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/sched.h>
  15 #include <linux/smp_lock.h>
  16 #include <linux/ext4_jbd2.h>
  17
  18 #include <linux/errno.h>
  19 #include <linux/slab.h>
  20
  21
  22 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  23 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  24
  25 static int verify_group_input(struct super_block *sb,
  26                               struct ext4_new_group_data *input)
  27 {
  28         struct ext4_sb_info *sbi = EXT4_SB(sb);
  29         struct ext4_super_block *es = sbi->s_es;
  30         ext4_fsblk_t start = ext4_blocks_count(es);
  31         ext4_fsblk_t end = start + input->blocks_count;
  32         unsigned group = input->group;
  33         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  34         unsigned overhead = ext4_bg_has_super(sb, group) ?
  35                 (1 + ext4_bg_num_gdb(sb, group) +
  36                  le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
  37         ext4_fsblk_t metaend = start + overhead;
  38         struct buffer_head *bh = NULL;
  39         ext4_grpblk_t free_blocks_count, offset;
  40         int err = -EINVAL;
  41
  42         input->free_blocks_count = free_blocks_count =
  43                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  44
  45         if (test_opt(sb, DEBUG))
  46                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
  47                        "(%d free, %u reserved)\n",
  48                        ext4_bg_has_super(sb, input->group) ? "normal" :
  49                        "no-super", input->group, input->blocks_count,
  50                        free_blocks_count, input->reserved_blocks);
  51
  52         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
  53         if (group != sbi->s_groups_count)
  54                 ext4_warning(sb, __FUNCTION__,
  55                              "Cannot add at group %u (only %lu groups)",
  56                              input->group, sbi->s_groups_count);
  57         else if (offset != 0)
  58                         ext4_warning(sb, __FUNCTION__, "Last group not full");
  59         else if (input->reserved_blocks > input->blocks_count / 5)
  60                 ext4_warning(sb, __FUNCTION__, "Reserved blocks too high (%u)",
  61                              input->reserved_blocks);
  62         else if (free_blocks_count < 0)
  63                 ext4_warning(sb, __FUNCTION__, "Bad blocks count %u",
  64                              input->blocks_count);
  65         else if (!(bh = sb_bread(sb, end - 1)))
  66                 ext4_warning(sb, __FUNCTION__,
  67                              "Cannot read last block (%llu)",
  68                              end - 1);
  69         else if (outside(input->block_bitmap, start, end))
  70                 ext4_warning(sb, __FUNCTION__,
  71                              "Block bitmap not in group (block %llu)",
  72                              (unsigned long long)input->block_bitmap);
  73         else if (outside(input->inode_bitmap, start, end))
  74                 ext4_warning(sb, __FUNCTION__,
  75                              "Inode bitmap not in group (block %llu)",
  76                              (unsigned long long)input->inode_bitmap);
  77         else if (outside(input->inode_table, start, end) ||
  78                  outside(itend - 1, start, end))
  79                 ext4_warning(sb, __FUNCTION__,
  80                              "Inode table not in group (blocks %llu-%llu)",
  81                              (unsigned long long)input->inode_table, itend - 1);
  82         else if (input->inode_bitmap == input->block_bitmap)
  83                 ext4_warning(sb, __FUNCTION__,
  84                              "Block bitmap same as inode bitmap (%llu)",
  85                              (unsigned long long)input->block_bitmap);
  86         else if (inside(input->block_bitmap, input->inode_table, itend))
  87                 ext4_warning(sb, __FUNCTION__,
  88                              "Block bitmap (%llu) in inode table (%llu-%llu)",
  89                              (unsigned long long)input->block_bitmap,
  90                              (unsigned long long)input->inode_table, itend - 1);
  91         else if (inside(input->inode_bitmap, input->inode_table, itend))
  92                 ext4_warning(sb, __FUNCTION__,
  93                              "Inode bitmap (%llu) in inode table (%llu-%llu)",
  94                              (unsigned long long)input->inode_bitmap,
  95                              (unsigned long long)input->inode_table, itend - 1);
  96         else if (inside(input->block_bitmap, start, metaend))
  97                 ext4_warning(sb, __FUNCTION__,
  98                              "Block bitmap (%llu) in GDT table"
  99                              " (%llu-%llu)",
 100                              (unsigned long long)input->block_bitmap,
 101                              start, metaend - 1);
 102         else if (inside(input->inode_bitmap, start, metaend))
 103                 ext4_warning(sb, __FUNCTION__,
 104                              "Inode bitmap (%llu) in GDT table"
 105                              " (%llu-%llu)",
 106                              (unsigned long long)input->inode_bitmap,
 107                              start, metaend - 1);
 108         else if (inside(input->inode_table, start, metaend) ||
 109                  inside(itend - 1, start, metaend))
 110                 ext4_warning(sb, __FUNCTION__,
 111                              "Inode table (%llu-%llu) overlaps"
 112                              "GDT table (%llu-%llu)",
 113                              (unsigned long long)input->inode_table,
 114                              itend - 1, start, metaend - 1);
 115         else
 116                 err = 0;
 117         brelse(bh);
 118
 119         return err;
 120 }
 121
 122 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 123                                   ext4_fsblk_t blk)
 124 {
 125         struct buffer_head *bh;
 126         int err;
 127
 128         bh = sb_getblk(sb, blk);
 129         if (!bh)
 130                 return ERR_PTR(-EIO);
 131         if ((err = ext4_journal_get_write_access(handle, bh))) {
 132                 brelse(bh);
 133                 bh = ERR_PTR(err);
 134         } else {
 135                 lock_buffer(bh);
 136                 memset(bh->b_data, 0, sb->s_blocksize);
 137                 set_buffer_uptodate(bh);
 138                 unlock_buffer(bh);
 139         }
 140
 141         return bh;
 142 }
 143
 144 /*
 145  * To avoid calling the atomic setbit hundreds or thousands of times, we only
 146  * need to use it within a single byte (to ensure we get endianness right).
