fs/ext3/resize.c

   1 /*
   2  *  linux/fs/ext3/resize.c
   3  *
   4  * Support for resizing an ext3 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 #include <linux/config.h>
  12
  13 #define EXT3FS_DEBUG
  14
  15 #include <linux/sched.h>
  16 #include <linux/smp_lock.h>
  17 #include <linux/ext3_jbd.h>
  18
  19 #include <linux/errno.h>
  20 #include <linux/slab.h>
  21
  22
  23 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  24 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  25
  26 static int verify_group_input(struct super_block *sb,
  27                               struct ext3_new_group_data *input)
  28 {
  29         struct ext3_sb_info *sbi = EXT3_SB(sb);
  30         struct ext3_super_block *es = sbi->s_es;
  31         unsigned start = le32_to_cpu(es->s_blocks_count);
  32         unsigned end = start + input->blocks_count;
  33         unsigned group = input->group;
  34         unsigned itend = input->inode_table + EXT3_SB(sb)->s_itb_per_group;
  35         unsigned overhead = ext3_bg_has_super(sb, group) ?
  36                 (1 + ext3_bg_num_gdb(sb, group) +
  37                  le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
  38         unsigned metaend = start + overhead;
  39         struct buffer_head *bh;
  40         int free_blocks_count;
  41         int err = -EINVAL;
  42
  43         input->free_blocks_count = free_blocks_count =
  44                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  45
  46         if (test_opt(sb, DEBUG))
  47                 printk("EXT3-fs: adding %s group %u: %u blocks "
  48                        "(%d free, %u reserved)\n",
  49                        ext3_bg_has_super(sb, input->group) ? "normal" :
  50                        "no-super", input->group, input->blocks_count,
  51                        free_blocks_count, input->reserved_blocks);
  52
  53         if (group != sbi->s_groups_count)
  54                 ext3_warning(sb, __FUNCTION__,
  55                              "Cannot add at group %u (only %lu groups)",
  56                              input->group, sbi->s_groups_count);
  57         else if ((start - le32_to_cpu(es->s_first_data_block)) %
  58                  EXT3_BLOCKS_PER_GROUP(sb))
  59                 ext3_warning(sb, __FUNCTION__, "Last group not full");
  60         else if (input->reserved_blocks > input->blocks_count / 5)
  61                 ext3_warning(sb, __FUNCTION__, "Reserved blocks too high (%u)",
  62                              input->reserved_blocks);
  63         else if (free_blocks_count < 0)
  64                 ext3_warning(sb, __FUNCTION__, "Bad blocks count %u",
  65                              input->blocks_count);
  66         else if (!(bh = sb_bread(sb, end - 1)))
  67                 ext3_warning(sb, __FUNCTION__, "Cannot read last block (%u)",
  68                              end - 1);
  69         else if (outside(input->block_bitmap, start, end))
  70                 ext3_warning(sb, __FUNCTION__,
  71                              "Block bitmap not in group (block %u)",
  72                              input->block_bitmap);
  73         else if (outside(input->inode_bitmap, start, end))
  74                 ext3_warning(sb, __FUNCTION__,
  75                              "Inode bitmap not in group (block %u)",
  76                              input->inode_bitmap);
  77         else if (outside(input->inode_table, start, end) ||
  78                  outside(itend - 1, start, end))
  79                 ext3_warning(sb, __FUNCTION__,
  80                              "Inode table not in group (blocks %u-%u)",
  81                              input->inode_table, itend - 1);
  82         else if (input->inode_bitmap == input->block_bitmap)
  83                 ext3_warning(sb, __FUNCTION__,
  84                              "Block bitmap same as inode bitmap (%u)",
  85                              input->block_bitmap);
  86         else if (inside(input->block_bitmap, input->inode_table, itend))
  87                 ext3_warning(sb, __FUNCTION__,
  88                              "Block bitmap (%u) in inode table (%u-%u)",
  89                              input->block_bitmap, input->inode_table, itend-1);
  90         else if (inside(input->inode_bitmap, input->inode_table, itend))
  91                 ext3_warning(sb, __FUNCTION__,
  92                              "Inode bitmap (%u) in inode table (%u-%u)",
  93                              input->inode_bitmap, input->inode_table, itend-1);
  94         else if (inside(input->block_bitmap, start, metaend))
  95                 ext3_warning(sb, __FUNCTION__,
  96                              "Block bitmap (%u) in GDT table (%u-%u)",
  97                              input->block_bitmap, start, metaend - 1);
  98         else if (inside(input->inode_bitmap, start, metaend))
  99                 ext3_warning(sb, __FUNCTION__,
 100                              "Inode bitmap (%u) in GDT table (%u-%u)",
 101                              input->inode_bitmap, start, metaend - 1);
 102         else if (inside(input->inode_table, start, metaend) ||
 103                  inside(itend - 1, start, metaend))
 104                 ext3_warning(sb, __FUNCTION__,
 105                              "Inode table (%u-%u) overlaps GDT table (%u-%u)",
 106                              input->inode_table, itend - 1, start, metaend - 1);
 107         else {
 108                 brelse(bh);
 109                 err = 0;
 110         }
 111
 112         return err;
 113 }
 114
 115 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 116                                   unsigned long blk)
 117 {
 118         struct buffer_head *bh;
 119         int err;
 120
 121         bh = sb_getblk(sb, blk);
 122         set_buffer_uptodate(bh);
 123         if ((err = ext3_journal_get_write_access(handle, bh))) {
 124                 brelse(bh);
 125                 bh = ERR_PTR(err);
 126         } else
 127                 memset(bh->b_data, 0, sb->s_blocksize);
 128
 129         return bh;
 130 }
 131
 132 /*
 133  * To avoid calling the atomic setbit hundreds or thousands of times, we only
 134  * need to use it within a single byte (to ensure we get endianness right).
