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 "ext4_jbd2.h"
  17 #include "ext4_extents.h"
  18
  19 /*
  20  * The contiguous blocks details which can be
  21  * represented by a single extent
  22  */
  23 struct list_blocks_struct {
  24         ext4_lblk_t first_block, last_block;
  25         ext4_fsblk_t first_pblock, last_pblock;
  26 };
  27
  28 static int finish_range(handle_t *handle, struct inode *inode,
  29                                 struct list_blocks_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);
  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, ext4_lblk_t blk_num,
  90                                 struct list_blocks_struct *lb)
  91 {
  92         int retval;
  93         /*
  94          * See if we can add on to the existing range (if it exists)
  95          */
  96         if (lb->first_pblock &&
  97                 (lb->last_pblock+1 == pblock) &&
  98                 (lb->last_block+1 == blk_num)) {
  99                 lb->last_pblock = pblock;
 100                 lb->last_block = blk_num;
 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 = blk_num;
 109
 110         return retval;
 111 }
 112
 113 static int update_ind_extent_range(handle_t *handle, struct inode *inode,
 114                                    ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 115                                    struct list_blocks_struct *lb)
 116 {
 117         struct buffer_head *bh;
 118         __le32 *i_data;
 119         int i, retval = 0;
 120         ext4_lblk_t blk_count = *blk_nump;
 121         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 122
 123         if (!pblock) {
 124                 /* Only update the file block number */
 125                 *blk_nump += max_entries;
 126                 return 0;
 127         }
 128
 129         bh = sb_bread(inode->i_sb, pblock);
 130         if (!bh)
 131                 return -EIO;
 132
 133         i_data = (__le32 *)bh->b_data;
 134         for (i = 0; i < max_entries; i++, blk_count++) {
 135                 if (i_data[i]) {
 136                         retval = update_extent_range(handle, inode,
 137                                                 le32_to_cpu(i_data[i]),
 138                                                 blk_count, lb);
 139                         if (retval)
 140                                 break;
 141                 }
 142         }
 143
 144         /* Update the file block number */
 145         *blk_nump = blk_count;
 146         put_bh(bh);
 147         return retval;
 148
 149 }
 150
 151 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
 152                                     ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 153                                     struct list_blocks_struct *lb)
 154 {
 155         struct buffer_head *bh;
 156         __le32 *i_data;
 157         int i, retval = 0;
 158         ext4_lblk_t blk_count = *blk_nump;
 159         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 160
 161         if (!pblock) {
 162                 /* Only update the file block number */
 163                 *blk_nump += max_entries * max_entries;
 164                 return 0;
 165         }
 166         bh = sb_bread(inode->i_sb, pblock);
 167         if (!bh)
 168                 return -EIO;
 169
 170         i_data = (__le32 *)bh->b_data;
 171         for (i = 0; i < max_entries; i++) {
 172                 if (i_data[i]) {
 173                         retval = update_ind_extent_range(handle, inode,
 174                                                 le32_to_cpu(i_data[i]),
 175                                                 &blk_count, lb);
 176                         if (retval)
 177                                 break;
 178                 } else {
 179                         /* Only update the file block number */
 180                         blk_count += max_entries;
 181                 }
 182         }
 183
 184         /* Update the file block number */
 185         *blk_nump = blk_count;
 186         put_bh(bh);
 187         return retval;
 188
 189 }
 190
 191 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
 192                                      ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 193                                      struct list_blocks_struct *lb)
 194 {
 195         struct buffer_head *bh;
 196         __le32 *i_data;
 197         int i, retval = 0;
 198         ext4_lblk_t blk_count = *blk_nump;
 199         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 200
 201         if (!pblock) {
 202                 /* Only update the file block number */
 203                 *blk_nump += max_entries * max_entries * max_entries;
 204                 return 0;
 205         }
 206         bh = sb_bread(inode->i_sb, pblock);
 207         if (!bh)
 208                 return -EIO;
 209
 210         i_data = (__le32 *)bh->b_data;
 211         for (i = 0; i < max_entries; i++) {
 212                 if (i_data[i]) {
 213                         retval = update_dind_extent_range(handle, inode,
 214                                                 le32_to_cpu(i_data[i]),
 215                                                 &blk_count, lb);
 216                         if (retval)
 217                                 break;
 218                 } else
 219                         /* Only update the file block number */
 220                         blk_count += max_entries * max_entries;
 221         }
 222         /* Update the file block number */
 223         *blk_nump = blk_count;
 224         put_bh(bh);
 225         return retval;
 226
 227 }
 228
 229 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
 230 {
 231         int retval = 0, needed;
 232
 233         if (ext4_handle_has_enough_credits(handle, EXT4_RESERVE_TRANS_BLOCKS+1))
 234                 return 0;
 235         /*
 236          * We are freeing a blocks. During this we touch
 237          * superblock, group descriptor and block bitmap.
