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, EXT4_EX_NOCACHE);
  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  * ext4_double_down_write_data_sem - Acquire two inodes' write lock
 148  *                                   of i_data_sem
 149  *
 150  * Acquire write lock of i_data_sem of the two inodes
 151  */
 152 void
 153 ext4_double_down_write_data_sem(struct inode *first, struct inode *second)
 154 {
 155         if (first < second) {
 156                 down_write(&EXT4_I(first)->i_data_sem);
 157                 down_write_nested(&EXT4_I(second)->i_data_sem, SINGLE_DEPTH_NESTING);
 158         } else {
 159                 down_write(&EXT4_I(second)->i_data_sem);
 160                 down_write_nested(&EXT4_I(first)->i_data_sem, SINGLE_DEPTH_NESTING);
 161
 162         }
 163 }
 164
 165 /**
 166  * ext4_double_up_write_data_sem - Release two inodes' write lock of i_data_sem
 167  *
 168  * @orig_inode:         original inode structure to be released its lock first
 169  * @donor_inode:        donor inode structure to be released its lock second
 170  * Release write lock of i_data_sem of two inodes (orig and donor).
 171  */
 172 void
 173 ext4_double_up_write_data_sem(struct inode *orig_inode,
 174                               struct inode *donor_inode)
 175 {
 176         up_write(&EXT4_I(orig_inode)->i_data_sem);
 177         up_write(&EXT4_I(donor_inode)->i_data_sem);
 178 }
 179
 180 /**
 181  * mext_insert_across_blocks - Insert extents across leaf block
 182  *
 183  * @handle:             journal handle
 184  * @orig_inode:         original inode
 185  * @o_start:            first original extent to be changed
 186  * @o_end:              last original extent to be changed
 187  * @start_ext:          first new extent to be inserted
 188  * @new_ext:            middle of new extent to be inserted
 189  * @end_ext:            last new extent to be inserted
 190  *
 191  * Allocate a new leaf block and insert extents into it. Return 0 on success,
 192  * or a negative error value on failure.
 193  */
 194 static int
 195 mext_insert_across_blocks(handle_t *handle, struct inode *orig_inode,
 196                 struct ext4_extent *o_start, struct ext4_extent *o_end,
 197                 struct ext4_extent *start_ext, struct ext4_extent *new_ext,
 198                 struct ext4_extent *end_ext)
 199 {
 200         struct ext4_ext_path *orig_path = NULL;
 201         ext4_lblk_t eblock = 0;
 202         int new_flag = 0;
 203         int end_flag = 0;
 204         int err = 0;
 205
 206         if (start_ext->ee_len && new_ext->ee_len && end_ext->ee_len) {
 207                 if (o_start == o_end) {
 208
 209                         /*       start_ext   new_ext    end_ext
 210                          * donor |---------|-----------|--------|
 211                          * orig  |------------------------------|
 212                          */
 213                         end_flag = 1;
 214                 } else {
 215
 216                         /*       start_ext   new_ext   end_ext
 217                          * donor |---------|----------|---------|
 218                          * orig  |---------------|--------------|
 219                          */
 220                         o_end->ee_block = end_ext->ee_block;
 221                         o_end->ee_len = end_ext->ee_len;
 222                         ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
 223                 }
 224
 225                 o_start->ee_len = start_ext->ee_len;
 226                 eblock = le32_to_cpu(start_ext->ee_block);
 227                 new_flag = 1;
 228
 229         } else if (start_ext->ee_len && new_ext->ee_len &&
 230                    !end_ext->ee_len && o_start == o_end) {
 231
 232                 /*       start_ext      new_ext
 233                  * donor |--------------|---------------|
 234                  * orig  |------------------------------|
 235                  */
 236                 o_start->ee_len = start_ext->ee_len;
 237                 eblock = le32_to_cpu(start_ext->ee_block);
 238                 new_flag = 1;
 239
 240         } else if (!start_ext->ee_len && new_ext->ee_len &&
 241                    end_ext->ee_len && o_start == o_end) {
 242
 243                 /*        new_ext       end_ext
 244                  * donor |--------------|---------------|
 245                  * orig  |------------------------------|
 246                  */
 247                 o_end->ee_block = end_ext->ee_block;
 248                 o_end->ee_len = end_ext->ee_len;
 249                 ext4_ext_store_pblock(o_end, ext4_ext_pblock(end_ext));
 250
 251                 /*
 252                  * Set 0 to the extent block if new_ext was
 253                  * the first block.
 254                  */
 255                 if (new_ext->ee_block)
 256                         eblock = le32_to_cpu(new_ext->ee_block);
 257
 258                 new_flag = 1;
 259         } else {
 260                 ext4_debug("ext4 move extent: Unexpected insert case\n");
 261                 return -EIO;
 262         }
 263
 264         if (new_flag) {
 265                 err = get_ext_path(orig_inode, eblock, &orig_path);
 266                 if (err)
 267                         goto out;
 268
 269                 if (ext4_ext_insert_extent(handle, orig_inode,
 270                                         orig_path, new_ext, 0))
 271                         goto out;
 272         }
 273
 274         if (end_flag) {
 275                 err = get_ext_path(orig_inode,
 276                                 le32_to_cpu(end_ext->ee_block) - 1, &orig_path);
 277                 if (err)
 278                         goto out;
 279
 280                 if (ext4_ext_insert_extent(handle, orig_inode,
 281                                            orig_path, end_ext, 0))
 282                         goto out;
 283         }
 284 out:
 285         if (orig_path) {
 286                 ext4_ext_drop_refs(orig_path);
 287                 kfree(orig_path);
 288         }
 289
 290         return err;
 291
 292 }
 293
 294 /**
 295  * mext_insert_inside_block - Insert new extent to the extent block
 296  *
 297  * @o_start:            first original extent to be moved
 298  * @o_end:              last original extent to be moved
 299  * @start_ext:          first new extent to be inserted
 300  * @new_ext:            middle of new extent to be inserted
 301  * @end_ext:            last new extent to be inserted
 302  * @eh:                 extent header of target leaf block
 303  * @range_to_move:      used to decide how to insert extent
 304  *
 305  * Insert extents into the leaf block. The extent (@o_start) is overwritten
 306  * by inserted extents.