 135  * We can use memset for the rest of the bitmap as there are no other users.
 136  */
 137 static void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
 138 {
 139         int i;
 140
 141         if (start_bit >= end_bit)
 142                 return;
 143
 144         ext3_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
 145         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
 146                 ext3_set_bit(i, bitmap);
 147         if (i < end_bit)
 148                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
 149 }
 150
 151 /*
 152  * Set up the block and inode bitmaps, and the inode table for the new group.
 153  * This doesn't need to be part of the main transaction, since we are only
 154  * changing blocks outside the actual filesystem.  We still do journaling to
 155  * ensure the recovery is correct in case of a failure just after resize.
 156  * If any part of this fails, we simply abort the resize.
 157  *
 158  * We only pass inode because of the ext3 journal wrappers.
 159  */
 160 static int setup_new_group_blocks(struct super_block *sb, struct inode *inode,
 161                                   struct ext3_new_group_data *input)
 162 {
 163         struct ext3_sb_info *sbi = EXT3_SB(sb);
 164         unsigned long start = input->group * sbi->s_blocks_per_group +
 165                 le32_to_cpu(sbi->s_es->s_first_data_block);
 166         int reserved_gdb = ext3_bg_has_super(sb, input->group) ?
 167                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 168         unsigned long gdblocks = ext3_bg_num_gdb(sb, input->group);
 169         struct buffer_head *bh;
 170         handle_t *handle;
 171         unsigned long block;
 172         int bit;
 173         int i;
 174         int err = 0, err2;
 175
 176         handle = ext3_journal_start(inode, reserved_gdb + gdblocks +
 177                                     2 + sbi->s_itb_per_group);
 178         if (IS_ERR(handle))
 179                 return PTR_ERR(handle);
 180
 181         lock_super(sb);
 182         if (input->group != sbi->s_groups_count) {
 183                 err = -EBUSY;
 184                 goto exit_journal;
 185         }
 186
 187         if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
 188                 err = PTR_ERR(bh);
 189                 goto exit_journal;
 190         }
 191
 192         if (ext3_bg_has_super(sb, input->group)) {
 193                 ext3_debug("mark backup superblock %#04lx (+0)\n", start);
 194                 ext3_set_bit(0, bh->b_data);
 195         }
 196
 197         /* Copy all of the GDT blocks into the backup in this group */
 198         for (i = 0, bit = 1, block = start + 1;
 199              i < gdblocks; i++, block++, bit++) {
 200                 struct buffer_head *gdb;
 201
 202                 ext3_debug("update backup group %#04lx (+%d)\n", block, bit);
 203
 204                 gdb = sb_getblk(sb, block);
 205                 set_buffer_uptodate(gdb);
 206                 if ((err = ext3_journal_get_write_access(handle, gdb))) {
 207                         brelse(gdb);
 208                         goto exit_bh;
 209                 }
 210                 memcpy(gdb->b_data, sbi->s_group_desc[i], bh->b_size);
 211                 ext3_journal_dirty_metadata(handle, gdb);
 212                 ext3_set_bit(bit, bh->b_data);
 213                 brelse(gdb);
 214         }
 215
 216         /* Zero out all of the reserved backup group descriptor table blocks */
 217         for (i = 0, bit = gdblocks + 1, block = start + bit;
 218              i < reserved_gdb; i++, block++, bit++) {
 219                 struct buffer_head *gdb;
 220
 221                 ext3_debug("clear reserved block %#04lx (+%d)\n", block, bit);
 222
 223                 if (IS_ERR(gdb = bclean(handle, sb, block))) {
 224                         err = PTR_ERR(bh);
 225                         goto exit_bh;
 226                 }
 227                 ext3_journal_dirty_metadata(handle, gdb);
 228                 ext3_set_bit(bit, bh->b_data);
 229                 brelse(gdb);
 230         }
 231         ext3_debug("mark block bitmap %#04x (+%ld)\n", input->block_bitmap,
 232                    input->block_bitmap - start);
 233         ext3_set_bit(input->block_bitmap - start, bh->b_data);
 234         ext3_debug("mark inode bitmap %#04x (+%ld)\n", input->inode_bitmap,
 235                    input->inode_bitmap - start);
 236         ext3_set_bit(input->inode_bitmap - start, bh->b_data);
 237
 238         /* Zero out all of the inode table blocks */
 239         for (i = 0, block = input->inode_table, bit = block - start;
 240              i < sbi->s_itb_per_group; i++, bit++, block++) {
 241                 struct buffer_head *it;
