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 #include "ext4_extents.h"
  22
  23 /**
  24  * get_ext_path - Find an extent path for designated logical block number.
  25  *
  26  * @inode:      an inode which is searched
  27  * @lblock:     logical block number to find an extent path
  28  * @path:       pointer to an extent path pointer (for output)
  29  *
  30  * ext4_ext_find_extent wrapper. Return 0 on success, or a negative error value
  31  * on failure.
  32  */
  33 static inline int
  34 get_ext_path(struct inode *inode, ext4_lblk_t lblock,
  35                 struct ext4_ext_path **orig_path)
  36 {
  37         int ret = 0;
  38         struct ext4_ext_path *path;
  39
  40         path = ext4_ext_find_extent(inode, lblock, *orig_path);
  41         if (IS_ERR(path))
  42                 ret = PTR_ERR(path);
  43         else if (path[ext_depth(inode)].p_ext == NULL)
  44                 ret = -ENODATA;
  45         else
  46                 *orig_path = path;
  47
  48         return ret;
  49 }
  50
  51 /**
  52  * copy_extent_status - Copy the extent's initialization status
  53  *
  54  * @src:        an extent for getting initialize status
  55  * @dest:       an extent to be set the status
  56  */
  57 static void
  58 copy_extent_status(struct ext4_extent *src, struct ext4_extent *dest)
  59 {
  60         if (ext4_ext_is_uninitialized(src))
  61                 ext4_ext_mark_uninitialized(dest);
  62         else
  63                 dest->ee_len = cpu_to_le16(ext4_ext_get_actual_len(dest));
  64 }
  65
  66 /**
  67  * mext_next_extent - Search for the next extent and set it to "extent"
  68  *
  69  * @inode:      inode which is searched
  70  * @path:       this will obtain data for the next extent
  71  * @extent:     pointer to the next extent we have just gotten
  72  *
  73  * Search the next extent in the array of ext4_ext_path structure (@path)
  74  * and set it to ext4_extent structure (@extent). In addition, the member of
  75  * @path (->p_ext) also points the next extent. Return 0 on success, 1 if
  76  * ext4_ext_path structure refers to the last extent, or a negative error
  77  * value on failure.
  78  */
  79 static int
  80 mext_next_extent(struct inode *inode, struct ext4_ext_path *path,
  81                       struct ext4_extent **extent)
  82 {
  83         struct ext4_extent_header *eh;
  84         int ppos, leaf_ppos = path->p_depth;
  85
  86         ppos = leaf_ppos;
  87         if (EXT_LAST_EXTENT(path[ppos].p_hdr) > path[ppos].p_ext) {
  88                 /* leaf block */
  89                 *extent = ++path[ppos].p_ext;
  90                 path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
  91                 return 0;
  92         }
  93
  94         while (--ppos >= 0) {
  95                 if (EXT_LAST_INDEX(path[ppos].p_hdr) >
  96                     path[ppos].p_idx) {
  97                         int cur_ppos = ppos;
  98
  99                         /* index block */
 100                         path[ppos].p_idx++;
 101                         path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
 102                         if (path[ppos+1].p_bh)
 103                                 brelse(path[ppos+1].p_bh);
 104                         path[ppos+1].p_bh =
 105                                 sb_bread(inode->i_sb, path[ppos].p_block);
 106                         if (!path[ppos+1].p_bh)
 107                                 return -EIO;
 108                         path[ppos+1].p_hdr =
 109                                 ext_block_hdr(path[ppos+1].p_bh);
 110
 111                         /* Halfway index block */
 112                         while (++cur_ppos < leaf_ppos) {
 113                                 path[cur_ppos].p_idx =
 114                                         EXT_FIRST_INDEX(path[cur_ppos].p_hdr);
 115                                 path[cur_ppos].p_block =
 116                                         ext4_idx_pblock(path[cur_ppos].p_idx);
 117                                 if (path[cur_ppos+1].p_bh)
 118                                         brelse(path[cur_ppos+1].p_bh);
 119                                 path[cur_ppos+1].p_bh = sb_bread(inode->i_sb,
 120                                         path[cur_ppos].p_block);
 121                                 if (!path[cur_ppos+1].p_bh)
 122                                         return -EIO;
 123                                 path[cur_ppos+1].p_hdr =
 124                                         ext_block_hdr(path[cur_ppos+1].p_bh);
 125                         }
 126
 127                         path[leaf_ppos].p_ext = *extent = NULL;
 128
 129                         eh = path[leaf_ppos].p_hdr;
 130                         if (le16_to_cpu(eh->eh_entries) == 0)
 131                                 /* empty leaf is found */
 132                                 return -ENODATA;
 133
 134                         /* leaf block */
 135                         path[leaf_ppos].p_ext = *extent =
 136                                 EXT_FIRST_EXTENT(path[leaf_ppos].p_hdr);
 137                         path[leaf_ppos].p_block =
 138                                         ext4_ext_pblock(path[leaf_ppos].p_ext);
 139                         return 0;
 140                 }
 141         }
 142         /* We found the last extent */
 143         return 1;
 144 }
 145
 146 /**
 147  * double_down_write_data_sem - Acquire two inodes' write lock of i_data_sem
 148  *
 149  * Acquire write lock of i_data_sem of the two inodes
 150  */
 151 static void
 152 double_down_write_data_sem(struct inode *first, struct inode *second)
 153 {
 154         if (first < second) {
 155                 down_write(&EXT4_I(first)->i_data_sem);
 156                 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
 157         } else {
 158                 down_write(&EXT4_I(second)->i_data_sem);
 159                 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING);
 160
 161         }
 162 }
 163
 164 /**
 165  * double_up_write_data_sem - Release two inodes' write lock of i_data_sem
 166  *
 167  * @orig_inode:         original inode structure to be released its lock first
 168  * @donor_inode:        donor inode structure to be released its lock second
 169  * Release write lock of i_data_sem of two inodes (orig and donor).
 170  */
 171 static void
 172 double_up_write_data_sem(struct inode *orig_inode, struct inode *donor_inode)
 173 {
 174         up_write(&EXT4_I(orig_inode)->i_data_sem);
 175         up_write(&EXT4_I(donor_inode)->i_data_sem);
 176 }
 177
 178 /**
 179  * mext_insert_across_blocks - Insert extents across leaf block
 180  *
 181  * @handle:             journal handle
 182  * @orig_inode:         original inode
 183  * @o_start:            first original extent to be changed
 184  * @o_end:              last original extent to be changed
 185  * @start_ext:          first new extent to be inserted
 186  * @new_ext:            middle of new extent to be inserted
 187  * @end_ext:            last new extent to be inserted
 188  *
 189  * Allocate a new leaf block and insert extents into it. Return 0 on success,
 190  * or a negative error value on failure.
