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/ext4_jbd2.h>
  17 #include <linux/ext4_fs_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_insert(inode, path);
  57
  58         /*
  59          * Make sure the credit we accumalated is not really high
  60          */
  61         if (needed && handle->h_buffer_credits >= EXT4_RESERVE_TRANS_BLOCKS) {
  62                 retval = ext4_journal_restart(handle, needed);
  63                 if (retval)
  64                         goto err_out;
  65         } else if (needed) {
  66                 retval = ext4_journal_extend(handle, needed);
  67                 if (retval) {
  68                         /*
  69                          * IF not able to extend the journal restart the journal
  70                          */
  71                         retval = ext4_journal_restart(handle, needed);
  72                         if (retval)
  73                                 goto err_out;
  74                 }
  75         }
  76         retval = ext4_ext_insert_extent(handle, inode, path, &newext);
  77 err_out:
  78         if (path) {
  79                 ext4_ext_drop_refs(path);
  80                 kfree(path);
  81         }
  82         lb->first_pblock = 0;
  83         return retval;
  84 }
  85
  86 static int update_extent_range(handle_t *handle, struct inode *inode,
  87                                 ext4_fsblk_t pblock, ext4_lblk_t blk_num,
  88                                 struct list_blocks_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 == blk_num)) {
  97                 lb->last_pblock = pblock;
  98                 lb->last_block = blk_num;
  99                 return 0;
 100         }
 101         /*
 102          * Start a new range.
 103          */
 104         retval = finish_range(handle, inode, lb);
 105         lb->first_pblock = lb->last_pblock = pblock;
 106         lb->first_block = lb->last_block = blk_num;
 107
 108         return retval;
 109 }
 110
 111 static int update_ind_extent_range(handle_t *handle, struct inode *inode,
 112                                    ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 113                                    struct list_blocks_struct *lb)
 114 {
 115         struct buffer_head *bh;
 116         __le32 *i_data;
 117         int i, retval = 0;
 118         ext4_lblk_t blk_count = *blk_nump;
 119         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 120
 121         if (!pblock) {
 122                 /* Only update the file block number */
 123                 *blk_nump += max_entries;
 124                 return 0;
 125         }
 126
 127         bh = sb_bread(inode->i_sb, pblock);
 128         if (!bh)
 129                 return -EIO;
 130
 131         i_data = (__le32 *)bh->b_data;
 132         for (i = 0; i < max_entries; i++, blk_count++) {
 133                 if (i_data[i]) {
 134                         retval = update_extent_range(handle, inode,
 135                                                 le32_to_cpu(i_data[i]),
 136                                                 blk_count, lb);
 137                         if (retval)
 138                                 break;
 139                 }
 140         }
 141
 142         /* Update the file block number */
 143         *blk_nump = blk_count;
 144         put_bh(bh);
 145         return retval;
 146
 147 }
 148
 149 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
 150                                     ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 151                                     struct list_blocks_struct *lb)
 152 {
 153         struct buffer_head *bh;
 154         __le32 *i_data;
 155         int i, retval = 0;
 156         ext4_lblk_t blk_count = *blk_nump;
 157         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 158
 159         if (!pblock) {
 160                 /* Only update the file block number */
 161                 *blk_nump += max_entries * max_entries;
 162                 return 0;
 163         }
 164         bh = sb_bread(inode->i_sb, pblock);
 165         if (!bh)
 166                 return -EIO;
 167
 168         i_data = (__le32 *)bh->b_data;
 169         for (i = 0; i < max_entries; i++) {
 170                 if (i_data[i]) {
 171                         retval = update_ind_extent_range(handle, inode,
 172                                                 le32_to_cpu(i_data[i]),
 173                                                 &blk_count, lb);
 174                         if (retval)
 175                                 break;
 176                 } else {
 177                         /* Only update the file block number */
 178                         blk_count += max_entries;
 179                 }
 180         }
 181
 182         /* Update the file block number */
 183         *blk_nump = blk_count;
 184         put_bh(bh);
 185         return retval;
 186
 187 }
 188
 189 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
 190                                      ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 191                                      struct list_blocks_struct *lb)
 192 {
 193         struct buffer_head *bh;
 194         __le32 *i_data;
 195         int i, retval = 0;
 196         ext4_lblk_t blk_count = *blk_nump;
 197         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 198
 199         if (!pblock) {
 200                 /* Only update the file block number */
 201                 *blk_nump += max_entries * max_entries * max_entries;
 202                 return 0;
 203         }
 204         bh = sb_bread(inode->i_sb, pblock);
 205         if (!bh)
 206                 return -EIO;
 207
 208         i_data = (__le32 *)bh->b_data;
 209         for (i = 0; i < max_entries; i++) {
 210                 if (i_data[i]) {
 211                         retval = update_dind_extent_range(handle, inode,
 212                                                 le32_to_cpu(i_data[i]),
 213                                                 &blk_count, lb);
 214                         if (retval)
 215                                 break;
 216                 } else
 217                         /* Only update the file block number */
 218                         blk_count += max_entries * max_entries;
 219         }
 220         /* Update the file block number */
 221         *blk_nump = blk_count;
 222         put_bh(bh);
 223         return retval;
 224
 225 }
 226
 227 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
 228 {
 229         int retval = 0, needed;
 230
 231         if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
 232                 return 0;
 233         /*
 234          * We are freeing a blocks. During this we touch
 235          * superblock, group descriptor and block bitmap.
