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  * 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         if (depth) {
 413                 ret = ext4_handle_dirty_metadata(handle, orig_inode,
 414                                                  orig_path->p_bh);
 415                 if (ret)
 416                         return ret;
 417         } else {
 418                 ret = ext4_mark_inode_dirty(handle, orig_inode);
 419                 if (ret < 0)
 420                         return ret;
 421         }
 422
 423         return 0;
 424 }
 425
 426 /**
 427  * mext_leaf_block - Move one leaf extent block into the inode.
 428  *
 429  * @handle:             journal handle
 430  * @orig_inode:         original inode
 431  * @orig_path:          path indicates first extent to be changed
 432  * @dext:               donor extent
 433  * @from:               start offset on the target file
 434  *
 435  * In order to insert extents into the leaf block, we must divide the extent
 436  * in the leaf block into three extents. The one is located to be inserted
 437  * extents, and the others are located around it.
 438  *
 439  * Therefore, this function creates structures to save extents of the leaf
 440  * block, and inserts extents by calling mext_insert_extents() with
 441  * created extents. Return 0 on success, or a negative error value on failure.
 442  */
 443 static int
 444 mext_leaf_block(handle_t *handle, struct inode *orig_inode,
 445                      struct ext4_ext_path *orig_path, struct ext4_extent *dext,
 446                      ext4_lblk_t *from)
 447 {
 448         struct ext4_extent *oext, *o_start, *o_end, *prev_ext;
 449         struct ext4_extent new_ext, start_ext, end_ext;
 450         ext4_lblk_t new_ext_end;
 451         int oext_alen, new_ext_alen, end_ext_alen;
 452         int depth = ext_depth(orig_inode);
 453         int ret;
 454
 455         start_ext.ee_block = end_ext.ee_block = 0;
 456         o_start = o_end = oext = orig_path[depth].p_ext;
 457         oext_alen = ext4_ext_get_actual_len(oext);
 458         start_ext.ee_len = end_ext.ee_len = 0;
 459
 460         new_ext.ee_block = cpu_to_le32(*from);
 461         ext4_ext_store_pblock(&new_ext, ext4_ext_pblock(dext));
 462         new_ext.ee_len = dext->ee_len;
 463         new_ext_alen = ext4_ext_get_actual_len(&new_ext);
 464         new_ext_end = le32_to_cpu(new_ext.ee_block) + new_ext_alen - 1;
 465
 466         /*
 467          * Case: original extent is first
 468          * oext      |--------|
 469          * new_ext      |--|
 470          * start_ext |--|
 471          */
 472         if (le32_to_cpu(oext->ee_block) < le32_to_cpu(new_ext.ee_block) &&
 473                 le32_to_cpu(new_ext.ee_block) <
 474                 le32_to_cpu(oext->ee_block) + oext_alen) {
 475                 start_ext.ee_len = cpu_to_le16(le32_to_cpu(new_ext.ee_block) -
 476                                                le32_to_cpu(oext->ee_block));
 477                 start_ext.ee_block = oext->ee_block;
 478                 copy_extent_status(oext, &start_ext);
 479         } else if (oext > EXT_FIRST_EXTENT(orig_path[depth].p_hdr)) {
 480                 prev_ext = oext - 1;
 481                 /*
 482                  * We can merge new_ext into previous extent,
 483                  * if these are contiguous and same extent type.
 484                  */
 485                 if (ext4_can_extents_be_merged(orig_inode, prev_ext,
 486                                                &new_ext)) {
 487                         o_start = prev_ext;
 488                         start_ext.ee_len = cpu_to_le16(
 489                                 ext4_ext_get_actual_len(prev_ext) +
 490                                 new_ext_alen);
 491                         start_ext.ee_block = oext->ee_block;
 492                         copy_extent_status(prev_ext, &start_ext);
 493                         new_ext.ee_len = 0;
 494                 }
 495         }
 496
 497         /*
 498          * Case: new_ext_end must be less than oext
 499          * oext      |-----------|
 500          * new_ext       |-------|
 501          */
 502         if (le32_to_cpu(oext->ee_block) + oext_alen - 1 < new_ext_end) {
 503                 EXT4_ERROR_INODE(orig_inode,
 504                         "new_ext_end(%u) should be less than or equal to "
 505                         "oext->ee_block(%u) + oext_alen(%d) - 1",
 506                         new_ext_end, le32_to_cpu(oext->ee_block),
 507                         oext_alen);
 508                 ret = -EIO;
 509                 goto out;
 510         }
 511
 512         /*
 513          * Case: new_ext is smaller than original extent
 514          * oext    |---------------|
 515          * new_ext |-----------|
 516          * end_ext             |---|
 517          */
 518         if (le32_to_cpu(oext->ee_block) <= new_ext_end &&
 519                 new_ext_end < le32_to_cpu(oext->ee_block) + oext_alen - 1) {
 520                 end_ext.ee_len =
 521                         cpu_to_le16(le32_to_cpu(oext->ee_block) +
 522                         oext_alen - 1 - new_ext_end);
 523                 copy_extent_status(oext, &end_ext);
 524                 end_ext_alen = ext4_ext_get_actual_len(&end_ext);
 525                 ext4_ext_store_pblock(&end_ext,
 526                         (ext4_ext_pblock(o_end) + oext_alen - end_ext_alen));
 527                 end_ext.ee_block =
 528                         cpu_to_le32(le32_to_cpu(o_end->ee_block) +
 529                         oext_alen - end_ext_alen);
 530         }
 531
 532         ret = mext_insert_extents(handle, orig_inode, orig_path, o_start,
 533                                 o_end, &start_ext, &new_ext, &end_ext);
 534 out:
 535         return ret;
 536 }
 537
 538 /**
 539  * mext_calc_swap_extents - Calculate extents for extent swapping.
 540  *
 541  * @tmp_dext:           the extent that will belong to the original inode
 542  * @tmp_oext:           the extent that will belong to the donor inode
 543  * @orig_off:           block offset of original inode
 544  * @donor_off:          block offset of donor inode
 545  * @max_count:          the maximum length of extents
 546  *
 547  * Return 0 on success, or a negative error value on failure.
