linux/fs/ext3/resize.c
<<
>>
Prefs
   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
  12#define EXT3FS_DEBUG
  13
  14#include "ext3.h"
  15
  16
  17#define outside(b, first, last) ((b) < (first) || (b) >= (last))
  18#define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  19
  20static int verify_group_input(struct super_block *sb,
  21                              struct ext3_new_group_data *input)
  22{
  23        struct ext3_sb_info *sbi = EXT3_SB(sb);
  24        struct ext3_super_block *es = sbi->s_es;
  25        ext3_fsblk_t start = le32_to_cpu(es->s_blocks_count);
  26        ext3_fsblk_t end = start + input->blocks_count;
  27        unsigned group = input->group;
  28        ext3_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  29        unsigned overhead = ext3_bg_has_super(sb, group) ?
  30                (1 + ext3_bg_num_gdb(sb, group) +
  31                 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
  32        ext3_fsblk_t metaend = start + overhead;
  33        struct buffer_head *bh = NULL;
  34        ext3_grpblk_t free_blocks_count;
  35        int err = -EINVAL;
  36
  37        input->free_blocks_count = free_blocks_count =
  38                input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  39
  40        if (test_opt(sb, DEBUG))
  41                printk(KERN_DEBUG "EXT3-fs: adding %s group %u: %u blocks "
  42                       "(%d free, %u reserved)\n",
  43                       ext3_bg_has_super(sb, input->group) ? "normal" :
  44                       "no-super", input->group, input->blocks_count,
  45                       free_blocks_count, input->reserved_blocks);
  46
  47        if (group != sbi->s_groups_count)
  48                ext3_warning(sb, __func__,
  49                             "Cannot add at group %u (only %lu groups)",
  50                             input->group, sbi->s_groups_count);
  51        else if ((start - le32_to_cpu(es->s_first_data_block)) %
  52                 EXT3_BLOCKS_PER_GROUP(sb))
  53                ext3_warning(sb, __func__, "Last group not full");
  54        else if (input->reserved_blocks > input->blocks_count / 5)
  55                ext3_warning(sb, __func__, "Reserved blocks too high (%u)",
  56                             input->reserved_blocks);
  57        else if (free_blocks_count < 0)
  58                ext3_warning(sb, __func__, "Bad blocks count %u",
  59                             input->blocks_count);
  60        else if (!(bh = sb_bread(sb, end - 1)))
  61                ext3_warning(sb, __func__,
  62                             "Cannot read last block ("E3FSBLK")",
  63                             end - 1);
  64        else if (outside(input->block_bitmap, start, end))
  65                ext3_warning(sb, __func__,
  66                             "Block bitmap not in group (block %u)",
  67                             input->block_bitmap);
  68        else if (outside(input->inode_bitmap, start, end))
  69                ext3_warning(sb, __func__,
  70                             "Inode bitmap not in group (block %u)",
  71                             input->inode_bitmap);
  72        else if (outside(input->inode_table, start, end) ||
  73                 outside(itend - 1, start, end))
  74                ext3_warning(sb, __func__,
  75                             "Inode table not in group (blocks %u-"E3FSBLK")",
  76                             input->inode_table, itend - 1);
  77        else if (input->inode_bitmap == input->block_bitmap)
  78                ext3_warning(sb, __func__,
  79                             "Block bitmap same as inode bitmap (%u)",
  80                             input->block_bitmap);
  81        else if (inside(input->block_bitmap, input->inode_table, itend))
  82                ext3_warning(sb, __func__,
  83                             "Block bitmap (%u) in inode table (%u-"E3FSBLK")",
  84                             input->block_bitmap, input->inode_table, itend-1);
  85        else if (inside(input->inode_bitmap, input->inode_table, itend))
  86                ext3_warning(sb, __func__,
  87                             "Inode bitmap (%u) in inode table (%u-"E3FSBLK")",
  88                             input->inode_bitmap, input->inode_table, itend-1);