 191  */
 192 static int
 193 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
 194                 struct ext4_extent *o_start, struct ext4_extent *o_end,
 195                 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
 196                 struct ext4_extent *end_ext)
 197 {
 198         struct ext4_ext_path *orig_path = NULL;
 199         ext4_lblk_t eblock = 0;
 200         int new_flag = 0;
 201         int end_flag = 0;
 202         int err = 0;
 203
 204         if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
 205                 if (o_start == o_end) {
 206
 207                         /*       start_ext   new_ext    end_ext
 208                          * donor |---------|-----------|--------|
 209                          * orig  |------------------------------|
 210                          */
 211                         end_flag = 1;
 212                 } else {
 213
 214                         /*       start_ext   new_ext   end_ext
 215                          * donor |---------|----------|---------|
 216                          * orig  |---------------|--------------|
 217                          */
 218                         o_end->ee_block = end_ext->ee_block;
 219                         o_end->ee_len = end_ext->ee_len;
 220                         ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
 221                 }
 222
 223                 o_start->ee_len = start_ext->ee_len;
 224                 eblock = le32_to_cpu(start_ext->ee_block);
 225                 new_flag = 1;
 226
 227         } else if (start_ext->ee_len && new_ext->ee_len &&
 228                    !end_ext->ee_len && o_start == o_end) {
 229
 230                 /*       start_ext      new_ext
 231                  * donor |--------------|---------------|
 232                  * orig  |------------------------------|
 233                  */
 234                 o_start->ee_len = start_ext->ee_len;
 235                 eblock = le32_to_cpu(start_ext->ee_block);
 236                 new_flag = 1;
 237
 238         } else if (!start_ext->ee_len && new_ext->ee_len &&
 239                    end_ext->ee_len && o_start == o_end) {
 240
 241                 /*        new_ext       end_ext
 242                  * donor |--------------|---------------|
 243                  * orig  |------------------------------|
 244                  */
 245                 o_end->ee_block = end_ext->ee_block;
 246                 o_end->ee_len = end_ext->ee_len;
 247                 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
 248
 249                 /*
 250                  * Set 0 to the extent block if new_ext was
 251                  * the first block.
 252                  */
 253                 if (new_ext->ee_block)
 254                         eblock = le32_to_cpu(new_ext->ee_block);
 255
 256                 new_flag = 1;
 257         } else {
 258                 ext4_debug("ext4 move extent: Unexpected insert case\n");
 259                 return -EIO;
 260         }
 261
 262         if (new_flag) {
 263                 err = get_ext_path(orig_inode, eblock, &orig_path);
 264                 if (err)
 265                         goto out;
 266
 267                 if (ext4_ext_insert_extent(handle, orig_inode,
 268                                         orig_path, new_ext, 0))
 269                         goto out;
 270         }
 271
 272         if (end_flag) {
 273                 err = get_ext_path(orig_inode,
 274                                 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
 275                 if (err)
 276                         goto out;
 277
 278                 if (ext4_ext_insert_extent(handle, orig_inode,
 279                                            orig_path, end_ext, 0))
 280                         goto out;
 281         }
 282 out:
 283         if (orig_path) {
 284                 ext4_ext_drop_refs(orig_path);
 285                 kfree(orig_path);
 286         }
 287
 288         return err;
 289
 290 }
 291
 292 /**
 293  * mext_insert_inside_block - Insert new extent to the extent block
 294  *
 295  * @o_start:            first original extent to be moved
 296  * @o_end:              last original extent to be moved
 297  * @start_ext:          first new extent to be inserted
 298  * @new_ext:            middle of new extent to be inserted
 299  * @end_ext:            last new extent to be inserted
 300  * @eh:                 extent header of target leaf block
 301  * @range_to_move:      used to decide how to insert extent
 302  *
 303  * Insert extents into the leaf block. The extent (@o_start) is overwritten
 304  * by inserted extents.
 305  */
 306 static void
 307 mext_insert_inside_block(struct ext4_extent *o_start,
 308                               struct ext4_extent *o_end,
 309                               struct ext4_extent *start_ext,
 310                               struct ext4_extent *new_ext,
 311                               struct ext4_extent *end_ext,
 312                               struct ext4_extent_header *eh,
 313                               int range_to_move)
 314 {
 315         int i = 0;
 316         unsigned long len;
 317
 318         /* Move the existing extents */
 319         if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
 320                 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
 321                         (unsigned long)(o_end + 1);
 322                 memmove(o_end + 1 + range_to_move, o_end + 1, len);
 323         }
 324
 325         /* Insert start entry */
 326         if (start_ext->ee_len)
 327                 o_start[i++].ee_len = start_ext->ee_len;
 328
 329         /* Insert new entry */
 330         if (new_ext->ee_len) {
 331                 o_start[i] = *new_ext;
 332                 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
 333         }
 334
 335         /* Insert end entry */
 336         if (end_ext->ee_len)
 337                 o_start[i] = *end_ext;
 338
 339         /* Increment the total entries counter on the extent block */
 340         le16_add_cpu(&eh->eh_entries, range_to_move);
 341 }
 342
 343 /**
 344  * mext_insert_extents - Insert new extent
 345  *
 346  * @handle:     journal handle
 347  * @orig_inode: original inode
 348  * @orig_path:  path indicates first extent to be changed
 349  * @o_start:    first original extent to be changed
 350  * @o_end:      last original extent to be changed
 351  * @start_ext:  first new extent to be inserted
 352  * @new_ext:    middle of new extent to be inserted
 353  * @end_ext:    last new extent to be inserted
 354  *
 355  * Call the function to insert extents. If we cannot add more extents into
 356  * the leaf block, we call mext_insert_across_blocks() to create a
 357  * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
 358  * on success, or a negative error value on failure.
 359  */
 360 static int
 361 mext_insert_extents(handle_t *handle, struct inode *orig_inode,
 362                          struct ext4_ext_path *orig_path,
 363                          struct ext4_extent *o_start,
 364                          struct ext4_extent *o_end,
 365                          struct ext4_extent *start_ext,
 366                          struct ext4_extent *new_ext,
 367                          struct ext4_extent *end_ext)
 368 {
 369         struct  ext4_extent_header *eh;
 370         unsigned long need_slots, slots_range;
 371         int     range_to_move, depth, ret;
 372
 373         /*
 374          * The extents need to be inserted
 375          * start_extent + new_extent + end_extent.