 147  * We can use memset for the rest of the bitmap as there are no other users.
 148  */
 149 static void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
 150 {
 151         int i;
 152
 153         if (start_bit >= end_bit)
 154                 return;
 155
 156         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
 157         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
 158                 ext4_set_bit(i, bitmap);
 159         if (i < end_bit)
 160                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
 161 }
 162
 163 /*
 164  * Set up the block and inode bitmaps, and the inode table for the new group.
 165  * This doesn't need to be part of the main transaction, since we are only
 166  * changing blocks outside the actual filesystem.  We still do journaling to
 167  * ensure the recovery is correct in case of a failure just after resize.
 168  * If any part of this fails, we simply abort the resize.
 169  */
 170 static int setup_new_group_blocks(struct super_block *sb,
 171                                   struct ext4_new_group_data *input)
 172 {
 173         struct ext4_sb_info *sbi = EXT4_SB(sb);
 174         ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
 175         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 176                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 177         unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
 178         struct buffer_head *bh;
 179         handle_t *handle;
 180         ext4_fsblk_t block;
 181         ext4_grpblk_t bit;
 182         int i;
 183         int err = 0, err2;
 184
 185         handle = ext4_journal_start_sb(sb, reserved_gdb + gdblocks +
 186                                        2 + sbi->s_itb_per_group);
 187         if (IS_ERR(handle))
 188                 return PTR_ERR(handle);
 189
 190         lock_super(sb);
 191         if (input->group != sbi->s_groups_count) {
 192                 err = -EBUSY;
 193                 goto exit_journal;
 194         }
 195
 196         if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
 197                 err = PTR_ERR(bh);
 198                 goto exit_journal;
 199         }
 200
 201         if (ext4_bg_has_super(sb, input->group)) {
 202                 ext4_debug("mark backup superblock %#04lx (+0)\n", start);
 203                 ext4_set_bit(0, bh->b_data);
 204         }
 205
 206         /* Copy all of the GDT blocks into the backup in this group */
 207         for (i = 0, bit = 1, block = start + 1;
 208              i < gdblocks; i++, block++, bit++) {
 209                 struct buffer_head *gdb;
 210
 211                 ext4_debug("update backup group %#04lx (+%d)\n", block, bit);
 212
 213                 gdb = sb_getblk(sb, block);
 214                 if (!gdb) {
 215                         err = -EIO;
 216                         goto exit_bh;
 217                 }
 218                 if ((err = ext4_journal_get_write_access(handle, gdb))) {
 219                         brelse(gdb);
 220                         goto exit_bh;
 221                 }
 222                 lock_buffer(bh);
 223                 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, bh->b_size);
 224                 set_buffer_uptodate(gdb);
 225                 unlock_buffer(bh);
 226                 ext4_journal_dirty_metadata(handle, gdb);
 227                 ext4_set_bit(bit, bh->b_data);
 228                 brelse(gdb);
 229         }
 230
 231         /* Zero out all of the reserved backup group descriptor table blocks */
 232         for (i = 0, bit = gdblocks + 1, block = start + bit;
 233              i < reserved_gdb; i++, block++, bit++) {
 234                 struct buffer_head *gdb;
 235
 236                 ext4_debug("clear reserved block %#04lx (+%d)\n", block, bit);
 237
 238                 if (IS_ERR(gdb = bclean(handle, sb, block))) {
 239                         err = PTR_ERR(bh);
 240                         goto exit_bh;
 241                 }
 242                 ext4_journal_dirty_metadata(handle, gdb);
 243                 ext4_set_bit(bit, bh->b_data);
 244                 brelse(gdb);
 245         }
 246         ext4_debug("mark block bitmap %#04x (+%ld)\n", input->block_bitmap,
 247                    input->block_bitmap - start);
 248         ext4_set_bit(input->block_bitmap - start, bh->b_data);