 242
 243                 ext3_debug("clear inode block %#04x (+%ld)\n", block, bit);
 244                 if (IS_ERR(it = bclean(handle, sb, block))) {
 245                         err = PTR_ERR(it);
 246                         goto exit_bh;
 247                 }
 248                 ext3_journal_dirty_metadata(handle, it);
 249                 brelse(it);
 250                 ext3_set_bit(bit, bh->b_data);
 251         }
 252         mark_bitmap_end(input->blocks_count, EXT3_BLOCKS_PER_GROUP(sb),
 253                         bh->b_data);
 254         ext3_journal_dirty_metadata(handle, bh);
 255         brelse(bh);
 256
 257         /* Mark unused entries in inode bitmap used */
 258         ext3_debug("clear inode bitmap %#04x (+%ld)\n",
 259                    input->inode_bitmap, input->inode_bitmap - start);
 260         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 261                 err = PTR_ERR(bh);
 262                 goto exit_journal;
 263         }
 264
 265         mark_bitmap_end(EXT3_INODES_PER_GROUP(sb), EXT3_BLOCKS_PER_GROUP(sb),
 266                         bh->b_data);
 267         ext3_journal_dirty_metadata(handle, bh);
 268 exit_bh:
 269         brelse(bh);
 270
 271 exit_journal:
 272         unlock_super(sb);
 273         if ((err2 = ext3_journal_stop(handle)) && !err)
 274                 err = err2;
 275
 276         return err;
 277 }
 278
 279 /*
 280  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 281  * ext3 filesystem.  The counters should be initialized to 1, 5, and 7 before
 282  * calling this for the first time.  In a sparse filesystem it will be the
 283  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 284  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 285  */
 286 unsigned ext3_list_backups(struct super_block *sb, unsigned *three,
 287                            unsigned *five, unsigned *seven)
 288 {
 289         unsigned *min = three;
 290         int mult = 3;
 291         unsigned ret;
 292
 293         if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
 294                                         EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 295                 ret = *min;
 296                 *min += 1;
 297                 return ret;
 298         }
 299
 300         if (*five < *min) {
 301                 min = five;
 302                 mult = 5;
 303         }
 304         if (*seven < *min) {
 305                 min = seven;
 306                 mult = 7;
 307         }
 308
 309         ret = *min;
 310         *min *= mult;
 311
 312         return ret;
 313 }
 314
 315 /*
 316  * Check that all of the backup GDT blocks are held in the primary GDT block.
 317  * It is assumed that they are stored in group order.  Returns the number of
 318  * groups in current filesystem that have BACKUPS, or -ve error code.
 319  */
 320 static int verify_reserved_gdb(struct super_block *sb,
 321                                struct buffer_head *primary)
 322 {
 323         const unsigned long blk = primary->b_blocknr;
 324         const unsigned long end = EXT3_SB(sb)->s_groups_count;
 325         unsigned three = 1;
 326         unsigned five = 5;
 327         unsigned seven = 7;
 328         unsigned grp;
 329         __u32 *p = (__u32 *)primary->b_data;
 330         int gdbackups = 0;
 331
 332         while ((grp = ext3_list_backups(sb, &three, &five, &seven)) < end) {
 333                 if (le32_to_cpu(*p++) != grp * EXT3_BLOCKS_PER_GROUP(sb) + blk){
 334                         ext3_warning(sb, __FUNCTION__,
 335                                      "reserved GDT %ld missing grp %d (%ld)\n",
 336                                      blk, grp,
 337                                      grp * EXT3_BLOCKS_PER_GROUP(sb) + blk);
 338                         return -EINVAL;
 339                 }
 340                 if (++gdbackups > EXT3_ADDR_PER_BLOCK(sb))
 341                         return -EFBIG;
 342         }
 343
 344         return gdbackups;
 345 }
 346
 347 /*
 348  * Called when we need to bring a reserved group descriptor table block into
 349  * use from the resize inode.  The primary copy of the new GDT block currently
 350  * is an indirect block (under the double indirect block in the resize inode).
 351  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 352  * block, in group order.  Even though we know all the block numbers we need,
 353  * we check to ensure that the resize inode has actually reserved these blocks.
 354  *
 355  * Don't need to update the block bitmaps because the blocks are still in use.
 356  *
 357  * We get all of the error cases out of the way, so that we are sure to not
 358  * fail once we start modifying the data on disk, because JBD has no rollback.
