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 && handle->h_buffer_credits >= 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);
  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, ext4_lblk_t blk_num,
  89                                 struct list_blocks_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 == blk_num)) {
  98                 lb->last_pblock = pblock;
  99                 lb->last_block = blk_num;
 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 = blk_num;
 108
 109         return retval;
 110 }
 111
 112 static int update_ind_extent_range(handle_t *handle, struct inode *inode,
 113                                    ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 114                                    struct list_blocks_struct *lb)
 115 {
 116         struct buffer_head *bh;
 117         __le32 *i_data;
 118         int i, retval = 0;
 119         ext4_lblk_t blk_count = *blk_nump;
 120         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 121
 122         if (!pblock) {
 123                 /* Only update the file block number */
 124                 *blk_nump += max_entries;
 125                 return 0;
 126         }
 127
 128         bh = sb_bread(inode->i_sb, pblock);
 129         if (!bh)
 130                 return -EIO;
 131
 132         i_data = (__le32 *)bh->b_data;
 133         for (i = 0; i < max_entries; i++, blk_count++) {
 134                 if (i_data[i]) {
 135                         retval = update_extent_range(handle, inode,
 136                                                 le32_to_cpu(i_data[i]),
 137                                                 blk_count, lb);
 138                         if (retval)
 139                                 break;
 140                 }
 141         }
 142
 143         /* Update the file block number */
 144         *blk_nump = blk_count;
 145         put_bh(bh);
 146         return retval;
 147
 148 }
 149
 150 static int update_dind_extent_range(handle_t *handle, struct inode *inode,
 151                                     ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 152                                     struct list_blocks_struct *lb)
 153 {
 154         struct buffer_head *bh;
 155         __le32 *i_data;
 156         int i, retval = 0;
 157         ext4_lblk_t blk_count = *blk_nump;
 158         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 159
 160         if (!pblock) {
 161                 /* Only update the file block number */
 162                 *blk_nump += max_entries * max_entries;
 163                 return 0;
 164         }
 165         bh = sb_bread(inode->i_sb, pblock);
 166         if (!bh)
 167                 return -EIO;
 168
 169         i_data = (__le32 *)bh->b_data;
 170         for (i = 0; i < max_entries; i++) {
 171                 if (i_data[i]) {
 172                         retval = update_ind_extent_range(handle, inode,
 173                                                 le32_to_cpu(i_data[i]),
 174                                                 &blk_count, lb);
 175                         if (retval)
 176                                 break;
 177                 } else {
 178                         /* Only update the file block number */
 179                         blk_count += max_entries;
 180                 }
 181         }
 182
 183         /* Update the file block number */
 184         *blk_nump = blk_count;
 185         put_bh(bh);
 186         return retval;
 187
 188 }
 189
 190 static int update_tind_extent_range(handle_t *handle, struct inode *inode,
 191                                      ext4_fsblk_t pblock, ext4_lblk_t *blk_nump,
 192                                      struct list_blocks_struct *lb)
 193 {
 194         struct buffer_head *bh;
 195         __le32 *i_data;
 196         int i, retval = 0;
 197         ext4_lblk_t blk_count = *blk_nump;
 198         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 199
 200         if (!pblock) {
 201                 /* Only update the file block number */
 202                 *blk_nump += max_entries * max_entries * max_entries;
 203                 return 0;
 204         }
 205         bh = sb_bread(inode->i_sb, pblock);
 206         if (!bh)
 207                 return -EIO;
 208
 209         i_data = (__le32 *)bh->b_data;
 210         for (i = 0; i < max_entries; i++) {
 211                 if (i_data[i]) {
 212                         retval = update_dind_extent_range(handle, inode,
 213                                                 le32_to_cpu(i_data[i]),
 214                                                 &blk_count, lb);
 215                         if (retval)
 216                                 break;
 217                 } else
 218                         /* Only update the file block number */
 219                         blk_count += max_entries * max_entries;
 220         }
 221         /* Update the file block number */
 222         *blk_nump = blk_count;
 223         put_bh(bh);
 224         return retval;
 225
 226 }
 227
 228 static int extend_credit_for_blkdel(handle_t *handle, struct inode *inode)
 229 {
 230         int retval = 0, needed;
 231
 232         if (handle->h_buffer_credits > EXT4_RESERVE_TRANS_BLOCKS)
 233                 return 0;
 234         /*
 235          * We are freeing a blocks. During this we touch
 236          * superblock, group descriptor and block bitmap.
 237          * So allocate a credit of 3. We may update
 238          * quota (user and group).
