside_block(struct ext4_extent *o_start,
 305                               struct ext4_extent *o_end,
 306                               struct ext4_extent *start_ext,
 307                               struct ext4_extent *new_ext,
 308                               struct ext4_extent *end_ext,
 309                               struct ext4_extent_header *eh,
 310                               int range_to_move)
 311 {
 312         int i = 0;
 313         unsigned long len;
 314
 315         /* Move the existing extents */
 316         if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
 317                 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
 318                         (unsigned long)(o_end + 1);
 319                 memmove(o_end + 1 + range_to_move, o_end + 1, len);
 320         }
 321
 322         /* Insert start entry */
 323         if (start_ext->ee_len)
 324                 o_start[i++].ee_len = start_ext->ee_len;
 325
 326         /* Insert new entry */
 327         if (new_ext->ee_len) {
 328                 o_start[i] = *new_ext;
 329                 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
 330         }
 331
 332         /* Insert end entry */
 333         if (end_ext->ee_len)
 334                 o_start[i] = *end_ext;
 335
 336         /* Increment the total entries counter on the extent block */
 337         le16_add_cpu(&eh->eh_entries, range_to_move);
 338 }
 339
 340 /**
 341  * mext_insert_extents - Insert new extent
 342  *
 343  * @handle:     journal handle
 344  * @orig_inode: original inode
 345  * @orig_path:  path indicates first extent to be changed
 346  * @o_start:    first original extent to be changed
 347  * @o_end:      last original extent to be changed
 348  * @start_ext:  first new extent to be inserted
 349  * @new_ext:    middle of new extent to be inserted
 350  * @end_ext:    last new extent to be inserted
 351  *
 352  * Call the function to insert extents. If we cannot add more extents into
 353  * the leaf block, we call mext_insert_across_blocks() to create a
 354  * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
 355  * on success, or a negative error value on failure.
 356  */
 357 static int
 358 mext_insert_extents(handle_t *handle, struct inode *orig_inode,
 359                          struct ext4_ext_path *orig_path,
 360                          struct ext4_extent *o_start,
 361                          struct ext4_extent *o_end,
 362                          struct ext4_extent *start_ext,
 363                          struct ext4_extent *new_ext,
 364                          struct ext4_extent *end_ext)
 365 {
 366         struct  ext4_extent_header *eh;
 367         unsigned long need_slots, slots_range;
 368         int     range_to_move, depth, ret;
 369
 370         /*
 371          * The extents need to be inserted
 372          * start_extent + new_extent + end_extent.
 373          */
 374         need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
 375                 (new_ext->ee_len ? 1 : 0);
 376
 377         /* The number of slots between start and end */
 378         slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
 379                 / sizeof(struct ext4_extent);
 380
 381         /* Range to move the end of extent */
 382         range_to_move = need_slots - slots_range;
 383         depth = orig_path->p_depth;
 384         orig_path += depth;
 385         eh = orig_path->p_hdr;
 386
 387         if (depth) {
 388                 /* Register to journal */
 389                 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
 390                 if (ret)
 391                         return ret;
 392         }
 393
 394         /* Expansion */
 395         if (range_to_move > 0 &&
 396                 (range_to_move > le16_to_cpu(eh->eh_max)
 397                         - le16_to_cpu(eh->eh_entries))) {
 398
 399                 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
 400                                         o_end, start_ext, new_ext, end_ext);
 401                 if (ret < 0)
 402                         return ret;
 403         } else
 404                 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
 405                                                 end_ext, eh, range_to_move);
 406
 407         if (depth) {
 408                 ret = ext4_handle_dirty_metadata(handle, orig_inode,
 409                                                  orig_path->p_bh);
 410                 if (ret)
 411                         return ret;
 412         } else {
 413                 ret = ext4_mark_inode_dirty(handle, orig_inode);
 414                 if (ret < 0)
 415                         return ret;
 416         }
 417
 418         return 0;
 419 }
 420
 421 /**
 422  * mext_leaf_block - Move one leaf extent block into the inode.
 423  *
 424  * @handle:             journal handle
 425  * @orig_inode:         original inode
 426  * @orig_path:          path indicates first extent to be changed
 427  * @dext:               donor extent
 428  * @from:               start offset on the target file
 429  *
 430  * In order to insert extents into the leaf block, we must divide the extent
 431  * in the leaf block into three extents. The one is located to be inserted
 432  * extents, and the others are located around it.
 433  *
 434  * Therefore, this function creates structures to save extents of the leaf
 435  * block, and inserts extents by calling mext_insert_extents() with
 436  * created extents. Return 0 on success, or a negative error value on failure.
 437  */
 438 static int
 439 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
 440                      struct ext4_ext_path *orig_path, struct ext4_extent *dext,
 441                      ext4_lblk_t *from)
 442 {
 443         struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
 444         struct ext4_extent new_ext, start_ext, end_ext;
 445         ext4_lblk_t new_ext_end;
 446         int oext_alen, new_ext_alen, end_ext_alen;
 447         int depth = ext_depth(orig_inode);
 448         int ret;
 449
 450         start_ext.ee_block = end_ext.ee_block = 0;
 451         o_start = o_end = oext = orig_path[depth].p_ext;
 452         oext_alen = ext4_ext_get_actual_len(oext);
 453         start_ext.ee_len = end_ext.ee_len = 0;
 454
 455         new_ext.ee_block = cpu_to_le32(*from);
 456         ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
 457         new_ext.ee_len = dext->ee_len;
 458         new_ext_alen = ext4_ext_get_actual_len(&new_ext);
 459         new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
 460
 461         /*
 462          * Case: original extent is first
 463          * oext      |--------|
 464          * new_ext      |--|
 465          * start_ext |--|
 466          */
 467         if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
 468                 le32_to_cpu(new_ext.ee_block) <
 469                 le32_to_cpu(oext->ee_block) + oext_alen) {
 470                 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
 471                                                le32_to_cpu(oext->ee_block));
 472                 start_ext.ee_block = oext->ee_block;
 473                 copy_extent_status(oext, &start_ext);
 474         } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
 475                 prev_ext = oext - 1;
 476                 /*
 477                  * We can merge new_ext into previous extent,
 478                  * if these are contiguous and same extent type.