 236          * So allocate a credit of 3. We may update
 237          * quota (user and group).
 238          */
 239         needed = 3 + 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
 240
 241         if (ext4_journal_extend(handle, needed) != 0)
 242                 retval = ext4_journal_restart(handle, needed);
 243
 244         return retval;
 245 }
 246
 247 static int free_dind_blocks(handle_t *handle,
 248                                 struct inode *inode, __le32 i_data)
 249 {
 250         int i;
 251         __le32 *tmp_idata;
 252         struct buffer_head *bh;
 253         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 254
 255         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 256         if (!bh)
 257                 return -EIO;
 258
 259         tmp_idata = (__le32 *)bh->b_data;
 260         for (i = 0; i < max_entries; i++) {
 261                 if (tmp_idata[i]) {
 262                         extend_credit_for_blkdel(handle, inode);
 263                         ext4_free_blocks(handle, inode,
 264                                         le32_to_cpu(tmp_idata[i]), 1, 1);
 265                 }
 266         }
 267         put_bh(bh);
 268         extend_credit_for_blkdel(handle, inode);
 269         ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 270         return 0;
 271 }
 272
 273 static int free_tind_blocks(handle_t *handle,
 274                                 struct inode *inode, __le32 i_data)
 275 {
 276         int i, retval = 0;
 277         __le32 *tmp_idata;
 278         struct buffer_head *bh;
 279         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 280
 281         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 282         if (!bh)
 283                 return -EIO;
 284
 285         tmp_idata = (__le32 *)bh->b_data;
 286         for (i = 0; i < max_entries; i++) {
 287                 if (tmp_idata[i]) {
 288                         retval = free_dind_blocks(handle,
 289                                         inode, tmp_idata[i]);
 290                         if (retval) {
 291                                 put_bh(bh);
 292                                 return retval;
 293                         }
 294                 }
 295         }
 296         put_bh(bh);
 297         extend_credit_for_blkdel(handle, inode);
 298         ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 299         return 0;
 300 }
 301
 302 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
 303 {
 304         int retval;
 305
 306         /* ei->i_data[EXT4_IND_BLOCK] */
 307         if (i_data[0]) {
 308                 extend_credit_for_blkdel(handle, inode);
 309                 ext4_free_blocks(handle, inode,
 310                                 le32_to_cpu(i_data[0]), 1, 1);
 311         }
 312
 313         /* ei->i_data[EXT4_DIND_BLOCK] */
 314         if (i_data[1]) {
 315                 retval = free_dind_blocks(handle, inode, i_data[1]);
 316                 if (retval)
 317                         return retval;
 318         }
 319
 320         /* ei->i_data[EXT4_TIND_BLOCK] */
 321         if (i_data[2]) {
 322                 retval = free_tind_blocks(handle, inode, i_data[2]);
 323                 if (retval)
 324                         return retval;
 325         }
 326         return 0;
 327 }
 328
 329 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 330                                 struct inode *tmp_inode)
 331 {
 332         int retval;
 333         __le32  i_data[3];
 334         struct ext4_inode_info *ei = EXT4_I(inode);
 335         struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
 336
 337         /*
 338          * One credit accounted for writing the
 339          * i_data field of the original inode
 340          */
 341         retval = ext4_journal_extend(handle, 1);
 342         if (retval != 0) {
 343                 retval = ext4_journal_restart(handle, 1);
 344                 if (retval)
 345                         goto err_out;
 346         }
 347
 348         i_data[0] = ei->i_data[EXT4_IND_BLOCK];
 349         i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
 350         i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
 351
 352         down_write(&EXT4_I(inode)->i_data_sem);
 353         /*
 354          * We have the extent map build with the tmp inode.
