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/module.h>
  16 #include <linux/slab.h>
  17 #include "ext4_jbd2.h"
  18
  19 /*
  20  * The contiguous blocks details which can be
  21  * represented by a single extent
  22  */
  23 struct migrate_struct {
  24         ext4_lblk_t first_block, last_block, curr_block;
  25         ext4_fsblk_t first_pblock, last_pblock;
  26 };
  27
  28 static int finish_range(handle_t *handle, struct inode *inode,
  29                                 struct migrate_struct *lb)
  30
  31 {
  32         int retval = 0, needed;
  33         struct ext4_extent newext;
  34         struct ext4_ext_path *path;
  35         if (lb->first_pblock == 0)
  36                 return 0;
  37
  38         /* Add the extent to temp inode*/
  39         newext.ee_block = cpu_to_le32(lb->first_block);
  40         newext.ee_len   = cpu_to_le16(lb->last_block - lb->first_block + 1);
  41         ext4_ext_store_pblock(&newext, lb->first_pblock);
  42         path = ext4_ext_find_extent(inode, lb->first_block, NULL);
  43
  44         if (IS_ERR(path)) {
  45                 retval = PTR_ERR(path);
  46                 path = NULL;
  47                 goto err_out;
  48         }
  49
  50         /*
  51          * Calculate the credit needed to inserting this extent
  52          * Since we are doing this in loop we may accumalate extra
  53          * credit. But below we try to not accumalate too much
  54          * of them by restarting the journal.
  55          */
  56         needed = ext4_ext_calc_credits_for_single_extent(inode,
  57                     lb->last_block - lb->first_block + 1, path);
  58
  59         /*
  60          * Make sure the credit we accumalated is not really high
  61          */
  62         if (needed && ext4_handle_has_enough_credits(handle,
  63                                                 EXT4_RESERVE_TRANS_BLOCKS)) {
  64                 retval = ext4_journal_restart(handle, needed);
  65                 if (retval)
  66                         goto err_out;
  67         } else if (needed) {
  68                 retval = ext4_journal_extend(handle, needed);
  69                 if (retval) {
  70                         /*
  71                          * IF not able to extend the journal restart the journal
  72                          */
  73                         retval = ext4_journal_restart(handle, needed);
  74                         if (retval)
  75                                 goto err_out;
  76                 }
  77         }
  78         retval = ext4_ext_insert_extent(handle, inode, path, &newext, 0);
  79 err_out:
  80         if (path) {
  81                 ext4_ext_drop_refs(path);
  82                 kfree(path);
  83         }
  84         lb->first_pblock = 0;
  85         return retval;
  86 }
  87
  88 static int update_extent_range(handle_t *handle, struct inode *inode,
  89                                ext4_fsblk_t pblock, struct migrate_struct *lb)
  90 {
  91         int retval;
  92         /*
  93          * See if we can add on to the existing range (if it exists)
  94          */
  95         if (lb->first_pblock &&
  96                 (lb->last_pblock+1 == pblock) &&
  97                 (lb->last_block+1 == lb->curr_block)) {
  98                 lb->last_pblock = pblock;
  99                 lb->last_block = lb->curr_block;
 100                 lb->curr_block++;
 101                 return 0;
 102         }
 103         /*
 104          * Start a new range.