 376          */
 377         need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
 378                 (new_ext->ee_len ? 1 : 0);
 379
 380         /* The number of slots between start and end */
 381         slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
 382                 / sizeof(struct ext4_extent);
 383
 384         /* Range to move the end of extent */
 385         range_to_move = need_slots - slots_range;
 386         depth = orig_path->p_depth;
 387         orig_path += depth;
 388         eh = orig_path->p_hdr;
 389
 390         if (depth) {
 391                 /* Register to journal */
 392                 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
 393                 if (ret)
 394                         return ret;
 395         }
 396
 397         /* Expansion */
 398         if (range_to_move > 0 &&
 399                 (range_to_move > le16_to_cpu(eh->eh_max)
 400                         - le16_to_cpu(eh->eh_entries))) {
 401
 402                 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
 403                                         o_end, start_ext, new_ext, end_ext);
 404                 if (ret < 0)
 405                         return ret;
 406         } else
 407                 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
 408                                                 end_ext, eh, range_to_move);
 409
 410         if (depth) {
 411                 ret = ext4_handle_dirty_metadata(handle, orig_inode,
 412                                                  orig_path->p_bh);
 413                 if (ret)
 414                         return ret;
 415         } else {
 416                 ret = ext4_mark_inode_dirty(handle, orig_inode);
 417                 if (ret < 0)
 418                         return ret;
 419         }
 420
 421         return 0;
 422 }
 423
 424 /**
 425  * mext_leaf_block - Move one leaf extent block into the inode.
 426  *
 427  * @handle:             journal handle
 428  * @orig_inode:         original inode
 429  * @orig_path:          path indicates first extent to be changed
 430  * @dext:               donor extent
 431  * @from:               start offset on the target file
 432  *
 433  * In order to insert extents into the leaf block, we must divide the extent
 434  * in the leaf block into three extents. The one is located to be inserted
 435  * extents, and the others are located around it.
 436  *
 437  * Therefore, this function creates structures to save extents of the leaf
 438  * block, and inserts extents by calling mext_insert_extents() with
 439  * created extents. Return 0 on success, or a negative error value on failure.
 440  */
 441 static int
 442 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
 443                      struct ext4_ext_path *orig_path, struct ext4_extent *dext,
 444                      ext4_lblk_t *from)
 445 {
 446         struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
 447         struct ext4_extent new_ext, start_ext, end_ext;
 448         ext4_lblk_t new_ext_end;
 449         int oext_alen, new_ext_alen, end_ext_alen;
 450         int depth = ext_depth(orig_inode);
 451         int ret;
 452
 453         start_ext.ee_block = end_ext.ee_block = 0;
 454         o_start = o_end = oext = orig_path[depth].p_ext;
 455         oext_alen = ext4_ext_get_actual_len(oext);
 456         start_ext.ee_len = end_ext.ee_len = 0;
 457
 458         new_ext.ee_block = cpu_to_le32(*from);
 459         ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
 460         new_ext.ee_len = dext->ee_len;
 461         new_ext_alen = ext4_ext_get_actual_len(&new_ext);
 462         new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
 463
 464         /*
 465          * Case: original extent is first
 466          * oext      |--------|
 467          * new_ext      |--|
 468          * start_ext |--|
 469          */
 470         if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
 471                 le32_to_cpu(new_ext.ee_block) <
 472                 le32_to_cpu(oext->ee_block) + oext_alen) {
 473                 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
 474                                                le32_to_cpu(oext->ee_block));
 475                 start_ext.ee_block = oext->ee_block;
 476                 copy_extent_status(oext, &start_ext);
 477         } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
 478                 prev_ext = oext - 1;
 479                 /*
 480                  * We can merge new_ext into previous extent,
 481                  * if these are contiguous and same extent type.
 482                  */
 483                 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
 484                                                &new_ext)) {
 485                         o_start = prev_ext;
 486                         start_ext.ee_len = cpu_to_le16(
 487                                 ext4_ext_get_actual_len(prev_ext) +
 488                                 new_ext_alen);
 489                         start_ext.ee_block = oext->ee_block;
 490                         copy_extent_status(prev_ext, &start_ext);
 491                         new_ext.ee_len = 0;
 492                 }
 493         }
 494
 495         /*
 496          * Case: new_ext_end must be less than oext
 497          * oext      |-----------|
 498          * new_ext       |-------|
 499          */
 500         if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
 501                 EXT4_ERROR_INODE(orig_inode,
 502                         "new_ext_end(%u) should be less than or equal to "
 503                         "oext->ee_block(%u) + oext_alen(%d) - 1",
 504                         new_ext_end, le32_to_cpu(oext->ee_block),
 505                         oext_alen);
 506                 ret = -EIO;
 507                 goto out;
 508         }
 509
 510         /*
 511          * Case: new_ext is smaller than original extent
 512          * oext    |---------------|
 513          * new_ext |-----------|
 514          * end_ext             |---|
 515          */
 516         if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
 517                 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
 518                 end_ext.ee_len =
 519                         cpu_to_le16(le32_to_cpu(oext->ee_block) +
 520                         oext_alen - 1 - new_ext_end);
 521                 copy_extent_status(oext, &end_ext);
 522                 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
 523                 ext4_ext_store_pblock(&end_ext,
 524                         (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
 525                 end_ext.ee_block =
 526                         cpu_to_le32(le32_to_cpu(o_end->ee_block) +
 527                         oext_alen - end_ext_alen);
 528         }
 529
 530         ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
 531                                 o_end, &start_ext, &new_ext, &end_ext);
 532 out:
 533         return ret;
 534 }
 535
 536 /**
 537  * mext_calc_swap_extents - Calculate extents for extent swapping.
 538  *
 539  * @tmp_dext:           the extent that will belong to the original inode
 540  * @tmp_oext:           the extent that will belong to the donor inode
 541  * @orig_off:           block offset of original inode
 542  * @donor_off:          block offset of donor inode
 543  * @max_count:          the maximum length of extents
 544  *
 545  * Return 0 on success, or a negative error value on failure.
 546  */
 547 static int
 548 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
 549                               struct ext4_extent *tmp_oext,
 550                               ext4_lblk_t orig_off, ext4_lblk_t donor_off,
 551                               ext4_lblk_t max_count)
 552 {
 553         ext4_lblk_t diff, orig_diff;
 554         struct ext4_extent dext_old, oext_old;
 555
 556         BUG_ON(orig_off != donor_off);
 557
 558         /* original and donor extents have to cover the same block offset */
 559         if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
 560             le32_to_cpu(tmp_oext->ee_block) +
 561                         ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
 562                 return -ENODATA;
 563
 564         if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
 565             le32_to_cpu(tmp_dext->ee_block) +
 566                         ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
 567                 return -ENODATA;
 568
 569         dext_old = *tmp_dext;
 570         oext_old = *tmp_oext;
 571
 572         /* When tmp_dext is too large, pick up the target range. */
 573         diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
 574
 575         ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
 576         le32_add_cpu(&tmp_dext->ee_block, diff);
 577         le16_add_cpu(&tmp_dext->ee_len, -diff);
 578
 579         if (max_count < ext4_ext_get_actual_len(tmp_dext))
 580                 tmp_dext->ee_len = cpu_to_le16(max_count);
 581
 582         orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
 583         ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
 584
 585         /* Adjust extent length if donor extent is larger than orig */
 586         if (ext4_ext_get_actual_len(tmp_dext) >
 587             ext4_ext_get_actual_len(tmp_oext) - orig_diff)
 588                 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
 589                                                 orig_diff);
 590
 591         tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
 592
 593         copy_extent_status(&oext_old, tmp_dext);
 594         copy_extent_status(&dext_old, tmp_oext);
 595
 596         return 0;
 597 }
 598
 599 /**
 600  * mext_check_coverage - Check that all extents in range has the same type
 601  *
 602  * @inode:              inode in question
 603  * @from:               block offset of inode
 604  * @count:              block count to be checked
 605  * @uninit:             extents expected to be uninitialized
 606  * @err:                pointer to save error value
 607  *
 608  * Return 1 if all extents in range has expected type, and zero otherwise.