  89        else if (inside(input->block_bitmap, start, metaend))
  90                ext3_warning(sb, __func__,
  91                             "Block bitmap (%u) in GDT table"
  92                             " ("E3FSBLK"-"E3FSBLK")",
  93                             input->block_bitmap, start, metaend - 1);
  94        else if (inside(input->inode_bitmap, start, metaend))
  95                ext3_warning(sb, __func__,
  96                             "Inode bitmap (%u) in GDT table"
  97                             " ("E3FSBLK"-"E3FSBLK")",
  98                             input->inode_bitmap, start, metaend - 1);
  99        else if (inside(input->inode_table, start, metaend) ||
 100                 inside(itend - 1, start, metaend))
 101                ext3_warning(sb, __func__,
 102                             "Inode table (%u-"E3FSBLK") overlaps"
 103                             "GDT table ("E3FSBLK"-"E3FSBLK")",
 104                             input->inode_table, itend - 1, start, metaend - 1);
 105        else
 106                err = 0;
 107        brelse(bh);
 108
 109        return err;
 110}
 111
 112static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 113                                  ext3_fsblk_t blk)
 114{
 115        struct buffer_head *bh;
 116        int err;
 117
 118        bh = sb_getblk(sb, blk);
 119        if (!bh)
 120                return ERR_PTR(-EIO);
 121        if ((err = ext3_journal_get_write_access(handle, bh))) {
 122                brelse(bh);
 123                bh = ERR_PTR(err);
 124        } else {
 125                lock_buffer(bh);
 126                memset(bh->b_data, 0, sb->s_blocksize);
 127                set_buffer_uptodate(bh);
 128                unlock_buffer(bh);
 129        }
 130
 131        return bh;
 132}
 133
 134/*
 135 * To avoid calling the atomic setbit hundreds or thousands of times, we only
 136 * need to use it within a single byte (to ensure we get endianness right).
 137 * We can use memset for the rest of the bitmap as there are no other users.
 138 */
 139static void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
 140{
 141        int i;
 142
 143        if (start_bit >= end_bit)
 144                return;
 145
 146        ext3_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
 147        for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
 148                ext3_set_bit(i, bitmap);
 149        if (i < end_bit)
 150                memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
 151}
 152
 153/*
 154 * If we have fewer than thresh credits, extend by EXT3_MAX_TRANS_DATA.
 155 * If that fails, restart the transaction & regain write access for the
 156 * buffer head which is used for block_bitmap modifications.
 157 */
 158static int extend_or_restart_transaction(handle_t *handle, int thresh,
 159                                         struct buffer_head *bh)
 160{
 161        int err;
 162
 163        if (handle->h_buffer_credits >= thresh)
 164                return 0;
 165
 166        err = ext3_journal_extend(handle, EXT3_MAX_TRANS_DATA);
 167        if (err < 0)
 168                return err;
 169        if (err) {
 170                err = ext3_journal_restart(handle, EXT3_MAX_TRANS_DATA);
 171                if (err)
 172                        return err;
 173                err = ext3_journal_get_write_access(handle, bh);
 174                if (err)
 175                        return err;
 176        }
 177
 178        return 0;
 179}
 180
 181/*
 182 * Set up the block and inode bitmaps, and the inode table for the new group.
 183 * This doesn't need to be part of the main transaction, since we are only
 184 * changing blocks outside the actual filesystem.  We still do journaling to
 185 * ensure the recovery is correct in case of a failure just after resize.
 186 * If any part of this fails, we simply abort the resize.
 187 */
 188static int setup_new_group_blocks(struct super_block *sb,
 189                                  struct ext3_new_group_data *input)
 190{
 191        struct ext3_sb_info *sbi = EXT3_SB(sb);
 192        ext3_fsblk_t start = ext3_group_first_block_no(sb, input->group);
 193        int reserved_gdb = ext3_bg_has_super(sb, input->group) ?