 307  */
 308 static void
 309 mext_insert_inside_block(struct ext4_extent *o_start,
 310                               struct ext4_extent *o_end,
 311                               struct ext4_extent *start_ext,
 312                               struct ext4_extent *new_ext,
 313                               struct ext4_extent *end_ext,
 314                               struct ext4_extent_header *eh,
 315                               int range_to_move)
 316 {
 317         int i = 0;
 318         unsigned long len;
 319
 320         /* Move the existing extents */
 321         if (range_to_move && o_end < EXT_LAST_EXTENT(eh)) {
 322                 len = (unsigned long)(EXT_LAST_EXTENT(eh) + 1) -
 323                         (unsigned long)(o_end + 1);
 324                 memmove(o_end + 1 + range_to_move, o_end + 1, len);
 325         }
 326
 327         /* Insert start entry */
 328         if (start_ext->ee_len)
 329                 o_start[i++].ee_len = start_ext->ee_len;
 330
 331         /* Insert new entry */
 332         if (new_ext->ee_len) {
 333                 o_start[i] = *new_ext;
 334                 ext4_ext_store_pblock(&o_start[i++], ext4_ext_pblock(new_ext));
 335         }
 336
 337         /* Insert end entry */
 338         if (end_ext->ee_len)
 339                 o_start[i] = *end_ext;
 340
 341         /* Increment the total entries counter on the extent block */
 342         le16_add_cpu(&eh->eh_entries, range_to_move);
 343 }
 344
 345 /**
 346  * mext_insert_extents - Insert new extent
 347  *
 348  * @handle:     journal handle
 349  * @orig_inode: original inode
 350  * @orig_path:  path indicates first extent to be changed
 351  * @o_start:    first original extent to be changed
 352  * @o_end:      last original extent to be changed
 353  * @start_ext:  first new extent to be inserted
 354  * @new_ext:    middle of new extent to be inserted
 355  * @end_ext:    last new extent to be inserted
 356  *
 357  * Call the function to insert extents. If we cannot add more extents into
 358  * the leaf block, we call mext_insert_across_blocks() to create a
 359  * new leaf block. Otherwise call mext_insert_inside_block(). Return 0
 360  * on success, or a negative error value on failure.
 361  */
 362 static int
 363 mext_insert_extents(handle_t *handle, struct inode *orig_inode,
 364                          struct ext4_ext_path *orig_path,
 365                          struct ext4_extent *o_start,
 366                          struct ext4_extent *o_end,
 367                          struct ext4_extent *start_ext,
 368                          struct ext4_extent *new_ext,
 369                          struct ext4_extent *end_ext)
 370 {
 371         struct  ext4_extent_header *eh;
 372         unsigned long need_slots, slots_range;
 373         int     range_to_move, depth, ret;
 374
 375         /*
 376          * The extents need to be inserted
 377          * start_extent + new_extent + end_extent.
 378          */
 379         need_slots = (start_ext->ee_len ? 1 : 0) + (end_ext->ee_len ? 1 : 0) +
 380                 (new_ext->ee_len ? 1 : 0);
 381
 382         /* The number of slots between start and end */
 383         slots_range = ((unsigned long)(o_end + 1) - (unsigned long)o_start + 1)
 384                 / sizeof(struct ext4_extent);
 385
 386         /* Range to move the end of extent */
 387         range_to_move = need_slots - slots_range;
 388         depth = orig_path->p_depth;
 389         orig_path += depth;
 390         eh = orig_path->p_hdr;
 391
 392         if (depth) {
 393                 /* Register to journal */
 394                 ret = ext4_journal_get_write_access(handle, orig_path->p_bh);
 395                 if (ret)
 396                         return ret;
 397         }
 398
 399         /* Expansion */
 400         if (range_to_move > 0 &&
 401                 (range_to_move > le16_to_cpu(eh->eh_max)
 402                         - le16_to_cpu(eh->eh_entries))) {
 403
 404                 ret = mext_insert_across_blocks(handle, orig_inode, o_start,
 405                                         o_end, start_ext, new_ext, end_ext);
 406                 if (ret < 0)
 407                         return ret;
 408         } else
 409                 mext_insert_inside_block(o_start, o_end, start_ext, new_ext,
 410                                                 end_ext, eh, range_to_move);
 411
 412         return ext4_ext_dirty(handle, orig_inode, orig_path);
 413 }
 414
 415 /**
 416  * mext_leaf_block - Move one leaf extent block into the inode.
 417  *
 418  * @handle:             journal handle
 419  * @orig_inode:         original inode
 420  * @orig_path:          path indicates first extent to be changed
 421  * @dext:               donor extent
 422  * @from:               start offset on the target file
 423  *
 424  * In order to insert extents into the leaf block, we must divide the extent
 425  * in the leaf block into three extents. The one is located to be inserted
 426  * extents, and the others are located around it.
 427  *
 428  * Therefore, this function creates structures to save extents of the leaf
 429  * block, and inserts extents by calling mext_insert_extents() with
 430  * created extents. Return 0 on success, or a negative error value on failure.
 431  */
 432 static int
 433 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
 434                      struct ext4_ext_path *orig_path, struct ext4_extent *dext,
 435                      ext4_lblk_t *from)
 436 {
 437         struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
 438         struct ext4_extent new_ext, start_ext, end_ext;
 439         ext4_lblk_t new_ext_end;
 440         int oext_alen, new_ext_alen, end_ext_alen;
 441         int depth = ext_depth(orig_inode);
 442         int ret;
 443
 444         start_ext.ee_block = end_ext.ee_block = 0;
 445         o_start = o_end = oext = orig_path[depth].p_ext;
 446         oext_alen = ext4_ext_get_actual_len(oext);
 447         start_ext.ee_len = end_ext.ee_len = 0;
 448
 449         new_ext.ee_block = cpu_to_le32(*from);
 450         ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
 451         new_ext.ee_len = dext->ee_len;
 452         new_ext_alen = ext4_ext_get_actual_len(&new_ext);
 453         new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
 454
 455         /*
 456          * Case: original extent is first
 457          * oext      |--------|
 458          * new_ext      |--|
 459          * start_ext |--|
 460          */
 461         if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
 462                 le32_to_cpu(new_ext.ee_block) <
 463                 le32_to_cpu(oext->ee_block) + oext_alen) {
 464                 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
 465                                                le32_to_cpu(oext->ee_block));
 466                 start_ext.ee_block = oext->ee_block;
 467                 copy_extent_status(oext, &start_ext);
 468         } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
 469                 prev_ext = oext - 1;
 470                 /*
 471                  * We can merge new_ext into previous extent,
 472                  * if these are contiguous and same extent type.