 359  */
 360 static int add_new_gdb(handle_t *handle, struct inode *inode,
 361                        struct ext3_new_group_data *input,
 362                        struct buffer_head **primary)
 363 {
 364         struct super_block *sb = inode->i_sb;
 365         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 366         unsigned long gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb);
 367         unsigned long gdb_off = input->group % EXT3_DESC_PER_BLOCK(sb);
 368         unsigned long gdblock = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 369         struct buffer_head **o_group_desc, **n_group_desc;
 370         struct buffer_head *dind;
 371         int gdbackups;
 372         struct ext3_iloc iloc;
 373         __u32 *data;
 374         int err;
 375
 376         if (test_opt(sb, DEBUG))
 377                 printk("EXT3-fs: ext3_add_new_gdb: adding group block %lu\n",
 378                        gdb_num);
 379
 380         /*
 381          * If we are not using the primary superblock/GDT copy don't resize,
 382          * because the user tools have no way of handling this.  Probably a
 383          * bad time to do it anyways.
 384          */
 385         if (EXT3_SB(sb)->s_sbh->b_blocknr !=
 386             le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) {
 387                 ext3_warning(sb, __FUNCTION__,
 388                              "won't resize using backup superblock at %lu\n",
 389                              EXT3_SB(sb)->s_sbh->b_blocknr);
 390                 return -EPERM;
 391         }
 392
 393         *primary = sb_bread(sb, gdblock);
 394         if (!*primary)
 395                 return -EIO;
 396
 397         if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
 398                 err = gdbackups;
 399                 goto exit_bh;
 400         }
 401
 402         data = EXT3_I(inode)->i_data + EXT3_DIND_BLOCK;
 403         dind = sb_bread(sb, le32_to_cpu(*data));
 404         if (!dind) {
 405                 err = -EIO;
 406                 goto exit_bh;
 407         }
 408
 409         data = (__u32 *)dind->b_data;
 410         if (le32_to_cpu(data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)]) != gdblock) {
 411                 ext3_warning(sb, __FUNCTION__,
 412                              "new group %u GDT block %lu not reserved\n",
 413                              input->group, gdblock);
 414                 err = -EINVAL;
 415                 goto exit_dind;
 416         }
 417
 418         if ((err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh)))
 419                 goto exit_dind;
 420
 421         if ((err = ext3_journal_get_write_access(handle, *primary)))
 422                 goto exit_sbh;
 423
 424         if ((err = ext3_journal_get_write_access(handle, dind)))
 425                 goto exit_primary;
 426
 427         /* ext3_reserve_inode_write() gets a reference on the iloc */
 428         if ((err = ext3_reserve_inode_write(handle, inode, &iloc)))
 429                 goto exit_dindj;
 430
 431         n_group_desc = (struct buffer_head **)kmalloc((gdb_num + 1) *
 432                                 sizeof(struct buffer_head *), GFP_KERNEL);
 433         if (!n_group_desc) {
 434                 err = -ENOMEM;
 435                 ext3_warning (sb, __FUNCTION__,
 436                               "not enough memory for %lu groups", gdb_num + 1);
 437                 goto exit_inode;
 438         }
 439
 440         /*
 441          * Finally, we have all of the possible failures behind us...
 442          *
 443          * Remove new GDT block from inode double-indirect block and clear out
 444          * the new GDT block for use (which also "frees" the backup GDT blocks
 445          * from the reserved inode).  We don't need to change the bitmaps for
 446          * these blocks, because they are marked as in-use from being in the
 447          * reserved inode, and will become GDT blocks (primary and backup).
 448          */
 449         /*
 450         printk("removing block %d = %ld from dindir %ld[%ld]\n",
 451                ((__u32 *)(dind->b_data))[gdb_off], gdblock, dind->b_blocknr,
 452                gdb_num); */
 453         data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)] = 0;
 454         ext3_journal_dirty_metadata(handle, dind);
 455         brelse(dind);
 456         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 457         ext3_mark_iloc_dirty(handle, inode, &iloc);
 458         memset((*primary)->b_data, 0, sb->s_blocksize);
 459         ext3_journal_dirty_metadata(handle, *primary);
 460
 461         o_group_desc = EXT3_SB(sb)->s_group_desc;
 462         memcpy(n_group_desc, o_group_desc,
 463                EXT3_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 464         n_group_desc[gdb_num] = *primary;
 465         EXT3_SB(sb)->s_group_desc = n_group_desc;
 466         EXT3_SB(sb)->s_gdb_count++;
 467         kfree(o_group_desc);
 468
 469         es->s_reserved_gdt_blocks =
 470                 cpu_to_le16(le16_to_cpu(es->s_reserved_gdt_blocks) - 1);
 471         ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 472
 473         return 0;
 474
 475 exit_inode:
 476         //ext3_journal_release_buffer(handle, iloc.bh);
 477         brelse(iloc.bh);
 478 exit_dindj:
 479         //ext3_journal_release_buffer(handle, dind);
 480 exit_primary:
 481         //ext3_journal_release_buffer(handle, *primary);
 482 exit_sbh:
 483         //ext3_journal_release_buffer(handle, *primary);
 484 exit_dind:
 485         brelse(dind);
 486 exit_bh:
 487         brelse(*primary);
 488
 489         ext3_debug("leaving with error %d\n", err);
 490         return err;
 491 }
 492
 493 /*
 494  * Called when we are adding a new group which has a backup copy of each of
 495  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 496  * We need to add these reserved backup GDT blocks to the resize inode, so
 497  * that they are kept for future resizing and not allocated to files.
