fs/ext4/resize.c

   1 /*
   2  *  linux/fs/ext4/resize.c
   3  *
   4  * Support for resizing an ext4 filesystem while it is mounted.
   5  *
   6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   7  *
   8  * This could probably be made into a module, because it is not often in use.
   9  */
  10
  11
  12 #define EXT4FS_DEBUG
  13
  14 #include <linux/errno.h>
  15 #include <linux/slab.h>
  16
  17 #include "ext4_jbd2.h"
  18
  19 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  20 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  21
  22 static int verify_group_input(struct super_block *sb,
  23                               struct ext4_new_group_data *input)
  24 {
  25         struct ext4_sb_info *sbi = EXT4_SB(sb);
  26         struct ext4_super_block *es = sbi->s_es;
  27         ext4_fsblk_t start = ext4_blocks_count(es);
  28         ext4_fsblk_t end = start + input->blocks_count;
  29         ext4_group_t group = input->group;
  30         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  31         unsigned overhead = ext4_bg_has_super(sb, group) ?
  32                 (1 + ext4_bg_num_gdb(sb, group) +
  33                  le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
  34         ext4_fsblk_t metaend = start + overhead;
  35         struct buffer_head *bh = NULL;
  36         ext4_grpblk_t free_blocks_count, offset;
  37         int err = -EINVAL;
  38
  39         input->free_blocks_count = free_blocks_count =
  40                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  41
  42         if (test_opt(sb, DEBUG))
  43                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
  44                        "(%d free, %u reserved)\n",
  45                        ext4_bg_has_super(sb, input->group) ? "normal" :
  46                        "no-super", input->group, input->blocks_count,
  47                        free_blocks_count, input->reserved_blocks);
  48
  49         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
  50         if (group != sbi->s_groups_count)
  51                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
  52                              input->group, sbi->s_groups_count);
  53         else if (offset != 0)
  54                         ext4_warning(sb, "Last group not full");
  55         else if (input->reserved_blocks > input->blocks_count / 5)
  56                 ext4_warning(sb, "Reserved blocks too high (%u)",
  57                              input->reserved_blocks);
  58         else if (free_blocks_count < 0)
  59                 ext4_warning(sb, "Bad blocks count %u",
  60                              input->blocks_count);
  61         else if (!(bh = sb_bread(sb, end - 1)))
  62                 ext4_warning(sb, "Cannot read last block (%llu)",
  63                              end - 1);
  64         else if (outside(input->block_bitmap, start, end))
  65                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
  66                              (unsigned long long)input->block_bitmap);
  67         else if (outside(input->inode_bitmap, start, end))
  68                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
  69                              (unsigned long long)input->inode_bitmap);
  70         else if (outside(input->inode_table, start, end) ||
  71                  outside(itend - 1, start, end))
  72                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
  73                              (unsigned long long)input->inode_table, itend - 1);
  74         else if (input->inode_bitmap == input->block_bitmap)
  75                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
  76                              (unsigned long long)input->block_bitmap);
  77         else if (inside(input->block_bitmap, input->inode_table, itend))
  78                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
  79                              "(%llu-%llu)",
  80                              (unsigned long long)input->block_bitmap,
  81                              (unsigned long long)input->inode_table, itend - 1);
  82         else if (inside(input->inode_bitmap, input->inode_table, itend))
  83                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
  84                              "(%llu-%llu)",
  85                              (unsigned long long)input->inode_bitmap,
  86                              (unsigned long long)input->inode_table, itend - 1);
  87         else if (inside(input->block_bitmap, start, metaend))
  88                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
  89                              (unsigned long long)input->block_bitmap,
  90                              start, metaend - 1);
  91         else if (inside(input->inode_bitmap, start, metaend))
  92                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
  93                              (unsigned long long)input->inode_bitmap,
  94                              start, metaend - 1);
  95         else if (inside(input->inode_table, start, metaend) ||
  96                  inside(itend - 1, start, metaend))
  97                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
  98                              "(%llu-%llu)",
  99                              (unsigned long long)input->inode_table,
 100                              itend - 1, start, metaend - 1);
 101         else
 102                 err = 0;
 103         brelse(bh);
 104
 105         return err;
 106 }
 107
 108 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 109                                   ext4_fsblk_t blk)
 110 {
 111         struct buffer_head *bh;
 112         int err;
 113
 114         bh = sb_getblk(sb, blk);
 115         if (!bh)
 116                 return ERR_PTR(-EIO);
 117         if ((err = ext4_journal_get_write_access(handle, bh))) {
 118                 brelse(bh);
 119                 bh = ERR_PTR(err);
 120         } else {
 121                 lock_buffer(bh);
 122                 memset(bh->b_data, 0, sb->s_blocksize);
 123                 set_buffer_uptodate(bh);
 124                 unlock_buffer(bh);
 125         }
 126
 127         return bh;
 128 }
 129
 130 /*
 131  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 132  * If that fails, restart the transaction & regain write access for the
 133  * buffer head which is used for block_bitmap modifications.