 249         ext4_debug("mark inode bitmap %#04x (+%ld)\n", input->inode_bitmap,
 250                    input->inode_bitmap - start);
 251         ext4_set_bit(input->inode_bitmap - start, bh->b_data);
 252
 253         /* Zero out all of the inode table blocks */
 254         for (i = 0, block = input->inode_table, bit = block - start;
 255              i < sbi->s_itb_per_group; i++, bit++, block++) {
 256                 struct buffer_head *it;
 257
 258                 ext4_debug("clear inode block %#04lx (+%d)\n", block, bit);
 259                 if (IS_ERR(it = bclean(handle, sb, block))) {
 260                         err = PTR_ERR(it);
 261                         goto exit_bh;
 262                 }
 263                 ext4_journal_dirty_metadata(handle, it);
 264                 brelse(it);
 265                 ext4_set_bit(bit, bh->b_data);
 266         }
 267         mark_bitmap_end(input->blocks_count, EXT4_BLOCKS_PER_GROUP(sb),
 268                         bh->b_data);
 269         ext4_journal_dirty_metadata(handle, bh);
 270         brelse(bh);
 271
 272         /* Mark unused entries in inode bitmap used */
 273         ext4_debug("clear inode bitmap %#04x (+%ld)\n",
 274                    input->inode_bitmap, input->inode_bitmap - start);
 275         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 276                 err = PTR_ERR(bh);
 277                 goto exit_journal;
 278         }
 279
 280         mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
 281                         bh->b_data);
 282         ext4_journal_dirty_metadata(handle, bh);
 283 exit_bh:
 284         brelse(bh);
 285
 286 exit_journal:
 287         unlock_super(sb);
 288         if ((err2 = ext4_journal_stop(handle)) && !err)
 289                 err = err2;
 290
 291         return err;
 292 }
 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 unsigned long 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, __FUNCTION__,
 352                                      "reserved GDT %llu"
 353                                      " missing grp %d (%llu)",
 354                                      blk, grp,
 355                                      grp *
 356                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 357                                      blk);
 358                         return -EINVAL;
 359                 }
 360                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 361                         return -EFBIG;
 362         }
 363
 364         return gdbackups;
 365 }
 366
 367 /*
 368  * Called when we need to bring a reserved group descriptor table block into
 369  * use from the resize inode.  The primary copy of the new GDT block currently
 370  * is an indirect block (under the double indirect block in the resize inode).
 371  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 372  * block, in group order.  Even though we know all the block numbers we need,
 373  * we check to ensure that the resize inode has actually reserved these blocks.
 374  *
 375  * Don't need to update the block bitmaps because the blocks are still in use.
 376  *
 377  * We get all of the error cases out of the way, so that we are sure to not
 378  * fail once we start modifying the data on disk, because JBD has no rollback.
 379  */
 380 static int add_new_gdb(handle_t *handle, struct inode *inode,
 381                        struct ext4_new_group_data *input,
 382                        struct buffer_head **primary)
 383 {
 384         struct super_block *sb = inode->i_sb;
 385         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 386         unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 387         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 388         struct buffer_head **o_group_desc, **n_group_desc;
 389         struct buffer_head *dind;
 390         int gdbackups;
 391         struct ext4_iloc iloc;
 392         __le32 *data;
 393         int err;
 394
 395         if (test_opt(sb, DEBUG))
 396                 printk(KERN_DEBUG
 397                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 398                        gdb_num);
 399
 400         /*
 401          * If we are not using the primary superblock/GDT copy don't resize,
 402          * because the user tools have no way of handling this.  Probably a
 403          * bad time to do it anyways.