 194                le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 195        unsigned long gdblocks = ext3_bg_num_gdb(sb, input->group);
 196        struct buffer_head *bh;
 197        handle_t *handle;
 198        ext3_fsblk_t block;
 199        ext3_grpblk_t bit;
 200        int i;
 201        int err = 0, err2;
 202
 203        /* This transaction may be extended/restarted along the way */
 204        handle = ext3_journal_start_sb(sb, EXT3_MAX_TRANS_DATA);
 205
 206        if (IS_ERR(handle))
 207                return PTR_ERR(handle);
 208
 209        mutex_lock(&sbi->s_resize_lock);
 210        if (input->group != sbi->s_groups_count) {
 211                err = -EBUSY;
 212                goto exit_journal;
 213        }
 214
 215        if (IS_ERR(bh = bclean(handle, sb, input->block_bitmap))) {
 216                err = PTR_ERR(bh);
 217                goto exit_journal;
 218        }
 219
 220        if (ext3_bg_has_super(sb, input->group)) {
 221                ext3_debug("mark backup superblock %#04lx (+0)\n", start);
 222                ext3_set_bit(0, bh->b_data);
 223        }
 224
 225        /* Copy all of the GDT blocks into the backup in this group */
 226        for (i = 0, bit = 1, block = start + 1;
 227             i < gdblocks; i++, block++, bit++) {
 228                struct buffer_head *gdb;
 229
 230                ext3_debug("update backup group %#04lx (+%d)\n", block, bit);
 231
 232                err = extend_or_restart_transaction(handle, 1, bh);
 233                if (err)
 234                        goto exit_bh;
 235
 236                gdb = sb_getblk(sb, block);
 237                if (!gdb) {
 238                        err = -EIO;
 239                        goto exit_bh;
 240                }
 241                if ((err = ext3_journal_get_write_access(handle, gdb))) {
 242                        brelse(gdb);
 243                        goto exit_bh;
 244                }
 245                lock_buffer(gdb);
 246                memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
 247                set_buffer_uptodate(gdb);
 248                unlock_buffer(gdb);
 249                err = ext3_journal_dirty_metadata(handle, gdb);
 250                if (err) {
 251                        brelse(gdb);
 252                        goto exit_bh;
 253                }
 254                ext3_set_bit(bit, bh->b_data);
 255                brelse(gdb);
 256        }
 257
 258        /* Zero out all of the reserved backup group descriptor table blocks */
 259        for (i = 0, bit = gdblocks + 1, block = start + bit;
 260             i < reserved_gdb; i++, block++, bit++) {
 261                struct buffer_head *gdb;
 262
 263                ext3_debug("clear reserved block %#04lx (+%d)\n", block, bit);
 264
 265                err = extend_or_restart_transaction(handle, 1, bh);
 266                if (err)
 267                        goto exit_bh;
 268
 269                if (IS_ERR(gdb = bclean(handle, sb, block))) {
 270                        err = PTR_ERR(gdb);
 271                        goto exit_bh;
 272                }
 273                err = ext3_journal_dirty_metadata(handle, gdb);
 274                if (err) {
 275                        brelse(gdb);
 276                        goto exit_bh;
 277                }
 278                ext3_set_bit(bit, bh->b_data);
 279                brelse(gdb);
 280        }
 281        ext3_debug("mark block bitmap %#04x (+%ld)\n", input->block_bitmap,
 282                   input->block_bitmap - start);
 283        ext3_set_bit(input->block_bitmap - start, bh->b_data);
 284        ext3_debug("mark inode bitmap %#04x (+%ld)\n", input->inode_bitmap,
 285                   input->inode_bitmap - start);
 286        ext3_set_bit(input->inode_bitmap - start, bh->b_data);
 287
 288        /* Zero out all of the inode table blocks */
 289        for (i = 0, block = input->inode_table, bit = block - start;
 290             i < sbi->s_itb_per_group; i++, bit++, block++) {
 291                struct buffer_head *it;
 292
 293                ext3_debug("clear inode block %#04lx (+%d)\n", block, bit);
 294
 295                err = extend_or_restart_transaction(handle, 1, bh);
 296                if (err)
 297                        goto exit_bh;
 298
 299                if (IS_ERR(it = bclean(handle, sb, block))) {
 300                        err = PTR_ERR(it);
