fs/ext4/migrate.c

   1 /*
   2  * Copyright IBM Corporation, 2007
   3  * Author Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of version 2.1 of the GNU Lesser General Public License
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful, but
  10  * WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  12  *
  13  */
  14
  15 #include <linux/slab.h>
  16 #include "ext4_jbd2.h"
  17
  18 /*
  19  * The contiguous blocks details which can be
  20  * represented by a single extent
  21  */
  22 struct migrate_struct {
  23         ext4_lblk_t first_block, last_block, curr_block;
  24         ext4_fsblk_t first_pblock, last_pblock;
  25 };
  26
  27 static int finish_range(handle_t *handle, struct inode *inode,
  28                                 struct migrate_struct *lb)
  29
  30 {
  31         int retval = 0, needed;
  32         struct ext4_extent newext;
  33         struct ext4_ext_path *path;
  34         if (lb->first_pblock == 0)
  35                 return 0;
  36
  37         /* Add the extent to temp inode*/
  38         newext.ee_block = cpu_to_le32(lb->first_block);
  39         newext.ee_len   = cpu_to_le16(lb->last_block - lb->first_block + 1);
  40         ext4_ext_store_pblock(&newext, lb->first_pblock);
  41         path = ext4_ext_find_extent(inode, lb->first_block, NULL);
  42
  43         if (IS_ERR(path)) {
  44                 retval = PTR_ERR(path);
  45                 path = NULL;
  46                 goto err_out;
  47         }
  48
  49         /*
  50          * Calculate the credit needed to inserting this extent
  51          * Since we are doing this in loop we may accumalate extra
  52          * credit. But below we try to not accumalate too much
  53          * of them by restarting the journal.
  54          */
  55         needed = ext4_ext_calc_credits_for_single_extent(inode,
  56                     lb->last_block - lb->first_block + 1, path);
  57
  58         /*
  59          * Make sure the credit we accumalated is not really high
  60          */
  61         if (needed && ext4_handle_has_enough_credits(handle,
  62                                                 EXT4_RESERVE_TRANS_BLOCKS)) {
  63                 retval = ext4_journal_restart(handle, needed);
  64                 if (retval)
  65                         goto err_out;
  66         } else if (needed) {
  67                 retval = ext4_journal_extend(handle, needed);
  68                 if (retval) {
  69                         /*
  70                          * IF not able to extend the journal restart the journal
  71                          */
  72                         retval = ext4_journal_restart(handle, needed);
  73                         if (retval)
  74                                 goto err_out;
  75                 }
  76         }
  77         retval = ext4_ext_insert_extent(handle, inode, path, &newext, 0);
  78 err_out:
  79         if (path) {
  80                 ext4_ext_drop_refs(path);
  81                 kfree(path);
  82         }
  83         lb->first_pblock = 0;
  84         return retval;
  85 }
  86
  87 static int update_extent_range(handle_t *handle, struct inode *inode,
  88                                ext4_fsblk_t pblock, struct migrate_struct *lb)
  89 {
  90         int retval;
  91         /*
  92          * See if we can add on to the existing range (if it exists)
  93          */
  94         if (lb->first_pblock &&
  95                 (lb->last_pblock+1 == pblock) &&
  96                 (lb->last_block+1 == lb->curr_block)) {
  97                 lb->last_pblock = pblock;
  98                 lb->last_block = lb->curr_block;
  99                 lb->curr_block++;
 100                 return 0;
 101         }
 102         /*
 103          * Start a new range.