 105          */
 106         retval = finish_range(handle, inode, lb);
 107         lb->first_pblock = lb->last_pblock = pblock;
 108         lb->first_block = lb->last_block = lb->curr_block;
 109         lb->curr_block++;
 110         return retval;
 111 }
 112
 113 static int update_ind_extent_range(handle_t *handle, struct inode *inode,
 114                                    ext4_fsblk_t pblock,
 115                                    struct migrate_struct *lb)
 116 {
 117         struct buffer_head *bh;
 118         __le32 *i_data;
 119         int i, retval = 0;
 120         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 121
 122         bh = sb_bread(inode->i_sb, pblock);
 123         if (!bh)
 124                 return -EIO;
 125
 126         i_data = (__le32 *)bh->b_data;
 127         for (i = 0; i < max_entries; i++) {
 128                 if (i_data[i]) {
 129                         retval = update_extent_range(handle, inode,
 130                                                 le32_to_cpu(i_data[i]), lb);
 131                         if (retval)
 132                                 break;
 133                 } else {
 134                         lb->curr_block++;
 135                 }
 136         }
 137         put_bh(bh);
 138         return retval;
 139
 140 }
 141
 142 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
 143                                     ext4_fsblk_t pblock,
 144                                     struct migrate_struct *lb)
 145 {
 146         struct buffer_head *bh;
 147         __le32 *i_data;
 148         int i, retval = 0;
 149         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 150
 151         bh = sb_bread(inode->i_sb, pblock);
 152         if (!bh)
 153                 return -EIO;
 154
 155         i_data = (__le32 *)bh->b_data;
 156         for (i = 0; i < max_entries; i++) {
 157                 if (i_data[i]) {
 158                         retval = update_ind_extent_range(handle, inode,
 159                                                 le32_to_cpu(i_data[i]), lb);
 160                         if (retval)
 161                                 break;
 162                 } else {
 163                         /* Only update the file block number */
 164                         lb->curr_block += max_entries;
 165                 }
 166         }
 167         put_bh(bh);
 168         return retval;
 169
 170 }
 171
 172 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
 173                                     ext4_fsblk_t pblock,
 174                                     struct migrate_struct *lb)
 175 {
 176         struct buffer_head *bh;
 177         __le32 *i_data;
 178         int i, retval = 0;
 179         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 180
 181         bh = sb_bread(inode->i_sb, pblock);
 182         if (!bh)
 183                 return -EIO;
 184
 185         i_data = (__le32 *)bh->b_data;
 186         for (i = 0; i < max_entries; i++) {
 187                 if (i_data[i]) {
 188                         retval = update_dind_extent_range(handle, inode,
 189                                                 le32_to_cpu(i_data[i]), lb);
 190                         if (retval)
 191                                 break;
 192                 } else {
 193                         /* Only update the file block number */
 194                         lb->curr_block += max_entries * max_entries;
 195                 }
 196         }
 197         put_bh(bh);
 198         return retval;
 199
 200 }
 201
 202 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
 203 {
 204         int retval = 0, needed;
 205
 206         if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
 207                 return 0;
 208         /*
 209          * We are freeing a blocks. During this we touch
 210          * superblock, group descriptor and block bitmap.
 211          * So allocate a credit of 3. We may update
 212          * quota (user and group).
 213          */
 214         needed = 3 + EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
 215
 216         if (ext4_journal_extend(handle, needed) != 0)
 217                 retval = ext4_journal_restart(handle, needed);
 218
 219         return retval;
 220 }
 221
 222 static int free_dind_blocks(handle_t *handle,
 223                                 struct inode *inode, __le32 i_data)
 224 {
 225         int i;
 226         __le32 *tmp_idata;
 227         struct buffer_head *bh;
 228         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 229
 230         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 231         if (!bh)
 232                 return -EIO;
 233
 234         tmp_idata = (__le32 *)bh->b_data;
 235         for (i = 0; i < max_entries; i++) {
 236                 if (tmp_idata[i]) {
 237                         extend_credit_for_blkdel(handle, inode);
 238                         ext4_free_blocks(handle, inode, NULL,