 609  */
 610 static int
 611 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
 612                           int uninit, int *err)
 613 {
 614         struct ext4_ext_path *path = NULL;
 615         struct ext4_extent *ext;
 616         int ret = 0;
 617         ext4_lblk_t last = from + count;
 618         while (from < last) {
 619                 *err = get_ext_path(inode, from, &path);
 620                 if (*err)
 621                         goto out;
 622                 ext = path[ext_depth(inode)].p_ext;
 623                 if (uninit != ext4_ext_is_uninitialized(ext))
 624                         goto out;
 625                 from += ext4_ext_get_actual_len(ext);
 626                 ext4_ext_drop_refs(path);
 627         }
 628         ret = 1;
 629 out:
 630         if (path) {
 631                 ext4_ext_drop_refs(path);
 632                 kfree(path);
 633         }
 634         return ret;
 635 }
 636
 637 /**
 638  * mext_replace_branches - Replace original extents with new extents
 639  *
 640  * @handle:             journal handle
 641  * @orig_inode:         original inode
 642  * @donor_inode:        donor inode
 643  * @from:               block offset of orig_inode
 644  * @count:              block count to be replaced
 645  * @err:                pointer to save return value
 646  *
 647  * Replace original inode extents and donor inode extents page by page.
 648  * We implement this replacement in the following three steps:
 649  * 1. Save the block information of original and donor inodes into
 650  *    dummy extents.
 651  * 2. Change the block information of original inode to point at the
 652  *    donor inode blocks.
 653  * 3. Change the block information of donor inode to point at the saved
 654  *    original inode blocks in the dummy extents.
 655  *
 656  * Return replaced block count.
 657  */
 658 static int
 659 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
 660                            struct inode *donor_inode, ext4_lblk_t from,
 661                            ext4_lblk_t count, int *err)
 662 {
 663         struct ext4_ext_path *orig_path = NULL;
 664         struct ext4_ext_path *donor_path = NULL;
 665         struct ext4_extent *oext, *dext;
 666         struct ext4_extent tmp_dext, tmp_oext;
 667         ext4_lblk_t orig_off = from, donor_off = from;
 668         int depth;
 669         int replaced_count = 0;
 670         int dext_alen;
 671
 672         *err = ext4_es_remove_extent(orig_inode, from, count);
 673         if (*err)
 674                 goto out;
 675
 676         *err = ext4_es_remove_extent(donor_inode, from, count);
 677         if (*err)
 678                 goto out;
 679
 680         /* Get the original extent for the block "orig_off" */
 681         *err = get_ext_path(orig_inode, orig_off, &orig_path);
 682         if (*err)
 683                 goto out;
 684
 685         /* Get the donor extent for the head */
 686         *err = get_ext_path(donor_inode, donor_off, &donor_path);
 687         if (*err)
 688                 goto out;
 689         depth = ext_depth(orig_inode);
 690         oext = orig_path[depth].p_ext;
 691         tmp_oext = *oext;
 692
 693         depth = ext_depth(donor_inode);
 694         dext = donor_path[depth].p_ext;
 695         if (unlikely(!dext))
 696                 goto missing_donor_extent;
 697         tmp_dext = *dext;
 698
 699         *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 700                                       donor_off, count);
 701         if (*err)
 702                 goto out;
 703
 704         /* Loop for the donor extents */
 705         while (1) {
 706                 /* The extent for donor must be found. */
 707                 if (unlikely(!dext)) {
 708                 missing_donor_extent:
 709                         EXT4_ERROR_INODE(donor_inode,
 710                                    "The extent for donor must be found");
 711                         *err = -EIO;
 712                         goto out;
 713                 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
 714                         EXT4_ERROR_INODE(donor_inode,
 715                                 "Donor offset(%u) and the first block of donor "
 716                                 "extent(%u) should be equal",
 717                                 donor_off,
 718                                 le32_to_cpu(tmp_dext.ee_block));
 719                         *err = -EIO;
 720                         goto out;
 721                 }
 722
 723                 /* Set donor extent to orig extent */
 724                 *err = mext_leaf_block(handle, orig_inode,
 725                                            orig_path, &tmp_dext, &orig_off);
 726                 if (*err)
 727                         goto out;
 728
 729                 /* Set orig extent to donor extent */
 730                 *err = mext_leaf_block(handle, donor_inode,
 731                                            donor_path, &tmp_oext, &donor_off);
 732                 if (*err)
 733                         goto out;
 734
 735                 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
 736                 replaced_count += dext_alen;
 737                 donor_off += dext_alen;
 738                 orig_off += dext_alen;
 739
 740                 /* Already moved the expected blocks */
 741                 if (replaced_count >= count)
 742                         break;
 743
 744                 if (orig_path)
 745                         ext4_ext_drop_refs(orig_path);
 746                 *err = get_ext_path(orig_inode, orig_off, &orig_path);
 747                 if (*err)
 748                         goto out;
 749                 depth = ext_depth(orig_inode);
 750                 oext = orig_path[depth].p_ext;
 751                 tmp_oext = *oext;
 752
 753                 if (donor_path)
 754                         ext4_ext_drop_refs(donor_path);
 755                 *err = get_ext_path(donor_inode, donor_off, &donor_path);
 756                 if (*err)
 757                         goto out;
 758                 depth = ext_depth(donor_inode);
 759                 dext = donor_path[depth].p_ext;
 760                 tmp_dext = *dext;
 761
 762                 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 763                                            donor_off, count - replaced_count);
 764                 if (*err)
 765                         goto out;
 766         }
 767
 768 out:
 769         if (orig_path) {
 770                 ext4_ext_drop_refs(orig_path);
 771                 kfree(orig_path);
 772         }
 773         if (donor_path) {
 774                 ext4_ext_drop_refs(donor_path);
 775                 kfree(donor_path);
 776         }
 777
 778         return replaced_count;
 779 }
 780
 781 /**
 782  * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
 783  *
 784  * @inode1:     the inode structure
 785  * @inode2:     the inode structure
 786  * @index:      page index
 787  * @page:       result page vector
 788  *
 789  * Grab two locked pages for inode's by inode order
 790  */
 791 static int
 792 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
 793                       pgoff_t index, struct page *page[2])
 794 {
 795         struct address_space *mapping[2];
 796         unsigned fl = AOP_FLAG_NOFS;
 797
 798         BUG_ON(!inode1 || !inode2);
 799         if (inode1 < inode2) {