 479                  */
 480                 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
 481                                                &new_ext)) {
 482                         o_start = prev_ext;
 483                         start_ext.ee_len = cpu_to_le16(
 484                                 ext4_ext_get_actual_len(prev_ext) +
 485                                 new_ext_alen);
 486                         start_ext.ee_block = oext->ee_block;
 487                         copy_extent_status(prev_ext, &start_ext);
 488                         new_ext.ee_len = 0;
 489                 }
 490         }
 491
 492         /*
 493          * Case: new_ext_end must be less than oext
 494          * oext      |-----------|
 495          * new_ext       |-------|
 496          */
 497         if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
 498                 EXT4_ERROR_INODE(orig_inode,
 499                         "new_ext_end(%u) should be less than or equal to "
 500                         "oext->ee_block(%u) + oext_alen(%d) - 1",
 501                         new_ext_end, le32_to_cpu(oext->ee_block),
 502                         oext_alen);
 503                 ret = -EIO;
 504                 goto out;
 505         }
 506
 507         /*
 508          * Case: new_ext is smaller than original extent
 509          * oext    |---------------|
 510          * new_ext |-----------|
 511          * end_ext             |---|
 512          */
 513         if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
 514                 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
 515                 end_ext.ee_len =
 516                         cpu_to_le16(le32_to_cpu(oext->ee_block) +
 517                         oext_alen - 1 - new_ext_end);
 518                 copy_extent_status(oext, &end_ext);
 519                 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
 520                 ext4_ext_store_pblock(&end_ext,
 521                         (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
 522                 end_ext.ee_block =
 523                         cpu_to_le32(le32_to_cpu(o_end->ee_block) +
 524                         oext_alen - end_ext_alen);
 525         }
 526
 527         ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
 528                                 o_end, &start_ext, &new_ext, &end_ext);
 529 out:
 530         return ret;
 531 }
 532
 533 /**
 534  * mext_calc_swap_extents - Calculate extents for extent swapping.
 535  *
 536  * @tmp_dext:           the extent that will belong to the original inode
 537  * @tmp_oext:           the extent that will belong to the donor inode
 538  * @orig_off:           block offset of original inode
 539  * @donor_off:          block offset of donor inode
 540  * @max_count:          the maximum length of extents
 541  *
 542  * Return 0 on success, or a negative error value on failure.
 543  */
 544 static int
 545 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
 546                               struct ext4_extent *tmp_oext,
 547                               ext4_lblk_t orig_off, ext4_lblk_t donor_off,
 548                               ext4_lblk_t max_count)
 549 {
 550         ext4_lblk_t diff, orig_diff;
 551         struct ext4_extent dext_old, oext_old;
 552
 553         BUG_ON(orig_off != donor_off);
 554
 555         /* original and donor extents have to cover the same block offset */
 556         if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
 557             le32_to_cpu(tmp_oext->ee_block) +
 558                         ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
 559                 return -ENODATA;
 560
 561         if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
 562             le32_to_cpu(tmp_dext->ee_block) +
 563                         ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
 564                 return -ENODATA;
 565
 566         dext_old = *tmp_dext;
 567         oext_old = *tmp_oext;
 568
 569         /* When tmp_dext is too large, pick up the target range. */
 570         diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
 571
 572         ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
 573         le32_add_cpu(&tmp_dext->ee_block, diff);
 574         le16_add_cpu(&tmp_dext->ee_len, -diff);
 575
 576         if (max_count < ext4_ext_get_actual_len(tmp_dext))
 577                 tmp_dext->ee_len = cpu_to_le16(max_count);
 578
 579         orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
 580         ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
 581
 582         /* Adjust extent length if donor extent is larger than orig */
 583         if (ext4_ext_get_actual_len(tmp_dext) >
 584             ext4_ext_get_actual_len(tmp_oext) - orig_diff)
 585                 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
 586                                                 orig_diff);
 587
 588         tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
 589
 590         copy_extent_status(&oext_old, tmp_dext);
 591         copy_extent_status(&dext_old, tmp_oext);
 592
 593         return 0;
 594 }
 595
 596 /**
 597  * mext_check_coverage - Check that all extents in range has the same type
 598  *
 599  * @inode:              inode in question
 600  * @from:               block offset of inode
 601  * @count:              block count to be checked
 602  * @uninit:             extents expected to be uninitialized
 603  * @err:                pointer to save error value
 604  *
 605  * Return 1 if all extents in range has expected type, and zero otherwise.