 134  */
 135 static int extend_or_restart_transaction(handle_t *handle, int thresh,
 136                                          struct buffer_head *bh)
 137 {
 138         int err;
 139
 140         if (ext4_handle_has_enough_credits(handle, thresh))
 141                 return 0;
 142
 143         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 144         if (err < 0)
 145                 return err;
 146         if (err) {
 147                 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 148                         return err;
 149                 if ((err = ext4_journal_get_write_access(handle, bh)))
 150                         return err;
 151         }
 152
 153         return 0;
 154 }
 155
 156 /*
 157  * Set up the block and inode bitmaps, and the inode table for the new group.
 158  * This doesn't need to be part of the main transaction, since we are only
 159  * changing blocks outside the actual filesystem.  We still do journaling to
 160  * ensure the recovery is correct in case of a failure just after resize.
 161  * If any part of this fails, we simply abort the resize.
 162  */
 163 static int setup_new_group_blocks(struct super_block *sb,
 164                                   struct ext4_new_group_data *input)
 165 {
 166         struct ext4_sb_info *sbi = EXT4_SB(sb);
 167         ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
 168         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 169                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 170         unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
 171         struct buffer_head *bh;
 172         handle_t *handle;
 173         ext4_fsblk_t block;
 174         ext4_grpblk_t bit;
 175         int i;
 176         int err = 0, err2;
 177
 178         /* This transaction may be extended/restarted along the way */
 179         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 180
 181         if (IS_ERR(handle))
 182                 return PTR_ERR(handle);
 183
 184         mutex_lock(&sbi->s_resize_lock);
 185         if (input->group != sbi->s_groups_count) {
 186                 err = -EBUSY;
 187                 goto exit_journal;
 188         }
 189
 190         if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
 191                 err = PTR_ERR(bh);
 192                 goto exit_journal;
 193         }
 194
 195         if (ext4_bg_has_super(sb, input->group)) {
 196                 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
 197                 ext4_set_bit(0, bh->b_data);
 198         }
 199
 200         /* Copy all of the GDT blocks into the backup in this group */
 201         for (i = 0, bit = 1, block = start + 1;
 202              i < gdblocks; i++, block++, bit++) {
 203                 struct buffer_head *gdb;
 204
 205                 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
 206
 207                 if ((err = extend_or_restart_transaction(handle, 1, bh)))
 208                         goto exit_bh;
 209
 210                 gdb = sb_getblk(sb, block);
 211                 if (!gdb) {
 212                         err = -EIO;
 213                         goto exit_bh;
 214                 }
 215                 if ((err = ext4_journal_get_write_access(handle, gdb))) {
 216                         brelse(gdb);
 217                         goto exit_bh;
 218                 }
 219                 lock_buffer(gdb);
 220                 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
 221                 set_buffer_uptodate(gdb);
 222                 unlock_buffer(gdb);
 223                 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 224                 if (unlikely(err)) {
 225                         brelse(gdb);
 226                         goto exit_bh;
 227                 }
 228                 ext4_set_bit(bit, bh->b_data);
 229                 brelse(gdb);
 230         }
 231
 232         /* Zero out all of the reserved backup group descriptor table blocks */
 233         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 234                         block, sbi->s_itb_per_group);
 235         err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
 236                                GFP_NOFS);
 237         if (err)
 238                 goto exit_bh;
 239         for (i = 0, bit = gdblocks + 1; i < reserved_gdb; i++, bit++)
 240                 ext4_set_bit(bit, bh->b_data);
 241
 242         ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
 243                    input->block_bitmap - start);
 244         ext4_set_bit(input->block_bitmap - start, bh->b_data);
 245         ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
 246                    input->inode_bitmap - start);
 247         ext4_set_bit(input->inode_bitmap - start, bh->b_data);
 248
 249         /* Zero out all of the inode table blocks */
 250         block = input->inode_table;
 251         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 252                         block, sbi->s_itb_per_group);
 253         err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
 254         if (err)
 255                 goto exit_bh;
 256         for (i = 0, bit = input->inode_table - start;
 257              i < sbi->s_itb_per_group; i++, bit++)
 258                 ext4_set_bit(bit, bh->b_data);
 259
 260         if ((err = extend_or_restart_transaction(handle, 2, bh)))
 261                 goto exit_bh;
 262
 263         ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
 264                              bh->b_data);