 238          * So allocate a credit of 3. We may update
 239          * quota (user and group).
 240          */
 241         needed = 3 + 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
 242
 243         if (ext4_journal_extend(handle, needed) != 0)
 244                 retval = ext4_journal_restart(handle, needed);
 245
 246         return retval;
 247 }
 248
 249 static int free_dind_blocks(handle_t *handle,
 250                                 struct inode *inode, __le32 i_data)
 251 {
 252         int i;
 253         __le32 *tmp_idata;
 254         struct buffer_head *bh;
 255         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 256
 257         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 258         if (!bh)
 259                 return -EIO;
 260
 261         tmp_idata = (__le32 *)bh->b_data;
 262         for (i = 0; i < max_entries; i++) {
 263                 if (tmp_idata[i]) {
 264                         extend_credit_for_blkdel(handle, inode);
 265                         ext4_free_blocks(handle, inode,
 266                                         le32_to_cpu(tmp_idata[i]), 1, 1);
 267                 }
 268         }
 269         put_bh(bh);
 270         extend_credit_for_blkdel(handle, inode);
 271         ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 272         return 0;
 273 }
 274
 275 static int free_tind_blocks(handle_t *handle,
 276                                 struct inode *inode, __le32 i_data)
 277 {
 278         int i, retval = 0;
 279         __le32 *tmp_idata;
 280         struct buffer_head *bh;
 281         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 282
 283         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 284         if (!bh)
 285                 return -EIO;
 286
 287         tmp_idata = (__le32 *)bh->b_data;
 288         for (i = 0; i < max_entries; i++) {
 289                 if (tmp_idata[i]) {
 290                         retval = free_dind_blocks(handle,
 291                                         inode, tmp_idata[i]);
 292                         if (retval) {
 293                                 put_bh(bh);
 294                                 return retval;
 295                         }
 296                 }
 297         }
 298         put_bh(bh);
 299         extend_credit_for_blkdel(handle, inode);
 300         ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 301         return 0;
 302 }
 303
 304 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
 305 {
 306         int retval;
 307
 308         /* ei->i_data[EXT4_IND_BLOCK] */
 309         if (i_data[0]) {
 310                 extend_credit_for_blkdel(handle, inode);
 311                 ext4_free_blocks(handle, inode,
 312                                 le32_to_cpu(i_data[0]), 1, 1);
 313         }
 314
 315         /* ei->i_data[EXT4_DIND_BLOCK] */
 316         if (i_data[1]) {
 317                 retval = free_dind_blocks(handle, inode, i_data[1]);
 318                 if (retval)
 319                         return retval;
 320         }
 321
 322         /* ei->i_data[EXT4_TIND_BLOCK] */
 323         if (i_data[2]) {
 324                 retval = free_tind_blocks(handle, inode, i_data[2]);
 325                 if (retval)
 326                         return retval;
 327         }
 328         return 0;
 329 }
 330
 331 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 332                                                 struct inode *tmp_inode)
 333 {
 334         int retval;
 335         __le32  i_data[3];
 336         struct ext4_inode_info *ei = EXT4_I(inode);
 337         struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
 338
 339         /*
 340          * One credit accounted for writing the
 341          * i_data field of the original inode
 342          */
 343         retval = ext4_journal_extend(handle, 1);
 344         if (retval) {
 345                 retval = ext4_journal_restart(handle, 1);
 346                 if (retval)
 347                         goto err_out;
 348         }
 349
 350         i_data[0] = ei->i_data[EXT4_IND_BLOCK];
 351         i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
 352         i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
 353
 354         down_write(&EXT4_I(inode)->i_data_sem);
 355         /*
 356          * if EXT4_EXT_MIGRATE is cleared a block allocation
 357          * happened after we started the migrate. We need to
 358          * fail the migrate
 359          */
 360         if (!(EXT4_I(inode)->i_flags & EXT4_EXT_MIGRATE)) {
 361                 retval = -EAGAIN;
 362                 up_write(&EXT4_I(inode)->i_data_sem);
 363                 goto err_out;
 364         } else
 365                 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
 366                                                         ~EXT4_EXT_MIGRATE;
 367         /*
 368          * We have the extent map build with the tmp inode.