 606  */
 607 static int
 608 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
 609                           int uninit, int *err)
 610 {
 611         struct ext4_ext_path *path = NULL;
 612         struct ext4_extent *ext;
 613         ext4_lblk_t last = from + count;
 614         while (from < last) {
 615                 *err = get_ext_path(inode, from, &path);
 616                 if (*err)
 617                         return 0;
 618                 ext = path[ext_depth(inode)].p_ext;
 619                 if (!ext) {
 620                         ext4_ext_drop_refs(path);
 621                         return 0;
 622                 }
 623                 if (uninit != ext4_ext_is_uninitialized(ext)) {
 624                         ext4_ext_drop_refs(path);
 625                         return 0;
 626                 }
 627                 from += ext4_ext_get_actual_len(ext);
 628                 ext4_ext_drop_refs(path);
 629         }
 630         return 1;
 631 }
 632
 633 /**
 634  * mext_replace_branches - Replace original extents with new extents
 635  *
 636  * @handle:             journal handle
 637  * @orig_inode:         original inode
 638  * @donor_inode:        donor inode
 639  * @from:               block offset of orig_inode
 640  * @count:              block count to be replaced
 641  * @err:                pointer to save return value
 642  *
 643  * Replace original inode extents and donor inode extents page by page.
 644  * We implement this replacement in the following three steps:
 645  * 1. Save the block information of original and donor inodes into
 646  *    dummy extents.
 647  * 2. Change the block information of original inode to point at the
 648  *    donor inode blocks.
 649  * 3. Change the block information of donor inode to point at the saved
 650  *    original inode blocks in the dummy extents.
 651  *
 652  * Return replaced block count.
 653  */
 654 static int
 655 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
 656                            struct inode *donor_inode, ext4_lblk_t from,
 657                            ext4_lblk_t count, int *err)
 658 {
 659         struct ext4_ext_path *orig_path = NULL;
 660         struct ext4_ext_path *donor_path = NULL;
 661         struct ext4_extent *oext, *dext;
 662         struct ext4_extent tmp_dext, tmp_oext;
 663         ext4_lblk_t orig_off = from, donor_off = from;
 664         int depth;
 665         int replaced_count = 0;
 666         int dext_alen;
 667
 668         /* Get the original extent for the block "orig_off" */
 669         *err = get_ext_path(orig_inode, orig_off, &orig_path);
 670         if (*err)
 671                 goto out;
 672
 673         /* Get the donor extent for the head */
 674         *err = get_ext_path(donor_inode, donor_off, &donor_path);
 675         if (*err)
 676                 goto out;
 677         depth = ext_depth(orig_inode);
 678         oext = orig_path[depth].p_ext;
 679         tmp_oext = *oext;
 680
 681         depth = ext_depth(donor_inode);
 682         dext = donor_path[depth].p_ext;
 683         tmp_dext = *dext;
 684
 685         *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 686                                       donor_off, count);
 687         if (*err)
 688                 goto out;
 689
 690         /* Loop for the donor extents */
 691         while (1) {
 692                 /* The extent for donor must be found. */
 693                 if (!dext) {
 694                         EXT4_ERROR_INODE(donor_inode,
 695                                    "The extent for donor must be found");
 696                         *err = -EIO;
 697                         goto out;
 698                 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
 699                         EXT4_ERROR_INODE(donor_inode,
 700                                 "Donor offset(%u) and the first block of donor "
 701                                 "extent(%u) should be equal",
 702                                 donor_off,
 703                                 le32_to_cpu(tmp_dext.ee_block));
 704                         *err = -EIO;
 705                         goto out;
 706                 }
 707
 708                 /* Set donor extent to orig extent */
 709                 *err = mext_leaf_block(handle, orig_inode,
 710                                            orig_path, &tmp_dext, &orig_off);
 711                 if (*err)
 712                         goto out;
 713
 714                 /* Set orig extent to donor extent */
 715                 *err = mext_leaf_block(handle, donor_inode,
 716                                            donor_path, &tmp_oext, &donor_off);
 717                 if (*err)
 718                         goto out;
 719
 720                 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
 721                 replaced_count += dext_alen;
 722                 donor_off += dext_alen;
 723                 orig_off += dext_alen;
 724
 725                 /* Already moved the expected blocks */
 726                 if (replaced_count >= count)
 727                         break;
 728
 729                 if (orig_path)
 730                         ext4_ext_drop_refs(orig_path);
 731                 *err = get_ext_path(orig_inode, orig_off, &orig_path);
 732                 if (*err)
 733                         goto out;
 734                 depth = ext_depth(orig_inode);
 735                 oext = orig_path[depth].p_ext;
 736                 tmp_oext = *oext;
 737
 738                 if (donor_path)
 739                         ext4_ext_drop_refs(donor_path);
 740                 *err = get_ext_path(donor_inode, donor_off, &donor_path);
 741                 if (*err)
 742                         goto out;
 743                 depth = ext_depth(donor_inode);
 744                 dext = donor_path[depth].p_ext;
 745                 tmp_dext = *dext;
 746
 747                 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 748                                            donor_off, count - replaced_count);
 749                 if (*err)
 750                         goto out;
 751         }
 752
 753 out:
 754         if (orig_path) {
 755                 ext4_ext_drop_refs(orig_path);
 756                 kfree(orig_path);
 757         }
 758         if (donor_path) {
 759                 ext4_ext_drop_refs(donor_path);
 760                 kfree(donor_path);
 761         }
 762
 763         ext4_ext_invalidate_cache(orig_inode);
 764         ext4_ext_invalidate_cache(donor_inode);
 765
 766         return replaced_count;
 767 }
 768
 769 /**
 770  * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
 771  *
 772  * @inode1:     the inode structure
 773  * @inode2:     the inode structure
 774  * @index:      page index
 775  * @page:       result page vector
 776  *