 548  */
 549 static int
 550 mext_calc_swap_extents(struct ext4_extent *tmp_dext,
 551                               struct ext4_extent *tmp_oext,
 552                               ext4_lblk_t orig_off, ext4_lblk_t donor_off,
 553                               ext4_lblk_t max_count)
 554 {
 555         ext4_lblk_t diff, orig_diff;
 556         struct ext4_extent dext_old, oext_old;
 557
 558         BUG_ON(orig_off != donor_off);
 559
 560         /* original and donor extents have to cover the same block offset */
 561         if (orig_off < le32_to_cpu(tmp_oext->ee_block) ||
 562             le32_to_cpu(tmp_oext->ee_block) +
 563                         ext4_ext_get_actual_len(tmp_oext) - 1 < orig_off)
 564                 return -ENODATA;
 565
 566         if (orig_off < le32_to_cpu(tmp_dext->ee_block) ||
 567             le32_to_cpu(tmp_dext->ee_block) +
 568                         ext4_ext_get_actual_len(tmp_dext) - 1 < orig_off)
 569                 return -ENODATA;
 570
 571         dext_old = *tmp_dext;
 572         oext_old = *tmp_oext;
 573
 574         /* When tmp_dext is too large, pick up the target range. */
 575         diff = donor_off - le32_to_cpu(tmp_dext->ee_block);
 576
 577         ext4_ext_store_pblock(tmp_dext, ext4_ext_pblock(tmp_dext) + diff);
 578         le32_add_cpu(&tmp_dext->ee_block, diff);
 579         le16_add_cpu(&tmp_dext->ee_len, -diff);
 580
 581         if (max_count < ext4_ext_get_actual_len(tmp_dext))
 582                 tmp_dext->ee_len = cpu_to_le16(max_count);
 583
 584         orig_diff = orig_off - le32_to_cpu(tmp_oext->ee_block);
 585         ext4_ext_store_pblock(tmp_oext, ext4_ext_pblock(tmp_oext) + orig_diff);
 586
 587         /* Adjust extent length if donor extent is larger than orig */
 588         if (ext4_ext_get_actual_len(tmp_dext) >
 589             ext4_ext_get_actual_len(tmp_oext) - orig_diff)
 590                 tmp_dext->ee_len = cpu_to_le16(le16_to_cpu(tmp_oext->ee_len) -
 591                                                 orig_diff);
 592
 593         tmp_oext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(tmp_dext));
 594
 595         copy_extent_status(&oext_old, tmp_dext);
 596         copy_extent_status(&dext_old, tmp_oext);
 597
 598         return 0;
 599 }
 600
 601 /**
 602  * mext_check_coverage - Check that all extents in range has the same type
 603  *
 604  * @inode:              inode in question
 605  * @from:               block offset of inode
 606  * @count:              block count to be checked
 607  * @uninit:             extents expected to be uninitialized
 608  * @err:                pointer to save error value
 609  *
 610  * Return 1 if all extents in range has expected type, and zero otherwise.
 611  */
 612 static int
 613 mext_check_coverage(struct inode *inode, ext4_lblk_t from, ext4_lblk_t count,
 614                           int uninit, int *err)
 615 {
 616         struct ext4_ext_path *path = NULL;
 617         struct ext4_extent *ext;
 618         int ret = 0;
 619         ext4_lblk_t last = from + count;
 620         while (from < last) {
 621                 *err = get_ext_path(inode, from, &path);
 622                 if (*err)
 623                         goto out;
 624                 ext = path[ext_depth(inode)].p_ext;
 625                 if (uninit != ext4_ext_is_uninitialized(ext))
 626                         goto out;
 627                 from += ext4_ext_get_actual_len(ext);
 628                 ext4_ext_drop_refs(path);
 629         }
 630         ret = 1;
 631 out:
 632         if (path) {
 633                 ext4_ext_drop_refs(path);
 634                 kfree(path);
 635         }
 636         return ret;
 637 }
 638
 639 /**
 640  * mext_replace_branches - Replace original extents with new extents
 641  *
 642  * @handle:             journal handle
 643  * @orig_inode:         original inode
 644  * @donor_inode:        donor inode
 645  * @from:               block offset of orig_inode
 646  * @count:              block count to be replaced
 647  * @err:                pointer to save return value
 648  *
 649  * Replace original inode extents and donor inode extents page by page.
 650  * We implement this replacement in the following three steps:
 651  * 1. Save the block information of original and donor inodes into
 652  *    dummy extents.
 653  * 2. Change the block information of original inode to point at the
 654  *    donor inode blocks.
 655  * 3. Change the block information of donor inode to point at the saved
 656  *    original inode blocks in the dummy extents.
 657  *
 658  * Return replaced block count.