 239                                          le32_to_cpu(tmp_idata[i]), 1,
 240                                          EXT4_FREE_BLOCKS_METADATA |
 241                                          EXT4_FREE_BLOCKS_FORGET);
 242                 }
 243         }
 244         put_bh(bh);
 245         extend_credit_for_blkdel(handle, inode);
 246         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
 247                          EXT4_FREE_BLOCKS_METADATA |
 248                          EXT4_FREE_BLOCKS_FORGET);
 249         return 0;
 250 }
 251
 252 static int free_tind_blocks(handle_t *handle,
 253                                 struct inode *inode, __le32 i_data)
 254 {
 255         int i, retval = 0;
 256         __le32 *tmp_idata;
 257         struct buffer_head *bh;
 258         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 259
 260         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 261         if (!bh)
 262                 return -EIO;
 263
 264         tmp_idata = (__le32 *)bh->b_data;
 265         for (i = 0; i < max_entries; i++) {
 266                 if (tmp_idata[i]) {
 267                         retval = free_dind_blocks(handle,
 268                                         inode, tmp_idata[i]);
 269                         if (retval) {
 270                                 put_bh(bh);
 271                                 return retval;
 272                         }
 273                 }
 274         }
 275         put_bh(bh);
 276         extend_credit_for_blkdel(handle, inode);
 277         ext4_free_blocks(handle, inode, NULL, le32_to_cpu(i_data), 1,
 278                          EXT4_FREE_BLOCKS_METADATA |
 279                          EXT4_FREE_BLOCKS_FORGET);
 280         return 0;
 281 }
 282
 283 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
 284 {
 285         int retval;
 286
 287         /* ei->i_data[EXT4_IND_BLOCK] */
 288         if (i_data[0]) {
 289                 extend_credit_for_blkdel(handle, inode);
 290                 ext4_free_blocks(handle, inode, NULL,
 291                                 le32_to_cpu(i_data[0]), 1,
 292                                  EXT4_FREE_BLOCKS_METADATA |
 293                                  EXT4_FREE_BLOCKS_FORGET);
 294         }
 295
 296         /* ei->i_data[EXT4_DIND_BLOCK] */
 297         if (i_data[1]) {
 298                 retval = free_dind_blocks(handle, inode, i_data[1]);
 299                 if (retval)
 300                         return retval;
 301         }
 302
 303         /* ei->i_data[EXT4_TIND_BLOCK] */
 304         if (i_data[2]) {
 305                 retval = free_tind_blocks(handle, inode, i_data[2]);
 306                 if (retval)
 307                         return retval;
 308         }
 309         return 0;
 310 }
 311
 312 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 313                                                 struct inode *tmp_inode)
 314 {
 315         int retval;
 316         __le32  i_data[3];
 317         struct ext4_inode_info *ei = EXT4_I(inode);
 318         struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
 319
 320         /*
 321          * One credit accounted for writing the
 322          * i_data field of the original inode
 323          */
 324         retval = ext4_journal_extend(handle, 1);
 325         if (retval) {
 326                 retval = ext4_journal_restart(handle, 1);
 327                 if (retval)
 328                         goto err_out;
 329         }
 330
 331         i_data[0] = ei->i_data[EXT4_IND_BLOCK];
 332         i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
 333         i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
 334
 335         down_write(&EXT4_I(inode)->i_data_sem);
 336         /*
 337          * if EXT4_STATE_EXT_MIGRATE is cleared a block allocation
 338          * happened after we started the migrate. We need to
 339          * fail the migrate
 340          */
 341         if (!ext4_test_inode_state(inode, EXT4_STATE_EXT_MIGRATE)) {
 342                 retval = -EAGAIN;
 343                 up_write(&EXT4_I(inode)->i_data_sem);
 344                 goto err_out;
 345         } else
 346                 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 347         /*
 348          * We have the extent map build with the tmp inode.
 349          * Now copy the i_data across
 350          */
 351         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
 352         memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
 353
 354         /*
 355          * Update i_blocks with the new blocks that got
 356          * allocated while adding extents for extent index
 357          * blocks.
 358          *
 359          * While converting to extents we need not
 360          * update the orignal inode i_blocks for extent blocks
 361          * via quota APIs. The quota update happened via tmp_inode already.