 777  * Grab two locked pages for inode's by inode order
 778  */
 779 static int
 780 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
 781                       pgoff_t index, struct page *page[2])
 782 {
 783         struct address_space *mapping[2];
 784         unsigned fl = AOP_FLAG_NOFS;
 785
 786         BUG_ON(!inode1 || !inode2);
 787         if (inode1 < inode2) {
 788                 mapping[0] = inode1->i_mapping;
 789                 mapping[1] = inode2->i_mapping;
 790         } else {
 791                 mapping[0] = inode2->i_mapping;
 792                 mapping[1] = inode1->i_mapping;
 793         }
 794
 795         page[0] = grab_cache_page_write_begin(mapping[0], index, fl);
 796         if (!page[0])
 797                 return -ENOMEM;
 798
 799         page[1] = grab_cache_page_write_begin(mapping[1], index, fl);
 800         if (!page[1]) {
 801                 unlock_page(page[0]);
 802                 page_cache_release(page[0]);
 803                 return -ENOMEM;
 804         }
 805
 806         if (inode1 > inode2) {
 807                 struct page *tmp;
 808                 tmp = page[0];
 809                 page[0] = page[1];
 810                 page[1] = tmp;
 811         }
 812         return 0;
 813 }
 814
 815 /* Force page buffers uptodate w/o dropping page's lock */
 816 static int
 817 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
 818 {
 819         struct inode *inode = page->mapping->host;
 820         sector_t block;
 821         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 822         unsigned int blocksize, block_start, block_end;
 823         int i, err,  nr = 0, partial = 0;
 824         BUG_ON(!PageLocked(page));
 825         BUG_ON(PageWriteback(page));
 826
 827         if (PageUptodate(page))
 828                 return 0;
 829
 830         blocksize = 1 << inode->i_blkbits;
 831         if (!page_has_buffers(page))
 832                 create_empty_buffers(page, blocksize, 0);
 833
 834         head = page_buffers(page);
 835         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 836         for (bh = head, block_start = 0; bh != head || !block_start;
 837              block++, block_start = block_end, bh = bh->b_this_page) {
 838                 block_end = block_start + blocksize;
 839                 if (block_end <= from || block_start >= to) {
 840                         if (!buffer_uptodate(bh))
 841                                 partial = 1;
 842                         continue;
 843                 }
 844                 if (buffer_uptodate(bh))
 845                         continue;
 846                 if (!buffer_mapped(bh)) {
 847                         int err = 0;
 848                         err = ext4_get_block(inode, block, bh, 0);
 849                         if (err) {
 850                                 SetPageError(page);
 851                                 return err;
 852                         }
 853                         if (!buffer_mapped(bh)) {
 854                                 zero_user(page, block_start, blocksize);
 855                                 if (!err)
 856                                         set_buffer_uptodate(bh);
 857                                 continue;
 858                         }
 859                 }
 860                 BUG_ON(nr >= MAX_BUF_PER_PAGE);
 861                 arr[nr++] = bh;
 862         }
 863         /* No io required */
 864         if (!nr)
 865                 goto out;
 866
 867         for (i = 0; i < nr; i++) {
 868                 bh = arr[i];
 869                 if (!bh_uptodate_or_lock(bh)) {
 870                         err = bh_submit_read(bh);
 871                         if (err)
 872                                 return err;
 873                 }
 874         }
 875 out:
 876         if (!partial)
 877                 SetPageUptodate(page);
 878         return 0;
 879 }
 880
 881 /**
 882  * move_extent_per_page - Move extent data per page
 883  *
 884  * @o_filp:                     file structure of original file
 885  * @donor_inode:                donor inode
 886  * @orig_page_offset:           page index on original file
 887  * @data_offset_in_page:        block index where data swapping starts
 888  * @block_len_in_page:          the number of blocks to be swapped
 889  * @uninit:                     orig extent is uninitialized or not
 890  * @err:                        pointer to save return value
 891  *
 892  * Save the data in original inode blocks and replace original inode extents
 893  * with donor inode extents by calling mext_replace_branches().
 894  * Finally, write out the saved data in new original inode blocks. Return
 895  * replaced block count.
 896  */
 897 static int
 898 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
 899                   pgoff_t orig_page_offset, int data_offset_in_page,
 900                   int block_len_in_page, int uninit, int *err)
 901 {
 902         struct inode *orig_inode = o_filp->f_dentry->d_inode;
 903         struct page *pagep[2] = {NULL, NULL};
 904         handle_t *handle;
 905         ext4_lblk_t orig_blk_offset;
 906         long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
 907         unsigned long blocksize = orig_inode->i_sb->s_blocksize;
 908         unsigned int w_flags = 0;
 909         unsigned int tmp_data_size, data_size, replaced_size;
 910         int err2, jblocks, retries = 0;
 911         int replaced_count = 0;
 912         int from = data_offset_in_page << orig_inode->i_blkbits;
 913         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
 914
 915         /*
 916          * It needs twice the amount of ordinary journal buffers because
 917          * inode and donor_inode may change each different metadata blocks.
 918          */
 919 again:
 920         *err = 0;
 921         jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
 922         handle = ext4_journal_start(orig_inode, jblocks);
 923         if (IS_ERR(handle)) {
 924                 *err = PTR_ERR(handle);
 925                 return 0;
 926         }
 927
 928         if (segment_eq(get_fs(), KERNEL_DS))
 929                 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
 930
 931         orig_blk_offset = orig_page_offset * blocks_per_page +
 932                 data_offset_in_page;
 933
 934         offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
 935
 936         /* Calculate data_size */
 937         if ((orig_blk_offset + block_len_in_page - 1) ==
 938             ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
 939                 /* Replace the last block */
 940                 tmp_data_size = orig_inode->i_size & (blocksize - 1);
 941                 /*
 942                  * If data_size equal zero, it shows data_size is multiples of
 943                  * blocksize. So we set appropriate value.