 104          */
 105         retval = finish_range(handle, inode, lb);
 106         lb->first_pblock = lb->last_pblock = pblock;
 107         lb->first_block = lb->last_block = lb->curr_block;
 108         lb->curr_block++;
 109         return retval;
 110 }
 111
 112 static int update_ind_extent_range(handle_t *handle, struct inode *inode,
 113                                    ext4_fsblk_t pblock,
 114                                    struct migrate_struct *lb)
 115 {
 116         struct buffer_head *bh;
 117         __le32 *i_data;
 118         int i, retval = 0;
 119         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 120
 121         bh = sb_bread(inode->i_sb, pblock);
 122         if (!bh)
 123                 return -EIO;
 124
 125         i_data = (__le32 *)bh->b_data;
 126         for (i = 0; i < max_entries; i++) {
 127                 if (i_data[i]) {
 128                         retval = update_extent_range(handle, inode,
 129                                                 le32_to_cpu(i_data[i]), lb);
 130                         if (retval)
 131                                 break;
 132                 } else {
 133                         lb->curr_block++;
 134                 }
 135         }
 136         put_bh(bh);
 137         return retval;
 138
 139 }
 140
 141 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
 142                                     ext4_fsblk_t pblock,
 143                                     struct migrate_struct *lb)
 144 {
 145         struct buffer_head *bh;
 146         __le32 *i_data;
 147         int i, retval = 0;
 148         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 149
 150         bh = sb_bread(inode->i_sb, pblock);
 151         if (!bh)
 152                 return -EIO;
 153
 154         i_data = (__le32 *)bh->b_data;
 155         for (i = 0; i < max_entries; i++) {
 156                 if (i_data[i]) {
 157                         retval = update_ind_extent_range(handle, inode,
 158                                                 le32_to_cpu(i_data[i]), lb);
 159                         if (retval)
 160                                 break;
 161                 } else {
 162                         /* Only update the file block number */
 163                         lb->curr_block += max_entries;
 164                 }
 165         }
 166         put_bh(bh);
 167         return retval;
 168
 169 }
 170
 171 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
 172                                     ext4_fsblk_t pblock,
 173                                     struct migrate_struct *lb)
 174 {
 175         struct buffer_head *bh;
 176         __le32 *i_data;
 177         int i, retval = 0;
 178         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 179
 180         bh = sb_bread(inode->i_sb, pblock);
 181         if (!bh)
 182                 return -EIO;
 183
 184         i_data = (__le32 *)bh->b_data;
 185         for (i = 0; i < max_entries; i++) {
 186                 if (i_data[i]) {
 187                         retval = update_dind_extent_range(handle, inode,
 188                                                 le32_to_cpu(i_data[i]), lb);
 189                         if (retval)
 190                                 break;
 191                 } else {
 192                         /* Only update the file block number */
 193                         lb->curr_block += max_entries * max_entries;
 194                 }
 195         }
 196         put_bh(bh);
 197         return retval;
 198
 199 }
 200
 201 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
 202 {
 203         int retval = 0, needed;
 204
 205         if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
 206                 return 0;
 207         /*
 208          * We are freeing a blocks. During this we touch
 209          * superblock, group descriptor and block bitmap.
 210          * So allocate a credit of 3. We may update
 211          * quota (user and group).