 800                 mapping[0] = inode1->i_mapping;
 801                 mapping[1] = inode2->i_mapping;
 802         } else {
 803                 mapping[0] = inode2->i_mapping;
 804                 mapping[1] = inode1->i_mapping;
 805         }
 806
 807         page[0] = grab_cache_page_write_begin(mapping[0], index, fl);
 808         if (!page[0])
 809                 return -ENOMEM;
 810
 811         page[1] = grab_cache_page_write_begin(mapping[1], index, fl);
 812         if (!page[1]) {
 813                 unlock_page(page[0]);
 814                 page_cache_release(page[0]);
 815                 return -ENOMEM;
 816         }
 817
 818         if (inode1 > inode2) {
 819                 struct page *tmp;
 820                 tmp = page[0];
 821                 page[0] = page[1];
 822                 page[1] = tmp;
 823         }
 824         return 0;
 825 }
 826
 827 /* Force page buffers uptodate w/o dropping page's lock */
 828 static int
 829 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
 830 {
 831         struct inode *inode = page->mapping->host;
 832         sector_t block;
 833         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 834         unsigned int blocksize, block_start, block_end;
 835         int i, err,  nr = 0, partial = 0;
 836         BUG_ON(!PageLocked(page));
 837         BUG_ON(PageWriteback(page));
 838
 839         if (PageUptodate(page))
 840                 return 0;
 841
 842         blocksize = 1 << inode->i_blkbits;
 843         if (!page_has_buffers(page))
 844                 create_empty_buffers(page, blocksize, 0);
 845
 846         head = page_buffers(page);
 847         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 848         for (bh = head, block_start = 0; bh != head || !block_start;
 849              block++, block_start = block_end, bh = bh->b_this_page) {
 850                 block_end = block_start + blocksize;
 851                 if (block_end <= from || block_start >= to) {
 852                         if (!buffer_uptodate(bh))
 853                                 partial = 1;
 854                         continue;
 855                 }
 856                 if (buffer_uptodate(bh))
 857                         continue;
 858                 if (!buffer_mapped(bh)) {
 859                         int err = 0;
 860                         err = ext4_get_block(inode, block, bh, 0);
 861                         if (err) {
 862                                 SetPageError(page);
 863                                 return err;
 864                         }
 865                         if (!buffer_mapped(bh)) {
 866                                 zero_user(page, block_start, blocksize);
 867                                 if (!err)
 868                                         set_buffer_uptodate(bh);
 869                                 continue;
 870                         }
 871                 }
 872                 BUG_ON(nr >= MAX_BUF_PER_PAGE);
 873                 arr[nr++] = bh;
 874         }
 875         /* No io required */
 876         if (!nr)
 877                 goto out;
 878
 879         for (i = 0; i < nr; i++) {
 880                 bh = arr[i];
 881                 if (!bh_uptodate_or_lock(bh)) {
 882                         err = bh_submit_read(bh);
 883                         if (err)
 884                                 return err;
 885                 }
 886         }
 887 out:
 888         if (!partial)
 889                 SetPageUptodate(page);
 890         return 0;
 891 }
 892
 893 /**
 894  * move_extent_per_page - Move extent data per page
 895  *
 896  * @o_filp:                     file structure of original file
 897  * @donor_inode:                donor inode
 898  * @orig_page_offset:           page index on original file
 899  * @data_offset_in_page:        block index where data swapping starts
 900  * @block_len_in_page:          the number of blocks to be swapped
 901  * @uninit:                     orig extent is uninitialized or not
 902  * @err:                        pointer to save return value
 903  *
 904  * Save the data in original inode blocks and replace original inode extents
 905  * with donor inode extents by calling mext_replace_branches().
 906  * Finally, write out the saved data in new original inode blocks. Return
 907  * replaced block count.
 908  */
 909 static int
 910 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
 911                   pgoff_t orig_page_offset, int data_offset_in_page,
 912                   int block_len_in_page, int uninit, int *err)
 913 {
 914         struct inode *orig_inode = file_inode(o_filp);
 915         struct page *pagep[2] = {NULL, NULL};
 916         handle_t *handle;
 917         ext4_lblk_t orig_blk_offset;
 918         long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
 919         unsigned long blocksize = orig_inode->i_sb->s_blocksize;
 920         unsigned int w_flags = 0;
 921         unsigned int tmp_data_size, data_size, replaced_size;
 922         int err2, jblocks, retries = 0;
 923         int replaced_count = 0;
 924         int from = data_offset_in_page << orig_inode->i_blkbits;
 925         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
 926
 927         /*
 928          * It needs twice the amount of ordinary journal buffers because
 929          * inode and donor_inode may change each different metadata blocks.
 930          */
 931 again:
 932         *err = 0;
 933         jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
 934         handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
 935         if (IS_ERR(handle)) {
 936                 *err = PTR_ERR(handle);
 937                 return 0;
 938         }
 939
 940         if (segment_eq(get_fs(), KERNEL_DS))
 941                 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
 942
 943         orig_blk_offset = orig_page_offset * blocks_per_page +
 944                 data_offset_in_page;
 945
 946         offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
 947
 948         /* Calculate data_size */
 949         if ((orig_blk_offset + block_len_in_page - 1) ==
 950             ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
 951                 /* Replace the last block */
 952                 tmp_data_size = orig_inode->i_size & (blocksize - 1);
 953                 /*
 954                  * If data_size equal zero, it shows data_size is multiples of
 955                  * blocksize. So we set appropriate value.
 956                  */
 957                 if (tmp_data_size == 0)
 958                         tmp_data_size = blocksize;
 959
 960                 data_size = tmp_data_size +
 961                         ((block_len_in_page - 1) << orig_inode->i_blkbits);
 962         } else
 963                 data_size = block_len_in_page << orig_inode->i_blkbits;
 964
 965         replaced_size = data_size;
 966
 967         *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
 968                                      pagep);
 969         if (unlikely(*err < 0))
 970                 goto stop_journal;
 971         /*
 972          * If orig extent was uninitialized it can become initialized
 973          * at any time after i_data_sem was dropped, in order to
 974          * serialize with delalloc we have recheck extent while we
 975          * hold page's lock, if it is still the case data copy is not
 976          * necessary, just swap data blocks between orig and donor.