 944                  */
 945                 if (tmp_data_size == 0)
 946                         tmp_data_size = blocksize;
 947
 948                 data_size = tmp_data_size +
 949                         ((block_len_in_page - 1) << orig_inode->i_blkbits);
 950         } else
 951                 data_size = block_len_in_page << orig_inode->i_blkbits;
 952
 953         replaced_size = data_size;
 954
 955         *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
 956                                      pagep);
 957         if (unlikely(*err < 0))
 958                 goto stop_journal;
 959         /*
 960          * If orig extent was uninitialized it can become initialized
 961          * at any time after i_data_sem was dropped, in order to
 962          * serialize with delalloc we have recheck extent while we
 963          * hold page's lock, if it is still the case data copy is not
 964          * necessary, just swap data blocks between orig and donor.
 965          */
 966         if (uninit) {
 967                 double_down_write_data_sem(orig_inode, donor_inode);
 968                 /* If any of extents in range became initialized we have to
 969                  * fallback to data copying */
 970                 uninit = mext_check_coverage(orig_inode, orig_blk_offset,
 971                                              block_len_in_page, 1, err);
 972                 if (*err)
 973                         goto drop_data_sem;
 974
 975                 uninit &= mext_check_coverage(donor_inode, orig_blk_offset,
 976                                               block_len_in_page, 1, err);
 977                 if (*err)
 978                         goto drop_data_sem;
 979
 980                 if (!uninit) {
 981                         double_up_write_data_sem(orig_inode, donor_inode);
 982                         goto data_copy;
 983                 }
 984                 if ((page_has_private(pagep[0]) &&
 985                      !try_to_release_page(pagep[0], 0)) ||
 986                     (page_has_private(pagep[1]) &&
 987                      !try_to_release_page(pagep[1], 0))) {
 988                         *err = -EBUSY;
 989                         goto drop_data_sem;
 990                 }
 991                 replaced_count = mext_replace_branches(handle, orig_inode,
 992                                                 donor_inode, orig_blk_offset,
 993                                                 block_len_in_page, err);
 994         drop_data_sem:
 995                 double_up_write_data_sem(orig_inode, donor_inode);
 996                 goto unlock_pages;
 997         }
 998 data_copy:
 999         *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
1000         if (*err)
1001                 goto unlock_pages;
1002
1003         /* At this point all buffers in range are uptodate, old mapping layout
1004          * is no longer required, try to drop it now. */
1005         if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
1006             (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
1007                 *err = -EBUSY;
1008                 goto unlock_pages;
1009         }
1010
1011         replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
1012                                                orig_blk_offset,
1013                                                block_len_in_page, err);
1014         if (*err) {
1015                 if (replaced_count) {
1016                         block_len_in_page = replaced_count;
1017                         replaced_size =
1018                                 block_len_in_page << orig_inode->i_blkbits;
1019                 } else
1020                         goto unlock_pages;
1021         }
1022         /* Perform all necessary steps similar write_begin()/write_end()
1023          * but keeping in mind that i_size will not change */
1024         *err = __block_write_begin(pagep[0], from, from + replaced_size,
1025                                    ext4_get_block);
1026         if (!*err)
1027                 *err = block_commit_write(pagep[0], from, from + replaced_size);
1028
1029         if (unlikely(*err < 0))
1030                 goto repair_branches;
1031
1032         /* Even in case of data=writeback it is reasonable to pin
1033          * inode to transaction, to prevent unexpected data loss */
1034         *err = ext4_jbd2_file_inode(handle, orig_inode);
1035
1036 unlock_pages:
1037         unlock_page(pagep[0]);
1038         page_cache_release(pagep[0]);
1039         unlock_page(pagep[1]);
1040         page_cache_release(pagep[1]);
1041 stop_journal:
1042         ext4_journal_stop(handle);
1043         /* Buffer was busy because probably is pinned to journal transaction,
1044          * force transaction commit may help to free it. */
1045         if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
1046                                                       &retries))
1047                 goto again;
1048         return replaced_count;
1049
1050 repair_branches:
1051         /*
1052          * This should never ever happen!
1053          * Extents are swapped already, but we are not able to copy data.
1054          * Try to swap extents to it's original places
1055          */
1056         double_down_write_data_sem(orig_inode, donor_inode);
1057         replaced_count = mext_replace_branches(handle, donor_inode, orig_inode,
1058                                                orig_blk_offset,
1059                                                block_len_in_page, &err2);
1060         double_up_write_data_sem(orig_inode, donor_inode);
1061         if (replaced_count != block_len_in_page) {
1062                 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
1063                                        "Unable to copy data block,"
1064                                        " data will be lost.");
1065                 *err = -EIO;
1066         }
1067         replaced_count = 0;
1068         goto unlock_pages;
1069 }
1070
1071 /**
1072  * mext_check_arguments - Check whether move extent can be done
1073  *
1074  * @orig_inode:         original inode
1075  * @donor_inode:        donor inode
1076  * @orig_start:         logical start offset in block for orig
1077  * @donor_start:        logical start offset in block for donor
1078  * @len:                the number of blocks to be moved
1079  *
1080  * Check the arguments of ext4_move_extents() whether the files can be
1081  * exchanged with each other.
1082  * Return 0 on success, or a negative error value on failure.