 404          */
 405         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 406             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 407                 ext4_warning(sb, __FUNCTION__,
 408                         "won't resize using backup superblock at %llu",
 409                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 410                 return -EPERM;
 411         }
 412
 413         *primary = sb_bread(sb, gdblock);
 414         if (!*primary)
 415                 return -EIO;
 416
 417         if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
 418                 err = gdbackups;
 419                 goto exit_bh;
 420         }
 421
 422         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 423         dind = sb_bread(sb, le32_to_cpu(*data));
 424         if (!dind) {
 425                 err = -EIO;
 426                 goto exit_bh;
 427         }
 428
 429         data = (__le32 *)dind->b_data;
 430         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 431                 ext4_warning(sb, __FUNCTION__,
 432                              "new group %u GDT block %llu not reserved",
 433                              input->group, gdblock);
 434                 err = -EINVAL;
 435                 goto exit_dind;
 436         }
 437
 438         if ((err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh)))
 439                 goto exit_dind;
 440
 441         if ((err = ext4_journal_get_write_access(handle, *primary)))
 442                 goto exit_sbh;
 443
 444         if ((err = ext4_journal_get_write_access(handle, dind)))
 445                 goto exit_primary;
 446
 447         /* ext4_reserve_inode_write() gets a reference on the iloc */
 448         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 449                 goto exit_dindj;
 450
 451         n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 452                         GFP_KERNEL);
 453         if (!n_group_desc) {
 454                 err = -ENOMEM;
 455                 ext4_warning (sb, __FUNCTION__,
 456                               "not enough memory for %lu groups", gdb_num + 1);
 457                 goto exit_inode;
 458         }
 459
 460         /*
 461          * Finally, we have all of the possible failures behind us...
 462          *
 463          * Remove new GDT block from inode double-indirect block and clear out
 464          * the new GDT block for use (which also "frees" the backup GDT blocks
 465          * from the reserved inode).  We don't need to change the bitmaps for
 466          * these blocks, because they are marked as in-use from being in the
 467          * reserved inode, and will become GDT blocks (primary and backup).
 468          */
 469         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 470         ext4_journal_dirty_metadata(handle, dind);
 471         brelse(dind);
 472         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 473         ext4_mark_iloc_dirty(handle, inode, &iloc);
 474         memset((*primary)->b_data, 0, sb->s_blocksize);
 475         ext4_journal_dirty_metadata(handle, *primary);
 476
 477         o_group_desc = EXT4_SB(sb)->s_group_desc;
 478         memcpy(n_group_desc, o_group_desc,
 479                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 480         n_group_desc[gdb_num] = *primary;
 481         EXT4_SB(sb)->s_group_desc = n_group_desc;
 482         EXT4_SB(sb)->s_gdb_count++;
 483         kfree(o_group_desc);
 484
 485         es->s_reserved_gdt_blocks =
 486                 cpu_to_le16(le16_to_cpu(es->s_reserved_gdt_blocks) - 1);
 487         ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
 488
 489         return 0;
 490
 491 exit_inode:
 492         //ext4_journal_release_buffer(handle, iloc.bh);
 493         brelse(iloc.bh);
 494 exit_dindj:
 495         //ext4_journal_release_buffer(handle, dind);
 496 exit_primary:
 497         //ext4_journal_release_buffer(handle, *primary);
 498 exit_sbh:
 499         //ext4_journal_release_buffer(handle, *primary);
 500 exit_dind:
 501         brelse(dind);
 502 exit_bh:
 503         brelse(*primary);
 504
 505         ext4_debug("leaving with error %d\n", err);
 506         return err;
 507 }
 508
 509 /*
 510  * Called when we are adding a new group which has a backup copy of each of
 511  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 512  * We need to add these reserved backup GDT blocks to the resize inode, so
 513  * that they are kept for future resizing and not allocated to files.
 514  *
 515  * Each reserved backup GDT block will go into a different indirect block.
 516  * The indirect blocks are actually the primary reserved GDT blocks,
 517  * so we know in advance what their block numbers are.  We only get the
 518  * double-indirect block to verify it is pointing to the primary reserved
 519  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 520  * backup GDT blocks are stored in their reserved primary GDT block.