 659  */
 660 static int
 661 mext_replace_branches(handle_t *handle, struct inode *orig_inode,
 662                            struct inode *donor_inode, ext4_lblk_t from,
 663                            ext4_lblk_t count, int *err)
 664 {
 665         struct ext4_ext_path *orig_path = NULL;
 666         struct ext4_ext_path *donor_path = NULL;
 667         struct ext4_extent *oext, *dext;
 668         struct ext4_extent tmp_dext, tmp_oext;
 669         ext4_lblk_t orig_off = from, donor_off = from;
 670         int depth;
 671         int replaced_count = 0;
 672         int dext_alen;
 673
 674         *err = ext4_es_remove_extent(orig_inode, from, count);
 675         if (*err)
 676                 goto out;
 677
 678         *err = ext4_es_remove_extent(donor_inode, from, count);
 679         if (*err)
 680                 goto out;
 681
 682         /* Get the original extent for the block "orig_off" */
 683         *err = get_ext_path(orig_inode, orig_off, &orig_path);
 684         if (*err)
 685                 goto out;
 686
 687         /* Get the donor extent for the head */
 688         *err = get_ext_path(donor_inode, donor_off, &donor_path);
 689         if (*err)
 690                 goto out;
 691         depth = ext_depth(orig_inode);
 692         oext = orig_path[depth].p_ext;
 693         tmp_oext = *oext;
 694
 695         depth = ext_depth(donor_inode);
 696         dext = donor_path[depth].p_ext;
 697         if (unlikely(!dext))
 698                 goto missing_donor_extent;
 699         tmp_dext = *dext;
 700
 701         *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 702                                       donor_off, count);
 703         if (*err)
 704                 goto out;
 705
 706         /* Loop for the donor extents */
 707         while (1) {
 708                 /* The extent for donor must be found. */
 709                 if (unlikely(!dext)) {
 710                 missing_donor_extent:
 711                         EXT4_ERROR_INODE(donor_inode,
 712                                    "The extent for donor must be found");
 713                         *err = -EIO;
 714                         goto out;
 715                 } else if (donor_off != le32_to_cpu(tmp_dext.ee_block)) {
 716                         EXT4_ERROR_INODE(donor_inode,
 717                                 "Donor offset(%u) and the first block of donor "
 718                                 "extent(%u) should be equal",
 719                                 donor_off,
 720                                 le32_to_cpu(tmp_dext.ee_block));
 721                         *err = -EIO;
 722                         goto out;
 723                 }
 724
 725                 /* Set donor extent to orig extent */
 726                 *err = mext_leaf_block(handle, orig_inode,
 727                                            orig_path, &tmp_dext, &orig_off);
 728                 if (*err)
 729                         goto out;
 730
 731                 /* Set orig extent to donor extent */
 732                 *err = mext_leaf_block(handle, donor_inode,
 733                                            donor_path, &tmp_oext, &donor_off);
 734                 if (*err)
 735                         goto out;
 736
 737                 dext_alen = ext4_ext_get_actual_len(&tmp_dext);
 738                 replaced_count += dext_alen;
 739                 donor_off += dext_alen;
 740                 orig_off += dext_alen;
 741
 742                 /* Already moved the expected blocks */
 743                 if (replaced_count >= count)
 744                         break;
 745
 746                 if (orig_path)
 747                         ext4_ext_drop_refs(orig_path);
 748                 *err = get_ext_path(orig_inode, orig_off, &orig_path);
 749                 if (*err)
 750                         goto out;
 751                 depth = ext_depth(orig_inode);
 752                 oext = orig_path[depth].p_ext;
 753                 tmp_oext = *oext;
 754
 755                 if (donor_path)
 756                         ext4_ext_drop_refs(donor_path);
 757                 *err = get_ext_path(donor_inode, donor_off, &donor_path);
 758                 if (*err)
 759                         goto out;
 760                 depth = ext_depth(donor_inode);
 761                 dext = donor_path[depth].p_ext;
 762                 tmp_dext = *dext;
 763
 764                 *err = mext_calc_swap_extents(&tmp_dext, &tmp_oext, orig_off,
 765                                            donor_off, count - replaced_count);
 766                 if (*err)
 767                         goto out;
 768         }
 769
 770 out:
 771         if (orig_path) {
 772                 ext4_ext_drop_refs(orig_path);
 773                 kfree(orig_path);
 774         }
 775         if (donor_path) {
 776                 ext4_ext_drop_refs(donor_path);
 777                 kfree(donor_path);
 778         }
 779
 780         return replaced_count;
 781 }
 782
 783 /**
 784  * mext_page_double_lock - Grab and lock pages on both @inode1 and @inode2
 785  *
 786  * @inode1:     the inode structure
 787  * @inode2:     the inode structure
 788  * @index:      page index
 789  * @page:       result page vector
 790  *
 791  * Grab two locked pages for inode's by inode order
 792  */
 793 static int
 794 mext_page_double_lock(struct inode *inode1, struct inode *inode2,
 795                       pgoff_t index, struct page *page[2])
 796 {
 797         struct address_space *mapping[2];
 798         unsigned fl = AOP_FLAG_NOFS;
 799
 800         BUG_ON(!inode1 || !inode2);
 801         if (inode1 < inode2) {
 802                 mapping[0] = inode1->i_mapping;
 803                 mapping[1] = inode2->i_mapping;
 804         } else {
 805                 mapping[0] = inode2->i_mapping;
 806                 mapping[1] = inode1->i_mapping;
 807         }
 808
 809         page[0] = grab_cache_page_write_begin(mapping[0], index, fl);
 810         if (!page[0])
 811                 return -ENOMEM;
 812
 813         page[1] = grab_cache_page_write_begin(mapping[1], index, fl);
 814         if (!page[1]) {
 815                 unlock_page(page[0]);
 816                 page_cache_release(page[0]);
 817                 return -ENOMEM;
 818         }
 819
 820         if (inode1 > inode2) {
 821                 struct page *tmp;
 822                 tmp = page[0];
 823                 page[0] = page[1];
 824                 page[1] = tmp;
 825         }
 826         return 0;
 827 }
 828
 829 /* Force page buffers uptodate w/o dropping page's lock */