 369          * Now copy the i_data across
 370          */
 371         ei->i_flags |= EXT4_EXTENTS_FL;
 372         memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
 373
 374         /*
 375          * Update i_blocks with the new blocks that got
 376          * allocated while adding extents for extent index
 377          * blocks.
 378          *
 379          * While converting to extents we need not
 380          * update the orignal inode i_blocks for extent blocks
 381          * via quota APIs. The quota update happened via tmp_inode already.
 382          */
 383         spin_lock(&inode->i_lock);
 384         inode->i_blocks += tmp_inode->i_blocks;
 385         spin_unlock(&inode->i_lock);
 386         up_write(&EXT4_I(inode)->i_data_sem);
 387
 388         /*
 389          * We mark the inode dirty after, because we decrement the
 390          * i_blocks when freeing the indirect meta-data blocks
 391          */
 392         retval = free_ind_block(handle, inode, i_data);
 393         ext4_mark_inode_dirty(handle, inode);
 394
 395 err_out:
 396         return retval;
 397 }
 398
 399 static int free_ext_idx(handle_t *handle, struct inode *inode,
 400                                         struct ext4_extent_idx *ix)
 401 {
 402         int i, retval = 0;
 403         ext4_fsblk_t block;
 404         struct buffer_head *bh;
 405         struct ext4_extent_header *eh;
 406
 407         block = idx_pblock(ix);
 408         bh = sb_bread(inode->i_sb, block);
 409         if (!bh)
 410                 return -EIO;
 411
 412         eh = (struct ext4_extent_header *)bh->b_data;
 413         if (eh->eh_depth != 0) {
 414                 ix = EXT_FIRST_INDEX(eh);
 415                 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 416                         retval = free_ext_idx(handle, inode, ix);
 417                         if (retval)
 418                                 break;
 419                 }
 420         }
 421         put_bh(bh);
 422         extend_credit_for_blkdel(handle, inode);
 423         ext4_free_blocks(handle, inode, block, 1, 1);
 424         return retval;
 425 }
 426
 427 /*
 428  * Free the extent meta data blocks only
 429  */
 430 static int free_ext_block(handle_t *handle, struct inode *inode)
 431 {
 432         int i, retval = 0;
 433         struct ext4_inode_info *ei = EXT4_I(inode);
 434         struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
 435         struct ext4_extent_idx *ix;
 436         if (eh->eh_depth == 0)
 437                 /*
 438                  * No extra blocks allocated for extent meta data
 439                  */
 440                 return 0;
 441         ix = EXT_FIRST_INDEX(eh);
 442         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 443                 retval = free_ext_idx(handle, inode, ix);
 444                 if (retval)
 445                         return retval;
 446         }
 447         return retval;
 448
 449 }
 450
 451 int ext4_ext_migrate(struct inode *inode)
 452 {
 453         handle_t *handle;
 454         int retval = 0, i;
 455         __le32 *i_data;
 456         ext4_lblk_t blk_count = 0;
 457         struct ext4_inode_info *ei;
 458         struct inode *tmp_inode = NULL;
 459         struct list_blocks_struct lb;
 460         unsigned long max_entries;
 461
 462         /*
 463          * If the filesystem does not support extents, or the inode
 464          * already is extent-based, error out.
 465          */
 466         if (!EXT4_HAS_INCOMPAT_FEATURE(inode->i_sb,
 467                                        EXT4_FEATURE_INCOMPAT_EXTENTS) ||
 468             (EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
 469                 return -EINVAL;
 470
 471         if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
 472                 /*
 473                  * don't migrate fast symlink
 474                  */
 475                 return retval;
 476
 477         handle = ext4_journal_start(inode,
 478                                         EXT4_DATA_TRANS_BLOCKS(inode->i_sb) +
 479                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
 480                                         2 * EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)
 481                                         + 1);
 482         if (IS_ERR(handle)) {
 483                 retval = PTR_ERR(handle);
 484                 return retval;
 485         }
 486         tmp_inode = ext4_new_inode(handle,
 487                                 inode->i_sb->s_root->d_inode,
 488                                 S_IFREG);
 489         if (IS_ERR(tmp_inode)) {
 490                 retval = -ENOMEM;
 491                 ext4_journal_stop(handle);
 492                 return retval;
 493         }
 494         i_size_write(tmp_inode, i_size_read(inode));
 495         /*
 496          * We don't want the inode to be reclaimed
 497          * if we got interrupted in between. We have
 498          * this tmp inode carrying reference to the
 499          * data blocks of the original file. We set
 500          * the i_nlink to zero at the last stage after
 501          * switching the original file to extent format
 502          */
 503         tmp_inode->i_nlink = 1;
 504
 505         ext4_ext_tree_init(handle, tmp_inode);
 506         ext4_orphan_add(handle, tmp_inode);
 507         ext4_journal_stop(handle);
 508
 509         /*
 510          * start with one credit accounted for
 511          * superblock modification.