 355          * Now copy the i_data across
 356          */
 357         ei->i_flags |= EXT4_EXTENTS_FL;
 358         memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
 359
 360         /*
 361          * Update i_blocks with the new blocks that got
 362          * allocated while adding extents for extent index
 363          * blocks.
 364          *
 365          * While converting to extents we need not
 366          * update the orignal inode i_blocks for extent blocks
 367          * via quota APIs. The quota update happened via tmp_inode already.
 368          */
 369         spin_lock(&inode->i_lock);
 370         inode->i_blocks += tmp_inode->i_blocks;
 371         spin_unlock(&inode->i_lock);
 372         up_write(&EXT4_I(inode)->i_data_sem);
 373
 374         /*
 375          * We mark the inode dirty after, because we decrement the
 376          * i_blocks when freeing the indirect meta-data blocks
 377          */
 378         retval = free_ind_block(handle, inode, i_data);
 379         ext4_mark_inode_dirty(handle, inode);
 380
 381 err_out:
 382         return retval;
 383 }
 384
 385 static int free_ext_idx(handle_t *handle, struct inode *inode,
 386                                         struct ext4_extent_idx *ix)
 387 {
 388         int i, retval = 0;
 389         ext4_fsblk_t block;
 390         struct buffer_head *bh;
 391         struct ext4_extent_header *eh;
 392
 393         block = idx_pblock(ix);
 394         bh = sb_bread(inode->i_sb, block);
 395         if (!bh)
 396                 return -EIO;
 397
 398         eh = (struct ext4_extent_header *)bh->b_data;
 399         if (eh->eh_depth != 0) {
 400                 ix = EXT_FIRST_INDEX(eh);
 401                 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 402                         retval = free_ext_idx(handle, inode, ix);
 403                         if (retval)
 404                                 break;
 405                 }
 406         }
 407         put_bh(bh);
 408         extend_credit_for_blkdel(handle, inode);
 409         ext4_free_blocks(handle, inode, block, 1, 1);
 410         return retval;
 411 }
 412
 413 /*
 414  * Free the extent meta data blocks only
 415  */
 416 static int free_ext_block(handle_t *handle, struct inode *inode)
 417 {
 418         int i, retval = 0;
 419         struct ext4_inode_info *ei = EXT4_I(inode);
 420         struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
 421         struct ext4_extent_idx *ix;
 422         if (eh->eh_depth == 0)
 423                 /*
 424                  * No extra blocks allocated for extent meta data
 425                  */
 426                 return 0;
 427         ix = EXT_FIRST_INDEX(eh);
 428         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 429                 retval = free_ext_idx(handle, inode, ix);
 430                 if (retval)
 431                         return retval;
 432         }
 433         return retval;
 434
 435 }
 436
 437 int ext4_ext_migrate(struct inode *inode, struct file *filp,
 438                                 unsigned int cmd, unsigned long arg)
 439 {
 440         handle_t *handle;
 441         int retval = 0, i;
 442         __le32 *i_data;
 443         ext4_lblk_t blk_count = 0;
 444         struct ext4_inode_info *ei;
 445         struct inode *tmp_inode = NULL;
 446         struct list_blocks_struct lb;
 447         unsigned long max_entries;
 448
 449         if (!test_opt(inode->i_sb, EXTENTS))
 450                 /*
 451                  * if mounted with noextents we don't allow the migrate
 452                  */
 453                 return -EINVAL;
 454
 455         if ((EXT4_I(inode)->i_flags & EXT4_EXTENTS_FL))
 456                 return -EINVAL;
 457
 458         if (S_ISLNK(inode->i_mode) && inode->i_blocks == 0)
 459                 /*
 460                  * don't migrate fast symlink
 461                  */
 462                 return retval;
 463
 464         handle = ext4_journal_start(inode,
 465                                         EXT4_DATA_TRANS_BLOCKS(inode->i_sb) +
 466                                         EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
 467                                         2 * EXT4_QUOTA_INIT_BLOCKS(inode->i_sb)
 468                                         + 1);
 469         if (IS_ERR(handle)) {
 470                 retval = PTR_ERR(handle);
 471                 goto err_out;
 472         }
 473         tmp_inode = ext4_new_inode(handle,
 474                                 inode->i_sb->s_root->d_inode,
 475                                 S_IFREG);
 476         if (IS_ERR(tmp_inode)) {
 477                 retval = -ENOMEM;
 478                 ext4_journal_stop(handle);
 479                 tmp_inode = NULL;
 480                 goto err_out;
 481         }
 482         i_size_write(tmp_inode, i_size_read(inode));
 483         /*
 484          * We don't want the inode to be reclaimed
 485          * if we got interrupted in between. We have
 486          * this tmp inode carrying reference to the
 487          * data blocks of the original file. We set
 488          * the i_nlink to zero at the last stage after
 489          * switching the original file to extent format
 490          */
 491         tmp_inode->i_nlink = 1;
 492
 493         ext4_ext_tree_init(handle, tmp_inode);
 494         ext4_orphan_add(handle, tmp_inode);
 495         ext4_journal_stop(handle);
 496
 497         /*
 498          * start with one credit accounted for
 499          * superblock modification.