 830 static int
 831 mext_page_mkuptodate(struct page *page, unsigned from, unsigned to)
 832 {
 833         struct inode *inode = page->mapping->host;
 834         sector_t block;
 835         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
 836         unsigned int blocksize, block_start, block_end;
 837         int i, err,  nr = 0, partial = 0;
 838         BUG_ON(!PageLocked(page));
 839         BUG_ON(PageWriteback(page));
 840
 841         if (PageUptodate(page))
 842                 return 0;
 843
 844         blocksize = 1 << inode->i_blkbits;
 845         if (!page_has_buffers(page))
 846                 create_empty_buffers(page, blocksize, 0);
 847
 848         head = page_buffers(page);
 849         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
 850         for (bh = head, block_start = 0; bh != head || !block_start;
 851              block++, block_start = block_end, bh = bh->b_this_page) {
 852                 block_end = block_start + blocksize;
 853                 if (block_end <= from || block_start >= to) {
 854                         if (!buffer_uptodate(bh))
 855                                 partial = 1;
 856                         continue;
 857                 }
 858                 if (buffer_uptodate(bh))
 859                         continue;
 860                 if (!buffer_mapped(bh)) {
 861                         int err = 0;
 862                         err = ext4_get_block(inode, block, bh, 0);
 863                         if (err) {
 864                                 SetPageError(page);
 865                                 return err;
 866                         }
 867                         if (!buffer_mapped(bh)) {
 868                                 zero_user(page, block_start, blocksize);
 869                                 if (!err)
 870                                         set_buffer_uptodate(bh);
 871                                 continue;
 872                         }
 873                 }
 874                 BUG_ON(nr >= MAX_BUF_PER_PAGE);
 875                 arr[nr++] = bh;
 876         }
 877         /* No io required */
 878         if (!nr)
 879                 goto out;
 880
 881         for (i = 0; i < nr; i++) {
 882                 bh = arr[i];
 883                 if (!bh_uptodate_or_lock(bh)) {
 884                         err = bh_submit_read(bh);
 885                         if (err)
 886                                 return err;
 887                 }
 888         }
 889 out:
 890         if (!partial)
 891                 SetPageUptodate(page);
 892         return 0;
 893 }
 894
 895 /**
 896  * move_extent_per_page - Move extent data per page
 897  *
 898  * @o_filp:                     file structure of original file
 899  * @donor_inode:                donor inode
 900  * @orig_page_offset:           page index on original file
 901  * @data_offset_in_page:        block index where data swapping starts
 902  * @block_len_in_page:          the number of blocks to be swapped
 903  * @uninit:                     orig extent is uninitialized or not
 904  * @err:                        pointer to save return value
 905  *
 906  * Save the data in original inode blocks and replace original inode extents
 907  * with donor inode extents by calling mext_replace_branches().
 908  * Finally, write out the saved data in new original inode blocks. Return
 909  * replaced block count.
 910  */
 911 static int
 912 move_extent_per_page(struct file *o_filp, struct inode *donor_inode,
 913                   pgoff_t orig_page_offset, int data_offset_in_page,
 914                   int block_len_in_page, int uninit, int *err)
 915 {
 916         struct inode *orig_inode = file_inode(o_filp);
 917         struct page *pagep[2] = {NULL, NULL};
 918         handle_t *handle;
 919         ext4_lblk_t orig_blk_offset;
 920         long long offs = orig_page_offset << PAGE_CACHE_SHIFT;
 921         unsigned long blocksize = orig_inode->i_sb->s_blocksize;
 922         unsigned int w_flags = 0;
 923         unsigned int tmp_data_size, data_size, replaced_size;
 924         int err2, jblocks, retries = 0;
 925         int replaced_count = 0;
 926         int from = data_offset_in_page << orig_inode->i_blkbits;
 927         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
 928
 929         /*
 930          * It needs twice the amount of ordinary journal buffers because
 931          * inode and donor_inode may change each different metadata blocks.
 932          */
 933 again:
 934         *err = 0;
 935         jblocks = ext4_writepage_trans_blocks(orig_inode) * 2;
 936         handle = ext4_journal_start(orig_inode, EXT4_HT_MOVE_EXTENTS, jblocks);
 937         if (IS_ERR(handle)) {
 938                 *err = PTR_ERR(handle);
 939                 return 0;
 940         }
 941
 942         if (segment_eq(get_fs(), KERNEL_DS))
 943                 w_flags |= AOP_FLAG_UNINTERRUPTIBLE;
 944
 945         orig_blk_offset = orig_page_offset * blocks_per_page +
 946                 data_offset_in_page;
 947
 948         offs = (long long)orig_blk_offset << orig_inode->i_blkbits;
 949
 950         /* Calculate data_size */
 951         if ((orig_blk_offset + block_len_in_page - 1) ==
 952             ((orig_inode->i_size - 1) >> orig_inode->i_blkbits)) {
 953                 /* Replace the last block */
 954                 tmp_data_size = orig_inode->i_size & (blocksize - 1);
 955                 /*
 956                  * If data_size equal zero, it shows data_size is multiples of
 957                  * blocksize. So we set appropriate value.
 958                  */
 959                 if (tmp_data_size == 0)
 960                         tmp_data_size = blocksize;
 961
 962                 data_size = tmp_data_size +
 963                         ((block_len_in_page - 1) << orig_inode->i_blkbits);
 964         } else
 965                 data_size = block_len_in_page << orig_inode->i_blkbits;
 966
 967         replaced_size = data_size;
 968
 969         *err = mext_page_double_lock(orig_inode, donor_inode, orig_page_offset,
 970                                      pagep);
 971         if (unlikely(*err < 0))
 972                 goto stop_journal;
 973         /*
 974          * If orig extent was uninitialized it can become initialized
 975          * at any time after i_data_sem was dropped, in order to
 976          * serialize with delalloc we have recheck extent while we
 977          * hold page's lock, if it is still the case data copy is not
 978          * necessary, just swap data blocks between orig and donor.