 362          */
 363         spin_lock(&inode->i_lock);
 364         inode->i_blocks += tmp_inode->i_blocks;
 365         spin_unlock(&inode->i_lock);
 366         up_write(&EXT4_I(inode)->i_data_sem);
 367
 368         /*
 369          * We mark the inode dirty after, because we decrement the
 370          * i_blocks when freeing the indirect meta-data blocks
 371          */
 372         retval = free_ind_block(handle, inode, i_data);
 373         ext4_mark_inode_dirty(handle, inode);
 374
 375 err_out:
 376         return retval;
 377 }
 378
 379 static int free_ext_idx(handle_t *handle, struct inode *inode,
 380                                         struct ext4_extent_idx *ix)
 381 {
 382         int i, retval = 0;
 383         ext4_fsblk_t block;
 384         struct buffer_head *bh;
 385         struct ext4_extent_header *eh;
 386
 387         block = ext4_idx_pblock(ix);
 388         bh = sb_bread(inode->i_sb, block);
 389         if (!bh)
 390                 return -EIO;
 391
 392         eh = (struct ext4_extent_header *)bh->b_data;
 393         if (eh->eh_depth != 0) {
 394                 ix = EXT_FIRST_INDEX(eh);
 395                 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 396                         retval = free_ext_idx(handle, inode, ix);
 397                         if (retval)
 398                                 break;
 399                 }
 400         }
 401         put_bh(bh);
 402         extend_credit_for_blkdel(handle, inode);
 403         ext4_free_blocks(handle, inode, NULL, block, 1,
 404                          EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
 405         return retval;
 406 }
 407
 408 /*
 409  * Free the extent meta data blocks only
 410  */
 411 static int free_ext_block(handle_t *handle, struct inode *inode)
 412 {
 413         int i, retval = 0;
 414         struct ext4_inode_info *ei = EXT4_I(inode);
 415         struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
 416         struct ext4_extent_idx *ix;
 417         if (eh->eh_depth == 0)
 418                 /*
 419                  * No extra blocks allocated for extent meta data
 420                  */
 421                 return 0;
 422         ix = EXT_FIRST_INDEX(eh);
 423         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 424                 retval = free_ext_idx(handle, inode, ix);
 425                 if (retval)
 426                         return retval;
 427         }
 428         return retval;
 429
 430 }
 431
 432 int ext4_ext_migrate(struct inode *inode)
 433 {
 434         handle_t *handle;
 435         int retval = 0, i;
 436         __le32 *i_data;
 437         struct ext4_inode_info *ei;
 438         struct inode *tmp_inode = NULL;
 439         struct migrate_struct lb;
 440         unsigned long max_entries;
 441         __u32 goal;
 442         uid_t owner[2];
 443
 444         /*
 445          * If the filesystem does not support extents, or the inode
 446          * already is extent-based, error out.
 447          */
 448         if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 449                                        EXT4_FEATURE_INCOMPAT_EXTENTS) ||
 450             (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
 451                 return -EINVAL;
 452
 453         if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
 454                 /*
 455                  * don't migrate fast symlink
 456                  */
 457                 return retval;
 458
 459         handle = ext4_journal_start(inode,
 460                                         EXT4_DATA_TRANS_BLOCKS(inode->i_sb) +
 461                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
 462                                         EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)
 463                                         + 1);
 464         if (IS_ERR(handle)) {
 465                 retval = PTR_ERR(handle);
 466                 return retval;
 467         }
 468         goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
 469                 EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;
 470         owner[0] = inode->i_uid;
 471         owner[1] = inode->i_gid;
 472         tmp_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
 473                                    S_IFREG, NULL, goal, owner);
 474         if (IS_ERR(tmp_inode)) {
 475                 retval = PTR_ERR(inode);
 476                 ext4_journal_stop(handle);
 477                 return retval;
 478         }
 479         i_size_write(tmp_inode, i_size_read(inode));
 480         /*
 481          * Set the i_nlink to zero so it will be deleted later
 482          * when we drop inode reference.
 483          */
 484         clear_nlink(tmp_inode);
 485
 486         ext4_ext_tree_init(handle, tmp_inode);
 487         ext4_orphan_add(handle, tmp_inode);
 488         ext4_journal_stop(handle);
 489
 490         /*
 491          * start with one credit accounted for
 492          * superblock modification.