 473                  */
 474                 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
 475                                                &new_ext)) {
 476                         o_start = prev_ext;
 477                         start_ext.ee_len = cpu_to_le16(
 478                                 ext4_ext_get_actual_len(prev_ext) +
 479                                 new_ext_alen);
 480                         start_ext.ee_block = oext->ee_block;
 481                         copy_extent_status(prev_ext, &start_ext);
 482                         new_ext.ee_len = 0;
 483                 }
 484         }
 485
 486         /*
 487          * Case: new_ext_end must be less than oext
 488          * oext      |-----------|
 489          * new_ext       |-------|
 490          */
 491         if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
 492                 EXT4_ERROR_INODE(orig_inode,
 493                         "new_ext_end(%u) should be less than or equal to "
 494                         "oext->ee_block(%u) + oext_alen(%d) - 1",
 495                         new_ext_end, le32_to_cpu(oext->ee_block),
 496                         oext_alen);
 497                 ret = -EIO;
 498                 goto out;
 499         }
 500
 501         /*
 502          * Case: new_ext is smaller than original extent
 503          * oext    |---------------|
 504          * new_ext |-----------|
 505          * end_ext             |---|
 506          */
 507         if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
 508                 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
 509                 end_ext.ee_len =
 510                         cpu_to_le16(le32_to_cpu(oext->ee_block) +
 511                         oext_alen - 1 - new_ext_end);
 512                 copy_extent_status(oext, &end_ext);
 513                 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
 514                 ext4_ext_store_pblock(&end_ext,
 515                         (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
 516                 end_ext.ee_block =
 517                         cpu_to_le32(le32_to_cpu(o_end->ee_block) +
 518                         oext_alen - end_ext_alen);
 519         }
 520
 521         ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
 522                                 o_end, &start_ext, &new_ext, &end_ext);
 523 out:
 524         return ret;
 525 }
 526
 527 /**
 528  * mext_calc_swap_extents - Calculate extents for extent swapping.
 529  *
 530  * @tmp_dext:           the extent that will belong to the original inode
 531  * @tmp_oext:           the extent that will belong to the donor inode
 532  * @orig_off:           block offset of original inode
 533  * @donor_off:          block offset of donor inode
 534  * @max_count:          the maximum length of extents
 535  *
 536  * Return 0 on success, or a negative error value on failure.
 537  */
 538 static int
 539 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
 540                               struct ext4_extent *tmp_oext,
 541                               ext4_lblk_t orig_off, ext4_lblk_t donor_off,
 542                               ext4_lblk_t max_count)
 543 {
 544         ext4_lblk_t diff, orig_diff;
 545         struct ext4_extent dext_old, oext_old;
 546
 547         BUG_ON(orig_off != donor_off);
 548
 549         /* original and donor extents have to cover the same block offset */
 550         if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
 551             le32_to_cpu(tmp_oext->ee_block) +
 552                         ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
 553                 return -ENODATA;
 554
 555         if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
 556             le32_to_cpu(tmp_dext->ee_block) +
 557                         ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
 558                 return -ENODATA;
 559
 560         dext_old = *tmp_dext;
 561         oext_old = *tmp_oext;
 562
 563         /* When tmp_dext is too large, pick up the target range. */
 564         diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
 565
 566         ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
 567         le32_add_cpu(&tmp_dext->ee_block, diff);
 568         le16_add_cpu(&tmp_dext->ee_len, -diff);
 569
 570         if (max_count < ext4_ext_get_actual_len(tmp_dext))
 571                 tmp_dext->ee_len = cpu_to_le16(max_count);
 572
 573         orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
 574         ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
 575
 576         /* Adjust extent length if donor extent is larger than orig */
 577         if (ext4_ext_get_actual_len(tmp_dext) >
 578             ext4_ext_get_actual_len(tmp_oext) - orig_diff)
 579                 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
 580                                                 orig_diff);
 581
 582         tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
 583
 584         copy_extent_status(&oext_old, tmp_dext);
 585         copy_extent_status(&dext_old, tmp_oext);
 586
 587         return 0;
 588 }
 589
 590 /**
 591  * mext_check_coverage - Check that all extents in range has the same type
 592  *
 593  * @inode:              inode in question
 594  * @from:               block offset of inode
 595  * @count:              block count to be checked
 596  * @uninit:             extents expected to be uninitialized
 597  * @err:                pointer to save error value
 598  *
 599  * Return 1 if all extents in range has expected type, and zero otherwise.
 600  */
 601 static int
 602 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
 603                           int uninit, int *err)
 604 {
 605         struct ext4_ext_path *path = NULL;
 606         struct ext4_extent *ext;
 607         int ret = 0;
 608         ext4_lblk_t last = from + count;
 609         while (from < last) {
 610                 *err = get_ext_path(inode, from, &path);
 611                 if (*err)
 612                         goto out;
 613                 ext = path[ext_depth(inode)].p_ext;
 614                 if (uninit != ext4_ext_is_uninitialized(ext))
 615                         goto out;
 616                 from += ext4_ext_get_actual_len(ext);
 617                 ext4_ext_drop_refs(path);
 618         }
 619         ret = 1;
 620 out:
 621         if (path) {
 622                 ext4_ext_drop_refs(path);
 623                 kfree(path);
 624         }
 625         return ret;
 626 }
 627
 628 /**
 629  * mext_replace_branches - Replace original extents with new extents
 630  *
 631  * @handle:             journal handle
 632  * @orig_inode:         original inode
 633  * @donor_inode:        donor inode
 634  * @from:               block offset of orig_inode
 635  * @count:              block count to be replaced
 636  * @err:                pointer to save return value
 637  *
 638  * Replace original inode extents and donor inode extents page by page.
 639  * We implement this replacement in the following three steps:
 640  * 1. Save the block information of original and donor inodes into
 641  *    dummy extents.
 642  * 2. Change the block information of original inode to point at the
 643  *    donor inode blocks.
 644  * 3. Change the block information of donor inode to point at the saved
 645  *    original inode blocks in the dummy extents.
 646  *
 647  * Return replaced block count.