 977          */
 978         if (uninit) {
 979                 double_down_write_data_sem(orig_inode, donor_inode);
 980                 /* If any of extents in range became initialized we have to
 981                  * fallback to data copying */
 982                 uninit = mext_check_coverage(orig_inode, orig_blk_offset,
 983                                              block_len_in_page, 1, err);
 984                 if (*err)
 985                         goto drop_data_sem;
 986
 987                 uninit &= mext_check_coverage(donor_inode, orig_blk_offset,
 988                                               block_len_in_page, 1, err);
 989                 if (*err)
 990                         goto drop_data_sem;
 991
 992                 if (!uninit) {
 993                         double_up_write_data_sem(orig_inode, donor_inode);
 994                         goto data_copy;
 995                 }
 996                 if ((page_has_private(pagep[0]) &&
 997                      !try_to_release_page(pagep[0], 0)) ||
 998                     (page_has_private(pagep[1]) &&
 999                      !try_to_release_page(pagep[1], 0))) {
1000                         *err = -EBUSY;
1001                         goto drop_data_sem;
1002                 }
1003                 replaced_count = mext_replace_branches(handle, orig_inode,
1004                                                 donor_inode, orig_blk_offset,
1005                                                 block_len_in_page, err);
1006         drop_data_sem:
1007                 double_up_write_data_sem(orig_inode, donor_inode);
1008                 goto unlock_pages;
1009         }
1010 data_copy:
1011         *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
1012         if (*err)
1013                 goto unlock_pages;
1014
1015         /* At this point all buffers in range are uptodate, old mapping layout
1016          * is no longer required, try to drop it now. */
1017         if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
1018             (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
1019                 *err = -EBUSY;
1020                 goto unlock_pages;
1021         }
1022
1023         replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
1024                                                orig_blk_offset,
1025                                                block_len_in_page, err);
1026         if (*err) {
1027                 if (replaced_count) {
1028                         block_len_in_page = replaced_count;
1029                         replaced_size =
1030                                 block_len_in_page << orig_inode->i_blkbits;
1031                 } else
1032                         goto unlock_pages;
1033         }
1034         /* Perform all necessary steps similar write_begin()/write_end()
1035          * but keeping in mind that i_size will not change */
1036         *err = __block_write_begin(pagep[0], from, from + replaced_size,
1037                                    ext4_get_block);
1038         if (!*err)
1039                 *err = block_commit_write(pagep[0], from, from + replaced_size);
1040
1041         if (unlikely(*err < 0))
1042                 goto repair_branches;
1043
1044         /* Even in case of data=writeback it is reasonable to pin
1045          * inode to transaction, to prevent unexpected data loss */
1046         *err = ext4_jbd2_file_inode(handle, orig_inode);
1047
1048 unlock_pages:
1049         unlock_page(pagep[0]);
1050         page_cache_release(pagep[0]);
1051         unlock_page(pagep[1]);
1052         page_cache_release(pagep[1]);
1053 stop_journal:
1054         ext4_journal_stop(handle);
1055         /* Buffer was busy because probably is pinned to journal transaction,
1056          * force transaction commit may help to free it. */
1057         if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
1058                                                       &retries))
1059                 goto again;
1060         return replaced_count;
1061
1062 repair_branches:
1063         /*
1064          * This should never ever happen!
1065          * Extents are swapped already, but we are not able to copy data.
1066          * Try to swap extents to it's original places
1067          */
1068         double_down_write_data_sem(orig_inode, donor_inode);
1069         replaced_count = mext_replace_branches(handle, donor_inode, orig_inode,
1070                                                orig_blk_offset,
1071                                                block_len_in_page, &err2);
1072         double_up_write_data_sem(orig_inode, donor_inode);
1073         if (replaced_count != block_len_in_page) {
1074                 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
1075                                        "Unable to copy data block,"
1076                                        " data will be lost.");
1077                 *err = -EIO;
1078         }
1079         replaced_count = 0;
1080         goto unlock_pages;
1081 }
1082
1083 /**
1084  * mext_check_arguments - Check whether move extent can be done
1085  *
1086  * @orig_inode:         original inode
1087  * @donor_inode:        donor inode
1088  * @orig_start:         logical start offset in block for orig
1089  * @donor_start:        logical start offset in block for donor
1090  * @len:                the number of blocks to be moved
1091  *
1092  * Check the arguments of ext4_move_extents() whether the files can be
1093  * exchanged with each other.
1094  * Return 0 on success, or a negative error value on failure.
1095  */
1096 static int
1097 mext_check_arguments(struct inode *orig_inode,
1098                      struct inode *donor_inode, __u64 orig_start,
1099                      __u64 donor_start, __u64 *len)
1100 {
1101         ext4_lblk_t orig_blocks, donor_blocks;
1102         unsigned int blkbits = orig_inode->i_blkbits;
1103         unsigned int blocksize = 1 << blkbits;
1104
1105         if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
1106                 ext4_debug("ext4 move extent: suid or sgid is set"
1107                            " to donor file [ino:orig %lu, donor %lu]\n",
1108                            orig_inode->i_ino, donor_inode->i_ino);
1109                 return -EINVAL;
1110         }
1111
1112         if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
1113                 return -EPERM;
1114
1115         /* Ext4 move extent does not support swapfile */
1116         if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
1117                 ext4_debug("ext4 move extent: The argument files should "
1118                         "not be swapfile [ino:orig %lu, donor %lu]\n",
1119                         orig_inode->i_ino, donor_inode->i_ino);
1120                 return -EINVAL;
1121         }
1122
1123         /* Ext4 move extent supports only extent based file */
1124         if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
1125                 ext4_debug("ext4 move extent: orig file is not extents "
1126                         "based file [ino:orig %lu]\n", orig_inode->i_ino);
1127                 return -EOPNOTSUPP;
1128         } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
1129                 ext4_debug("ext4 move extent: donor file is not extents "
1130                         "based file [ino:donor %lu]\n", donor_inode->i_ino);
1131                 return -EOPNOTSUPP;
1132         }
1133
1134         if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1135                 ext4_debug("ext4 move extent: File size is 0 byte\n");
1136                 return -EINVAL;
1137         }
1138
1139         /* Start offset should be same */
1140         if (orig_start != donor_start) {
1141                 ext4_debug("ext4 move extent: orig and donor's start "
1142                         "offset are not same [ino:orig %lu, donor %lu]\n",
1143                         orig_inode->i_ino, donor_inode->i_ino);
1144                 return -EINVAL;
1145         }
1146
1147         if ((orig_start >= EXT_MAX_BLOCKS) ||
1148             (*len > EXT_MAX_BLOCKS) ||
1149             (orig_start + *len >= EXT_MAX_BLOCKS))  {
1150                 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1151                         "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1152                         orig_inode->i_ino, donor_inode->i_ino);
1153                 return -EINVAL;
1154         }
1155
1156         if (orig_inode->i_size > donor_inode->i_size) {
1157                 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1158                 /* TODO: eliminate this artificial restriction */
1159                 if (orig_start >= donor_blocks) {
1160                         ext4_debug("ext4 move extent: orig start offset "
1161                         "[%llu] should be less than donor file blocks "
1162                         "[%u] [ino:orig %lu, donor %lu]\n",
1163                         orig_start, donor_blocks,
1164                         orig_inode->i_ino, donor_inode->i_ino);
1165                         return -EINVAL;
1166                 }
1167
1168                 /* TODO: eliminate this artificial restriction */
1169                 if (orig_start + *len > donor_blocks) {
1170                         ext4_debug("ext4 move extent: End offset [%llu] should "
1171                                 "be less than donor file blocks [%u]."