 212          */
 213         needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
 214
 215         if (ext4_journal_extend(handle, needed) != 0)
 216                 retval = ext4_journal_restart(handle, needed);
 217
 218         return retval;
 219 }
 220
 221 static int free_dind_blocks(handle_t *handle,
 222                                 struct inode *inode, __le32 i_data)
 223 {
 224         int i;
 225         __le32 *tmp_idata;
 226         struct buffer_head *bh;
 227         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 228
 229         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 230         if (!bh)
 231                 return -EIO;
 232
 233         tmp_idata = (__le32 *)bh->b_data;
 234         for (i = 0; i < max_entries; i++) {
 235                 if (tmp_idata[i]) {
 236                         extend_credit_for_blkdel(handle, inode);
 237                         ext4_free_blocks(handle, inode, NULL,
 238                                          le32_to_cpu(tmp_idata[i]), 1,
 239                                          EXT4_FREE_BLOCKS_METADATA |
 240                                          EXT4_FREE_BLOCKS_FORGET);
 241                 }
 242         }
 243         put_bh(bh);
 244         extend_credit_for_blkdel(handle, inode);
 245         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
 246                          EXT4_FREE_BLOCKS_METADATA |
 247                          EXT4_FREE_BLOCKS_FORGET);
 248         return 0;
 249 }
 250
 251 static int free_tind_blocks(handle_t *handle,
 252                                 struct inode *inode, __le32 i_data)
 253 {
 254         int i, retval = 0;
 255         __le32 *tmp_idata;
 256         struct buffer_head *bh;
 257         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 258
 259         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 260         if (!bh)
 261                 return -EIO;
 262
 263         tmp_idata = (__le32 *)bh->b_data;
 264         for (i = 0; i < max_entries; i++) {
 265                 if (tmp_idata[i]) {
 266                         retval = free_dind_blocks(handle,
 267                                         inode, tmp_idata[i]);
 268                         if (retval) {
 269                                 put_bh(bh);
 270                                 return retval;
 271                         }
 272                 }
 273         }
 274         put_bh(bh);
 275         extend_credit_for_blkdel(handle, inode);
 276         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
 277                          EXT4_FREE_BLOCKS_METADATA |
 278                          EXT4_FREE_BLOCKS_FORGET);
 279         return 0;
 280 }
 281
 282 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
 283 {
 284         int retval;
 285
 286         /* ei->i_data[EXT4_IND_BLOCK] */
 287         if (i_data[0]) {
 288                 extend_credit_for_blkdel(handle, inode);
 289                 ext4_free_blocks(handle, inode, NULL,
 290                                 le32_to_cpu(i_data[0]), 1,
 291                                  EXT4_FREE_BLOCKS_METADATA |
 292                                  EXT4_FREE_BLOCKS_FORGET);
 293         }
 294
 295         /* ei->i_data[EXT4_DIND_BLOCK] */
 296         if (i_data[1]) {
 297                 retval = free_dind_blocks(handle, inode, i_data[1]);
 298                 if (retval)
 299                         return retval;
 300         }
 301
 302         /* ei->i_data[EXT4_TIND_BLOCK] */
 303         if (i_data[2]) {
 304                 retval = free_tind_blocks(handle, inode, i_data[2]);
 305                 if (retval)
 306                         return retval;
 307         }
 308         return 0;
 309 }
 310
 311 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 312                                                 struct inode *tmp_inode)
 313 {
 314         int retval;
 315         __le32  i_data[3];
 316         struct ext4_inode_info *ei = EXT4_I(inode);
 317         struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
 318
 319         /*
 320          * One credit accounted for writing the
 321          * i_data field of the original inode
 322          */
 323         retval = ext4_journal_extend(handle, 1);
 324         if (retval) {
 325                 retval = ext4_journal_restart(handle, 1);
 326                 if (retval)
 327                         goto err_out;
 328         }
 329
 330         i_data[0] = ei->i_data[EXT4_IND_BLOCK];
 331         i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
 332         i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
 333
 334         down_write(&EXT4_I(inode)->i_data_sem);
 335         /*
 336          * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
 337          * happened after we started the migrate. We need to
 338          * fail the migrate
 339          */
 340         if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
 341                 retval = -EAGAIN;
 342                 up_write(&EXT4_I(inode)->i_data_sem);
 343                 goto err_out;
 344         } else
 345                 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 346         /*
 347          * We have the extent map build with the tmp inode.
 348          * Now copy the i_data across
 349          */
 350         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
 351         memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
 352
 353         /*
 354          * Update i_blocks with the new blocks that got
 355          * allocated while adding extents for extent index
 356          * blocks.
 357          *
 358          * While converting to extents we need not
 359          * update the orignal inode i_blocks for extent blocks
 360          * via quota APIs. The quota update happened via tmp_inode already.