 648  */
 649 static int
 650 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
 651                            struct inode *donor_inode, ext4_lblk_t from,
 652                            ext4_lblk_t count, int *err)
 653 {
 654         struct ext4_ext_path *orig_path = NULL;
 655         struct ext4_ext_path *donor_path = NULL;
 656         struct ext4_extent *oext, *dext;
 657         struct ext4_extent tmp_dext, tmp_oext;
 658         ext4_lblk_t orig_off = from, donor_off = from;
 659         int depth;
 660         int replaced_count = 0;
 661         int dext_alen;
 662
 663         *err = ext4_es_remove_extent(orig_inode, from, count);
 664         if (*err)
 665                 goto out;
 666
 667         *err = ext4_es_remove_extent(donor_inode, from, count);
 668         if (*err)
 669                 goto out;
 670
 671         /* Get the original extent for the block "orig_off" */
 672         *err = get_ext_path(orig_inode, orig_off, &orig_path);
 673         if (*err)
 674                 goto out;
 675
 676         /* Get the donor extent for the head */
 677         *err = get_ext_path(donor_inode, donor_off, &donor_path);
 678         if (*err)
 679                 goto out;
 680         depth = ext_depth(orig_inode);
 681         oext = orig_path[depth].p_ext;
 682         tmp_oext = *oext;
 683
 684         depth = ext_depth(donor_inode);
 685         dext = donor_path[depth].p_ext;
 686         if (unlikely(!dext))
 687                 goto missing_donor_extent;
 688         tmp_dext = *dext;
 689
 690         *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 691                                       donor_off, count);
 692         if (*err)
 693                 goto out;
 694
 695         /* Loop for the donor extents */
 696         while (1) {
 697                 /* The extent for donor must be found. */
 698                 if (unlikely(!dext)) {
 699                 missing_donor_extent:
 700                         EXT4_ERROR_INODE(donor_inode,
 701                                    "The extent for donor must be found");
 702                         *err = -EIO;
 703                         goto out;
 704                 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
 705                         EXT4_ERROR_INODE(donor_inode,
 706                                 "Donor offset(%u) and the first block of donor "
 707                                 "extent(%u) should be equal",
 708                                 donor_off,
 709                                 le32_to_cpu(tmp_dext.ee_block));
 710                         *err = -EIO;
 711                         goto out;
 712                 }
 713
 714                 /* Set donor extent to orig extent */
 715                 *err = mext_leaf_block(handle, orig_inode,
 716                                            orig_path, &tmp_dext, &orig_off);
 717                 if (*err)
 718                         goto out;
 719
 720                 /* Set orig extent to donor extent */
 721                 *err = mext_leaf_block(handle, donor_inode,
 722                                            donor_path, &tmp_oext, &donor_off);
 723                 if (*err)
 724                         goto out;
 725
 726                 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
 727                 replaced_count += dext_alen;
 728                 donor_off += dext_alen;
 729                 orig_off += dext_alen;
 730
 731                 BUG_ON(replaced_count > count);
 732                 /* Already moved the expected blocks */
 733                 if (replaced_count >= count)
 734                         break;
 735
 736                 if (orig_path)
 737                         ext4_ext_drop_refs(orig_path);
 738                 *err = get_ext_path(orig_inode, orig_off, &orig_path);
 739                 if (*err)
 740                         goto out;
 741                 depth = ext_depth(orig_inode);
 742                 oext = orig_path[depth].p_ext;
 743                 tmp_oext = *oext;
 744
 745                 if (donor_path)
 746                         ext4_ext_drop_refs(donor_path);
 747                 *err = get_ext_path(donor_inode, donor_off, &donor_path);
 748                 if (*err)
 749                         goto out;
 750                 depth = ext_depth(donor_inode);
 751                 dext = donor_path[depth].p_ext;
 752                 tmp_dext = *dext;
 753
 754                 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 755                                            donor_off, count - replaced_count);
 756                 if (*err)
 757                         goto out;
 758         }
 759
 760 out:
 761         if (orig_path) {
 762                 ext4_ext_drop_refs(orig_path);
 763                 kfree(orig_path);
 764         }
 765         if (donor_path) {
 766                 ext4_ext_drop_refs(donor_path);
 767                 kfree(donor_path);
 768         }
 769
 770         return replaced_count;
 771 }
 772
 773 /**
 774  * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
 775  *
 776  * @inode1:     the inode structure
 777  * @inode2:     the inode structure
 778  * @index:      page index
 779  * @page:       result page vector
 780  *
 781  * Grab two locked pages for inode's by inode order
 782  */
 783 static int
 784 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
 785                       pgoff_t index, struct page *page[2])
 786 {
 787         struct address_space *mapping[2];
 788         unsigned fl = AOP_FLAG_NOFS;
 789
 790         BUG_ON(!inode1 || !inode2);
 791         if (inode1 < inode2) {
 792                 mapping[0] = inode1->i_mapping;
 793                 mapping[1] = inode2->i_mapping;
 794         } else {
 795                 mapping[0] = inode2->i_mapping;
 796                 mapping[1] = inode1->i_mapping;
 797         }
 798
 799         page[0] = grab_cache_page_write_begin(mapping[0], index, fl);
 800         if (!page[0])
 801                 return -ENOMEM;
 802
 803         page[1] = grab_cache_page_write_begin(mapping[1], index, fl);
 804         if (!page[1]) {
 805                 unlock_page(page[0]);
 806                 page_cache_release(page[0]);
 807                 return -ENOMEM;
 808         }
 809         /*
 810          * grab_cache_page_write_begin() may not wait on page's writeback if
 811          * BDI not demand that. But it is reasonable to be very conservative
 812          * here and explicitly wait on page's writeback
 813          */
 814         wait_on_page_writeback(page[0]);
 815         wait_on_page_writeback(page[1]);
 816         if (inode1 > inode2) {
 817                 struct page *tmp;
 818                 tmp = page[0];
 819                 page[0] = page[1];
 820                 page[1] = tmp;
 821         }
 822         return 0;
 823 }
 824
 825 /* Force page buffers uptodate w/o dropping page's lock */
 826 static int
 827 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
 828 {
 829         struct inode *inode = page->mapping->host;
 830         sector_t block;
 831         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 832         unsigned int blocksize, block_start, block_end;
 833         int i, err,  nr = 0, partial = 0;
 834         BUG_ON(!PageLocked(page));
 835         BUG_ON(PageWriteback(page));
 836
 837         if (PageUptodate(page))
 838                 return 0;
 839
 840         blocksize = 1 << inode->i_blkbits;
 841         if (!page_has_buffers(page))
 842                 create_empty_buffers(page, blocksize, 0);
 843
 844         head = page_buffers(page);
 845         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 846         for (bh = head, block_start = 0; bh != head || !block_start;
 847              block++, block_start = block_end, bh = bh->b_this_page) {
 848                 block_end = block_start + blocksize;
 849                 if (block_end <= from || block_start >= to) {
 850                         if (!buffer_uptodate(bh))
 851                                 partial = 1;
 852                         continue;
 853                 }
 854                 if (buffer_uptodate(bh))
 855                         continue;
 856                 if (!buffer_mapped(bh)) {
 857                         err = ext4_get_block(inode, block, bh, 0);
 858                         if (err) {
 859                                 SetPageError(page);
 860                                 return err;
 861                         }
 862                         if (!buffer_mapped(bh)) {
 863                                 zero_user(page, block_start, blocksize);
 864                                 if (!err)
 865                                         set_buffer_uptodate(bh);
 866                                 continue;
 867                         }
 868                 }
 869                 BUG_ON(nr >= MAX_BUF_PER_PAGE);
 870                 arr[nr++] = bh;
 871         }
 872         /* No io required */
 873         if (!nr)
 874                 goto out;
 875
 876         for (i = 0; i < nr; i++) {
 877                 bh = arr[i];
 878                 if (!bh_uptodate_or_lock(bh)) {
 879                         err = bh_submit_read(bh);
 880                         if (err)
 881                                 return err;
 882                 }
 883         }
 884 out:
 885         if (!partial)
 886                 SetPageUptodate(page);
 887         return 0;
 888 }
 889
 890 /**
 891  * move_extent_per_page - Move extent data per page
 892  *
 893  * @o_filp:                     file structure of original file
 894  * @donor_inode:                donor inode
 895  * @orig_page_offset:           page index on original file
 896  * @data_offset_in_page:        block index where data swapping starts
 897  * @block_len_in_page:          the number of blocks to be swapped
 898  * @uninit:                     orig extent is uninitialized or not
 899  * @err:                        pointer to save return value
 900  *
 901  * Save the data in original inode blocks and replace original inode extents
 902  * with donor inode extents by calling mext_replace_branches().