 498  *
 499  * Each reserved backup GDT block will go into a different indirect block.
 500  * The indirect blocks are actually the primary reserved GDT blocks,
 501  * so we know in advance what their block numbers are.  We only get the
 502  * double-indirect block to verify it is pointing to the primary reserved
 503  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 504  * backup GDT blocks are stored in their reserved primary GDT block.
 505  */
 506 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 507                               struct ext3_new_group_data *input)
 508 {
 509         struct super_block *sb = inode->i_sb;
 510         int reserved_gdb =le16_to_cpu(EXT3_SB(sb)->s_es->s_reserved_gdt_blocks);
 511         struct buffer_head **primary;
 512         struct buffer_head *dind;
 513         struct ext3_iloc iloc;
 514         unsigned long blk;
 515         __u32 *data, *end;
 516         int gdbackups = 0;
 517         int res, i;
 518         int err;
 519
 520         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_KERNEL);
 521         if (!primary)
 522                 return -ENOMEM;
 523
 524         data = EXT3_I(inode)->i_data + EXT3_DIND_BLOCK;
 525         dind = sb_bread(sb, le32_to_cpu(*data));
 526         if (!dind) {
 527                 err = -EIO;
 528                 goto exit_free;
 529         }
 530
 531         blk = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + EXT3_SB(sb)->s_gdb_count;
 532         data = (__u32 *)dind->b_data + EXT3_SB(sb)->s_gdb_count;
 533         end = (__u32 *)dind->b_data + EXT3_ADDR_PER_BLOCK(sb);
 534
 535         /* Get each reserved primary GDT block and verify it holds backups */
 536         for (res = 0; res < reserved_gdb; res++, blk++) {
 537                 if (le32_to_cpu(*data) != blk) {
 538                         ext3_warning(sb, __FUNCTION__,
 539                                      "reserved block %lu not at offset %ld\n",
 540                                      blk, (long)(data - (__u32 *)dind->b_data));
 541                         err = -EINVAL;
 542                         goto exit_bh;
 543                 }
 544                 primary[res] = sb_bread(sb, blk);
 545                 if (!primary[res]) {
 546                         err = -EIO;
 547                         goto exit_bh;
 548                 }
 549                 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
 550                         brelse(primary[res]);
 551                         err = gdbackups;
 552                         goto exit_bh;
 553                 }
 554                 if (++data >= end)
 555                         data = (__u32 *)dind->b_data;
 556         }
 557
 558         for (i = 0; i < reserved_gdb; i++) {
 559                 if ((err = ext3_journal_get_write_access(handle, primary[i]))) {
 560                         /*
 561                         int j;
 562                         for (j = 0; j < i; j++)
 563                                 ext3_journal_release_buffer(handle, primary[j]);
 564                          */
 565                         goto exit_bh;
 566                 }
 567         }
 568
 569         if ((err = ext3_reserve_inode_write(handle, inode, &iloc)))
 570                 goto exit_bh;
 571
 572         /*
 573          * Finally we can add each of the reserved backup GDT blocks from
 574          * the new group to its reserved primary GDT block.
 575          */
 576         blk = input->group * EXT3_BLOCKS_PER_GROUP(sb);
 577         for (i = 0; i < reserved_gdb; i++) {
 578                 int err2;
 579                 data = (__u32 *)primary[i]->b_data;
 580                 /* printk("reserving backup %lu[%u] = %lu\n",
 581                        primary[i]->b_blocknr, gdbackups,
 582                        blk + primary[i]->b_blocknr); */
 583                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 584                 err2 = ext3_journal_dirty_metadata(handle, primary[i]);
 585                 if (!err)
 586                         err = err2;
 587         }
 588         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 589         ext3_mark_iloc_dirty(handle, inode, &iloc);
 590
 591 exit_bh:
 592         while (--res >= 0)
 593                 brelse(primary[res]);
 594         brelse(dind);
 595
 596 exit_free:
 597         kfree(primary);
 598
 599         return err;
 600 }
 601
 602 /*
 603  * Update the backup copies of the ext3 metadata.  These don't need to be part
 604  * of the main resize transaction, because e2fsck will re-write them if there
 605  * is a problem (basically only OOM will cause a problem).  However, we
 606  * _should_ update the backups if possible, in case the primary gets trashed
 607  * for some reason and we need to run e2fsck from a backup superblock.  The
 608  * important part is that the new block and inode counts are in the backup
 609  * superblocks, and the location of the new group metadata in the GDT backups.
 610  *
 611  * We do not need lock_super() for this, because these blocks are not
 612  * otherwise touched by the filesystem code when it is mounted.  We don't
 613  * need to worry about last changing from sbi->s_groups_count, because the
 614  * worst that can happen is that we do not copy the full number of backups
 615  * at this time.  The resize which changed s_groups_count will backup again.