 979          */
 980         if (uninit) {
 981                 ext4_double_down_write_data_sem(orig_inode, donor_inode);
 982                 /* If any of extents in range became initialized we have to
 983                  * fallback to data copying */
 984                 uninit = mext_check_coverage(orig_inode, orig_blk_offset,
 985                                              block_len_in_page, 1, err);
 986                 if (*err)
 987                         goto drop_data_sem;
 988
 989                 uninit &= mext_check_coverage(donor_inode, orig_blk_offset,
 990                                               block_len_in_page, 1, err);
 991                 if (*err)
 992                         goto drop_data_sem;
 993
 994                 if (!uninit) {
 995                         ext4_double_up_write_data_sem(orig_inode, donor_inode);
 996                         goto data_copy;
 997                 }
 998                 if ((page_has_private(pagep[0]) &&
 999                      !try_to_release_page(pagep[0], 0)) ||
1000                     (page_has_private(pagep[1]) &&
1001                      !try_to_release_page(pagep[1], 0))) {
1002                         *err = -EBUSY;
1003                         goto drop_data_sem;
1004                 }
1005                 replaced_count = mext_replace_branches(handle, orig_inode,
1006                                                 donor_inode, orig_blk_offset,
1007                                                 block_len_in_page, err);
1008         drop_data_sem:
1009                 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1010                 goto unlock_pages;
1011         }
1012 data_copy:
1013         *err = mext_page_mkuptodate(pagep[0], from, from + replaced_size);
1014         if (*err)
1015                 goto unlock_pages;
1016
1017         /* At this point all buffers in range are uptodate, old mapping layout
1018          * is no longer required, try to drop it now. */
1019         if ((page_has_private(pagep[0]) && !try_to_release_page(pagep[0], 0)) ||
1020             (page_has_private(pagep[1]) && !try_to_release_page(pagep[1], 0))) {
1021                 *err = -EBUSY;
1022                 goto unlock_pages;
1023         }
1024
1025         replaced_count = mext_replace_branches(handle, orig_inode, donor_inode,
1026                                                orig_blk_offset,
1027                                                block_len_in_page, err);
1028         if (*err) {
1029                 if (replaced_count) {
1030                         block_len_in_page = replaced_count;
1031                         replaced_size =
1032                                 block_len_in_page << orig_inode->i_blkbits;
1033                 } else
1034                         goto unlock_pages;
1035         }
1036         /* Perform all necessary steps similar write_begin()/write_end()
1037          * but keeping in mind that i_size will not change */
1038         *err = __block_write_begin(pagep[0], from, from + replaced_size,
1039                                    ext4_get_block);
1040         if (!*err)
1041                 *err = block_commit_write(pagep[0], from, from + replaced_size);
1042
1043         if (unlikely(*err < 0))
1044                 goto repair_branches;
1045
1046         /* Even in case of data=writeback it is reasonable to pin
1047          * inode to transaction, to prevent unexpected data loss */
1048         *err = ext4_jbd2_file_inode(handle, orig_inode);
1049
1050 unlock_pages:
1051         unlock_page(pagep[0]);
1052         page_cache_release(pagep[0]);
1053         unlock_page(pagep[1]);
1054         page_cache_release(pagep[1]);
1055 stop_journal:
1056         ext4_journal_stop(handle);
1057         /* Buffer was busy because probably is pinned to journal transaction,
1058          * force transaction commit may help to free it. */
1059         if (*err == -EBUSY && ext4_should_retry_alloc(orig_inode->i_sb,
1060                                                       &retries))
1061                 goto again;
1062         return replaced_count;
1063
1064 repair_branches:
1065         /*
1066          * This should never ever happen!
1067          * Extents are swapped already, but we are not able to copy data.
1068          * Try to swap extents to it's original places
1069          */
1070         ext4_double_down_write_data_sem(orig_inode, donor_inode);
1071         replaced_count = mext_replace_branches(handle, donor_inode, orig_inode,
1072                                                orig_blk_offset,
1073                                                block_len_in_page, &err2);
1074         ext4_double_up_write_data_sem(orig_inode, donor_inode);
1075         if (replaced_count != block_len_in_page) {
1076                 EXT4_ERROR_INODE_BLOCK(orig_inode, (sector_t)(orig_blk_offset),
1077                                        "Unable to copy data block,"
1078                                        " data will be lost.");
1079                 *err = -EIO;
1080         }
1081         replaced_count = 0;
1082         goto unlock_pages;
1083 }
1084
1085 /**
1086  * mext_check_arguments - Check whether move extent can be done
1087  *
1088  * @orig_inode:         original inode
1089  * @donor_inode:        donor inode
1090  * @orig_start:         logical start offset in block for orig
1091  * @donor_start:        logical start offset in block for donor
1092  * @len:                the number of blocks to be moved
1093  *
1094  * Check the arguments of ext4_move_extents() whether the files can be
1095  * exchanged with each other.
1096  * Return 0 on success, or a negative error value on failure.
1097  */
1098 static int
1099 mext_check_arguments(struct inode *orig_inode,
1100                      struct inode *donor_inode, __u64 orig_start,
1101                      __u64 donor_start, __u64 *len)
1102 {
1103         ext4_lblk_t orig_blocks, donor_blocks;
1104         unsigned int blkbits = orig_inode->i_blkbits;
1105         unsigned int blocksize = 1 << blkbits;
1106
1107         if (donor_inode->i_mode & (S_ISUID|S_ISGID)) {
1108                 ext4_debug("ext4 move extent: suid or sgid is set"
1109                            " to donor file [ino:orig %lu, donor %lu]\n",
1110                            orig_inode->i_ino, donor_inode->i_ino);
1111                 return -EINVAL;
1112         }
1113
1114         if (IS_IMMUTABLE(donor_inode) || IS_APPEND(donor_inode))
1115                 return -EPERM;
1116
1117         /* Ext4 move extent does not support swapfile */
1118         if (IS_SWAPFILE(orig_inode) || IS_SWAPFILE(donor_inode)) {
1119                 ext4_debug("ext4 move extent: The argument files should "
1120                         "not be swapfile [ino:orig %lu, donor %lu]\n",
1121                         orig_inode->i_ino, donor_inode->i_ino);
1122                 return -EINVAL;
1123         }
1124