 903  * Finally, write out the saved data in new original inode blocks. Return
 904  * replaced block count.
 905  */
 906 static int
 907 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
 908                   pgoff_t orig_page_offset, int data_offset_in_page,
 909                   int block_len_in_page, int uninit, int *err)
 910 {
 911         struct inode *orig_inode = file_inode(o_filp);
 912         struct page *pagep[2] = {NULL, NULL};
 913         handle_t *handle;
 914         ext4_lblk_t orig_blk_offset;
 915         unsigned long blocksize = orig_inode->i_sb->s_blocksize;
 916         unsigned int w_flags = 0;
 917         unsigned int tmp_data_size, data_size, replaced_size;
 918         int err2, jblocks, retries = 0;
 919         int replaced_count = 0;
 920         int from = data_offset_in_page << orig_inode->i_blkbits;
 921         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
 922
 923         /*
 924          * It needs twice the amount of ordinary journal buffers because
 925          * inode and donor_inode may change each different metadata blocks.
 926          */
 927 again:
 928         *err = 0;
 929         jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
 930         handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
 931         if (IS_ERR(handle)) {
 932                 *err = PTR_ERR(handle);
 933                 return 0;
 934         }
 935
 936         if (segment_eq(get_fs(), KERNEL_DS))
 937                 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
 938
 939         orig_blk_offset = orig_page_offset * blocks_per_page +
 940                 data_offset_in_page;
 941
 942         /* Calculate data_size */
 943         if ((orig_blk_offset + block_len_in_page - 1) ==
 944             ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
 945                 /* Replace the last block */
 946                 tmp_data_size = orig_inode->i_size & (blocksize - 1);
 947                 /*
 948                  * If data_size equal zero, it shows data_size is multiples of
 949                  * blocksize. So we set appropriate value.
 950                  */
 951                 if (tmp_data_size == 0)
 952                         tmp_data_size = blocksize;
 953
 954                 data_size = tmp_data_size +
 955                         ((block_len_in_page - 1) << orig_inode->i_blkbits);
 956         } else
 957                 data_size = block_len_in_page << orig_inode->i_blkbits;
 958
 959         replaced_size = data_size;
 960
 961         *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
 962                                      pagep);
 963         if (unlikely(*err < 0))
 964                 goto stop_journal;
 965         /*
 966          * If orig extent was uninitialized it can become initialized
 967          * at any time after i_data_sem was dropped, in order to
 968          * serialize with delalloc we have recheck extent while we
 969          * hold page's lock, if it is still the case data copy is not
 970          * necessary, just swap data blocks between orig and donor.
 971          */
 972         if (uninit) {
 973                 ext4_double_down_write_data_sem(orig_inode, donor_inode);
 974                 /* If any of extents in range became initialized we have to
 975                  * fallback to data copying */
 976                 uninit = mext_check_coverage(orig_inode, orig_blk_offset,
 977                                              block_len_in_page, 1, err);
 978                 if (*err)
 979                         goto drop_data_sem;
 980
 981                 uninit &= mext_check_coverage(donor_inode, orig_blk_offset,
 982                                               block_len_in_page, 1, err);
 983                 if (*err)
 984                         goto drop_data_sem;
 985
 986                 if (!uninit) {
 987                         ext4_double_up_write_data_sem(orig_inode, donor_inode);
 988                         goto data_copy;
 989                 }
 990                 if ((page_has_private(pagep[0]) &&
 991                      !try_to_release_page(pagep[0], 0)) ||
 992                     (page_has_private(pagep[1]) &&
 993                      !try_to_release_page(pagep[1], 0))) {
 994                         *err = -EBUSY;
 995                         goto drop_data_sem;
 996                 }
 997                 replaced_count = mext_replace_branches(handle, orig_inode,
 998                                                 donor_inode, orig_blk_offset,
 999                                                 block_len_in_page, err);
1000         drop_data_sem:
1001                 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1002                 goto unlock_pages;
1003         }
1004 data_copy:
1005         *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
1006         if (*err)
1007                 goto unlock_pages;
1008
1009         /* At this point all buffers in range are uptodate, old mapping layout
1010          * is no longer required, try to drop it now. */
1011         if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
1012             (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
1013                 *err = -EBUSY;
1014                 goto unlock_pages;
1015         }
1016
1017         replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
1018                                                orig_blk_offset,
1019                                                block_len_in_page, err);
1020         if (*err) {
1021                 if (replaced_count) {
1022                         block_len_in_page = replaced_count;
1023                         replaced_size =
1024                                 block_len_in_page << orig_inode->i_blkbits;
1025                 } else
1026                         goto unlock_pages;
1027         }
1028         /* Perform all necessary steps similar write_begin()/write_end()
1029          * but keeping in mind that i_size will not change */
1030         *err = __block_write_begin(pagep[0], from, replaced_size,
1031                                    ext4_get_block);
1032         if (!*err)
1033                 *err = block_commit_write(pagep[0], from, from + replaced_size);
1034
1035         if (unlikely(*err < 0))
1036                 goto repair_branches;
1037
1038         /* Even in case of data=writeback it is reasonable to pin
1039          * inode to transaction, to prevent unexpected data loss */
1040         *err = ext4_jbd2_file_inode(handle, orig_inode);
1041
1042 unlock_pages:
1043         unlock_page(pagep[0]);
1044         page_cache_release(pagep[0]);
1045         unlock_page(pagep[1]);
1046         page_cache_release(pagep[1]);
1047 stop_journal:
1048         ext4_journal_stop(handle);
1049         /* Buffer was busy because probably is pinned to journal transaction,
1050          * force transaction commit may help to free it. */
1051         if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
1052                                                       &retries))
1053                 goto again;
1054         return replaced_count;
1055
1056 repair_branches:
1057         /*
1058          * This should never ever happen!