 616  *
 617  * We only pass inode because of the ext3 journal wrappers.
 618  */
 619 static void update_backups(struct super_block *sb, struct inode *inode,
 620                            int blk_off, char *data, int size)
 621 {
 622         struct ext3_sb_info *sbi = EXT3_SB(sb);
 623         const unsigned long last = sbi->s_groups_count;
 624         const int bpg = EXT3_BLOCKS_PER_GROUP(sb);
 625         unsigned three = 1;
 626         unsigned five = 5;
 627         unsigned seven = 7;
 628         unsigned group;
 629         int rest = sb->s_blocksize - size;
 630         handle_t *handle;
 631         int err = 0, err2;
 632
 633         handle = ext3_journal_start(inode, EXT3_MAX_TRANS_DATA);
 634         if (IS_ERR(handle)) {
 635                 group = 1;
 636                 err = PTR_ERR(handle);
 637                 goto exit_err;
 638         }
 639
 640         while ((group = ext3_list_backups(sb, &three, &five, &seven)) < last) {
 641                 struct buffer_head *bh;
 642
 643                 /* Out of journal space, and can't get more - abort - so sad */
 644                 if (handle->h_buffer_credits == 0 &&
 645                     ext3_journal_extend(handle, EXT3_MAX_TRANS_DATA) &&
 646                     (err = ext3_journal_restart(handle, EXT3_MAX_TRANS_DATA)))
 647                         break;
 648
 649                 bh = sb_getblk(sb, group * bpg + blk_off);
 650                 set_buffer_uptodate(bh);
 651                 ext3_debug(sb, __FUNCTION__, "update metadata backup %#04lx\n",
 652                            bh->b_blocknr);
 653                 if ((err = ext3_journal_get_write_access(handle, bh)))
 654                         break;
 655                 memcpy(bh->b_data, data, size);
 656                 if (rest)
 657                         memset(bh->b_data + size, 0, rest);
 658                 ext3_journal_dirty_metadata(handle, bh);
 659                 brelse(bh);
 660         }
 661         if ((err2 = ext3_journal_stop(handle)) && !err)
 662                 err = err2;
 663
 664         /*
 665          * Ugh! Need to have e2fsck write the backup copies.  It is too
 666          * late to revert the resize, we shouldn't fail just because of
 667          * the backup copies (they are only needed in case of corruption).
 668          *
 669          * However, if we got here we have a journal problem too, so we
 670          * can't really start a transaction to mark the superblock.
 671          * Chicken out and just set the flag on the hope it will be written
 672          * to disk, and if not - we will simply wait until next fsck.
 673          */
 674 exit_err:
 675         if (err) {
 676                 ext3_warning(sb, __FUNCTION__,
 677                              "can't update backup for group %d (err %d), "
 678                              "forcing fsck on next reboot\n", group, err);
 679                 sbi->s_mount_state &= ~EXT3_VALID_FS;
 680                 sbi->s_es->s_state &= ~cpu_to_le16(EXT3_VALID_FS);
 681                 mark_buffer_dirty(sbi->s_sbh);
 682         }
 683 }
 684
 685 /* Add group descriptor data to an existing or new group descriptor block.
 686  * Ensure we handle all possible error conditions _before_ we start modifying
 687  * the filesystem, because we cannot abort the transaction and not have it
 688  * write the data to disk.
 689  *
 690  * If we are on a GDT block boundary, we need to get the reserved GDT block.
 691  * Otherwise, we may need to add backup GDT blocks for a sparse group.
 692  *
 693  * We only need to hold the superblock lock while we are actually adding
 694  * in the new group's counts to the superblock.  Prior to that we have
 695  * not really "added" the group at all.  We re-check that we are still
 696  * adding in the last group in case things have changed since verifying.
 697  */
 698 int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
 699 {
 700         struct ext3_sb_info *sbi = EXT3_SB(sb);
 701         struct ext3_super_block *es = sbi->s_es;
 702         int reserved_gdb = ext3_bg_has_super(sb, input->group) ?