 301                        goto exit_bh;
 302                }
 303                err = ext3_journal_dirty_metadata(handle, it);
 304                if (err) {
 305                        brelse(it);
 306                        goto exit_bh;
 307                }
 308                brelse(it);
 309                ext3_set_bit(bit, bh->b_data);
 310        }
 311
 312        err = extend_or_restart_transaction(handle, 2, bh);
 313        if (err)
 314                goto exit_bh;
 315
 316        mark_bitmap_end(input->blocks_count, EXT3_BLOCKS_PER_GROUP(sb),
 317                        bh->b_data);
 318        err = ext3_journal_dirty_metadata(handle, bh);
 319        if (err)
 320                goto exit_bh;
 321        brelse(bh);
 322
 323        /* Mark unused entries in inode bitmap used */
 324        ext3_debug("clear inode bitmap %#04x (+%ld)\n",
 325                   input->inode_bitmap, input->inode_bitmap - start);
 326        if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 327                err = PTR_ERR(bh);
 328                goto exit_journal;
 329        }
 330
 331        mark_bitmap_end(EXT3_INODES_PER_GROUP(sb), EXT3_BLOCKS_PER_GROUP(sb),
 332                        bh->b_data);
 333        err = ext3_journal_dirty_metadata(handle, bh);
 334exit_bh:
 335        brelse(bh);
 336
 337exit_journal:
 338        mutex_unlock(&sbi->s_resize_lock);
 339        if ((err2 = ext3_journal_stop(handle)) && !err)
 340                err = err2;
 341
 342        return err;
 343}
 344
 345/*
 346 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 347 * ext3 filesystem.  The counters should be initialized to 1, 5, and 7 before
 348 * calling this for the first time.  In a sparse filesystem it will be the
 349 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 350 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 351 */
 352static unsigned ext3_list_backups(struct super_block *sb, unsigned *three,
 353                                  unsigned *five, unsigned *seven)
 354{
 355        unsigned *min = three;
 356        int mult = 3;
 357        unsigned ret;
 358
 359        if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
 360                                        EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 361                ret = *min;
 362                *min += 1;
 363                return ret;
 364        }
 365
 366        if (*five < *min) {
 367                min = five;
 368                mult = 5;
 369        }
 370        if (*seven < *min) {
 371                min = seven;
 372                mult = 7;
 373        }
 374
 375        ret = *min;
 376        *min *= mult;
 377
 378        return ret;
 379}
 380
 381/*
 382 * Check that all of the backup GDT blocks are held in the primary GDT block.
 383 * It is assumed that they are stored in group order.  Returns the number of
 384 * groups in current filesystem that have BACKUPS, or -ve error code.
 385 */
 386static int verify_reserved_gdb(struct super_block *sb,
 387                               struct buffer_head *primary)
 388{
 389        const ext3_fsblk_t blk = primary->b_blocknr;
 390        const unsigned long end = EXT3_SB(sb)->s_groups_count;
 391        unsigned three = 1;
 392        unsigned five = 5;
 393        unsigned seven = 7;
 394        unsigned grp;
 395        __le32 *p = (__le32 *)primary->b_data;
 396        int gdbackups = 0;
 397
 398        while ((grp = ext3_list_backups(sb, &three, &five, &seven)) < end) {
 399                if (le32_to_cpu(*p++) != grp * EXT3_BLOCKS_PER_GROUP(sb) + blk){
 400                        ext3_warning(sb, __func__,
 401                                     "reserved GDT "E3FSBLK
 402                                     " missing grp %d ("E3FSBLK")",
 403                                     blk, grp,
 404                                     grp * EXT3_BLOCKS_PER_GROUP(sb) + blk);
 405                        return -EINVAL;
 406                }
 407                if (++gdbackups > EXT3_ADDR_PER_BLOCK(sb))
 408                        return -EFBIG;
 409        }
 410
 411        return gdbackups;
 412}
 413
 414/*
 415 * Called when we need to bring a reserved group descriptor table block into
 416 * use from the resize inode.  The primary copy of the new GDT block currently
 417 * is an indirect block (under the double indirect block in the resize inode).