 239          */
 240         needed = 3 + 2*EXT4_QUOTA_TRANS_BLOCKS(inode->i_sb);
 241
 242         if (ext4_journal_extend(handle, needed) != 0)
 243                 retval = ext4_journal_restart(handle, needed);
 244
 245         return retval;
 246 }
 247
 248 static int free_dind_blocks(handle_t *handle,
 249                                 struct inode *inode, __le32 i_data)
 250 {
 251         int i;
 252         __le32 *tmp_idata;
 253         struct buffer_head *bh;
 254         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 255
 256         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 257         if (!bh)
 258                 return -EIO;
 259
 260         tmp_idata = (__le32 *)bh->b_data;
 261         for (i = 0; i < max_entries; i++) {
 262                 if (tmp_idata[i]) {
 263                         extend_credit_for_blkdel(handle, inode);
 264                         ext4_free_blocks(handle, inode,
 265                                         le32_to_cpu(tmp_idata[i]), 1, 1);
 266                 }
 267         }
 268         put_bh(bh);
 269         extend_credit_for_blkdel(handle, inode);
 270         ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 271         return 0;
 272 }
 273
 274 static int free_tind_blocks(handle_t *handle,
 275                                 struct inode *inode, __le32 i_data)
 276 {
 277         int i, retval = 0;
 278         __le32 *tmp_idata;
 279         struct buffer_head *bh;
 280         unsigned long max_entries = inode->i_sb->s_blocksize >> 2;
 281
 282         bh = sb_bread(inode->i_sb, le32_to_cpu(i_data));
 283         if (!bh)
 284                 return -EIO;
 285
 286         tmp_idata = (__le32 *)bh->b_data;
 287         for (i = 0; i < max_entries; i++) {
 288                 if (tmp_idata[i]) {
 289                         retval = free_dind_blocks(handle,
 290                                         inode, tmp_idata[i]);
 291                         if (retval) {
 292                                 put_bh(bh);
 293                                 return retval;
 294                         }
 295                 }
 296         }
 297         put_bh(bh);
 298         extend_credit_for_blkdel(handle, inode);
 299         ext4_free_blocks(handle, inode, le32_to_cpu(i_data), 1, 1);
 300         return 0;
 301 }
 302
 303 static int free_ind_block(handle_t *handle, struct inode *inode, __le32 *i_data)
 304 {
 305         int retval;
 306
 307         /* ei->i_data[EXT4_IND_BLOCK] */
 308         if (i_data[0]) {
 309                 extend_credit_for_blkdel(handle, inode);
 310                 ext4_free_blocks(handle, inode,
 311                                 le32_to_cpu(i_data[0]), 1, 1);
 312         }
 313
 314         /* ei->i_data[EXT4_DIND_BLOCK] */
 315         if (i_data[1]) {
 316                 retval = free_dind_blocks(handle, inode, i_data[1]);
 317                 if (retval)
 318                         return retval;
 319         }
 320
 321         /* ei->i_data[EXT4_TIND_BLOCK] */
 322         if (i_data[2]) {
 323                 retval = free_tind_blocks(handle, inode, i_data[2]);
 324                 if (retval)
 325                         return retval;
 326         }
 327         return 0;
 328 }
 329
 330 static int ext4_ext_swap_inode_data(handle_t *handle, struct inode *inode,
 331                                                 struct inode *tmp_inode)
 332 {
 333         int retval;
 334         __le32  i_data[3];
 335         struct ext4_inode_info *ei = EXT4_I(inode);
 336         struct ext4_inode_info *tmp_ei = EXT4_I(tmp_inode);
 337
 338         /*
 339          * One credit accounted for writing the
 340          * i_data field of the original inode
 341          */
 342         retval = ext4_journal_extend(handle, 1);
 343         if (retval) {
 344                 retval = ext4_journal_restart(handle, 1);
 345                 if (retval)
 346                         goto err_out;
 347         }
 348
 349         i_data[0] = ei->i_data[EXT4_IND_BLOCK];
 350         i_data[1] = ei->i_data[EXT4_DIND_BLOCK];
 351         i_data[2] = ei->i_data[EXT4_TIND_BLOCK];
 352
 353         down_write(&EXT4_I(inode)->i_data_sem);
 354         /*
 355          * if EXT4_EXT_MIGRATE is cleared a block allocation
 356          * happened after we started the migrate. We need to
 357          * fail the migrate
 358          */
 359         if (!(EXT4_I(inode)->i_flags & EXT4_EXT_MIGRATE)) {
 360                 retval = -EAGAIN;
 361                 up_write(&EXT4_I(inode)->i_data_sem);
 362                 goto err_out;
 363         } else
 364                 EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags &
 365                                                         ~EXT4_EXT_MIGRATE;
 366         /*
 367          * We have the extent map build with the tmp inode.