 500          *
 501          * For the tmp_inode we already have commited the
 502          * trascation that created the inode. Later as and
 503          * when we add extents we extent the journal
 504          */
 505         /*
 506          * inode_mutex prevent write and truncate on the file. Read still goes
 507          * through. We take i_data_sem in ext4_ext_swap_inode_data before we
 508          * switch the inode format to prevent read.
 509          */
 510         mutex_lock(&(inode->i_mutex));
 511         handle = ext4_journal_start(inode, 1);
 512
 513         ei = EXT4_I(inode);
 514         i_data = ei->i_data;
 515         memset(&lb, 0, sizeof(lb));
 516
 517         /* 32 bit block address 4 bytes */
 518         max_entries = inode->i_sb->s_blocksize >> 2;
 519         for (i = 0; i < EXT4_NDIR_BLOCKS; i++, blk_count++) {
 520                 if (i_data[i]) {
 521                         retval = update_extent_range(handle, tmp_inode,
 522                                                 le32_to_cpu(i_data[i]),
 523                                                 blk_count, &lb);
 524                         if (retval)
 525                                 goto err_out;
 526                 }
 527         }
 528         if (i_data[EXT4_IND_BLOCK]) {
 529                 retval = update_ind_extent_range(handle, tmp_inode,
 530                                         le32_to_cpu(i_data[EXT4_IND_BLOCK]),
 531                                         &blk_count, &lb);
 532                         if (retval)
 533                                 goto err_out;
 534         } else
 535                 blk_count +=  max_entries;
 536         if (i_data[EXT4_DIND_BLOCK]) {
 537                 retval = update_dind_extent_range(handle, tmp_inode,
 538                                         le32_to_cpu(i_data[EXT4_DIND_BLOCK]),
 539                                         &blk_count, &lb);
 540                         if (retval)
 541                                 goto err_out;
 542         } else
 543                 blk_count += max_entries * max_entries;
 544         if (i_data[EXT4_TIND_BLOCK]) {
 545                 retval = update_tind_extent_range(handle, tmp_inode,
 546                                         le32_to_cpu(i_data[EXT4_TIND_BLOCK]),
 547                                         &blk_count, &lb);
 548                         if (retval)
 549                                 goto err_out;
 550         }
 551         /*
 552          * Build the last extent
 553          */
 554         retval = finish_range(handle, tmp_inode, &lb);
 555 err_out:
 556         if (retval)
 557                 /*
 558                  * Failure case delete the extent information with the
 559                  * tmp_inode
 560                  */
 561                 free_ext_block(handle, tmp_inode);
 562         else
 563                 retval = ext4_ext_swap_inode_data(handle, inode,
 564                                                         tmp_inode);
 565
 566         /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
 567         if (ext4_journal_extend(handle, 1) != 0)
 568                 ext4_journal_restart(handle, 1);
 569
 570         /*
 571          * Mark the tmp_inode as of size zero
 572          */
 573         i_size_write(tmp_inode, 0);
 574
 575         /*
 576          * set the  i_blocks count to zero
 577          * so that the ext4_delete_inode does the
 578          * right job
 579          *
 580          * We don't need to take the i_lock because
 581          * the inode is not visible to user space.
 582          */
 583         tmp_inode->i_blocks = 0;
 584
 585         /* Reset the extent details */
 586         ext4_ext_tree_init(handle, tmp_inode);
 587
 588         /*
 589          * Set the i_nlink to zero so that
 590          * generic_drop_inode really deletes the
 591          * inode
 592          */
 593         tmp_inode->i_nlink = 0;
 594
 595         ext4_journal_stop(handle);
 596         mutex_unlock(&(inode->i_mutex));
 597
 598         if (tmp_inode)
 599                 iput(tmp_inode);
 600
 601         return retval;
 602 }