1059          * Extents are swapped already, but we are not able to copy data.
1060          * Try to swap extents to it's original places
1061          */
1062         ext4_double_down_write_data_sem(orig_inode, donor_inode);
1063         replaced_count = mext_replace_branches(handle, donor_inode, orig_inode,
1064                                                orig_blk_offset,
1065                                                block_len_in_page, &err2);
1066         ext4_double_up_write_data_sem(orig_inode, donor_inode);
1067         if (replaced_count != block_len_in_page) {
1068                 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
1069                                        "Unable to copy data block,"
1070                                        " data will be lost.");
1071                 *err = -EIO;
1072         }
1073         replaced_count = 0;
1074         goto unlock_pages;
1075 }
1076
1077 /**
1078  * mext_check_arguments - Check whether move extent can be done
1079  *
1080  * @orig_inode:         original inode
1081  * @donor_inode:        donor inode
1082  * @orig_start:         logical start offset in block for orig
1083  * @donor_start:        logical start offset in block for donor
1084  * @len:                the number of blocks to be moved
1085  *
1086  * Check the arguments of ext4_move_extents() whether the files can be
1087  * exchanged with each other.
1088  * Return 0 on success, or a negative error value on failure.
1089  */
1090 static int
1091 mext_check_arguments(struct inode *orig_inode,
1092                      struct inode *donor_inode, __u64 orig_start,
1093                      __u64 donor_start, __u64 *len)
1094 {
1095         ext4_lblk_t orig_blocks, donor_blocks;
1096         unsigned int blkbits = orig_inode->i_blkbits;
1097         unsigned int blocksize = 1 << blkbits;
1098
1099         if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
1100                 ext4_debug("ext4 move extent: suid or sgid is set"
1101                            " to donor file [ino:orig %lu, donor %lu]\n",
1102                            orig_inode->i_ino, donor_inode->i_ino);
1103                 return -EINVAL;
1104         }
1105
1106         if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
1107                 return -EPERM;
1108
1109         /* Ext4 move extent does not support swapfile */
1110         if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
1111                 ext4_debug("ext4 move extent: The argument files should "
1112                         "not be swapfile [ino:orig %lu, donor %lu]\n",
1113                         orig_inode->i_ino, donor_inode->i_ino);
1114                 return -EINVAL;
1115         }
1116
1117         /* Ext4 move extent supports only extent based file */
1118         if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
1119                 ext4_debug("ext4 move extent: orig file is not extents "
1120                         "based file [ino:orig %lu]\n", orig_inode->i_ino);
1121                 return -EOPNOTSUPP;
1122         } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
1123                 ext4_debug("ext4 move extent: donor file is not extents "
1124                         "based file [ino:donor %lu]\n", donor_inode->i_ino);
1125                 return -EOPNOTSUPP;
1126         }
1127
1128         if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1129                 ext4_debug("ext4 move extent: File size is 0 byte\n");
1130                 return -EINVAL;
1131         }
1132
1133         /* Start offset should be same */
1134         if (orig_start != donor_start) {
1135                 ext4_debug("ext4 move extent: orig and donor's start "
1136                         "offset are not same [ino:orig %lu, donor %lu]\n",
1137                         orig_inode->i_ino, donor_inode->i_ino);
1138                 return -EINVAL;
1139         }
1140
1141         if ((orig_start >= EXT_MAX_BLOCKS) ||
1142             (*len > EXT_MAX_BLOCKS) ||
1143             (orig_start + *len >= EXT_MAX_BLOCKS))  {
1144                 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1145                         "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1146                         orig_inode->i_ino, donor_inode->i_ino);
1147                 return -EINVAL;
1148         }
1149
1150         if (orig_inode->i_size > donor_inode->i_size) {
1151                 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1152                 /* TODO: eliminate this artificial restriction */
1153                 if (orig_start >= donor_blocks) {
1154                         ext4_debug("ext4 move extent: orig start offset "
1155                         "[%llu] should be less than donor file blocks "
1156                         "[%u] [ino:orig %lu, donor %lu]\n",
1157                         orig_start, donor_blocks,
1158                         orig_inode->i_ino, donor_inode->i_ino);
1159                         return -EINVAL;
1160                 }
1161
1162                 /* TODO: eliminate this artificial restriction */
1163                 if (orig_start + *len > donor_blocks) {
1164                         ext4_debug("ext4 move extent: End offset [%llu] should "
1165                                 "be less than donor file blocks [%u]."