 521  */
 522 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 523                               struct ext4_new_group_data *input)
 524 {
 525         struct super_block *sb = inode->i_sb;
 526         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 527         struct buffer_head **primary;
 528         struct buffer_head *dind;
 529         struct ext4_iloc iloc;
 530         ext4_fsblk_t blk;
 531         __le32 *data, *end;
 532         int gdbackups = 0;
 533         int res, i;
 534         int err;
 535
 536         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_KERNEL);
 537         if (!primary)
 538                 return -ENOMEM;
 539
 540         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 541         dind = sb_bread(sb, le32_to_cpu(*data));
 542         if (!dind) {
 543                 err = -EIO;
 544                 goto exit_free;
 545         }
 546
 547         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 548         data = (__le32 *)dind->b_data + EXT4_SB(sb)->s_gdb_count;
 549         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 550
 551         /* Get each reserved primary GDT block and verify it holds backups */
 552         for (res = 0; res < reserved_gdb; res++, blk++) {
 553                 if (le32_to_cpu(*data) != blk) {
 554                         ext4_warning(sb, __FUNCTION__,
 555                                      "reserved block %llu"
 556                                      " not at offset %ld",
 557                                      blk,
 558                                      (long)(data - (__le32 *)dind->b_data));
 559                         err = -EINVAL;
 560                         goto exit_bh;
 561                 }
 562                 primary[res] = sb_bread(sb, blk);
 563                 if (!primary[res]) {
 564                         err = -EIO;
 565                         goto exit_bh;
 566                 }
 567                 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
 568                         brelse(primary[res]);
 569                         err = gdbackups;
 570                         goto exit_bh;
 571                 }
 572                 if (++data >= end)
 573                         data = (__le32 *)dind->b_data;
 574         }
 575
 576         for (i = 0; i < reserved_gdb; i++) {
 577                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
 578                         /*
 579                         int j;
 580                         for (j = 0; j < i; j++)
 581                                 ext4_journal_release_buffer(handle, primary[j]);
 582                          */
 583                         goto exit_bh;
 584                 }
 585         }
 586
 587         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 588                 goto exit_bh;
 589
 590         /*
 591          * Finally we can add each of the reserved backup GDT blocks from
 592          * the new group to its reserved primary GDT block.
 593          */
 594         blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
 595         for (i = 0; i < reserved_gdb; i++) {
 596                 int err2;
 597                 data = (__le32 *)primary[i]->b_data;
 598                 /* printk("reserving backup %lu[%u] = %lu\n",
 599                        primary[i]->b_blocknr, gdbackups,
 600                        blk + primary[i]->b_blocknr); */
 601                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 602                 err2 = ext4_journal_dirty_metadata(handle, primary[i]);
 603                 if (!err)
 604                         err = err2;
 605         }
 606         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 607         ext4_mark_iloc_dirty(handle, inode, &iloc);
 608
 609 exit_bh:
 610         while (--res >= 0)
 611                 brelse(primary[res]);
 612         brelse(dind);
 613
 614 exit_free:
 615         kfree(primary);
 616
 617         return err;
 618 }
 619
 620 /*
 621  * Update the backup copies of the ext4 metadata.  These don't need to be part
 622  * of the main resize transaction, because e2fsck will re-write them if there
 623  * is a problem (basically only OOM will cause a problem).  However, we
 624  * _should_ update the backups if possible, in case the primary gets trashed
 625  * for some reason and we need to run e2fsck from a backup superblock.  The
 626  * important part is that the new block and inode counts are in the backup
 627  * superblocks, and the location of the new group metadata in the GDT backups.
 628  *
 629  * We do not need lock_super() for this, because these blocks are not
 630  * otherwise touched by the filesystem code when it is mounted.  We don't
 631  * need to worry about last changing from sbi->s_groups_count, because the
 632  * worst that can happen is that we do not copy the full number of backups
 633  * at this time.  The resize which changed s_groups_count will backup again.
 634  */
 635 static void update_backups(struct super_block *sb,
 636                            int blk_off, char *data, int size)
 637 {
 638         struct ext4_sb_info *sbi = EXT4_SB(sb);
 639         const unsigned long last = sbi->s_groups_count;
 640         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
 641         unsigned three = 1;
 642         unsigned five = 5;
 643         unsigned seven = 7;
 644         unsigned group;
 645         int rest = sb->s_blocksize - size;
 646         handle_t *handle;
 647         int err = 0, err2;
 648
 649         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 650         if (IS_ERR(handle)) {
 651                 group = 1;
 652                 err = PTR_ERR(handle);
 653                 goto exit_err;
 654         }
 655
 656         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
 657                 struct buffer_head *bh;
 658
 659                 /* Out of journal space, and can't get more - abort - so sad */
 660                 if (handle->h_buffer_credits == 0 &&
 661                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
 662                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 663                         break;
 664
 665                 bh = sb_getblk(sb, group * bpg + blk_off);
 666                 if (!bh) {
 667                         err = -EIO;
 668                         break;
 669                 }
 670                 ext4_debug("update metadata backup %#04lx\n",
 671                           (unsigned long)bh->b_blocknr);
 672                 if ((err = ext4_journal_get_write_access(handle, bh)))
 673                         break;
 674                 lock_buffer(bh);
 675                 memcpy(bh->b_data, data, size);
 676                 if (rest)
 677                         memset(bh->b_data + size, 0, rest);
 678                 set_buffer_uptodate(bh);
 679                 unlock_buffer(bh);
 680                 ext4_journal_dirty_metadata(handle, bh);
 681                 brelse(bh);
 682         }
 683         if ((err2 = ext4_journal_stop(handle)) && !err)
 684                 err = err2;
 685
 686         /*
 687          * Ugh! Need to have e2fsck write the backup copies.  It is too
 688          * late to revert the resize, we shouldn't fail just because of
 689          * the backup copies (they are only needed in case of corruption).