1172                                 "So adjust length from %llu to %llu "
1173                                 "[ino:orig %lu, donor %lu]\n",
1174                                 orig_start + *len, donor_blocks,
1175                                 *len, donor_blocks - orig_start,
1176                                 orig_inode->i_ino, donor_inode->i_ino);
1177                         *len = donor_blocks - orig_start;
1178                 }
1179         } else {
1180                 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1181                 if (orig_start >= orig_blocks) {
1182                         ext4_debug("ext4 move extent: start offset [%llu] "
1183                                 "should be less than original file blocks "
1184                                 "[%u] [ino:orig %lu, donor %lu]\n",
1185                                  orig_start, orig_blocks,
1186                                 orig_inode->i_ino, donor_inode->i_ino);
1187                         return -EINVAL;
1188                 }
1189
1190                 if (orig_start + *len > orig_blocks) {
1191                         ext4_debug("ext4 move extent: Adjust length "
1192                                 "from %llu to %llu. Because it should be "
1193                                 "less than original file blocks "
1194                                 "[ino:orig %lu, donor %lu]\n",
1195                                 *len, orig_blocks - orig_start,
1196                                 orig_inode->i_ino, donor_inode->i_ino);
1197                         *len = orig_blocks - orig_start;
1198                 }
1199         }
1200
1201         if (!*len) {
1202                 ext4_debug("ext4 move extent: len should not be 0 "
1203                         "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1204                         donor_inode->i_ino);
1205                 return -EINVAL;
1206         }
1207
1208         return 0;
1209 }
1210
1211 /**
1212  * mext_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1213  *
1214  * @inode1:     the inode structure
1215  * @inode2:     the inode structure
1216  *
1217  * Lock two inodes' i_mutex
1218  */
1219 static void
1220 mext_inode_double_lock(struct inode *inode1, struct inode *inode2)
1221 {
1222         BUG_ON(inode1 == inode2);
1223         if (inode1 < inode2) {
1224                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1225                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1226         } else {
1227                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1228                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1229         }
1230 }
1231
1232 /**
1233  * mext_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1234  *
1235  * @inode1:     the inode that is released first
1236  * @inode2:     the inode that is released second
1237  *
1238  */
1239
1240 static void
1241 mext_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1242 {
1243         mutex_unlock(&inode1->i_mutex);
1244         mutex_unlock(&inode2->i_mutex);
1245 }
1246
1247 /**
1248  * ext4_move_extents - Exchange the specified range of a file
1249  *
1250  * @o_filp:             file structure of the original file
1251  * @d_filp:             file structure of the donor file
1252  * @orig_start:         start offset in block for orig
1253  * @donor_start:        start offset in block for donor
1254  * @len:                the number of blocks to be moved
1255  * @moved_len:          moved block length
1256  *
1257  * This function returns 0 and moved block length is set in moved_len
1258  * if succeed, otherwise returns error value.
1259  *
1260  * Note: ext4_move_extents() proceeds the following order.
1261  * 1:ext4_move_extents() calculates the last block number of moving extent
1262  *   function by the start block number (orig_start) and the number of blocks
1263  *   to be moved (len) specified as arguments.
1264  *   If the {orig, donor}_start points a hole, the extent's start offset
1265  *   pointed by ext_cur (current extent), holecheck_path, orig_path are set
1266  *   after hole behind.
1267  * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1268  *   or the ext_cur exceeds the block_end which is last logical block number.
1269  * 3:To get the length of continues area, call mext_next_extent()
1270  *   specified with the ext_cur (initial value is holecheck_path) re-cursive,
1271  *   until find un-continuous extent, the start logical block number exceeds
1272  *   the block_end or the extent points to the last extent.
1273  * 4:Exchange the original inode data with donor inode data
1274  *   from orig_page_offset to seq_end_page.
1275  *   The start indexes of data are specified as arguments.
1276  *   That of the original inode is orig_page_offset,
1277  *   and the donor inode is also orig_page_offset
1278  *   (To easily handle blocksize != pagesize case, the offset for the
1279  *   donor inode is block unit).
1280  * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1281  *   then returns to step 2.
1282  * 6:Release holecheck_path, orig_path and set the len to moved_len
1283  *   which shows the number of moved blocks.
1284  *   The moved_len is useful for the command to calculate the file offset
1285  *   for starting next move extent ioctl.
1286  * 7:Return 0 on success, or a negative error value on failure.
1287  */
1288 int
1289 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1290                  __u64 orig_start, __u64 donor_start, __u64 len,
1291                  __u64 *moved_len)
1292 {
1293         struct inode *orig_inode = file_inode(o_filp);
1294         struct inode *donor_inode = file_inode(d_filp);
1295         struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1296         struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1297         ext4_lblk_t block_start = orig_start;
1298         ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1299         ext4_lblk_t rest_blocks;
1300         pgoff_t orig_page_offset = 0, seq_end_page;
1301         int ret, depth, last_extent = 0;
1302         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1303         int data_offset_in_page;
1304         int block_len_in_page;
1305         int uninit;
1306
1307         if (orig_inode->i_sb != donor_inode->i_sb) {
1308                 ext4_debug("ext4 move extent: The argument files "
1309                         "should be in same FS [ino:orig %lu, donor %lu]\n",
1310                         orig_inode->i_ino, donor_inode->i_ino);
1311                 return -EINVAL;
1312         }
1313
1314         /* orig and donor should be different inodes */
1315         if (orig_inode == donor_inode) {
1316                 ext4_debug("ext4 move extent: The argument files should not "
1317                         "be same inode [ino:orig %lu, donor %lu]\n",
1318                         orig_inode->i_ino, donor_inode->i_ino);
1319                 return -EINVAL;
1320         }
1321
1322         /* Regular file check */
1323         if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1324                 ext4_debug("ext4 move extent: The argument files should be "
1325                         "regular file [ino:orig %lu, donor %lu]\n",
1326                         orig_inode->i_ino, donor_inode->i_ino);
1327                 return -EINVAL;
1328         }
1329         /* TODO: This is non obvious task to swap blocks for inodes with full
1330            jornaling enabled */
1331         if (ext4_should_journal_data(orig_inode) ||
1332             ext4_should_journal_data(donor_inode)) {
1333                 return -EINVAL;
1334         }
1335         /* Protect orig and donor inodes against a truncate */
1336         mext_inode_double_lock(orig_inode, donor_inode);
1337
1338         /* Wait for all existing dio workers */
1339         ext4_inode_block_unlocked_dio(orig_inode);
1340         ext4_inode_block_unlocked_dio(donor_inode);
1341         inode_dio_wait(orig_inode);
1342         inode_dio_wait(donor_inode);
1343
1344         /* Protect extent tree against block allocations via delalloc */
1345         double_down_write_data_sem(orig_inode, donor_inode);
1346         /* Check the filesystem environment whether move_extent can be done */
1347         ret = mext_check_arguments(orig_inode, donor_inode, orig_start,
1348                                     donor_start, &len);
1349         if (ret)
1350                 goto out;
1351
1352         file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1353         block_end = block_start + len - 1;
1354         if (file_end < block_end)
1355                 len -= block_end - file_end;
1356
1357         ret = get_ext_path(orig_inode, block_start, &orig_path);
1358         if (ret)
1359                 goto out;
1360
1361         /* Get path structure to check the hole */
1362         ret = get_ext_path(orig_inode, block_start, &holecheck_path);
1363         if (ret)
1364                 goto out;
1365
1366         depth = ext_depth(orig_inode);
1367         ext_cur = holecheck_path[depth].p_ext;
1368
1369         /*
1370          * Get proper starting location of block replacement if block_start was
1371          * within the hole.