 265         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 266         if (unlikely(err)) {
 267                 ext4_std_error(sb, err);
 268                 goto exit_bh;
 269         }
 270         brelse(bh);
 271         /* Mark unused entries in inode bitmap used */
 272         ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
 273                    input->inode_bitmap, input->inode_bitmap - start);
 274         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 275                 err = PTR_ERR(bh);
 276                 goto exit_journal;
 277         }
 278
 279         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
 280                              bh->b_data);
 281         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 282         if (unlikely(err))
 283                 ext4_std_error(sb, err);
 284 exit_bh:
 285         brelse(bh);
 286
 287 exit_journal:
 288         mutex_unlock(&sbi->s_resize_lock);
 289         if ((err2 = ext4_journal_stop(handle)) && !err)
 290                 err = err2;
 291
 292         return err;
 293 }
 294
 295 /*
 296  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 297  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 298  * calling this for the first time.  In a sparse filesystem it will be the
 299  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 300  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 301  */
 302 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 303                                   unsigned *five, unsigned *seven)
 304 {
 305         unsigned *min = three;
 306         int mult = 3;
 307         unsigned ret;
 308
 309         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 310                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 311                 ret = *min;
 312                 *min += 1;
 313                 return ret;
 314         }
 315
 316         if (*five < *min) {
 317                 min = five;
 318                 mult = 5;
 319         }
 320         if (*seven < *min) {
 321                 min = seven;
 322                 mult = 7;
 323         }
 324
 325         ret = *min;
 326         *min *= mult;
 327
 328         return ret;
 329 }
 330
 331 /*
 332  * Check that all of the backup GDT blocks are held in the primary GDT block.
 333  * It is assumed that they are stored in group order.  Returns the number of
 334  * groups in current filesystem that have BACKUPS, or -ve error code.
 335  */
 336 static int verify_reserved_gdb(struct super_block *sb,
 337                                struct buffer_head *primary)
 338 {
 339         const ext4_fsblk_t blk = primary->b_blocknr;
 340         const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
 341         unsigned three = 1;
 342         unsigned five = 5;
 343         unsigned seven = 7;
 344         unsigned grp;
 345         __le32 *p = (__le32 *)primary->b_data;
 346         int gdbackups = 0;
 347
 348         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 349                 if (le32_to_cpu(*p++) !=
 350                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 351                         ext4_warning(sb, "reserved GDT %llu"
 352                                      " missing grp %d (%llu)",
 353                                      blk, grp,
 354                                      grp *
 355                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 356                                      blk);
 357                         return -EINVAL;
 358                 }
 359                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 360                         return -EFBIG;
 361         }
 362
 363         return gdbackups;
 364 }
 365
 366 /*
 367  * Called when we need to bring a reserved group descriptor table block into
 368  * use from the resize inode.  The primary copy of the new GDT block currently
 369  * is an indirect block (under the double indirect block in the resize inode).
 370  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 371  * block, in group order.  Even though we know all the block numbers we need,
 372  * we check to ensure that the resize inode has actually reserved these blocks.
 373  *
 374  * Don't need to update the block bitmaps because the blocks are still in use.
 375  *
 376  * We get all of the error cases out of the way, so that we are sure to not
 377  * fail once we start modifying the data on disk, because JBD has no rollback.
 378  */
 379 static int add_new_gdb(handle_t *handle, struct inode *inode,
 380                        struct ext4_new_group_data *input,
 381                        struct buffer_head **primary)
 382 {
 383         struct super_block *sb = inode->i_sb;
 384         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 385         unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 386         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 387         struct buffer_head **o_group_desc, **n_group_desc;
 388         struct buffer_head *dind;
 389         int gdbackups;
 390         struct ext4_iloc iloc;
 391         __le32 *data;
 392         int err;
 393
 394         if (test_opt(sb, DEBUG))
 395                 printk(KERN_DEBUG
 396                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 397                        gdb_num);
 398
 399         /*
 400          * If we are not using the primary superblock/GDT copy don't resize,
 401          * because the user tools have no way of handling this.  Probably a
 402          * bad time to do it anyways.