 690          *
 691          * However, if we got here we have a journal problem too, so we
 692          * can't really start a transaction to mark the superblock.
 693          * Chicken out and just set the flag on the hope it will be written
 694          * to disk, and if not - we will simply wait until next fsck.
 695          */
 696 exit_err:
 697         if (err) {
 698                 ext4_warning(sb, __FUNCTION__,
 699                              "can't update backup for group %d (err %d), "
 700                              "forcing fsck on next reboot", group, err);
 701                 sbi->s_mount_state &= ~EXT4_VALID_FS;
 702                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
 703                 mark_buffer_dirty(sbi->s_sbh);
 704         }
 705 }
 706
 707 /* Add group descriptor data to an existing or new group descriptor block.
 708  * Ensure we handle all possible error conditions _before_ we start modifying
 709  * the filesystem, because we cannot abort the transaction and not have it
 710  * write the data to disk.
 711  *
 712  * If we are on a GDT block boundary, we need to get the reserved GDT block.
 713  * Otherwise, we may need to add backup GDT blocks for a sparse group.
 714  *
 715  * We only need to hold the superblock lock while we are actually adding
 716  * in the new group's counts to the superblock.  Prior to that we have
 717  * not really "added" the group at all.  We re-check that we are still
 718  * adding in the last group in case things have changed since verifying.
 719  */
 720 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
 721 {
 722         struct ext4_sb_info *sbi = EXT4_SB(sb);
 723         struct ext4_super_block *es = sbi->s_es;
 724         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 725                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
 726         struct buffer_head *primary = NULL;
 727         struct ext4_group_desc *gdp;
 728         struct inode *inode = NULL;
 729         handle_t *handle;
 730         int gdb_off, gdb_num;
 731         int err, err2;
 732
 733         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 734         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
 735
 736         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
 737                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 738                 ext4_warning(sb, __FUNCTION__,
 739                              "Can't resize non-sparse filesystem further");
 740                 return -EPERM;
 741         }
 742
 743         if (ext4_blocks_count(es) + input->blocks_count <
 744             ext4_blocks_count(es)) {
 745                 ext4_warning(sb, __FUNCTION__, "blocks_count overflow\n");
 746                 return -EINVAL;
 747         }
 748
 749         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
 750             le32_to_cpu(es->s_inodes_count)) {
 751                 ext4_warning(sb, __FUNCTION__, "inodes_count overflow\n");
 752                 return -EINVAL;
 753         }
 754
 755         if (reserved_gdb || gdb_off == 0) {
 756                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
 757                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)){
 758                         ext4_warning(sb, __FUNCTION__,
 759                                      "No reserved GDT blocks, can't resize");
 760                         return -EPERM;
 761                 }
 762                 inode = iget(sb, EXT4_RESIZE_INO);
 763                 if (!inode || is_bad_inode(inode)) {
 764                         ext4_warning(sb, __FUNCTION__,
 765                                      "Error opening resize inode");
 766                         iput(inode);
 767                         return -ENOENT;
 768                 }
 769         }
 770
 771         if ((err = verify_group_input(sb, input)))
 772                 goto exit_put;
 773
 774         if ((err = setup_new_group_blocks(sb, input)))
 775                 goto exit_put;
 776
 777         /*
 778          * We will always be modifying at least the superblock and a GDT
 779          * block.  If we are adding a group past the last current GDT block,
 780          * we will also modify the inode and the dindirect block.  If we
 781          * are adding a group with superblock/GDT backups  we will also
 782          * modify each of the reserved GDT dindirect blocks.
 783          */
 784         handle = ext4_journal_start_sb(sb,
 785                                        ext4_bg_has_super(sb, input->group) ?
 786                                        3 + reserved_gdb : 4);
 787         if (IS_ERR(handle)) {
 788                 err = PTR_ERR(handle);
 789                 goto exit_put;
 790         }
 791
 792         lock_super(sb);
 793         if (input->group != sbi->s_groups_count) {
 794                 ext4_warning(sb, __FUNCTION__,
 795                              "multiple resizers run on filesystem!");
 796                 err = -EBUSY;
 797                 goto exit_journal;
 798         }
 799
 800         if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
 801                 goto exit_journal;
 802
 803         /*
 804          * We will only either add reserved group blocks to a backup group
 805          * or remove reserved blocks for the first group in a new group block.