 418 * The new backup GDT blocks will be stored as leaf blocks in this indirect
 419 * block, in group order.  Even though we know all the block numbers we need,
 420 * we check to ensure that the resize inode has actually reserved these blocks.
 421 *
 422 * Don't need to update the block bitmaps because the blocks are still in use.
 423 *
 424 * We get all of the error cases out of the way, so that we are sure to not
 425 * fail once we start modifying the data on disk, because JBD has no rollback.
 426 */
 427static int add_new_gdb(handle_t *handle, struct inode *inode,
 428                       struct ext3_new_group_data *input,
 429                       struct buffer_head **primary)
 430{
 431        struct super_block *sb = inode->i_sb;
 432        struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 433        unsigned long gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb);
 434        ext3_fsblk_t gdblock = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 435        struct buffer_head **o_group_desc, **n_group_desc;
 436        struct buffer_head *dind;
 437        int gdbackups;
 438        struct ext3_iloc iloc;
 439        __le32 *data;
 440        int err;
 441
 442        if (test_opt(sb, DEBUG))
 443                printk(KERN_DEBUG
 444                       "EXT3-fs: ext3_add_new_gdb: adding group block %lu\n",
 445                       gdb_num);
 446
 447        /*
 448         * If we are not using the primary superblock/GDT copy don't resize,
 449         * because the user tools have no way of handling this.  Probably a
 450         * bad time to do it anyways.
 451         */
 452        if (EXT3_SB(sb)->s_sbh->b_blocknr !=
 453            le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) {
 454                ext3_warning(sb, __func__,
 455                        "won't resize using backup superblock at %llu",
 456                        (unsigned long long)EXT3_SB(sb)->s_sbh->b_blocknr);
 457                return -EPERM;
 458        }
 459
 460        *primary = sb_bread(sb, gdblock);
 461        if (!*primary)
 462                return -EIO;
 463
 464        if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
 465                err = gdbackups;
 466                goto exit_bh;
 467        }
 468
 469        data = EXT3_I(inode)->i_data + EXT3_DIND_BLOCK;
 470        dind = sb_bread(sb, le32_to_cpu(*data));
 471        if (!dind) {
 472                err = -EIO;
 473                goto exit_bh;
 474        }
 475
 476        data = (__le32 *)dind->b_data;
 477        if (le32_to_cpu(data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)]) != gdblock) {
 478                ext3_warning(sb, __func__,
 479                             "new group %u GDT block "E3FSBLK" not reserved",
 480                             input->group, gdblock);
 481                err = -EINVAL;
 482                goto exit_dind;
 483        }
 484
 485        if ((err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh)))
 486                goto exit_dind;
 487
 488        if ((err = ext3_journal_get_write_access(handle, *primary)))
 489                goto exit_sbh;
 490
 491        if ((err = ext3_journal_get_write_access(handle, dind)))
 492                goto exit_primary;
 493
 494        /* ext3_reserve_inode_write() gets a reference on the iloc */
 495        if ((err = ext3_reserve_inode_write(handle, inode, &iloc)))
 496                goto exit_dindj;
 497
 498        n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 499                        GFP_NOFS);
 500        if (!n_group_desc) {
 501                err = -ENOMEM;
 502                ext3_warning (sb, __func__,
 503                              "not enough memory for %lu groups", gdb_num + 1);
 504                goto exit_inode;
 505        }
 506
 507        /*
 508         * Finally, we have all of the possible failures behind us...
 509         *
 510         * Remove new GDT block from inode double-indirect block and clear out
 511         * the new GDT block for use (which also "frees" the backup GDT blocks
 512         * from the reserved inode).  We don't need to change the bitmaps for
 513         * these blocks, because they are marked as in-use from being in the
 514         * reserved inode, and will become GDT blocks (primary and backup).