 403          */
 404         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 405             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 406                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 407                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 408                 return -EPERM;
 409         }
 410
 411         *primary = sb_bread(sb, gdblock);
 412         if (!*primary)
 413                 return -EIO;
 414
 415         if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
 416                 err = gdbackups;
 417                 goto exit_bh;
 418         }
 419
 420         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 421         dind = sb_bread(sb, le32_to_cpu(*data));
 422         if (!dind) {
 423                 err = -EIO;
 424                 goto exit_bh;
 425         }
 426
 427         data = (__le32 *)dind->b_data;
 428         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 429                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 430                              input->group, gdblock);
 431                 err = -EINVAL;
 432                 goto exit_dind;
 433         }
 434
 435         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 436         if (unlikely(err))
 437                 goto exit_dind;
 438
 439         err = ext4_journal_get_write_access(handle, *primary);
 440         if (unlikely(err))
 441                 goto exit_sbh;
 442
 443         err = ext4_journal_get_write_access(handle, dind);
 444         if (unlikely(err))
 445                 ext4_std_error(sb, err);
 446
 447         /* ext4_reserve_inode_write() gets a reference on the iloc */
 448         err = ext4_reserve_inode_write(handle, inode, &iloc);
 449         if (unlikely(err))
 450                 goto exit_dindj;
 451
 452         n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 453                         GFP_NOFS);
 454         if (!n_group_desc) {
 455                 err = -ENOMEM;
 456                 ext4_warning(sb,
 457                               "not enough memory for %lu groups", gdb_num + 1);
 458                 goto exit_inode;
 459         }
 460
 461         /*
 462          * Finally, we have all of the possible failures behind us...
 463          *
 464          * Remove new GDT block from inode double-indirect block and clear out
 465          * the new GDT block for use (which also "frees" the backup GDT blocks
 466          * from the reserved inode).  We don't need to change the bitmaps for
 467          * these blocks, because they are marked as in-use from being in the
 468          * reserved inode, and will become GDT blocks (primary and backup).
 469          */
 470         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 471         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 472         if (unlikely(err)) {
 473                 ext4_std_error(sb, err);
 474                 goto exit_inode;
 475         }
 476         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 477         ext4_mark_iloc_dirty(handle, inode, &iloc);
 478         memset((*primary)->b_data, 0, sb->s_blocksize);
 479         err = ext4_handle_dirty_metadata(handle, NULL, *primary);
 480         if (unlikely(err)) {
 481                 ext4_std_error(sb, err);
 482                 goto exit_inode;
 483         }
 484         brelse(dind);
 485
 486         o_group_desc = EXT4_SB(sb)->s_group_desc;
 487         memcpy(n_group_desc, o_group_desc,
 488                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 489         n_group_desc[gdb_num] = *primary;
 490         EXT4_SB(sb)->s_group_desc = n_group_desc;
 491         EXT4_SB(sb)->s_gdb_count++;
 492         kfree(o_group_desc);
 493
 494         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 495         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 496         if (err)
 497                 ext4_std_error(sb, err);
 498
 499         return err;
 500
 501 exit_inode:
 502         /* ext4_handle_release_buffer(handle, iloc.bh); */
 503         brelse(iloc.bh);
 504 exit_dindj:
 505         /* ext4_handle_release_buffer(handle, dind); */
 506 exit_sbh:
 507         /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
 508 exit_dind:
 509         brelse(dind);
 510 exit_bh:
 511         brelse(*primary);
 512
 513         ext4_debug("leaving with error %d\n", err);
 514         return err;
 515 }
 516
 517 /*
 518  * Called when we are adding a new group which has a backup copy of each of
 519  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 520  * We need to add these reserved backup GDT blocks to the resize inode, so
 521  * that they are kept for future resizing and not allocated to files.
 522  *
 523  * Each reserved backup GDT block will go into a different indirect block.
 524  * The indirect blocks are actually the primary reserved GDT blocks,
 525  * so we know in advance what their block numbers are.  We only get the
 526  * double-indirect block to verify it is pointing to the primary reserved
 527  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 528  * backup GDT blocks are stored in their reserved primary GDT block.