 806          * Doing both would be mean more complex code, and sane people don't
 807          * use non-sparse filesystems anymore.  This is already checked above.
 808          */
 809         if (gdb_off) {
 810                 primary = sbi->s_group_desc[gdb_num];
 811                 if ((err = ext4_journal_get_write_access(handle, primary)))
 812                         goto exit_journal;
 813
 814                 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
 815                     (err = reserve_backup_gdb(handle, inode, input)))
 816                         goto exit_journal;
 817         } else if ((err = add_new_gdb(handle, inode, input, &primary)))
 818                 goto exit_journal;
 819
 820         /*
 821          * OK, now we've set up the new group.  Time to make it active.
 822          *
 823          * Current kernels don't lock all allocations via lock_super(),
 824          * so we have to be safe wrt. concurrent accesses the group
 825          * data.  So we need to be careful to set all of the relevant
 826          * group descriptor data etc. *before* we enable the group.
 827          *
 828          * The key field here is sbi->s_groups_count: as long as
 829          * that retains its old value, nobody is going to access the new
 830          * group.
 831          *
 832          * So first we update all the descriptor metadata for the new
 833          * group; then we update the total disk blocks count; then we
 834          * update the groups count to enable the group; then finally we
 835          * update the free space counts so that the system can start
 836          * using the new disk blocks.
 837          */
 838
 839         /* Update group descriptor block for new group */
 840         gdp = (struct ext4_group_desc *)primary->b_data + gdb_off;
 841
 842         ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
 843         ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
 844         ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
 845         gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
 846         gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
 847
 848         /*
 849          * Make the new blocks and inodes valid next.  We do this before
 850          * increasing the group count so that once the group is enabled,
 851          * all of its blocks and inodes are already valid.
 852          *
 853          * We always allocate group-by-group, then block-by-block or
 854          * inode-by-inode within a group, so enabling these
 855          * blocks/inodes before the group is live won't actually let us
 856          * allocate the new space yet.
 857          */
 858         ext4_blocks_count_set(es, ext4_blocks_count(es) +
 859                 input->blocks_count);
 860         es->s_inodes_count = cpu_to_le32(le32_to_cpu(es->s_inodes_count) +
 861                 EXT4_INODES_PER_GROUP(sb));
 862
 863         /*
 864          * We need to protect s_groups_count against other CPUs seeing
 865          * inconsistent state in the superblock.
 866          *
 867          * The precise rules we use are:
 868          *
 869          * * Writers of s_groups_count *must* hold lock_super
 870          * AND
 871          * * Writers must perform a smp_wmb() after updating all dependent
 872          *   data and before modifying the groups count
 873          *
 874          * * Readers must hold lock_super() over the access
 875          * OR
 876          * * Readers must perform an smp_rmb() after reading the groups count
 877          *   and before reading any dependent data.
 878          *
 879          * NB. These rules can be relaxed when checking the group count
 880          * while freeing data, as we can only allocate from a block
 881          * group after serialising against the group count, and we can
 882          * only then free after serialising in turn against that
 883          * allocation.
 884          */
 885         smp_wmb();
 886
 887         /* Update the global fs size fields */
 888         sbi->s_groups_count++;
 889
 890         ext4_journal_dirty_metadata(handle, primary);
 891
 892         /* Update the reserved block counts only once the new group is
 893          * active. */
 894         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
 895                 input->reserved_blocks);
 896
 897         /* Update the free space counts */
 898         percpu_counter_mod(&sbi->s_freeblocks_counter,
 899                            input->free_blocks_count);
 900         percpu_counter_mod(&sbi->s_freeinodes_counter,
 901                            EXT4_INODES_PER_GROUP(sb));
 902
 903         ext4_journal_dirty_metadata(handle, sbi->s_sbh);
 904         sb->s_dirt = 1;
 905
 906 exit_journal:
 907         unlock_super(sb);
 908         if ((err2 = ext4_journal_stop(handle)) && !err)
 909                 err = err2;
 910         if (!err) {
 911                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
 912                                sizeof(struct ext4_super_block));
 913                 update_backups(sb, primary->b_blocknr, primary->b_data,
 914                                primary->b_size);
 915         }
 916 exit_put:
 917         iput(inode);
 918         return err;
 919 } /* ext4_group_add */
 920
 921 /* Extend the filesystem to the new number of blocks specified.  This entry
 922  * point is only used to extend the current filesystem to the end of the last
 923  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
 924  * for emergencies (because it has no dependencies on reserved blocks).