1166                                 "So adjust length from %llu to %llu "
1167                                 "[ino:orig %lu, donor %lu]\n",
1168                                 orig_start + *len, donor_blocks,
1169                                 *len, donor_blocks - orig_start,
1170                                 orig_inode->i_ino, donor_inode->i_ino);
1171                         *len = donor_blocks - orig_start;
1172                 }
1173         } else {
1174                 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1175                 if (orig_start >= orig_blocks) {
1176                         ext4_debug("ext4 move extent: start offset [%llu] "
1177                                 "should be less than original file blocks "
1178                                 "[%u] [ino:orig %lu, donor %lu]\n",
1179                                  orig_start, orig_blocks,
1180                                 orig_inode->i_ino, donor_inode->i_ino);
1181                         return -EINVAL;
1182                 }
1183
1184                 if (orig_start + *len > orig_blocks) {
1185                         ext4_debug("ext4 move extent: Adjust length "
1186                                 "from %llu to %llu. Because it should be "
1187                                 "less than original file blocks "
1188                                 "[ino:orig %lu, donor %lu]\n",
1189                                 *len, orig_blocks - orig_start,
1190                                 orig_inode->i_ino, donor_inode->i_ino);
1191                         *len = orig_blocks - orig_start;
1192                 }
1193         }
1194
1195         if (!*len) {
1196                 ext4_debug("ext4 move extent: len should not be 0 "
1197                         "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1198                         donor_inode->i_ino);
1199                 return -EINVAL;
1200         }
1201
1202         return 0;
1203 }
1204
1205 /**
1206  * ext4_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1207  *
1208  * @inode1:     the inode structure
1209  * @inode2:     the inode structure
1210  *
1211  * Lock two inodes' i_mutex
1212  */
1213 void
1214 ext4_inode_double_lock(struct inode *inode1, struct inode *inode2)
1215 {
1216         BUG_ON(inode1 == inode2);
1217         if (inode1 < inode2) {
1218                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1219                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1220         } else {
1221                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1222                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1223         }
1224 }
1225
1226 /**
1227  * ext4_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1228  *
1229  * @inode1:     the inode that is released first
1230  * @inode2:     the inode that is released second
1231  *
1232  */
1233
1234 void
1235 ext4_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1236 {
1237         mutex_unlock(&inode1->i_mutex);
1238         mutex_unlock(&inode2->i_mutex);
1239 }
1240
1241 /**
1242  * ext4_move_extents - Exchange the specified range of a file
1243  *
1244  * @o_filp:             file structure of the original file
1245  * @d_filp:             file structure of the donor file
1246  * @orig_start:         start offset in block for orig
1247  * @donor_start:        start offset in block for donor
1248  * @len:                the number of blocks to be moved
1249  * @moved_len:          moved block length
1250  *
1251  * This function returns 0 and moved block length is set in moved_len
1252  * if succeed, otherwise returns error value.
1253  *
1254  * Note: ext4_move_extents() proceeds the following order.
1255  * 1:ext4_move_extents() calculates the last block number of moving extent
1256  *   function by the start block number (orig_start) and the number of blocks
1257  *   to be moved (len) specified as arguments.
1258  *   If the {orig, donor}_start points a hole, the extent's start offset
1259  *   pointed by ext_cur (current extent), holecheck_path, orig_path are set
1260  *   after hole behind.
1261  * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1262  *   or the ext_cur exceeds the block_end which is last logical block number.
1263  * 3:To get the length of continues area, call mext_next_extent()
1264  *   specified with the ext_cur (initial value is holecheck_path) re-cursive,
1265  *   until find un-continuous extent, the start logical block number exceeds
1266  *   the block_end or the extent points to the last extent.
1267  * 4:Exchange the original inode data with donor inode data
1268  *   from orig_page_offset to seq_end_page.
1269  *   The start indexes of data are specified as arguments.
1270  *   That of the original inode is orig_page_offset,
1271  *   and the donor inode is also orig_page_offset
1272  *   (To easily handle blocksize != pagesize case, the offset for the
1273  *   donor inode is block unit).
1274  * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1275  *   then returns to step 2.
1276  * 6:Release holecheck_path, orig_path and set the len to moved_len
1277  *   which shows the number of moved blocks.
1278  *   The moved_len is useful for the command to calculate the file offset
1279  *   for starting next move extent ioctl.
1280  * 7:Return 0 on success, or a negative error value on failure.
1281  */
1282 int
1283 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1284                  __u64 orig_start, __u64 donor_start, __u64 len,
1285                  __u64 *moved_len)
1286 {
1287         struct inode *orig_inode = file_inode(o_filp);
1288         struct inode *donor_inode = file_inode(d_filp);
1289         struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1290         struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1291         ext4_lblk_t block_start = orig_start;
1292         ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1293         ext4_lblk_t rest_blocks;
1294         pgoff_t orig_page_offset = 0, seq_end_page;
1295         int ret, depth, last_extent = 0;
1296         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1297         int data_offset_in_page;
1298         int block_len_in_page;
1299         int uninit;
1300
1301         if (orig_inode->i_sb != donor_inode->i_sb) {
1302                 ext4_debug("ext4 move extent: The argument files "
1303                         "should be in same FS [ino:orig %lu, donor %lu]\n",
1304                         orig_inode->i_ino, donor_inode->i_ino);
1305                 return -EINVAL;
1306         }
1307
1308         /* orig and donor should be different inodes */
1309         if (orig_inode == donor_inode) {
1310                 ext4_debug("ext4 move extent: The argument files should not "
1311                         "be same inode [ino:orig %lu, donor %lu]\n",
1312                         orig_inode->i_ino, donor_inode->i_ino);
1313                 return -EINVAL;
1314         }
1315
1316         /* Regular file check */
1317         if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1318                 ext4_debug("ext4 move extent: The argument files should be "
1319                         "regular file [ino:orig %lu, donor %lu]\n",
1320                         orig_inode->i_ino, donor_inode->i_ino);
1321                 return -EINVAL;
1322         }
1323         /* TODO: This is non obvious task to swap blocks for inodes with full
1324            jornaling enabled */
1325         if (ext4_should_journal_data(orig_inode) ||
1326             ext4_should_journal_data(donor_inode)) {
1327                 return -EINVAL;
1328         }
1329         /* Protect orig and donor inodes against a truncate */
1330         ext4_inode_double_lock(orig_inode, donor_inode);
1331
1332         /* Wait for all existing dio workers */
1333         ext4_inode_block_unlocked_dio(orig_inode);
1334         ext4_inode_block_unlocked_dio(donor_inode);
1335         inode_dio_wait(orig_inode);
1336         inode_dio_wait(donor_inode);
1337
1338         /* Protect extent tree against block allocations via delalloc */
1339         ext4_double_down_write_data_sem(orig_inode, donor_inode);
1340         /* Check the filesystem environment whether move_extent can be done */
1341         ret = mext_check_arguments(orig_inode, donor_inode, orig_start,
1342                                     donor_start, &len);
1343         if (ret)
1344                 goto out;
1345
1346         file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1347         block_end = block_start + len - 1;
1348         if (file_end < block_end)
1349                 len -= block_end - file_end;
1350
1351         ret = get_ext_path(orig_inode, block_start, &orig_path);
1352         if (ret)
1353                 goto out;
1354
1355         /* Get path structure to check the hole */
1356         ret = get_ext_path(orig_inode, block_start, &holecheck_path);
1357         if (ret)
1358                 goto out;
1359
1360         depth = ext_depth(orig_inode);
1361         ext_cur = holecheck_path[depth].p_ext;
1362
1363         /*
1364          * Get proper starting location of block replacement if block_start was
1365          * within the hole.