 703                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
 704         struct buffer_head *primary = NULL;
 705         struct ext3_group_desc *gdp;
 706         struct inode *inode = NULL;
 707         struct inode bogus;
 708         handle_t *handle;
 709         int gdb_off, gdb_num;
 710         int err, err2;
 711
 712         gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb);
 713         gdb_off = input->group % EXT3_DESC_PER_BLOCK(sb);
 714
 715         if (gdb_off == 0 && !EXT3_HAS_RO_COMPAT_FEATURE(sb,
 716                                         EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 717                 ext3_warning(sb, __FUNCTION__,
 718                              "Can't resize non-sparse filesystem further\n");
 719                 return -EPERM;
 720         }
 721
 722         if (reserved_gdb || gdb_off == 0) {
 723                 if (!EXT3_HAS_COMPAT_FEATURE(sb,
 724                                              EXT3_FEATURE_COMPAT_RESIZE_INODE)){
 725                         ext3_warning(sb, __FUNCTION__,
 726                                      "No reserved GDT blocks, can't resize\n");
 727                         return -EPERM;
 728                 }
 729                 inode = iget(sb, EXT3_RESIZE_INO);
 730                 if (!inode || is_bad_inode(inode)) {
 731                         ext3_warning(sb, __FUNCTION__,
 732                                      "Error opening resize inode\n");
 733                         iput(inode);
 734                         return -ENOENT;
 735                 }
 736         } else {
 737                 /* Used only for ext3 journal wrapper functions to get sb */
 738                 inode = &bogus;
 739                 bogus.i_sb = sb;
 740         }
 741
 742         if ((err = verify_group_input(sb, input)))
 743                 goto exit_put;
 744
 745         if ((err = setup_new_group_blocks(sb, inode, input)))
 746                 goto exit_put;
 747
 748         /*
 749          * We will always be modifying at least the superblock and a GDT
 750          * block.  If we are adding a group past the last current GDT block,
 751          * we will also modify the inode and the dindirect block.  If we
 752          * are adding a group with superblock/GDT backups  we will also
 753          * modify each of the reserved GDT dindirect blocks.
 754          */
 755         handle = ext3_journal_start(inode, ext3_bg_has_super(sb, input->group) ?
 756                                     3 + reserved_gdb : 4);
 757         if (IS_ERR(handle)) {
 758                 err = PTR_ERR(handle);
 759                 goto exit_put;
 760         }
 761
 762         lock_super(sb);
 763         if (input->group != EXT3_SB(sb)->s_groups_count) {
 764                 ext3_warning(sb, __FUNCTION__,
 765                              "multiple resizers run on filesystem!\n");
 766                 goto exit_journal;
 767         }
 768
 769         if ((err = ext3_journal_get_write_access(handle, sbi->s_sbh)))
 770                 goto exit_journal;
 771
 772         /*
 773          * We will only either add reserved group blocks to a backup group
 774          * or remove reserved blocks for the first group in a new group block.
 775          * Doing both would be mean more complex code, and sane people don't
 776          * use non-sparse filesystems anymore.  This is already checked above.
 777          */
 778         if (gdb_off) {
 779                 primary = sbi->s_group_desc[gdb_num];
 780                 if ((err = ext3_journal_get_write_access(handle, primary)))
 781                         goto exit_journal;
 782
 783                 if (reserved_gdb && ext3_bg_num_gdb(sb, input->group) &&
 784                     (err = reserve_backup_gdb(handle, inode, input)))
 785                         goto exit_journal;
 786         } else if ((err = add_new_gdb(handle, inode, input, &primary)))
 787                 goto exit_journal;
 788
 789         /* Finally update group descriptor block for new group */
 790         gdp = (struct ext3_group_desc *)primary->b_data + gdb_off;
 791
 792         gdp->bg_block_bitmap = cpu_to_le32(input->block_bitmap);
 793         gdp->bg_inode_bitmap = cpu_to_le32(input->inode_bitmap);
 794         gdp->bg_inode_table = cpu_to_le32(input->inode_table);
 795         gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
 796         gdp->bg_free_inodes_count = cpu_to_le16(EXT3_INODES_PER_GROUP(sb));
 797
 798         EXT3_SB(sb)->s_groups_count++;
 799         ext3_journal_dirty_metadata(handle, primary);
 800
 801         /* Update superblock with new block counts */
 802         es->s_blocks_count = cpu_to_le32(le32_to_cpu(es->s_blocks_count) +
 803                 input->blocks_count);
 804         es->s_free_blocks_count =
 805                 cpu_to_le32(le32_to_cpu(es->s_free_blocks_count) +
 806                             input->free_blocks_count);
 807         es->s_r_blocks_count = cpu_to_le32(le32_to_cpu(es->s_r_blocks_count) +
 808                 input->reserved_blocks);
 809         es->s_inodes_count = cpu_to_le32(le32_to_cpu(es->s_inodes_count) +
 810                 EXT3_INODES_PER_GROUP(sb));
 811         es->s_free_inodes_count =
 812                 cpu_to_le32(le32_to_cpu(es->s_free_inodes_count) +
 813                             EXT3_INODES_PER_GROUP(sb));
 814         ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 815         sb->s_dirt = 1;
 816
 817 exit_journal:
 818         unlock_super(sb);
 819         handle->h_sync = 1;
 820         if ((err2 = ext3_journal_stop(handle)) && !err)
 821                 err = err2;
 822         if (!err) {
 823                 update_backups(sb, inode, sbi->s_sbh->b_blocknr, (char *)es,
 824                                sizeof(struct ext3_super_block));
 825                 update_backups(sb, inode, primary->b_blocknr, primary->b_data,
 826                                primary->b_size);
 827         }
 828 exit_put:
 829         if (inode != &bogus)
 830                 iput(inode);
 831         return err;
 832 } /* ext3_group_add */
 833
 834 /* Extend the filesystem to the new number of blocks specified.  This entry
 835  * point is only used to extend the current filesystem to the end of the last
 836  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
 837  * for emergencies (because it has no dependencies on reserved blocks).