 512          *
 513          * For the tmp_inode we already have commited the
 514          * trascation that created the inode. Later as and
 515          * when we add extents we extent the journal
 516          */
 517         /*
 518          * Even though we take i_mutex we can still cause block allocation
 519          * via mmap write to holes. If we have allocated new blocks we fail
 520          * migrate.  New block allocation will clear EXT4_EXT_MIGRATE flag.
 521          * The flag is updated with i_data_sem held to prevent racing with
 522          * block allocation.
 523          */
 524         down_read((&EXT4_I(inode)->i_data_sem));
 525         EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags | EXT4_EXT_MIGRATE;
 526         up_read((&EXT4_I(inode)->i_data_sem));
 527
 528         handle = ext4_journal_start(inode, 1);
 529
 530         ei = EXT4_I(inode);
 531         i_data = ei->i_data;
 532         memset(&lb, 0, sizeof(lb));
 533
 534         /* 32 bit block address 4 bytes */
 535         max_entries = inode->i_sb->s_blocksize >> 2;
 536         for (i = 0; i < EXT4_NDIR_BLOCKS; i++, blk_count++) {
 537                 if (i_data[i]) {
 538                         retval = update_extent_range(handle, tmp_inode,
 539                                                 le32_to_cpu(i_data[i]),
 540                                                 blk_count, &lb);
 541                         if (retval)
 542                                 goto err_out;
 543                 }
 544         }
 545         if (i_data[EXT4_IND_BLOCK]) {
 546                 retval = update_ind_extent_range(handle, tmp_inode,
 547                                         le32_to_cpu(i_data[EXT4_IND_BLOCK]),
 548                                         &blk_count, &lb);
 549                         if (retval)
 550                                 goto err_out;
 551         } else
 552                 blk_count +=  max_entries;
 553         if (i_data[EXT4_DIND_BLOCK]) {
 554                 retval = update_dind_extent_range(handle, tmp_inode,
 555                                         le32_to_cpu(i_data[EXT4_DIND_BLOCK]),
 556                                         &blk_count, &lb);
 557                         if (retval)
 558                                 goto err_out;
 559         } else
 560                 blk_count += max_entries * max_entries;
 561         if (i_data[EXT4_TIND_BLOCK]) {
 562                 retval = update_tind_extent_range(handle, tmp_inode,
 563                                         le32_to_cpu(i_data[EXT4_TIND_BLOCK]),
 564                                         &blk_count, &lb);
 565                         if (retval)
 566                                 goto err_out;
 567         }
 568         /*
 569          * Build the last extent
 570          */
 571         retval = finish_range(handle, tmp_inode, &lb);
 572 err_out:
 573         if (retval)
 574                 /*
 575                  * Failure case delete the extent information with the
 576                  * tmp_inode
 577                  */
 578                 free_ext_block(handle, tmp_inode);
 579         else {
 580                 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
 581                 if (retval)
 582                         /*
 583                          * if we fail to swap inode data free the extent
 584                          * details of the tmp inode
 585                          */
 586                         free_ext_block(handle, tmp_inode);
 587         }
 588
 589         /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
 590         if (ext4_journal_extend(handle, 1) != 0)
 591                 ext4_journal_restart(handle, 1);
 592
 593         /*
 594          * Mark the tmp_inode as of size zero
 595          */
 596         i_size_write(tmp_inode, 0);
 597
 598         /*
 599          * set the  i_blocks count to zero
 600          * so that the ext4_delete_inode does the
 601          * right job
 602          *
 603          * We don't need to take the i_lock because
 604          * the inode is not visible to user space.
 605          */
 606         tmp_inode->i_blocks = 0;
 607
 608         /* Reset the extent details */
 609         ext4_ext_tree_init(handle, tmp_inode);
 610
 611         /*
 612          * Set the i_nlink to zero so that
 613          * generic_drop_inode really deletes the
 614          * inode
 615          */
 616         tmp_inode->i_nlink = 0;
 617
 618         ext4_journal_stop(handle);
 619
 620         iput(tmp_inode);
 621
 622         return retval;
 623 }