1125         /* Ext4 move extent supports only extent based file */
1126         if (!(ext4_test_inode_flag(orig_inode, EXT4_INODE_EXTENTS))) {
1127                 ext4_debug("ext4 move extent: orig file is not extents "
1128                         "based file [ino:orig %lu]\n", orig_inode->i_ino);
1129                 return -EOPNOTSUPP;
1130         } else if (!(ext4_test_inode_flag(donor_inode, EXT4_INODE_EXTENTS))) {
1131                 ext4_debug("ext4 move extent: donor file is not extents "
1132                         "based file [ino:donor %lu]\n", donor_inode->i_ino);
1133                 return -EOPNOTSUPP;
1134         }
1135
1136         if ((!orig_inode->i_size) || (!donor_inode->i_size)) {
1137                 ext4_debug("ext4 move extent: File size is 0 byte\n");
1138                 return -EINVAL;
1139         }
1140
1141         /* Start offset should be same */
1142         if (orig_start != donor_start) {
1143                 ext4_debug("ext4 move extent: orig and donor's start "
1144                         "offset are not same [ino:orig %lu, donor %lu]\n",
1145                         orig_inode->i_ino, donor_inode->i_ino);
1146                 return -EINVAL;
1147         }
1148
1149         if ((orig_start >= EXT_MAX_BLOCKS) ||
1150             (*len > EXT_MAX_BLOCKS) ||
1151             (orig_start + *len >= EXT_MAX_BLOCKS))  {
1152                 ext4_debug("ext4 move extent: Can't handle over [%u] blocks "
1153                         "[ino:orig %lu, donor %lu]\n", EXT_MAX_BLOCKS,
1154                         orig_inode->i_ino, donor_inode->i_ino);
1155                 return -EINVAL;
1156         }
1157
1158         if (orig_inode->i_size > donor_inode->i_size) {
1159                 donor_blocks = (donor_inode->i_size + blocksize - 1) >> blkbits;
1160                 /* TODO: eliminate this artificial restriction */
1161                 if (orig_start >= donor_blocks) {
1162                         ext4_debug("ext4 move extent: orig start offset "
1163                         "[%llu] should be less than donor file blocks "
1164                         "[%u] [ino:orig %lu, donor %lu]\n",
1165                         orig_start, donor_blocks,
1166                         orig_inode->i_ino, donor_inode->i_ino);
1167                         return -EINVAL;
1168                 }
1169
1170                 /* TODO: eliminate this artificial restriction */
1171                 if (orig_start + *len > donor_blocks) {
1172                         ext4_debug("ext4 move extent: End offset [%llu] should "
1173                                 "be less than donor file blocks [%u]."
1174                                 "So adjust length from %llu to %llu "
1175                                 "[ino:orig %lu, donor %lu]\n",
1176                                 orig_start + *len, donor_blocks,
1177                                 *len, donor_blocks - orig_start,
1178                                 orig_inode->i_ino, donor_inode->i_ino);
1179                         *len = donor_blocks - orig_start;
1180                 }
1181         } else {
1182                 orig_blocks = (orig_inode->i_size + blocksize - 1) >> blkbits;
1183                 if (orig_start >= orig_blocks) {
1184                         ext4_debug("ext4 move extent: start offset [%llu] "
1185                                 "should be less than original file blocks "
1186                                 "[%u] [ino:orig %lu, donor %lu]\n",
1187                                  orig_start, orig_blocks,
1188                                 orig_inode->i_ino, donor_inode->i_ino);
1189                         return -EINVAL;
1190                 }
1191
1192                 if (orig_start + *len > orig_blocks) {
1193                         ext4_debug("ext4 move extent: Adjust length "
1194                                 "from %llu to %llu. Because it should be "
1195                                 "less than original file blocks "
1196                                 "[ino:orig %lu, donor %lu]\n",
1197                                 *len, orig_blocks - orig_start,
1198                                 orig_inode->i_ino, donor_inode->i_ino);
1199                         *len = orig_blocks - orig_start;
1200                 }
1201         }
1202
1203         if (!*len) {
1204                 ext4_debug("ext4 move extent: len should not be 0 "
1205                         "[ino:orig %lu, donor %lu]\n", orig_inode->i_ino,
1206                         donor_inode->i_ino);
1207                 return -EINVAL;
1208         }
1209
1210         return 0;
1211 }
1212
1213 /**
1214  * ext4_inode_double_lock - Lock i_mutex on both @inode1 and @inode2
1215  *
1216  * @inode1:     the inode structure
1217  * @inode2:     the inode structure
1218  *
1219  * Lock two inodes' i_mutex
1220  */
1221 void
1222 ext4_inode_double_lock(struct inode *inode1, struct inode *inode2)
1223 {
1224         BUG_ON(inode1 == inode2);
1225         if (inode1 < inode2) {
1226                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_PARENT);
1227                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_CHILD);
1228         } else {
1229                 mutex_lock_nested(&inode2->i_mutex, I_MUTEX_PARENT);
1230                 mutex_lock_nested(&inode1->i_mutex, I_MUTEX_CHILD);
1231         }
1232 }
1233
1234 /**
1235  * ext4_inode_double_unlock - Release i_mutex on both @inode1 and @inode2
1236  *
1237  * @inode1:     the inode that is released first
1238  * @inode2:     the inode that is released second
1239  *
1240  */
1241
1242 void
1243 ext4_inode_double_unlock(struct inode *inode1, struct inode *inode2)
1244 {
1245         mutex_unlock(&inode1->i_mutex);
1246         mutex_unlock(&inode2->i_mutex);
1247 }
1248
1249 /**
1250  * ext4_move_extents - Exchange the specified range of a file
1251  *
1252  * @o_filp:             file structure of the original file
1253  * @d_filp:             file structure of the donor file
1254  * @orig_start:         start offset in block for orig
1255  * @donor_start:        start offset in block for donor
1256  * @len:                the number of blocks to be moved
1257  * @moved_len:          moved block length
1258  *
1259  * This function returns 0 and moved block length is set in moved_len
1260  * if succeed, otherwise returns error value.
1261  *
1262  * Note: ext4_move_extents() proceeds the following order.
1263  * 1:ext4_move_extents() calculates the last block number of moving extent
1264  *   function by the start block number (orig_start) and the number of blocks
1265  *   to be moved (len) specified as arguments.
1266  *   If the {orig, donor}_start points a hole, the extent's start offset
1267  *   pointed by ext_cur (current extent), holecheck_path, orig_path are set
1268  *   after hole behind.
1269  * 2:Continue step 3 to step 5, until the holecheck_path points to last_extent
1270  *   or the ext_cur exceeds the block_end which is last logical block number.