 925  *
 926  * If we _really_ wanted, we could use default values to call ext4_group_add()
 927  * allow the "remount" trick to work for arbitrary resizing, assuming enough
 928  * GDT blocks are reserved to grow to the desired size.
 929  */
 930 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
 931                       ext4_fsblk_t n_blocks_count)
 932 {
 933         ext4_fsblk_t o_blocks_count;
 934         unsigned long o_groups_count;
 935         ext4_grpblk_t last;
 936         ext4_grpblk_t add;
 937         struct buffer_head * bh;
 938         handle_t *handle;
 939         int err;
 940         unsigned long freed_blocks;
 941
 942         /* We don't need to worry about locking wrt other resizers just
 943          * yet: we're going to revalidate es->s_blocks_count after
 944          * taking lock_super() below. */
 945         o_blocks_count = ext4_blocks_count(es);
 946         o_groups_count = EXT4_SB(sb)->s_groups_count;
 947
 948         if (test_opt(sb, DEBUG))
 949                 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
 950                        o_blocks_count, n_blocks_count);
 951
 952         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
 953                 return 0;
 954
 955         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
 956                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
 957                         " too large to resize to %llu blocks safely\n",
 958                         sb->s_id, n_blocks_count);
 959                 if (sizeof(sector_t) < 8)
 960                         ext4_warning(sb, __FUNCTION__,
 961                         "CONFIG_LBD not enabled\n");
 962                 return -EINVAL;
 963         }
 964
 965         if (n_blocks_count < o_blocks_count) {
 966                 ext4_warning(sb, __FUNCTION__,
 967                              "can't shrink FS - resize aborted");
 968                 return -EBUSY;
 969         }
 970
 971         /* Handle the remaining blocks in the last group only. */
 972         ext4_get_group_no_and_offset(sb, o_blocks_count, NULL, &last);
 973
 974         if (last == 0) {
 975                 ext4_warning(sb, __FUNCTION__,
 976                              "need to use ext2online to resize further");
 977                 return -EPERM;
 978         }
 979
 980         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
 981
 982         if (o_blocks_count + add < o_blocks_count) {
 983                 ext4_warning(sb, __FUNCTION__, "blocks_count overflow");
 984                 return -EINVAL;
 985         }
 986
 987         if (o_blocks_count + add > n_blocks_count)
 988                 add = n_blocks_count - o_blocks_count;
 989
 990         if (o_blocks_count + add < n_blocks_count)
 991                 ext4_warning(sb, __FUNCTION__,
 992                              "will only finish group (%llu"
 993                              " blocks, %u new)",
 994                              o_blocks_count + add, add);
 995
 996         /* See if the device is actually as big as what was requested */
 997         bh = sb_bread(sb, o_blocks_count + add -1);
 998         if (!bh) {
 999                 ext4_warning(sb, __FUNCTION__,
1000                              "can't read last block, resize aborted");
1001                 return -ENOSPC;
1002         }
1003         brelse(bh);
1004
1005         /* We will update the superblock, one block bitmap, and
1006          * one group descriptor via ext4_free_blocks().
1007          */
1008         handle = ext4_journal_start_sb(sb, 3);
1009         if (IS_ERR(handle)) {
1010                 err = PTR_ERR(handle);
1011                 ext4_warning(sb, __FUNCTION__, "error %d on journal start",err);
1012                 goto exit_put;
1013         }
1014
1015         lock_super(sb);
1016         if (o_blocks_count != ext4_blocks_count(es)) {
1017                 ext4_warning(sb, __FUNCTION__,
1018                              "multiple resizers run on filesystem!");
1019                 unlock_super(sb);
1020                 err = -EBUSY;
1021                 goto exit_put;
1022         }
1023
1024         if ((err = ext4_journal_get_write_access(handle,
1025                                                  EXT4_SB(sb)->s_sbh))) {
1026                 ext4_warning(sb, __FUNCTION__,
1027                              "error %d on journal write access", err);
1028                 unlock_super(sb);
1029                 ext4_journal_stop(handle);
1030                 goto exit_put;
1031         }
1032         ext4_blocks_count_set(es, o_blocks_count + add);
1033         ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1034         sb->s_dirt = 1;
1035         unlock_super(sb);
1036         ext4_debug("freeing blocks %lu through %llu\n", o_blocks_count,
1037                    o_blocks_count + add);
1038         ext4_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
1039         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1040                    o_blocks_count + add);
1041         if ((err = ext4_journal_stop(handle)))
1042                 goto exit_put;
1043         if (test_opt(sb, DEBUG))
1044                 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1045                        ext4_blocks_count(es));
1046         update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1047                        sizeof(struct ext4_super_block));
1048 exit_put:
1049         return err;
1050 } /* ext4_group_extend */