 515         */
 516        data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)] = 0;
 517        err = ext3_journal_dirty_metadata(handle, dind);
 518        if (err)
 519                goto exit_group_desc;
 520        brelse(dind);
 521        dind = NULL;
 522        inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 523        err = ext3_mark_iloc_dirty(handle, inode, &iloc);
 524        if (err)
 525                goto exit_group_desc;
 526        memset((*primary)->b_data, 0, sb->s_blocksize);
 527        err = ext3_journal_dirty_metadata(handle, *primary);
 528        if (err)
 529                goto exit_group_desc;
 530
 531        o_group_desc = EXT3_SB(sb)->s_group_desc;
 532        memcpy(n_group_desc, o_group_desc,
 533               EXT3_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 534        n_group_desc[gdb_num] = *primary;
 535        EXT3_SB(sb)->s_group_desc = n_group_desc;
 536        EXT3_SB(sb)->s_gdb_count++;
 537        kfree(o_group_desc);
 538
 539        le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 540        err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
 541        if (err)
 542                goto exit_inode;
 543
 544        return 0;
 545
 546exit_group_desc:
 547        kfree(n_group_desc);
 548exit_inode:
 549        //ext3_journal_release_buffer(handle, iloc.bh);
 550        brelse(iloc.bh);
 551exit_dindj:
 552        //ext3_journal_release_buffer(handle, dind);
 553exit_primary:
 554        //ext3_journal_release_buffer(handle, *primary);
 555exit_sbh:
 556        //ext3_journal_release_buffer(handle, *primary);
 557exit_dind:
 558        brelse(dind);
 559exit_bh:
 560        brelse(*primary);
 561
 562        ext3_debug("leaving with error %d\n", err);
 563        return err;
 564}
 565
 566/*
 567 * Called when we are adding a new group which has a backup copy of each of
 568 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 569 * We need to add these reserved backup GDT blocks to the resize inode, so
 570 * that they are kept for future resizing and not allocated to files.
 571 *
 572 * Each reserved backup GDT block will go into a different indirect block.
 573 * The indirect blocks are actually the primary reserved GDT blocks,
 574 * so we know in advance what their block numbers are.  We only get the
 575 * double-indirect block to verify it is pointing to the primary reserved
 576 * GDT blocks so we don't overwrite a data block by accident.  The reserved
 577 * backup GDT blocks are stored in their reserved primary GDT block.
 578 */
 579static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 580                              struct ext3_new_group_data *input)
 581{
 582        struct super_block *sb = inode->i_sb;
 583        int reserved_gdb =le16_to_cpu(EXT3_SB(sb)->s_es->s_reserved_gdt_blocks);
 584        struct buffer_head **primary;
 585        struct buffer_head *dind;
 586        struct ext3_iloc iloc;
 587        ext3_fsblk_t blk;
 588        __le32 *data, *end;
 589        int gdbackups = 0;
 590        int res, i;
 591        int err;
 592
 593        primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 594        if (!primary)
 595                return -ENOMEM;
 596
 597        data = EXT3_I(inode)->i_data + EXT3_DIND_BLOCK;
 598        dind = sb_bread(sb, le32_to_cpu(*data));
 599        if (!dind) {
 600                err = -EIO;
 601                goto exit_free;
 602        }
 603
 604        blk = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + EXT3_SB(sb)->s_gdb_count;
 605        data = (__le32 *)dind->b_data + (EXT3_SB(sb)->s_gdb_count %
 606                                         EXT3_ADDR_PER_BLOCK(sb));
 607        end = (__le32 *)dind->b_data + EXT3_ADDR_PER_BLOCK(sb);
 608
 609        /* Get each reserved primary GDT block and verify it holds backups */
 610        for (res = 0; res < reserved_gdb; res++, blk++) {
 611                if (le32_to_cpu(*data) != blk) {
 612                        ext3_warning(sb, __func__,
 613                                     "reserved block "E3FSBLK
 614                                     " not at offset %ld",
 615                                     blk,
 616                                     (long)(data - (__le32 *)dind->b_data));
 617                        err = -EINVAL;
 618                        goto exit_bh;
 619                }
 620                primary[res] = sb_bread(sb, blk);
 621                if (!primary[res]) {
 622                        err = -EIO;
 623                        goto exit_bh;
 624                }
 625                if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
 626                        brelse(primary[res]);
 627                        err = gdbackups;
 628                        goto exit_bh;
 629                }
 630                if (++data >= end)
 631                        data = (__le32 *)dind->b_data;
 632        }
 633
 634        for (i = 0; i < reserved_gdb; i++) {
 635                if ((err = ext3_journal_get_write_access(handle, primary[i]))) {
 636                        /*
 637                        int j;
 638                        for (j = 0; j < i; j++)
 639                                ext3_journal_release_buffer(handle, primary[j]);
 640                         */
 641                        goto exit_bh;
 642                }
 643        }
 644
 645        if ((err = ext3_reserve_inode_write(handle, inode, &iloc)))
 646                goto exit_bh;
 647
 648        /*
 649         * Finally we can add each of the reserved backup GDT blocks from
 650         * the new group to its reserved primary GDT block.