1271  * 3:To get the length of continues area, call mext_next_extent()
1272  *   specified with the ext_cur (initial value is holecheck_path) re-cursive,
1273  *   until find un-continuous extent, the start logical block number exceeds
1274  *   the block_end or the extent points to the last extent.
1275  * 4:Exchange the original inode data with donor inode data
1276  *   from orig_page_offset to seq_end_page.
1277  *   The start indexes of data are specified as arguments.
1278  *   That of the original inode is orig_page_offset,
1279  *   and the donor inode is also orig_page_offset
1280  *   (To easily handle blocksize != pagesize case, the offset for the
1281  *   donor inode is block unit).
1282  * 5:Update holecheck_path and orig_path to points a next proceeding extent,
1283  *   then returns to step 2.
1284  * 6:Release holecheck_path, orig_path and set the len to moved_len
1285  *   which shows the number of moved blocks.
1286  *   The moved_len is useful for the command to calculate the file offset
1287  *   for starting next move extent ioctl.
1288  * 7:Return 0 on success, or a negative error value on failure.
1289  */
1290 int
1291 ext4_move_extents(struct file *o_filp, struct file *d_filp,
1292                  __u64 orig_start, __u64 donor_start, __u64 len,
1293                  __u64 *moved_len)
1294 {
1295         struct inode *orig_inode = file_inode(o_filp);
1296         struct inode *donor_inode = file_inode(d_filp);
1297         struct ext4_ext_path *orig_path = NULL, *holecheck_path = NULL;
1298         struct ext4_extent *ext_prev, *ext_cur, *ext_dummy;
1299         ext4_lblk_t block_start = orig_start;
1300         ext4_lblk_t block_end, seq_start, add_blocks, file_end, seq_blocks = 0;
1301         ext4_lblk_t rest_blocks;
1302         pgoff_t orig_page_offset = 0, seq_end_page;
1303         int ret, depth, last_extent = 0;
1304         int blocks_per_page = PAGE_CACHE_SIZE >> orig_inode->i_blkbits;
1305         int data_offset_in_page;
1306         int block_len_in_page;
1307         int uninit;
1308
1309         if (orig_inode->i_sb != donor_inode->i_sb) {
1310                 ext4_debug("ext4 move extent: The argument files "
1311                         "should be in same FS [ino:orig %lu, donor %lu]\n",
1312                         orig_inode->i_ino, donor_inode->i_ino);
1313                 return -EINVAL;
1314         }
1315
1316         /* orig and donor should be different inodes */
1317         if (orig_inode == donor_inode) {
1318                 ext4_debug("ext4 move extent: The argument files should not "
1319                         "be same inode [ino:orig %lu, donor %lu]\n",
1320                         orig_inode->i_ino, donor_inode->i_ino);
1321                 return -EINVAL;
1322         }
1323
1324         /* Regular file check */
1325         if (!S_ISREG(orig_inode->i_mode) || !S_ISREG(donor_inode->i_mode)) {
1326                 ext4_debug("ext4 move extent: The argument files should be "
1327                         "regular file [ino:orig %lu, donor %lu]\n",
1328                         orig_inode->i_ino, donor_inode->i_ino);
1329                 return -EINVAL;
1330         }
1331         /* TODO: This is non obvious task to swap blocks for inodes with full
1332            jornaling enabled */
1333         if (ext4_should_journal_data(orig_inode) ||
1334             ext4_should_journal_data(donor_inode)) {
1335                 return -EINVAL;
1336         }
1337         /* Protect orig and donor inodes against a truncate */
1338         ext4_inode_double_lock(orig_inode, donor_inode);
1339
1340         /* Wait for all existing dio workers */
1341         ext4_inode_block_unlocked_dio(orig_inode);
1342         ext4_inode_block_unlocked_dio(donor_inode);
1343         inode_dio_wait(orig_inode);
1344         inode_dio_wait(donor_inode);
1345
1346         /* Protect extent tree against block allocations via delalloc */
1347         ext4_double_down_write_data_sem(orig_inode, donor_inode);
1348         /* Check the filesystem environment whether move_extent can be done */
1349         ret = mext_check_arguments(orig_inode, donor_inode, orig_start,
1350                                     donor_start, &len);
1351         if (ret)
1352                 goto out;
1353
1354         file_end = (i_size_read(orig_inode) - 1) >> orig_inode->i_blkbits;
1355         block_end = block_start + len - 1;
1356         if (file_end < block_end)
1357                 len -= block_end - file_end;
1358
1359         ret = get_ext_path(orig_inode, block_start, &orig_path);
1360         if (ret)
1361                 goto out;
1362
1363         /* Get path structure to check the hole */
1364         ret = get_ext_path(orig_inode, block_start, &holecheck_path);
1365         if (ret)
1366                 goto out;
1367
1368         depth = ext_depth(orig_inode);
1369         ext_cur = holecheck_path[depth].p_ext;
1370
1371         /*
1372          * Get proper starting location of block replacement if block_start was
1373          * within the hole.
1374          */
1375         if (le32_to_cpu(ext_cur->ee_block) +
1376                 ext4_ext_get_actual_len(ext_cur) - 1 < block_start) {
1377                 /*
1378                  * The hole exists between extents or the tail of
1379                  * original file.