 493          *
 494          * For the tmp_inode we already have committed the
 495          * trascation that created the inode. Later as and
 496          * when we add extents we extent the journal
 497          */
 498         /*
 499          * Even though we take i_mutex we can still cause block
 500          * allocation via mmap write to holes. If we have allocated
 501          * new blocks we fail migrate.  New block allocation will
 502          * clear EXT4_STATE_EXT_MIGRATE flag.  The flag is updated
 503          * with i_data_sem held to prevent racing with block
 504          * allocation.
 505          */
 506         down_read((&EXT4_I(inode)->i_data_sem));
 507         ext4_set_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
 508         up_read((&EXT4_I(inode)->i_data_sem));
 509
 510         handle = ext4_journal_start(inode, 1);
 511         if (IS_ERR(handle)) {
 512                 /*
 513                  * It is impossible to update on-disk structures without
 514                  * a handle, so just rollback in-core changes and live other
 515                  * work to orphan_list_cleanup()
 516                  */
 517                 ext4_orphan_del(NULL, tmp_inode);
 518                 retval = PTR_ERR(handle);
 519                 goto out;
 520         }
 521
 522         ei = EXT4_I(inode);
 523         i_data = ei->i_data;
 524         memset(&lb, 0, sizeof(lb));
 525
 526         /* 32 bit block address 4 bytes */
 527         max_entries = inode->i_sb->s_blocksize >> 2;
 528         for (i = 0; i < EXT4_NDIR_BLOCKS; i++) {
 529                 if (i_data[i]) {
 530                         retval = update_extent_range(handle, tmp_inode,
 531                                                 le32_to_cpu(i_data[i]), &lb);
 532                         if (retval)
 533                                 goto err_out;
 534                 } else
 535                         lb.curr_block++;
 536         }
 537         if (i_data[EXT4_IND_BLOCK]) {
 538                 retval = update_ind_extent_range(handle, tmp_inode,
 539                                 le32_to_cpu(i_data[EXT4_IND_BLOCK]), &lb);
 540                         if (retval)
 541                                 goto err_out;
 542         } else
 543                 lb.curr_block += max_entries;
 544         if (i_data[EXT4_DIND_BLOCK]) {
 545                 retval = update_dind_extent_range(handle, tmp_inode,
 546                                 le32_to_cpu(i_data[EXT4_DIND_BLOCK]), &lb);
 547                         if (retval)
 548                                 goto err_out;
 549         } else
 550                 lb.curr_block += max_entries * max_entries;
 551         if (i_data[EXT4_TIND_BLOCK]) {
 552                 retval = update_tind_extent_range(handle, tmp_inode,
 553                                 le32_to_cpu(i_data[EXT4_TIND_BLOCK]), &lb);
 554                         if (retval)
 555                                 goto err_out;
 556         }
 557         /*
 558          * Build the last extent
 559          */
 560         retval = finish_range(handle, tmp_inode, &lb);
 561 err_out:
 562         if (retval)
 563                 /*
 564                  * Failure case delete the extent information with the
 565                  * tmp_inode
 566                  */
 567                 free_ext_block(handle, tmp_inode);
 568         else {
 569                 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
 570                 if (retval)
 571                         /*
 572                          * if we fail to swap inode data free the extent
 573                          * details of the tmp inode
 574                          */
 575                         free_ext_block(handle, tmp_inode);
 576         }
 577
 578         /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
 579         if (ext4_journal_extend(handle, 1) != 0)
 580                 ext4_journal_restart(handle, 1);
 581
 582         /*
 583          * Mark the tmp_inode as of size zero
 584          */
 585         i_size_write(tmp_inode, 0);
 586
 587         /*
 588          * set the  i_blocks count to zero
 589          * so that the ext4_delete_inode does the
 590          * right job
 591          *
 592          * We don't need to take the i_lock because
 593          * the inode is not visible to user space.
 594          */
 595         tmp_inode->i_blocks = 0;
 596
 597         /* Reset the extent details */
 598         ext4_ext_tree_init(handle, tmp_inode);
 599         ext4_journal_stop(handle);
 600 out:
 601         unlock_new_inode(tmp_inode);
 602         iput(tmp_inode);
 603
 604         return retval;
 605 }