 838  *
 839  * If we _really_ wanted, we could use default values to call ext3_group_add()
 840  * allow the "remount" trick to work for arbitrary resizing, assuming enough
 841  * GDT blocks are reserved to grow to the desired size.
 842  */
 843 int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
 844                       unsigned long n_blocks_count)
 845 {
 846         unsigned long o_blocks_count;
 847         unsigned long o_groups_count;
 848         unsigned long last;
 849         int add;
 850         struct inode *inode;
 851         struct buffer_head * bh;
 852         handle_t *handle;
 853         int err;
 854
 855         o_blocks_count = le32_to_cpu(es->s_blocks_count);
 856         o_groups_count = EXT3_SB(sb)->s_groups_count;
 857
 858         if (test_opt(sb, DEBUG))
 859                 printk("EXT3-fs: extending last group from %lu to %lu blocks\n",
 860                        o_blocks_count, n_blocks_count);
 861
 862         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
 863                 return 0;
 864
 865         if (n_blocks_count < o_blocks_count) {
 866                 ext3_warning(sb, __FUNCTION__,
 867                              "can't shrink FS - resize aborted");
 868                 return -EBUSY;
 869         }
 870
 871         /* Handle the remaining blocks in the last group only. */
 872         last = (o_blocks_count - le32_to_cpu(es->s_first_data_block)) %
 873                 EXT3_BLOCKS_PER_GROUP(sb);
 874
 875         if (last == 0) {
 876                 ext3_warning(sb, __FUNCTION__,
 877                              "need to use ext2online to resize further\n");
 878                 return -EPERM;
 879         }
 880
 881         add = EXT3_BLOCKS_PER_GROUP(sb) - last;
 882
 883         if (o_blocks_count + add > n_blocks_count)
 884                 add = n_blocks_count - o_blocks_count;
 885
 886         if (o_blocks_count + add < n_blocks_count)
 887                 ext3_warning(sb, __FUNCTION__,
 888                              "will only finish group (%lu blocks, %u new)",
 889                              o_blocks_count + add, add);
 890
 891         /* See if the device is actually as big as what was requested */
 892         bh = sb_bread(sb, o_blocks_count + add -1);
 893         if (!bh) {
 894                 ext3_warning(sb, __FUNCTION__,
 895                              "can't read last block, resize aborted");
 896                 return -ENOSPC;
 897         }
 898         brelse(bh);
 899
 900         /* Get a bogus inode to "free" the new blocks in this group. */
 901         if (!(inode = new_inode(sb))) {
 902                 ext3_warning(sb, __FUNCTION__,
 903                              "error getting dummy resize inode");
 904                 return -ENOMEM;
 905         }
 906         inode->i_ino = 0;
 907
 908         EXT3_I(inode)->i_state = EXT3_STATE_RESIZE;
 909
 910         /* We will update the superblock, one block bitmap, and
 911          * one group descriptor via ext3_free_blocks().
 912          */
 913         handle = ext3_journal_start(inode, 3);
 914         if (IS_ERR(handle)) {
 915                 err = PTR_ERR(handle);
 916                 ext3_warning(sb, __FUNCTION__, "error %d on journal start",err);
 917                 goto exit_put;
 918         }
 919
 920         lock_super(sb);
 921         if (o_blocks_count != le32_to_cpu(es->s_blocks_count)) {
 922                 ext3_warning(sb, __FUNCTION__,
 923                              "multiple resizers run on filesystem!\n");
 924                 err = -EBUSY;
 925                 goto exit_put;
 926         }
 927
 928         if ((err = ext3_journal_get_write_access(handle,
 929                                                  EXT3_SB(sb)->s_sbh))) {
 930                 ext3_warning(sb, __FUNCTION__,
 931                              "error %d on journal write access", err);
 932                 unlock_super(sb);
 933                 ext3_journal_stop(handle);
 934                 goto exit_put;
 935         }
 936         es->s_blocks_count = cpu_to_le32(o_blocks_count + add);
 937         ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 938         sb->s_dirt = 1;
 939         unlock_super(sb);
 940         ext3_debug("freeing blocks %ld through %ld\n", o_blocks_count,
 941                    o_blocks_count + add);
 942         ext3_free_blocks(handle, inode, o_blocks_count, add);
 943         ext3_debug("freed blocks %ld through %ld\n", o_blocks_count,
 944                    o_blocks_count + add);
 945         if ((err = ext3_journal_stop(handle)))
 946                 goto exit_put;
 947         if (test_opt(sb, DEBUG))
 948                 printk("EXT3-fs: extended group to %u blocks\n",
 949                        le32_to_cpu(es->s_blocks_count));
 950         update_backups(sb, inode, EXT3_SB(sb)->s_sbh->b_blocknr, (char *)es,
 951                        sizeof(struct ext3_super_block));
 952 exit_put:
 953         iput(inode);
 954
 955         return err;
 956 } /* ext3_group_extend */