1083  */
1084 static int
1085 mext_check_arguments(struct inode *orig_inode,
1086                      struct inode *donor_inode, __u64 orig_start,
1087                      __u64 donor_start, __u64 *len)
1088 {
1089         ext4_lblk_t orig_blocks, donor_blocks;
1090         unsigned int blkbits = orig_inode->i_blkbits;
1091         unsigned int blocksize = 1 << blkbits;
1092
1093         if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
1094                 ext4_debug("ext4 move extent: suid or sgid is set"
1095                            " to donor file [ino:orig %lu, donor %lu]\n",
1096                            orig_inode->i_ino, donor_inode->i_ino);
1097                 return -EINVAL;
1098         }
1099
1100         if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
1101                 return -EPERM;
1102
1103         /* Ext4 move extent does not support swapfile */
1104         if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
1105                 ext4_debug("ext4 move extent: The argument files should "
1106                         "not be swapfile [ino:orig %lu, donor %lu]\n",
1107                         orig_inode->i_ino, donor_inode->i_ino);
1108                 return -EINVAL;
1109         }
1110
1111         /* Ext4 move extent supports only extent based file */
1112         if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
1113                 ext4_debug("ext4 move extent: orig file is not extents "
1114                         "based file [ino:orig %lu]\n", orig_inode->i_ino);
1115                 return -EOPNOTSUPP;
1116         } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
1117                 ext4_debug("ext4 move extent: donor file is not extents "
1118                         "based file [ino:donor %lu]\n", donor_inode->i_ino);
1119                 return -EOPNOTSUPP;
1120         }
1121
1122         if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1123                 ext4_debug("ext4 move extent: File size is 0 byte\n");
1124                 return -EINVAL;
1125         }
1126
1127         /* Start offset should be same */
1128         if (orig_start != donor_start) {
1129                 ext4_debug("ext4 move extent: orig and donor's start "
1130                         "offset are not same [ino:orig %lu, donor %lu]\n",
1131                         orig_inode->i_ino, donor_inode->i_ino);
1132                 return -EINVAL;
1133         }
1134
1135         if ((orig_start >= EXT_MAX_BLOCKS) ||
1136             (*len > EXT_MAX_BLOCKS) ||
1137             (orig_start + *len >= EXT_MAX_BLOCKS))  {
1138                 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1139                         "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1140                         orig_inode->i_ino, donor_inode->i_ino);
1141                 return -EINVAL;
1142         }
1143
1144         if (orig_inode->i_size > donor_inode->i_size) {
1145                 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1146                 /* TODO: eliminate this artificial restriction */
1147                 if (orig_start >= donor_blocks) {
1148                         ext4_debug("ext4 move extent: orig start offset "
1149                         "[%llu] should be less than donor file blocks "
1150                         "[%u] [ino:orig %lu, donor %lu]\n",
1151                         orig_start, donor_blocks,
1152                         orig_inode->i_ino, donor_inode->i_ino);
1153                         return -EINVAL;
1154                 }
1155
1156                 /* TODO: eliminate this artificial restriction */
1157                 if (orig_start + *len > donor_blocks) {
1158                         ext4_debug("ext4 move extent: End offset [%llu] should "
1159                                 "be less than donor file blocks [%u]."
1160                                 "So adjust length from %llu to %llu "
1161                                 "[ino:orig %lu, donor %lu]\n",
1162                                 orig_start + *len, donor_blocks,
1163                                 *len, donor_blocks - orig_start,
1164                                 orig_inode->i_ino, donor_inode->i_ino);
1165                         *len = donor_blocks - orig_start;
1166                 }
1167         } else {
1168                 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1169                 if (orig_start >= orig_blocks) {
1170                         ext4_debug("ext4 move extent: start offset [%llu] "
1171                                 "should be less than original file blocks "
1172                                 "[%u] [ino:orig %lu, donor %lu]\n",
1173                                  orig_start, orig_blocks,
1174                                 orig_inode->i_ino, donor_inode->i_ino);
1175                         return -EINVAL;
1176                 }
1177
1178                 if (orig_start + *len > orig_blocks) {
1179                         ext4_debug("ext4 move extent: Adjust length "
1180                                 "from %llu to %llu. Because it should be "
1181                                 "less than original file blocks "
1182                                 "[ino:orig %lu, donor %lu]\n",
1183                                 *len, orig_blocks - orig_start,
1184                                 orig_inode->i_ino, donor_inode->i_ino);
1185                         *len = orig_blocks - orig_start;
1186                 }
1187         }
1188
1189         if (!*len) {
1190                 ext4_debug("ext4 move extent: len should not be 0 "
1191                         "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1192                         donor_inode->i_ino);
1193                 return -EINVAL;
1194         }
1195
1196         return 0;
1197 }
1198
1199 /**
1200  * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1201  *
1202  * @inode1:     the inode structure
1203  * @inode2:     the inode structure
1204  *
1205  * Lock two inodes' i_mutex
1206  */
1207 static void
1208 mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1209 {
1210         BUG_ON(inode1 == inode2);
1211         if (inode1 < inode2) {
1212                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1213                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1214         } else {
1215                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1216                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1217         }
1218 }
1219
1220 /**
1221  * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1222  *
1223  * @inode1:     the inode that is released first
1224  * @inode2:     the inode that is released second
1225  *
1226  */
1227
1228 static void
1229 mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1230 {
1231         mutex_unlock(&inode1->i_mutex);
1232         mutex_unlock(&inode2->i_mutex);
1233 }
1234
1235 /**
1236  * ext4_move_extents - Exchange the specified range of a file
1237  *
1238  * @o_filp:             file structure of the original file
1239  * @d_filp:             file structure of the donor file
1240  * @orig_start:         start offset in block for orig
1241  * @donor_start:        start offset in block for donor
1242  * @len:                the number of blocks to be moved
1243  * @moved_len:          moved block length
1244  *
1245  * This function returns 0 and moved block length is set in moved_len
1246  * if succeed, otherwise returns error value.