1366          */
1367         if (le32_to_cpu(ext_cur->ee_block) +
1368                 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1369                 /*
1370                  * The hole exists between extents or the tail of
1371                  * original file.
1372                  */
1373                 last_extent = mext_next_extent(orig_inode,
1374                                         holecheck_path, &ext_cur);
1375                 if (last_extent < 0) {
1376                         ret = last_extent;
1377                         goto out;
1378                 }
1379                 last_extent = mext_next_extent(orig_inode, orig_path,
1380                                                         &ext_dummy);
1381                 if (last_extent < 0) {
1382                         ret = last_extent;
1383                         goto out;
1384                 }
1385                 seq_start = le32_to_cpu(ext_cur->ee_block);
1386         } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1387                 /* The hole exists at the beginning of original file. */
1388                 seq_start = le32_to_cpu(ext_cur->ee_block);
1389         else
1390                 seq_start = block_start;
1391
1392         /* No blocks within the specified range. */
1393         if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1394                 ext4_debug("ext4 move extent: The specified range of file "
1395                                                         "may be the hole\n");
1396                 ret = -EINVAL;
1397                 goto out;
1398         }
1399
1400         /* Adjust start blocks */
1401         add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1402                          ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1403                      max(le32_to_cpu(ext_cur->ee_block), block_start);
1404
1405         while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1406                 seq_blocks += add_blocks;
1407
1408                 /* Adjust tail blocks */
1409                 if (seq_start + seq_blocks - 1 > block_end)
1410                         seq_blocks = block_end - seq_start + 1;
1411
1412                 ext_prev = ext_cur;
1413                 last_extent = mext_next_extent(orig_inode, holecheck_path,
1414                                                 &ext_cur);
1415                 if (last_extent < 0) {
1416                         ret = last_extent;
1417                         break;
1418                 }
1419                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1420
1421                 /*
1422                  * Extend the length of contiguous block (seq_blocks)
1423                  * if extents are contiguous.
1424                  */
1425                 if (ext4_can_extents_be_merged(orig_inode,
1426                                                ext_prev, ext_cur) &&
1427                     block_end >= le32_to_cpu(ext_cur->ee_block) &&
1428                     !last_extent)
1429                         continue;
1430
1431                 /* Is original extent is uninitialized */
1432                 uninit = ext4_ext_is_uninitialized(ext_prev);
1433
1434                 data_offset_in_page = seq_start % blocks_per_page;
1435
1436                 /*
1437                  * Calculate data blocks count that should be swapped
1438                  * at the first page.
1439                  */
1440                 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1441                         /* Swapped blocks are across pages */
1442                         block_len_in_page =
1443                                         blocks_per_page - data_offset_in_page;
1444                 } else {
1445                         /* Swapped blocks are in a page */
1446                         block_len_in_page = seq_blocks;
1447                 }
1448
1449                 orig_page_offset = seq_start >>
1450                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1451                 seq_end_page = (seq_start + seq_blocks - 1) >>
1452                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1453                 seq_start = le32_to_cpu(ext_cur->ee_block);
1454                 rest_blocks = seq_blocks;
1455
1456                 /*
1457                  * Up semaphore to avoid following problems:
1458                  * a. transaction deadlock among ext4_journal_start,
1459                  *    ->write_begin via pagefault, and jbd2_journal_commit
1460                  * b. racing with ->readpage, ->write_begin, and ext4_get_block
1461                  *    in move_extent_per_page
1462                  */
1463                 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1464
1465                 while (orig_page_offset <= seq_end_page) {
1466
1467                         /* Swap original branches with new branches */
1468                         block_len_in_page = move_extent_per_page(
1469                                                 o_filp, donor_inode,
1470                                                 orig_page_offset,
1471                                                 data_offset_in_page,
1472                                                 block_len_in_page, uninit,
1473                                                 &ret);
1474
1475                         /* Count how many blocks we have exchanged */
1476                         *moved_len += block_len_in_page;
1477                         if (ret < 0)
1478                                 break;
1479                         if (*moved_len > len) {
1480                                 EXT4_ERROR_INODE(orig_inode,
1481                                         "We replaced blocks too much! "
1482                                         "sum of replaced: %llu requested: %llu",
1483                                         *moved_len, len);
1484                                 ret = -EIO;
1485                                 break;
1486                         }
1487
1488                         orig_page_offset++;
1489                         data_offset_in_page = 0;
1490                         rest_blocks -= block_len_in_page;
1491                         if (rest_blocks > blocks_per_page)
1492                                 block_len_in_page = blocks_per_page;
1493                         else
1494                                 block_len_in_page = rest_blocks;
1495                 }
1496
1497                 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1498                 if (ret < 0)
1499                         break;
1500
1501                 /* Decrease buffer counter */
1502                 if (holecheck_path)
1503                         ext4_ext_drop_refs(holecheck_path);
1504                 ret = get_ext_path(orig_inode, seq_start, &holecheck_path);
1505                 if (ret)
1506                         break;
1507                 depth = holecheck_path->p_depth;
1508
1509                 /* Decrease buffer counter */
1510                 if (orig_path)
1511                         ext4_ext_drop_refs(orig_path);
1512                 ret = get_ext_path(orig_inode, seq_start, &orig_path);
1513                 if (ret)
1514                         break;
1515
1516                 ext_cur = holecheck_path[depth].p_ext;
1517                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1518                 seq_blocks = 0;
1519
1520         }
1521 out:
1522         if (*moved_len) {
1523                 ext4_discard_preallocations(orig_inode);
1524                 ext4_discard_preallocations(donor_inode);
1525         }
1526
1527         if (orig_path) {
1528                 ext4_ext_drop_refs(orig_path);
1529                 kfree(orig_path);
1530         }
1531         if (holecheck_path) {
1532                 ext4_ext_drop_refs(holecheck_path);
1533                 kfree(holecheck_path);
1534         }
1535         ext4_double_up_write_data_sem(orig_inode, donor_inode);
1536         ext4_inode_resume_unlocked_dio(orig_inode);
1537         ext4_inode_resume_unlocked_dio(donor_inode);
1538         ext4_inode_double_unlock(orig_inode, donor_inode);
1539
1540         return ret;
1541 }