 529  */
 530 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 531                               struct ext4_new_group_data *input)
 532 {
 533         struct super_block *sb = inode->i_sb;
 534         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 535         struct buffer_head **primary;
 536         struct buffer_head *dind;
 537         struct ext4_iloc iloc;
 538         ext4_fsblk_t blk;
 539         __le32 *data, *end;
 540         int gdbackups = 0;
 541         int res, i;
 542         int err;
 543
 544         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 545         if (!primary)
 546                 return -ENOMEM;
 547
 548         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 549         dind = sb_bread(sb, le32_to_cpu(*data));
 550         if (!dind) {
 551                 err = -EIO;
 552                 goto exit_free;
 553         }
 554
 555         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 556         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 557                                          EXT4_ADDR_PER_BLOCK(sb));
 558         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 559
 560         /* Get each reserved primary GDT block and verify it holds backups */
 561         for (res = 0; res < reserved_gdb; res++, blk++) {
 562                 if (le32_to_cpu(*data) != blk) {
 563                         ext4_warning(sb, "reserved block %llu"
 564                                      " not at offset %ld",
 565                                      blk,
 566                                      (long)(data - (__le32 *)dind->b_data));
 567                         err = -EINVAL;
 568                         goto exit_bh;
 569                 }
 570                 primary[res] = sb_bread(sb, blk);
 571                 if (!primary[res]) {
 572                         err = -EIO;
 573                         goto exit_bh;
 574                 }
 575                 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
 576                         brelse(primary[res]);
 577                         err = gdbackups;
 578                         goto exit_bh;
 579                 }
 580                 if (++data >= end)
 581                         data = (__le32 *)dind->b_data;
 582         }
 583
 584         for (i = 0; i < reserved_gdb; i++) {
 585                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
 586                         /*
 587                         int j;
 588                         for (j = 0; j < i; j++)
 589                                 ext4_handle_release_buffer(handle, primary[j]);
 590                          */
 591                         goto exit_bh;
 592                 }
 593         }
 594
 595         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 596                 goto exit_bh;
 597
 598         /*
 599          * Finally we can add each of the reserved backup GDT blocks from
 600          * the new group to its reserved primary GDT block.
 601          */
 602         blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
 603         for (i = 0; i < reserved_gdb; i++) {
 604                 int err2;
 605                 data = (__le32 *)primary[i]->b_data;
 606                 /* printk("reserving backup %lu[%u] = %lu\n",
 607                        primary[i]->b_blocknr, gdbackups,
 608                        blk + primary[i]->b_blocknr); */
 609                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 610                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 611                 if (!err)
 612                         err = err2;
 613         }
 614         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 615         ext4_mark_iloc_dirty(handle, inode, &iloc);
 616
 617 exit_bh:
 618         while (--res >= 0)
 619                 brelse(primary[res]);
 620         brelse(dind);
 621
 622 exit_free:
 623         kfree(primary);
 624
 625         return err;
 626 }
 627
 628 /*
 629  * Update the backup copies of the ext4 metadata.  These don't need to be part
 630  * of the main resize transaction, because e2fsck will re-write them if there
 631  * is a problem (basically only OOM will cause a problem).  However, we
 632  * _should_ update the backups if possible, in case the primary gets trashed
 633  * for some reason and we need to run e2fsck from a backup superblock.  The
 634  * important part is that the new block and inode counts are in the backup
 635  * superblocks, and the location of the new group metadata in the GDT backups.
 636  *
 637  * We do not need take the s_resize_lock for this, because these
 638  * blocks are not otherwise touched by the filesystem code when it is
 639  * mounted.  We don't need to worry about last changing from
 640  * sbi->s_groups_count, because the worst that can happen is that we
 641  * do not copy the full number of backups at this time.  The resize
 642  * which changed s_groups_count will backup again.
 643  */
 644 static void update_backups(struct super_block *sb,
 645                            int blk_off, char *data, int size)
 646 {
 647         struct ext4_sb_info *sbi = EXT4_SB(sb);
 648         const ext4_group_t last = sbi->s_groups_count;
 649         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
 650         unsigned three = 1;
 651         unsigned five = 5;
 652         unsigned seven = 7;
 653         ext4_group_t group;
 654         int rest = sb->s_blocksize - size;
 655         handle_t *handle;
 656         int err = 0, err2;
 657
 658         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 659         if (IS_ERR(handle)) {
 660                 group = 1;
 661                 err = PTR_ERR(handle);
 662                 goto exit_err;
 663         }
 664
 665         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
 666                 struct buffer_head *bh;
 667
 668                 /* Out of journal space, and can't get more - abort - so sad */
 669                 if (ext4_handle_valid(handle) &&
 670                     handle->h_buffer_credits == 0 &&
 671                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
 672                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 673                         break;
 674
 675                 bh = sb_getblk(sb, group * bpg + blk_off);
 676                 if (!bh) {
 677                         err = -EIO;
 678                         break;
 679                 }
 680                 ext4_debug("update metadata backup %#04lx\n",
 681                           (unsigned long)bh->b_blocknr);
 682                 if ((err = ext4_journal_get_write_access(handle, bh)))
 683                         break;
 684                 lock_buffer(bh);
 685                 memcpy(bh->b_data, data, size);
 686                 if (rest)
 687                         memset(bh->b_data + size, 0, rest);
 688                 set_buffer_uptodate(bh);
 689                 unlock_buffer(bh);
 690                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 691                 if (unlikely(err))
 692                         ext4_std_error(sb, err);
 693                 brelse(bh);
 694         }
 695         if ((err2 = ext4_journal_stop(handle)) && !err)
 696                 err = err2;
 697
 698         /*
 699          * Ugh! Need to have e2fsck write the backup copies.  It is too
 700          * late to revert the resize, we shouldn't fail just because of
 701          * the backup copies (they are only needed in case of corruption).
