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