 368          * Now copy the i_data across
 369          */
 370         ei->i_flags |= EXT4_EXTENTS_FL;
 371         memcpy(ei->i_data, tmp_ei->i_data, sizeof(ei->i_data));
 372
 373         /*
 374          * Update i_blocks with the new blocks that got
 375          * allocated while adding extents for extent index
 376          * blocks.
 377          *
 378          * While converting to extents we need not
 379          * update the orignal inode i_blocks for extent blocks
 380          * via quota APIs. The quota update happened via tmp_inode already.
 381          */
 382         spin_lock(&inode->i_lock);
 383         inode->i_blocks += tmp_inode->i_blocks;
 384         spin_unlock(&inode->i_lock);
 385         up_write(&EXT4_I(inode)->i_data_sem);
 386
 387         /*
 388          * We mark the inode dirty after, because we decrement the
 389          * i_blocks when freeing the indirect meta-data blocks
 390          */
 391         retval = free_ind_block(handle, inode, i_data);
 392         ext4_mark_inode_dirty(handle, inode);
 393
 394 err_out:
 395         return retval;
 396 }
 397
 398 static int free_ext_idx(handle_t *handle, struct inode *inode,
 399                                         struct ext4_extent_idx *ix)
 400 {
 401         int i, retval = 0;
 402         ext4_fsblk_t block;
 403         struct buffer_head *bh;
 404         struct ext4_extent_header *eh;
 405
 406         block = idx_pblock(ix);
 407         bh = sb_bread(inode->i_sb, block);
 408         if (!bh)
 409                 return -EIO;
 410
 411         eh = (struct ext4_extent_header *)bh->b_data;
 412         if (eh->eh_depth != 0) {
 413                 ix = EXT_FIRST_INDEX(eh);
 414                 for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 415                         retval = free_ext_idx(handle, inode, ix);
 416                         if (retval)
 417                                 break;
 418                 }
 419         }
 420         put_bh(bh);
 421         extend_credit_for_blkdel(handle, inode);
 422         ext4_free_blocks(handle, inode, block, 1, 1);
 423         return retval;
 424 }
 425
 426 /*
 427  * Free the extent meta data blocks only
 428  */
 429 static int free_ext_block(handle_t *handle, struct inode *inode)
 430 {
 431         int i, retval = 0;
 432         struct ext4_inode_info *ei = EXT4_I(inode);
 433         struct ext4_extent_header *eh = (struct ext4_extent_header *)ei->i_data;
 434         struct ext4_extent_idx *ix;
 435         if (eh->eh_depth == 0)
 436                 /*
 437                  * No extra blocks allocated for extent meta data
 438                  */
 439                 return 0;
 440         ix = EXT_FIRST_INDEX(eh);
 441         for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ix++) {
 442                 retval = free_ext_idx(handle, inode, ix);
 443                 if (retval)
 444                         return retval;
 445         }
 446         return retval;
 447
 448 }
 449
 450 int ext4_ext_migrate(struct inode *inode, struct file *filp,
 451                                 unsigned int cmd, unsigned long arg)
 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         if (!test_opt(inode->i_sb, EXTENTS))
 463                 /*
 464                  * if mounted with noextents we don't allow the migrate
 465                  */
 466                 return -EINVAL;
 467
 468         if ((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                 goto err_out;
 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                 tmp_inode = NULL;
 493                 goto err_out;
 494         }
 495         i_size_write(tmp_inode, i_size_read(inode));
 496         /*
 497          * We don't want the inode to be reclaimed
 498          * if we got interrupted in between. We have
 499          * this tmp inode carrying reference to the
 500          * data blocks of the original file. We set
 501          * the i_nlink to zero at the last stage after
 502          * switching the original file to extent format
 503          */
 504         tmp_inode->i_nlink = 1;
 505
 506         ext4_ext_tree_init(handle, tmp_inode);
 507         ext4_orphan_add(handle, tmp_inode);
 508         ext4_journal_stop(handle);
 509
 510         /*
 511          * start with one credit accounted for
 512          * superblock modification.