 702          *
 703          * However, if we got here we have a journal problem too, so we
 704          * can't really start a transaction to mark the superblock.
 705          * Chicken out and just set the flag on the hope it will be written
 706          * to disk, and if not - we will simply wait until next fsck.
 707          */
 708 exit_err:
 709         if (err) {
 710                 ext4_warning(sb, "can't update backup for group %u (err %d), "
 711                              "forcing fsck on next reboot", group, err);
 712                 sbi->s_mount_state &= ~EXT4_VALID_FS;
 713                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
 714                 mark_buffer_dirty(sbi->s_sbh);
 715         }
 716 }
 717
 718 /* Add group descriptor data to an existing or new group descriptor block.
 719  * Ensure we handle all possible error conditions _before_ we start modifying
 720  * the filesystem, because we cannot abort the transaction and not have it
 721  * write the data to disk.
 722  *
 723  * If we are on a GDT block boundary, we need to get the reserved GDT block.
 724  * Otherwise, we may need to add backup GDT blocks for a sparse group.
 725  *
 726  * We only need to hold the superblock lock while we are actually adding
 727  * in the new group's counts to the superblock.  Prior to that we have
 728  * not really "added" the group at all.  We re-check that we are still
 729  * adding in the last group in case things have changed since verifying.
 730  */
 731 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
 732 {
 733         struct ext4_sb_info *sbi = EXT4_SB(sb);
 734         struct ext4_super_block *es = sbi->s_es;
 735         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 736                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
 737         struct buffer_head *primary = NULL;
 738         struct ext4_group_desc *gdp;
 739         struct inode *inode = NULL;
 740         handle_t *handle;
 741         int gdb_off, gdb_num;
 742         int err, err2;
 743
 744         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 745         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
 746
 747         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
 748                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 749                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
 750                 return -EPERM;
 751         }
 752
 753         if (ext4_blocks_count(es) + input->blocks_count <
 754             ext4_blocks_count(es)) {
 755                 ext4_warning(sb, "blocks_count overflow");
 756                 return -EINVAL;
 757         }
 758
 759         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
 760             le32_to_cpu(es->s_inodes_count)) {
 761                 ext4_warning(sb, "inodes_count overflow");
 762                 return -EINVAL;
 763         }
 764
 765         if (reserved_gdb || gdb_off == 0) {
 766                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
 767                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
 768                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
 769                         ext4_warning(sb,
 770                                      "No reserved GDT blocks, can't resize");
 771                         return -EPERM;
 772                 }
 773                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
 774                 if (IS_ERR(inode)) {
 775                         ext4_warning(sb, "Error opening resize inode");
 776                         return PTR_ERR(inode);
 777                 }
 778         }
 779
 780
 781         if ((err = verify_group_input(sb, input)))
 782                 goto exit_put;
 783
 784         if ((err = setup_new_group_blocks(sb, input)))
 785                 goto exit_put;
 786
 787         /*
 788          * We will always be modifying at least the superblock and a GDT
 789          * block.  If we are adding a group past the last current GDT block,
 790          * we will also modify the inode and the dindirect block.  If we
 791          * are adding a group with superblock/GDT backups  we will also
 792          * modify each of the reserved GDT dindirect blocks.
 793          */
 794         handle = ext4_journal_start_sb(sb,
 795                                        ext4_bg_has_super(sb, input->group) ?
 796                                        3 + reserved_gdb : 4);
 797         if (IS_ERR(handle)) {
 798                 err = PTR_ERR(handle);
 799                 goto exit_put;
 800         }
 801
 802         mutex_lock(&sbi->s_resize_lock);
 803         if (input->group != sbi->s_groups_count) {
 804                 ext4_warning(sb, "multiple resizers run on filesystem!");
 805                 err = -EBUSY;
 806                 goto exit_journal;
 807         }
 808
 809         if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
 810                 goto exit_journal;
 811
 812         /*
 813          * We will only either add reserved group blocks to a backup group
 814          * or remove reserved blocks for the first group in a new group block.
 815          * Doing both would be mean more complex code, and sane people don't
 816          * use non-sparse filesystems anymore.  This is already checked above.