 361          */
 362         spin_lock(&inode->i_lock);
 363         inode->i_blocks += tmp_inode->i_blocks;
 364         spin_unlock(&inode->i_lock);
 365         up_write(&EXT4_I(inode)->i_data_sem);
 366
 367         /*
 368          * We mark the inode dirty after, because we decrement the
 369          * i_blocks when freeing the indirect meta-data blocks
 370          */
 371         retval = free_ind_block(handle, inode, i_data);
 372         ext4_mark_inode_dirty(handle, inode);
 373
 374 err_out:
 375         return retval;
 376 }
 377
 378 static int free_ext_idx(handle_t *handle, struct inode *inode,
 379                                         struct ext4_extent_idx *ix)
 380 {
 381         int i, retval = 0;
 382         ext4_fsblk_t block;
 383         struct buffer_head *bh;
 384         struct ext4_extent_header *eh;
 385
 386         block = ext4_idx_pblock(ix);
 387         bh = sb_bread(inode->i_sb, block);
 388         if (!bh)
 389                 return -EIO;
 390
 391         eh = (struct ext4_extent_header *)bh->b_data;
 392         if (eh->eh_depth != 0) {
 393                 ix = EXT_FIRST_INDEX(eh);
 394                 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 395                         retval = free_ext_idx(handle, inode, ix);
 396                         if (retval)
 397                                 break;
 398                 }
 399         }
 400         put_bh(bh);
 401         extend_credit_for_blkdel(handle, inode);
 402         ext4_free_blocks(handle, inode, NULL, block, 1,
 403                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
 404         return retval;
 405 }
 406
 407 /*
 408  * Free the extent meta data blocks only
 409  */
 410 static int free_ext_block(handle_t *handle, struct inode *inode)
 411 {
 412         int i, retval = 0;
 413         struct ext4_inode_info *ei = EXT4_I(inode);
 414         struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
 415         struct ext4_extent_idx *ix;
 416         if (eh->eh_depth == 0)
 417                 /*
 418                  * No extra blocks allocated for extent meta data
 419                  */
 420                 return 0;
 421         ix = EXT_FIRST_INDEX(eh);
 422         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 423                 retval = free_ext_idx(handle, inode, ix);
 424                 if (retval)
 425                         return retval;
 426         }
 427         return retval;
 428
 429 }
 430
 431 int ext4_ext_migrate(struct inode *inode)
 432 {
 433         handle_t *handle;
 434         int retval = 0, i;
 435         __le32 *i_data;
 436         struct ext4_inode_info *ei;
 437         struct inode *tmp_inode = NULL;
 438         struct migrate_struct lb;
 439         unsigned long max_entries;
 440         __u32 goal;
 441         uid_t owner[2];
 442
 443         /*
 444          * If the filesystem does not support extents, or the inode
 445          * already is extent-based, error out.
 446          */
 447         if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 448                                        EXT4_FEATURE_INCOMPAT_EXTENTS) ||
 449             (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 450                 return -EINVAL;
 451
 452         if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
 453                 /*
 454                  * don't migrate fast symlink
 455                  */
 456                 return retval;
 457
 458         handle = ext4_journal_start(inode,
 459                                         EXT4_DATA_TRANS_BLOCKS(inode->i_sb) +
 460                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
 461                                         EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)
 462                                         + 1);
 463         if (IS_ERR(handle)) {
 464                 retval = PTR_ERR(handle);
 465                 return retval;
 466         }
 467         goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
 468                 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
 469         owner[0] = inode->i_uid;
 470         owner[1] = inode->i_gid;
 471         tmp_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
 472                                    S_IFREG, NULL, goal, owner);
 473         if (IS_ERR(tmp_inode)) {
 474                 retval = PTR_ERR(tmp_inode);
 475                 ext4_journal_stop(handle);
 476                 return retval;
 477         }
 478         i_size_write(tmp_inode, i_size_read(inode));
 479         /*
 480          * Set the i_nlink to zero so it will be deleted later
 481          * when we drop inode reference.
 482          */
 483         clear_nlink(tmp_inode);
 484
 485         ext4_ext_tree_init(handle, tmp_inode);
 486         ext4_orphan_add(handle, tmp_inode);
 487         ext4_journal_stop(handle);
 488
 489         /*
 490          * start with one credit accounted for
 491          * superblock modification.