 513          *
 514          * For the tmp_inode we already have commited the
 515          * trascation that created the inode. Later as and
 516          * when we add extents we extent the journal
 517          */
 518         /*
 519          * inode_mutex prevent write and truncate on the file. Read still goes
 520          * through. We take i_data_sem in ext4_ext_swap_inode_data before we
 521          * switch the inode format to prevent read.
 522          */
 523         mutex_lock(&(inode->i_mutex));
 524         /*
 525          * Even though we take i_mutex we can still cause block allocation
 526          * via mmap write to holes. If we have allocated new blocks we fail
 527          * migrate.  New block allocation will clear EXT4_EXT_MIGRATE flag.
 528          * The flag is updated with i_data_sem held to prevent racing with
 529          * block allocation.
 530          */
 531         down_read((&EXT4_I(inode)->i_data_sem));
 532         EXT4_I(inode)->i_flags = EXT4_I(inode)->i_flags | EXT4_EXT_MIGRATE;
 533         up_read((&EXT4_I(inode)->i_data_sem));
 534
 535         handle = ext4_journal_start(inode, 1);
 536
 537         ei = EXT4_I(inode);
 538         i_data = ei->i_data;
 539         memset(&lb, 0, sizeof(lb));
 540
 541         /* 32 bit block address 4 bytes */
 542         max_entries = inode->i_sb->s_blocksize >> 2;
 543         for (i = 0; i < EXT4_NDIR_BLOCKS; i++, blk_count++) {
 544                 if (i_data[i]) {
 545                         retval = update_extent_range(handle, tmp_inode,
 546                                                 le32_to_cpu(i_data[i]),
 547                                                 blk_count, &lb);
 548                         if (retval)
 549                                 goto err_out;
 550                 }
 551         }
 552         if (i_data[EXT4_IND_BLOCK]) {
 553                 retval = update_ind_extent_range(handle, tmp_inode,
 554                                         le32_to_cpu(i_data[EXT4_IND_BLOCK]),
 555                                         &blk_count, &lb);
 556                         if (retval)
 557                                 goto err_out;
 558         } else
 559                 blk_count +=  max_entries;
 560         if (i_data[EXT4_DIND_BLOCK]) {
 561                 retval = update_dind_extent_range(handle, tmp_inode,
 562                                         le32_to_cpu(i_data[EXT4_DIND_BLOCK]),
 563                                         &blk_count, &lb);
 564                         if (retval)
 565                                 goto err_out;
 566         } else
 567                 blk_count += max_entries * max_entries;
 568         if (i_data[EXT4_TIND_BLOCK]) {
 569                 retval = update_tind_extent_range(handle, tmp_inode,
 570                                         le32_to_cpu(i_data[EXT4_TIND_BLOCK]),
 571                                         &blk_count, &lb);
 572                         if (retval)
 573                                 goto err_out;
 574         }
 575         /*
 576          * Build the last extent
 577          */
 578         retval = finish_range(handle, tmp_inode, &lb);
 579 err_out:
 580         if (retval)
 581                 /*
 582                  * Failure case delete the extent information with the
 583                  * tmp_inode
 584                  */
 585                 free_ext_block(handle, tmp_inode);
 586         else {
 587                 retval = ext4_ext_swap_inode_data(handle, inode, tmp_inode);
 588                 if (retval)
 589                         /*
 590                          * if we fail to swap inode data free the extent
 591                          * details of the tmp inode
 592                          */
 593                         free_ext_block(handle, tmp_inode);
 594         }
 595
 596         /* We mark the tmp_inode dirty via ext4_ext_tree_init. */
 597         if (ext4_journal_extend(handle, 1) != 0)
 598                 ext4_journal_restart(handle, 1);
 599
 600         /*
 601          * Mark the tmp_inode as of size zero
 602          */
 603         i_size_write(tmp_inode, 0);
 604
 605         /*
 606          * set the  i_blocks count to zero
 607          * so that the ext4_delete_inode does the
 608          * right job
 609          *
 610          * We don't need to take the i_lock because
 611          * the inode is not visible to user space.
 612          */
 613         tmp_inode->i_blocks = 0;
 614
 615         /* Reset the extent details */
 616         ext4_ext_tree_init(handle, tmp_inode);
 617
 618         /*
 619          * Set the i_nlink to zero so that
 620          * generic_drop_inode really deletes the
 621          * inode
 622          */
 623         tmp_inode->i_nlink = 0;
 624
 625         ext4_journal_stop(handle);
 626         mutex_unlock(&(inode->i_mutex));
 627
 628         if (tmp_inode)
 629                 iput(tmp_inode);
 630
 631         return retval;
 632 }