1380                  */
1381                 last_extent = mext_next_extent(orig_inode,
1382                                         holecheck_path, &ext_cur);
1383                 if (last_extent < 0) {
1384                         ret = last_extent;
1385                         goto out;
1386                 }
1387                 last_extent = mext_next_extent(orig_inode, orig_path,
1388                                                         &ext_dummy);
1389                 if (last_extent < 0) {
1390                         ret = last_extent;
1391                         goto out;
1392                 }
1393                 seq_start = le32_to_cpu(ext_cur->ee_block);
1394         } else if (le32_to_cpu(ext_cur->ee_block) > block_start)
1395                 /* The hole exists at the beginning of original file. */
1396                 seq_start = le32_to_cpu(ext_cur->ee_block);
1397         else
1398                 seq_start = block_start;
1399
1400         /* No blocks within the specified range. */
1401         if (le32_to_cpu(ext_cur->ee_block) > block_end) {
1402                 ext4_debug("ext4 move extent: The specified range of file "
1403                                                         "may be the hole\n");
1404                 ret = -EINVAL;
1405                 goto out;
1406         }
1407
1408         /* Adjust start blocks */
1409         add_blocks = min(le32_to_cpu(ext_cur->ee_block) +
1410                          ext4_ext_get_actual_len(ext_cur), block_end + 1) -
1411                      max(le32_to_cpu(ext_cur->ee_block), block_start);
1412
1413         while (!last_extent && le32_to_cpu(ext_cur->ee_block) <= block_end) {
1414                 seq_blocks += add_blocks;
1415
1416                 /* Adjust tail blocks */
1417                 if (seq_start + seq_blocks - 1 > block_end)
1418                         seq_blocks = block_end - seq_start + 1;
1419
1420                 ext_prev = ext_cur;
1421                 last_extent = mext_next_extent(orig_inode, holecheck_path,
1422                                                 &ext_cur);
1423                 if (last_extent < 0) {
1424                         ret = last_extent;
1425                         break;
1426                 }
1427                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1428
1429                 /*
1430                  * Extend the length of contiguous block (seq_blocks)
1431                  * if extents are contiguous.
1432                  */
1433                 if (ext4_can_extents_be_merged(orig_inode,
1434                                                ext_prev, ext_cur) &&
1435                     block_end >= le32_to_cpu(ext_cur->ee_block) &&
1436                     !last_extent)
1437                         continue;
1438
1439                 /* Is original extent is uninitialized */
1440                 uninit = ext4_ext_is_uninitialized(ext_prev);
1441
1442                 data_offset_in_page = seq_start % blocks_per_page;
1443
1444                 /*
1445                  * Calculate data blocks count that should be swapped
1446                  * at the first page.
1447                  */
1448                 if (data_offset_in_page + seq_blocks > blocks_per_page) {
1449                         /* Swapped blocks are across pages */
1450                         block_len_in_page =
1451                                         blocks_per_page - data_offset_in_page;
1452                 } else {
1453                         /* Swapped blocks are in a page */
1454                         block_len_in_page = seq_blocks;
1455                 }
1456
1457                 orig_page_offset = seq_start >>
1458                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1459                 seq_end_page = (seq_start + seq_blocks - 1) >>
1460                                 (PAGE_CACHE_SHIFT - orig_inode->i_blkbits);
1461                 seq_start = le32_to_cpu(ext_cur->ee_block);
1462                 rest_blocks = seq_blocks;
1463
1464                 /*
1465                  * Up semaphore to avoid following problems:
1466                  * a. transaction deadlock among ext4_journal_start,
1467                  *    ->write_begin via pagefault, and jbd2_journal_commit
1468                  * b. racing with ->readpage, ->write_begin, and ext4_get_block
1469                  *    in move_extent_per_page
1470                  */
1471                 ext4_double_up_write_data_sem(orig_inode, donor_inode);
1472
1473                 while (orig_page_offset <= seq_end_page) {
1474
1475                         /* Swap original branches with new branches */
1476                         block_len_in_page = move_extent_per_page(
1477                                                 o_filp, donor_inode,
1478                                                 orig_page_offset,
1479                                                 data_offset_in_page,
1480                                                 block_len_in_page, uninit,
1481                                                 &ret);
1482
1483                         /* Count how many blocks we have exchanged */
1484                         *moved_len += block_len_in_page;
1485                         if (ret < 0)
1486                                 break;
1487                         if (*moved_len > len) {
1488                                 EXT4_ERROR_INODE(orig_inode,
1489                                         "We replaced blocks too much! "
1490                                         "sum of replaced: %llu requested: %llu",
1491                                         *moved_len, len);
1492                                 ret = -EIO;
1493                                 break;
1494                         }
1495
1496                         orig_page_offset++;
1497                         data_offset_in_page = 0;
1498                         rest_blocks -= block_len_in_page;
1499                         if (rest_blocks > blocks_per_page)
1500                                 block_len_in_page = blocks_per_page;
1501                         else
1502                                 block_len_in_page = rest_blocks;
1503                 }
1504
1505                 ext4_double_down_write_data_sem(orig_inode, donor_inode);
1506                 if (ret < 0)
1507                         break;
1508
1509                 /* Decrease buffer counter */
1510                 if (holecheck_path)
1511                         ext4_ext_drop_refs(holecheck_path);
1512                 ret = get_ext_path(orig_inode, seq_start, &holecheck_path);
1513                 if (ret)
1514                         break;
1515                 depth = holecheck_path->p_depth;
1516
1517                 /* Decrease buffer counter */
1518                 if (orig_path)
1519                         ext4_ext_drop_refs(orig_path);
1520                 ret = get_ext_path(orig_inode, seq_start, &orig_path);
1521                 if (ret)
1522                         break;
1523
1524                 ext_cur = holecheck_path[depth].p_ext;
1525                 add_blocks = ext4_ext_get_actual_len(ext_cur);
1526                 seq_blocks = 0;
1527
1528         }
1529 out:
1530         if (*moved_len) {
1531                 ext4_discard_preallocations(orig_inode);
1532                 ext4_discard_preallocations(donor_inode);
1533         }
1534
1535         if (orig_path) {
1536                 ext4_ext_drop_refs(orig_path);
1537                 kfree(orig_path);
1538         }
1539         if (holecheck_path) {
1540                 ext4_ext_drop_refs(holecheck_path);
1541                 kfree(holecheck_path);
1542         }
1543         ext4_double_up_write_data_sem(orig_inode, donor_inode);
1544         ext4_inode_resume_unlocked_dio(orig_inode);
1545         ext4_inode_resume_unlocked_dio(donor_inode);
1546         ext4_inode_double_unlock(orig_inode, donor_inode);
1547
1548         return ret;
1549 }