 492          *
 493          * For the tmp_inode we already have committed the
 494          * trascation that created the inode. Later as and
 495          * when we add extents we extent the journal
 496          */
 497         /*
 498          * Even though we take i_mutex we can still cause block
 499          * allocation via mmap write to holes. If we have allocated
 500          * new blocks we fail migrate.  New block allocation will
 501          * clear EXT4_STATE_EXT_MIGRATE flag.  The flag is updated
 502          * with i_data_sem held to prevent racing with block
 503          * allocation.
 504          */
 505         down_read((&EXT4_I(inode)->i_data_sem));
 506         ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 507         up_read((&EXT4_I(inode)->i_data_sem));
 508
 509         handle = ext4_journal_start(inode, 1);
 510         if (IS_ERR(handle)) {
 511                 /*
 512                  * It is impossible to update on-disk structures without
 513                  * a handle, so just rollback in-core changes and live other
 514                  * work to orphan_list_cleanup()
 515                  */
 516                 ext4_orphan_del(NULL, tmp_inode);
 517                 retval = PTR_ERR(handle);
 518                 goto out;
 519         }
 520
 521         ei = EXT4_I(inode);
 522         i_data = ei->i_data;
 523         memset(&lb, 0, sizeof(lb));
 524
 525         /* 32 bit block address 4 bytes */
 526         max_entries = inode->i_sb->s_blocksize >> 2;
 527         for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
 528                 if (i_data[i]) {
 529                         retval = update_extent_range(handle, tmp_inode,
 530                                                 le32_to_cpu(i_data[i]), &lb);
 531                         if (retval)
 532                                 goto err_out;
 533                 } else
 534                         lb.curr_block++;
 535         }
 536         if (i_data[EXT4_IND_BLOCK]) {
 537                 retval = update_ind_extent_range(handle, tmp_inode,
 538                                 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
 539                         if (retval)
 540                                 goto err_out;
 541         } else
 542                 lb.curr_block += max_entries;
 543         if (i_data[EXT4_DIND_BLOCK]) {
 544                 retval = update_dind_extent_range(handle, tmp_inode,
 545                                 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
 546                         if (retval)
 547                                 goto err_out;
 548         } else
 549                 lb.curr_block += max_entries * max_entries;
 550         if (i_data[EXT4_TIND_BLOCK]) {
 551                 retval = update_tind_extent_range(handle, tmp_inode,
 552                                 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
 553                         if (retval)
 554                                 goto err_out;
 555         }
 556         /*
 557          * Build the last extent
 558          */
 559         retval = finish_range(handle, tmp_inode, &lb);
 560 err_out:
 561         if (retval)
 562                 /*
 563                  * Failure case delete the extent information with the
 564                  * tmp_inode
 565                  */
 566                 free_ext_block(handle, tmp_inode);
 567         else {
 568                 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
 569                 if (retval)
 570                         /*
 571                          * if we fail to swap inode data free the extent
 572                          * details of the tmp inode
 573                          */
 574                         free_ext_block(handle, tmp_inode);
 575         }
 576
 577         /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
 578         if (ext4_journal_extend(handle, 1) != 0)
 579                 ext4_journal_restart(handle, 1);
 580
 581         /*
 582          * Mark the tmp_inode as of size zero
 583          */
 584         i_size_write(tmp_inode, 0);
 585
 586         /*
 587          * set the  i_blocks count to zero
 588          * so that the ext4_delete_inode does the
 589          * right job
 590          *
 591          * We don't need to take the i_lock because
 592          * the inode is not visible to user space.
 593          */
 594         tmp_inode->i_blocks = 0;
 595
 596         /* Reset the extent details */
 597         ext4_ext_tree_init(handle, tmp_inode);
 598         ext4_journal_stop(handle);
 599 out:
 600         unlock_new_inode(tmp_inode);
 601         iput(tmp_inode);
 602
 603         return retval;
 604 }