1247  *
1248  * Note: ext4_move_extents() proceeds the following order.
1249  * 1:ext4_move_extents() calculates the last block number of moving extent
1250  *   function by the start block number (orig_start) and the number of blocks
1251  *   to be moved (len) specified as arguments.
1252  *   If the {orig, donor}_start points a hole, the extent's start offset
1253  *   pointed by ext_cur (current extent), holecheck_path, orig_path are set
1254  *   after hole behind.
1255  * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1256  *   or the ext_cur exceeds the block_end which is last logical block number.
1257  * 3:To get the length of continues area, call mext_next_extent()
1258  *   specified with the ext_cur (initial value is holecheck_path) re-cursive,
1259  *   until find un-continuous extent, the start logical block number exceeds
1260  *   the block_end or the extent points to the last extent.
1261  * 4:Exchange the original inode data with donor inode data
1262  *   from orig_page_offset to seq_end_page.
1263  *   The start indexes of data are specified as arguments.
1264  *   That of the original inode is orig_page_offset,
1265  *   and the donor inode is also orig_page_offset
1266  *   (To easily handle blocksize != pagesize case, the offset for the
1267  *   donor inode is block unit).
1268  * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1269  *   then returns to step 2.
1270  * 6:Release holecheck_path, orig_path and set the len to moved_len
1271  *   which shows the number of moved blocks.
1272  *   The moved_len is useful for the command to calculate the file offset
1273  *   for starting next move extent ioctl.
1274  * 7:Return 0 on success, or a negative error value on failure.
1275  */
1276 int
1277 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1278                  __u64 orig_start, __u64 donor_start, __u64 len,
1279                  __u64 *moved_len)
1280 {
1281         struct inode *orig_inode = o_filp->f_dentry->d_inode;
1282         struct inode *donor_inode = d_filp->f_dentry->d_inode;
1283         struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1284         struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1285         ext4_lblk_t block_start = orig_start;
1286         ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1287         ext4_lblk_t rest_blocks;
1288         pgoff_t orig_page_offset = 0, seq_end_page;
1289         int ret, depth, last_extent = 0;
1290         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1291         int data_offset_in_page;
1292         int block_len_in_page;
1293         int uninit;
1294
1295         if (orig_inode->i_sb != donor_inode->i_sb) {
1296                 ext4_debug("ext4 move extent: The argument files "
1297                         "should be in same FS [ino:orig %lu, donor %lu]\n",
1298                         orig_inode->i_ino, donor_inode->i_ino);
1299                 return -EINVAL;
1300         }
1301
1302         /* orig and donor should be different inodes */
1303         if (orig_inode == donor_inode) {
1304                 ext4_debug("ext4 move extent: The argument files should not "
1305                         "be same inode [ino:orig %lu, donor %lu]\n",
1306                         orig_inode->i_ino, donor_inode->i_ino);
1307                 return -EINVAL;
1308         }
1309
1310         /* Regular file check */
1311         if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1312                 ext4_debug("ext4 move extent: The argument files should be "
1313                         "regular file [ino:orig %lu, donor %lu]\n",
1314                         orig_inode->i_ino, donor_inode->i_ino);
1315                 return -EINVAL;
1316         }
1317         /* TODO: This is non obvious task to swap blocks for inodes with full
1318            jornaling enabled */
1319         if (ext4_should_journal_data(orig_inode) ||
1320             ext4_should_journal_data(donor_inode)) {
1321                 return -EINVAL;
1322         }
1323         /* Protect orig and donor inodes against a truncate */
1324         mext_inode_double_lock(orig_inode, donor_inode);
1325
1326         /* Wait for all existing dio workers */
1327         ext4_inode_block_unlocked_dio(orig_inode);
1328         ext4_inode_block_unlocked_dio(donor_inode);
1329         inode_dio_wait(orig_inode);
1330         inode_dio_wait(donor_inode);
1331
1332         /* Protect extent tree against block allocations via delalloc */
1333         double_down_write_data_sem(orig_inode, donor_inode);
1334         /* Check the filesystem environment whether move_extent can be done */
1335         ret = mext_check_arguments(orig_inode, donor_inode, orig_start,
1336                                     donor_start, &len);
1337         if (ret)
1338                 goto out;
1339
1340         file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1341         block_end = block_start + len - 1;
1342         if (file_end < block_end)
1343                 len -= block_end - file_end;
1344
1345         ret = get_ext_path(orig_inode, block_start, &orig_path);
1346         if (ret)
1347                 goto out;
1348
1349         /* Get path structure to check the hole */
1350         ret = get_ext_path(orig_inode, block_start, &holecheck_path);
1351         if (ret)
1352                 goto out;
1353
1354         depth = ext_depth(orig_inode);
1355         ext_cur = holecheck_path[depth].p_ext;
1356
1357         /*
1358          * Get proper starting location of block replacement if block_start was
1359          * within the hole.
1360          */
1361         if (le32_to_cpu(ext_cur->ee_block) +
1362                 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1363                 /*
1364                  * The hole exists between extents or the tail of
1365                  * original file.