1372          */
1373         if (le32_to_cpu(ext_cur->ee_block) +
1374                 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1375                 /*
1376                  * The hole exists between extents or the tail of
1377                  * original file.
1378                  */
1379                 last_extent = mext_next_extent(orig_inode,
1380                                         holecheck_path, &ext_cur);
1381                 if (last_extent < 0) {
1382                         ret = last_extent;
1383                         goto out;
1384                 }
1385                 last_extent = mext_next_extent(orig_inode, orig_path,
1386                                                         &ext_dummy);
1387                 if (last_extent < 0) {
1388                         ret = last_extent;
1389                         goto out;
1390                 }
1391                 seq_start = le32_to_cpu(ext_cur->ee_block);
1392         } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1393                 /* The hole exists at the beginning of original file. */
1394                 seq_start = le32_to_cpu(ext_cur->ee_block);
1395         else
1396                 seq_start = block_start;
1397
1398         /* No blocks within the specified range. */
1399         if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1400                 ext4_debug("ext4 move extent: The specified range of file "
1401                                                         "may be the hole\n");
1402                 ret = -EINVAL;
1403                 goto out;
1404         }
1405
1406         /* Adjust start blocks */
1407         add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1408                          ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1409                      max(le32_to_cpu(ext_cur->ee_block), block_start);
1410
1411         while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1412                 seq_blocks += add_blocks;
1413
1414                 /* Adjust tail blocks */
1415                 if (seq_start + seq_blocks - 1 > block_end)
1416                         seq_blocks = block_end - seq_start + 1;
1417
1418                 ext_prev = ext_cur;
1419                 last_extent = mext_next_extent(orig_inode, holecheck_path,
1420                                                 &ext_cur);
1421                 if (last_extent < 0) {
1422                         ret = last_extent;
1423                         break;
1424                 }
1425                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1426
1427                 /*
1428                  * Extend the length of contiguous block (seq_blocks)
1429                  * if extents are contiguous.
1430                  */
1431                 if (ext4_can_extents_be_merged(orig_inode,
1432                                                ext_prev, ext_cur) &&
1433                     block_end >= le32_to_cpu(ext_cur->ee_block) &&
1434                     !last_extent)
1435                         continue;
1436
1437                 /* Is original extent is uninitialized */
1438                 uninit = ext4_ext_is_uninitialized(ext_prev);
1439
1440                 data_offset_in_page = seq_start % blocks_per_page;
1441
1442                 /*
1443                  * Calculate data blocks count that should be swapped
1444                  * at the first page.
1445                  */
1446                 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1447                         /* Swapped blocks are across pages */
1448                         block_len_in_page =
1449                                         blocks_per_page - data_offset_in_page;
1450                 } else {
1451                         /* Swapped blocks are in a page */
1452                         block_len_in_page = seq_blocks;
1453                 }
1454
1455                 orig_page_offset = seq_start >>
1456                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1457                 seq_end_page = (seq_start + seq_blocks - 1) >>
1458                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1459                 seq_start = le32_to_cpu(ext_cur->ee_block);
1460                 rest_blocks = seq_blocks;
1461
1462                 /*
1463                  * Up semaphore to avoid following problems:
1464                  * a. transaction deadlock among ext4_journal_start,
1465                  *    ->write_begin via pagefault, and jbd2_journal_commit
1466                  * b. racing with ->readpage, ->write_begin, and ext4_get_block
1467                  *    in move_extent_per_page
1468                  */
1469                 double_up_write_data_sem(orig_inode, donor_inode);
1470
1471                 while (orig_page_offset <= seq_end_page) {
1472
1473                         /* Swap original branches with new branches */
1474                         block_len_in_page = move_extent_per_page(
1475                                                 o_filp, donor_inode,
1476                                                 orig_page_offset,
1477                                                 data_offset_in_page,
1478                                                 block_len_in_page, uninit,
1479                                                 &ret);
1480
1481                         /* Count how many blocks we have exchanged */
1482                         *moved_len += block_len_in_page;
1483                         if (ret < 0)
1484                                 break;
1485                         if (*moved_len > len) {
1486                                 EXT4_ERROR_INODE(orig_inode,
1487                                         "We replaced blocks too much! "
1488                                         "sum of replaced: %llu requested: %llu",
1489                                         *moved_len, len);
1490                                 ret = -EIO;
1491                                 break;
1492                         }
1493
1494                         orig_page_offset++;
1495                         data_offset_in_page = 0;
1496                         rest_blocks -= block_len_in_page;
1497                         if (rest_blocks > blocks_per_page)
1498                                 block_len_in_page = blocks_per_page;
1499                         else
1500                                 block_len_in_page = rest_blocks;
1501                 }
1502
1503                 double_down_write_data_sem(orig_inode, donor_inode);
1504                 if (ret < 0)
1505                         break;
1506
1507                 /* Decrease buffer counter */
1508                 if (holecheck_path)
1509                         ext4_ext_drop_refs(holecheck_path);
1510                 ret = get_ext_path(orig_inode, seq_start, &holecheck_path);
1511                 if (ret)
1512                         break;
1513                 depth = holecheck_path->p_depth;
1514
1515                 /* Decrease buffer counter */
1516                 if (orig_path)
1517                         ext4_ext_drop_refs(orig_path);
1518                 ret = get_ext_path(orig_inode, seq_start, &orig_path);
1519                 if (ret)
1520                         break;
1521
1522                 ext_cur = holecheck_path[depth].p_ext;
1523                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1524                 seq_blocks = 0;
1525
1526         }
1527 out:
1528         if (*moved_len) {
1529                 ext4_discard_preallocations(orig_inode);
1530                 ext4_discard_preallocations(donor_inode);
1531         }
1532
1533         if (orig_path) {
1534                 ext4_ext_drop_refs(orig_path);
1535                 kfree(orig_path);
1536         }
1537         if (holecheck_path) {
1538                 ext4_ext_drop_refs(holecheck_path);
1539                 kfree(holecheck_path);
1540         }
1541         double_up_write_data_sem(orig_inode, donor_inode);
1542         ext4_inode_resume_unlocked_dio(orig_inode);
1543         ext4_inode_resume_unlocked_dio(donor_inode);
1544         mext_inode_double_unlock(orig_inode, donor_inode);
1545
1546         return ret;
1547 }