 817          */
 818         if (gdb_off) {
 819                 primary = sbi->s_group_desc[gdb_num];
 820                 if ((err = ext4_journal_get_write_access(handle, primary)))
 821                         goto exit_journal;
 822
 823                 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
 824                     (err = reserve_backup_gdb(handle, inode, input)))
 825                         goto exit_journal;
 826         } else if ((err = add_new_gdb(handle, inode, input, &primary)))
 827                 goto exit_journal;
 828
 829         /*
 830          * OK, now we've set up the new group.  Time to make it active.
 831          *
 832          * We do not lock all allocations via s_resize_lock
 833          * so we have to be safe wrt. concurrent accesses the group
 834          * data.  So we need to be careful to set all of the relevant
 835          * group descriptor data etc. *before* we enable the group.
 836          *
 837          * The key field here is sbi->s_groups_count: as long as
 838          * that retains its old value, nobody is going to access the new
 839          * group.
 840          *
 841          * So first we update all the descriptor metadata for the new
 842          * group; then we update the total disk blocks count; then we
 843          * update the groups count to enable the group; then finally we
 844          * update the free space counts so that the system can start
 845          * using the new disk blocks.
 846          */
 847
 848         /* Update group descriptor block for new group */
 849         gdp = (struct ext4_group_desc *)((char *)primary->b_data +
 850                                          gdb_off * EXT4_DESC_SIZE(sb));
 851
 852         memset(gdp, 0, EXT4_DESC_SIZE(sb));
 853         ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
 854         ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
 855         ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
 856         ext4_free_blks_set(sb, gdp, input->free_blocks_count);
 857         ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
 858         gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
 859         gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
 860
 861         /*
 862          * We can allocate memory for mb_alloc based on the new group
 863          * descriptor
 864          */
 865         err = ext4_mb_add_groupinfo(sb, input->group, gdp);
 866         if (err)
 867                 goto exit_journal;
 868
 869         /*
 870          * Make the new blocks and inodes valid next.  We do this before
 871          * increasing the group count so that once the group is enabled,
 872          * all of its blocks and inodes are already valid.
 873          *
 874          * We always allocate group-by-group, then block-by-block or
 875          * inode-by-inode within a group, so enabling these
 876          * blocks/inodes before the group is live won't actually let us
 877          * allocate the new space yet.
 878          */
 879         ext4_blocks_count_set(es, ext4_blocks_count(es) +
 880                 input->blocks_count);
 881         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
 882
 883         /*
 884          * We need to protect s_groups_count against other CPUs seeing
 885          * inconsistent state in the superblock.
 886          *
 887          * The precise rules we use are:
 888          *
 889          * * Writers of s_groups_count *must* hold s_resize_lock
 890          * AND
 891          * * Writers must perform a smp_wmb() after updating all dependent
 892          *   data and before modifying the groups count
 893          *
 894          * * Readers must hold s_resize_lock over the access
 895          * OR
 896          * * Readers must perform an smp_rmb() after reading the groups count
 897          *   and before reading any dependent data.
 898          *
 899          * NB. These rules can be relaxed when checking the group count
 900          * while freeing data, as we can only allocate from a block
 901          * group after serialising against the group count, and we can
 902          * only then free after serialising in turn against that
 903          * allocation.
 904          */
 905         smp_wmb();
 906
 907         /* Update the global fs size fields */
 908         sbi->s_groups_count++;
 909
 910         err = ext4_handle_dirty_metadata(handle, NULL, primary);
 911         if (unlikely(err)) {
 912                 ext4_std_error(sb, err);
 913                 goto exit_journal;
 914         }
 915
 916         /* Update the reserved block counts only once the new group is
 917          * active. */
 918         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
 919                 input->reserved_blocks);
 920
 921         /* Update the free space counts */
 922         percpu_counter_add(&sbi->s_freeblocks_counter,
 923                            input->free_blocks_count);
 924         percpu_counter_add(&sbi->s_freeinodes_counter,
 925                            EXT4_INODES_PER_GROUP(sb));
 926
 927         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
 928             sbi->s_log_groups_per_flex) {
 929                 ext4_group_t flex_group;
 930                 flex_group = ext4_flex_group(sbi, input->group);
 931                 atomic_add(input->free_blocks_count,
 932                            &sbi->s_flex_groups[flex_group].free_blocks);
 933                 atomic_add(EXT4_INODES_PER_GROUP(sb),
 934                            &sbi->s_flex_groups[flex_group].free_inodes);
 935         }
 936
 937         ext4_handle_dirty_super(handle, sb);
 938
 939 exit_journal:
 940         mutex_unlock(&sbi->s_resize_lock);
 941         if ((err2 = ext4_journal_stop(handle)) && !err)
 942                 err = err2;
 943         if (!err) {
 944                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
 945                                sizeof(struct ext4_super_block));
 946                 update_backups(sb, primary->b_blocknr, primary->b_data,
 947                                primary->b_size);
 948         }
 949 exit_put:
 950         iput(inode);
 951         return err;
 952 } /* ext4_group_add */
 953
 954 /*
 955  * Extend the filesystem to the new number of blocks specified.  This entry
 956  * point is only used to extend the current filesystem to the end of the last
 957  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
 958  * for emergencies (because it has no dependencies on reserved blocks).