 651         */
 652        blk = input->group * EXT3_BLOCKS_PER_GROUP(sb);
 653        for (i = 0; i < reserved_gdb; i++) {
 654                int err2;
 655                data = (__le32 *)primary[i]->b_data;
 656                /* printk("reserving backup %lu[%u] = %lu\n",
 657                       primary[i]->b_blocknr, gdbackups,
 658                       blk + primary[i]->b_blocknr); */
 659                data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 660                err2 = ext3_journal_dirty_metadata(handle, primary[i]);
 661                if (!err)
 662                        err = err2;
 663        }
 664        inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 665        ext3_mark_iloc_dirty(handle, inode, &iloc);
 666
 667exit_bh:
 668        while (--res >= 0)
 669                brelse(primary[res]);
 670        brelse(dind);
 671
 672exit_free:
 673        kfree(primary);
 674
 675        return err;
 676}
 677
 678/*
 679 * Update the backup copies of the ext3 metadata.  These don't need to be part
 680 * of the main resize transaction, because e2fsck will re-write them if there
 681 * is a problem (basically only OOM will cause a problem).  However, we
 682 * _should_ update the backups if possible, in case the primary gets trashed
 683 * for some reason and we need to run e2fsck from a backup superblock.  The
 684 * important part is that the new block and inode counts are in the backup
 685 * superblocks, and the location of the new group metadata in the GDT backups.
 686 *
 687 * We do not need take the s_resize_lock for this, because these
 688 * blocks are not otherwise touched by the filesystem code when it is
 689 * mounted.  We don't need to worry about last changing from
 690 * sbi->s_groups_count, because the worst that can happen is that we
 691 * do not copy the full number of backups at this time.  The resize
 692 * which changed s_groups_count will backup again.