1366                  */
1367                 last_extent = mext_next_extent(orig_inode,
1368                                         holecheck_path, &ext_cur);
1369                 if (last_extent < 0) {
1370                         ret = last_extent;
1371                         goto out;
1372                 }
1373                 last_extent = mext_next_extent(orig_inode, orig_path,
1374                                                         &ext_dummy);
1375                 if (last_extent < 0) {
1376                         ret = last_extent;
1377                         goto out;
1378                 }
1379                 seq_start = le32_to_cpu(ext_cur->ee_block);
1380         } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1381                 /* The hole exists at the beginning of original file. */
1382                 seq_start = le32_to_cpu(ext_cur->ee_block);
1383         else
1384                 seq_start = block_start;
1385
1386         /* No blocks within the specified range. */
1387         if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1388                 ext4_debug("ext4 move extent: The specified range of file "
1389                                                         "may be the hole\n");
1390                 ret = -EINVAL;
1391                 goto out;
1392         }
1393
1394         /* Adjust start blocks */
1395         add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1396                          ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1397                      max(le32_to_cpu(ext_cur->ee_block), block_start);
1398
1399         while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1400                 seq_blocks += add_blocks;
1401
1402                 /* Adjust tail blocks */
1403                 if (seq_start + seq_blocks - 1 > block_end)
1404                         seq_blocks = block_end - seq_start + 1;
1405
1406                 ext_prev = ext_cur;
1407                 last_extent = mext_next_extent(orig_inode, holecheck_path,
1408                                                 &ext_cur);
1409                 if (last_extent < 0) {
1410                         ret = last_extent;
1411                         break;
1412                 }
1413                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1414
1415                 /*
1416                  * Extend the length of contiguous block (seq_blocks)
1417                  * if extents are contiguous.
1418                  */
1419                 if (ext4_can_extents_be_merged(orig_inode,
1420                                                ext_prev, ext_cur) &&
1421                     block_end >= le32_to_cpu(ext_cur->ee_block) &&
1422                     !last_extent)
1423                         continue;
1424
1425                 /* Is original extent is uninitialized */
1426                 uninit = ext4_ext_is_uninitialized(ext_prev);
1427
1428                 data_offset_in_page = seq_start % blocks_per_page;
1429
1430                 /*
1431                  * Calculate data blocks count that should be swapped
1432                  * at the first page.
1433                  */
1434                 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1435                         /* Swapped blocks are across pages */
1436                         block_len_in_page =
1437                                         blocks_per_page - data_offset_in_page;
1438                 } else {
1439                         /* Swapped blocks are in a page */
1440                         block_len_in_page = seq_blocks;
1441                 }
1442
1443                 orig_page_offset = seq_start >>
1444                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1445                 seq_end_page = (seq_start + seq_blocks - 1) >>
1446                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1447                 seq_start = le32_to_cpu(ext_cur->ee_block);
1448                 rest_blocks = seq_blocks;
1449
1450                 /*
1451                  * Up semaphore to avoid following problems:
1452                  * a. transaction deadlock among ext4_journal_start,
1453                  *    ->write_begin via pagefault, and jbd2_journal_commit
1454                  * b. racing with ->readpage, ->write_begin, and ext4_get_block
1455                  *    in move_extent_per_page
1456                  */
1457                 double_up_write_data_sem(orig_inode, donor_inode);
1458
1459                 while (orig_page_offset <= seq_end_page) {
1460
1461                         /* Swap original branches with new branches */
1462                         block_len_in_page = move_extent_per_page(
1463                                                 o_filp, donor_inode,
1464                                                 orig_page_offset,
1465                                                 data_offset_in_page,
1466                                                 block_len_in_page, uninit,
1467                                                 &ret);
1468
1469                         /* Count how many blocks we have exchanged */
1470                         *moved_len += block_len_in_page;
1471                         if (ret < 0)
1472                                 break;
1473                         if (*moved_len > len) {
1474                                 EXT4_ERROR_INODE(orig_inode,
1475                                         "We replaced blocks too much! "
1476                                         "sum of replaced: %llu requested: %llu",
1477                                         *moved_len, len);
1478                                 ret = -EIO;
1479                                 break;
1480                         }
1481
1482                         orig_page_offset++;
1483                         data_offset_in_page = 0;
1484                         rest_blocks -= block_len_in_page;
1485                         if (rest_blocks > blocks_per_page)
1486                                 block_len_in_page = blocks_per_page;
1487                         else
1488                                 block_len_in_page = rest_blocks;
1489                 }
1490
1491                 double_down_write_data_sem(orig_inode, donor_inode);
1492                 if (ret < 0)
1493                         break;
1494
1495                 /* Decrease buffer counter */
1496                 if (holecheck_path)
1497                         ext4_ext_drop_refs(holecheck_path);
1498                 ret = get_ext_path(orig_inode, seq_start, &holecheck_path);
1499                 if (ret)
1500                         break;
1501                 depth = holecheck_path->p_depth;
1502
1503                 /* Decrease buffer counter */
1504                 if (orig_path)
1505                         ext4_ext_drop_refs(orig_path);
1506                 ret = get_ext_path(orig_inode, seq_start, &orig_path);
1507                 if (ret)
1508                         break;
1509
1510                 ext_cur = holecheck_path[depth].p_ext;
1511                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1512                 seq_blocks = 0;
1513
1514         }
1515 out:
1516         if (*moved_len) {
1517                 ext4_discard_preallocations(orig_inode);
1518                 ext4_discard_preallocations(donor_inode);
1519         }
1520
1521         if (orig_path) {
1522                 ext4_ext_drop_refs(orig_path);
1523                 kfree(orig_path);
1524         }
1525         if (holecheck_path) {
1526                 ext4_ext_drop_refs(holecheck_path);
1527                 kfree(holecheck_path);
1528         }
1529         double_up_write_data_sem(orig_inode, donor_inode);
1530         ext4_inode_resume_unlocked_dio(orig_inode);
1531         ext4_inode_resume_unlocked_dio(donor_inode);
1532         mext_inode_double_unlock(orig_inode, donor_inode);
1533
1534         return ret;
1535 }