 959  *
 960  * If we _really_ wanted, we could use default values to call ext4_group_add()
 961  * allow the "remount" trick to work for arbitrary resizing, assuming enough
 962  * GDT blocks are reserved to grow to the desired size.
 963  */
 964 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
 965                       ext4_fsblk_t n_blocks_count)
 966 {
 967         ext4_fsblk_t o_blocks_count;
 968         ext4_grpblk_t last;
 969         ext4_grpblk_t add;
 970         struct buffer_head *bh;
 971         handle_t *handle;
 972         int err;
 973         ext4_group_t group;
 974
 975         /* We don't need to worry about locking wrt other resizers just
 976          * yet: we're going to revalidate es->s_blocks_count after
 977          * taking the s_resize_lock below. */
 978         o_blocks_count = ext4_blocks_count(es);
 979
 980         if (test_opt(sb, DEBUG))
 981                 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
 982                        o_blocks_count, n_blocks_count);
 983
 984         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
 985                 return 0;
 986
 987         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
 988                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
 989                         " too large to resize to %llu blocks safely\n",
 990                         sb->s_id, n_blocks_count);
 991                 if (sizeof(sector_t) < 8)
 992                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
 993                 return -EINVAL;
 994         }
 995
 996         if (n_blocks_count < o_blocks_count) {
 997                 ext4_warning(sb, "can't shrink FS - resize aborted");
 998                 return -EBUSY;
 999         }
1000
1001         /* Handle the remaining blocks in the last group only. */
1002         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1003
1004         if (last == 0) {
1005                 ext4_warning(sb, "need to use ext2online to resize further");
1006                 return -EPERM;
1007         }
1008
1009         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1010
1011         if (o_blocks_count + add < o_blocks_count) {
1012                 ext4_warning(sb, "blocks_count overflow");
1013                 return -EINVAL;
1014         }
1015
1016         if (o_blocks_count + add > n_blocks_count)
1017                 add = n_blocks_count - o_blocks_count;
1018
1019         if (o_blocks_count + add < n_blocks_count)
1020                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1021                              o_blocks_count + add, add);
1022
1023         /* See if the device is actually as big as what was requested */
1024         bh = sb_bread(sb, o_blocks_count + add - 1);
1025         if (!bh) {
1026                 ext4_warning(sb, "can't read last block, resize aborted");
1027                 return -ENOSPC;
1028         }
1029         brelse(bh);
1030
1031         /* We will update the superblock, one block bitmap, and
1032          * one group descriptor via ext4_free_blocks().
1033          */
1034         handle = ext4_journal_start_sb(sb, 3);
1035         if (IS_ERR(handle)) {
1036                 err = PTR_ERR(handle);
1037                 ext4_warning(sb, "error %d on journal start", err);
1038                 goto exit_put;
1039         }
1040
1041         mutex_lock(&EXT4_SB(sb)->s_resize_lock);
1042         if (o_blocks_count != ext4_blocks_count(es)) {
1043                 ext4_warning(sb, "multiple resizers run on filesystem!");
1044                 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1045                 ext4_journal_stop(handle);
1046                 err = -EBUSY;
1047                 goto exit_put;
1048         }
1049
1050         if ((err = ext4_journal_get_write_access(handle,
1051                                                  EXT4_SB(sb)->s_sbh))) {
1052                 ext4_warning(sb, "error %d on journal write access", err);
1053                 mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1054                 ext4_journal_stop(handle);
1055                 goto exit_put;
1056         }
1057         ext4_blocks_count_set(es, o_blocks_count + add);
1058         mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
1059         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1060                    o_blocks_count + add);
1061         /* We add the blocks to the bitmap and set the group need init bit */
1062         ext4_add_groupblocks(handle, sb, o_blocks_count, add);
1063         ext4_handle_dirty_super(handle, sb);
1064         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1065                    o_blocks_count + add);
1066         if ((err = ext4_journal_stop(handle)))
1067                 goto exit_put;
1068
1069         if (test_opt(sb, DEBUG))
1070                 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1071                        ext4_blocks_count(es));
1072         update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1073                        sizeof(struct ext4_super_block));
1074 exit_put:
1075         return err;
1076 } /* ext4_group_extend */