 693 */
 694static void update_backups(struct super_block *sb,
 695                           int blk_off, char *data, int size)
 696{
 697        struct ext3_sb_info *sbi = EXT3_SB(sb);
 698        const unsigned long last = sbi->s_groups_count;
 699        const int bpg = EXT3_BLOCKS_PER_GROUP(sb);
 700        unsigned three = 1;
 701        unsigned five = 5;
 702        unsigned seven = 7;
 703        unsigned group;
 704        int rest = sb->s_blocksize - size;
 705        handle_t *handle;
 706        int err = 0, err2;
 707
 708        handle = ext3_journal_start_sb(sb, EXT3_MAX_TRANS_DATA);
 709        if (IS_ERR(handle)) {
 710                group = 1;
 711                err = PTR_ERR(handle);
 712                goto exit_err;
 713        }
 714
 715        while ((group = ext3_list_backups(sb, &three, &five, &seven)) < last) {
 716                struct buffer_head *bh;
 717
 718                /* Out of journal space, and can't get more - abort - so sad */
 719                if (handle->h_buffer_credits == 0 &&
 720                    ext3_journal_extend(handle, EXT3_MAX_TRANS_DATA) &&
 721                    (err = ext3_journal_restart(handle, EXT3_MAX_TRANS_DATA)))
 722                        break;
 723
 724                bh = sb_getblk(sb, group * bpg + blk_off);
 725                if (!bh) {
 726                        err = -EIO;
 727                        break;
 728                }
 729                ext3_debug("update metadata backup %#04lx\n",
 730                          (unsigned long)bh->b_blocknr);
 731                if ((err = ext3_journal_get_write_access(handle, bh))) {
 732                        brelse(bh);
 733                        break;
 734                }
 735                lock_buffer(bh);
 736                memcpy(bh->b_data, data, size);
 737                if (rest)
 738                        memset(bh->b_data + size, 0, rest);
 739                set_buffer_uptodate(bh);
 740                unlock_buffer(bh);
 741                err = ext3_journal_dirty_metadata(handle, bh);
 742                brelse(bh);
 743                if (err)
 744                        break;
 745        }
 746        if ((err2 = ext3_journal_stop(handle)) && !err)
 747                err = err2;
 748
 749        /*
 750         * Ugh! Need to have e2fsck write the backup copies.  It is too
 751         * late to revert the resize, we shouldn't fail just because of
 752         * the backup copies (they are only needed in case of corruption).
 753         *
 754         * However, if we got here we have a journal problem too, so we
 755         * can't really start a transaction to mark the superblock.
 756         * Chicken out and just set the flag on the hope it will be written
 757         * to disk, and if not - we will simply wait until next fsck.
 758         */
 759exit_err:
 760        if (err) {
 761                ext3_warning(sb, __func__,
 762                             "can't update backup for group %d (err %d), "
 763                             "forcing fsck on next reboot", group, err);
 764                sbi->s_mount_state &= ~EXT3_VALID_FS;
 765                sbi->s_es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
 766                mark_buffer_dirty(sbi->s_sbh);
 767        }
 768}
 769
 770/* Add group descriptor data to an existing or new group descriptor block.
 771 * Ensure we handle all possible error conditions _before_ we start modifying
 772 * the filesystem, because we cannot abort the transaction and not have it
 773 * write the data to disk.
 774 *
 775 * If we are on a GDT block boundary, we need to get the reserved GDT block.
 776 * Otherwise, we may need to add backup GDT blocks for a sparse group.
 777 *
 778 * We only need to hold the superblock lock while we are actually adding
 779 * in the new group's counts to the superblock.  Prior to that we have
 780 * not really "added" the group at all.  We re-check that we are still
 781 * adding in the last group in case things have changed since verifying.
 782 */
 783int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
 784{
 785        struct ext3_sb_info *sbi = EXT3_SB(sb);
 786        struct ext3_super_block *es = sbi->s_es;
 787        int reserved_gdb = ext3_bg_has_super(sb, input->group) ?
 788                le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
 789        struct buffer_head *primary = NULL;
 790        struct ext3_group_desc *gdp;
 791        struct inode *inode = NULL;
 792        handle_t *handle;
 793        int gdb_off, gdb_num;
 794        int err, err2;
 795
 796        gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb);
 797        gdb_off = input->group % EXT3_DESC_PER_BLOCK(sb);
 798
 799        if (gdb_off == 0 && !EXT3_HAS_RO_COMPAT_FEATURE(sb,
 800                                        EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 801                ext3_warning(sb, __func__,
 802                             "Can't resize non-sparse filesystem further");
 803                return -EPERM;
 804        }
 805
 806        if (le32_to_cpu(es->s_blocks_count) + input->blocks_count <
 807            le32_to_cpu(es->s_blocks_count)) {
 808                ext3_warning(sb, __func__, "blocks_count overflow\n");
 809                return -EINVAL;
 810        }
 811
 812        if (le32_to_cpu(es->s_inodes_count) + EXT3_INODES_PER_GROUP(sb) <
 813            le32_to_cpu(es->s_inodes_count)) {
 814                ext3_warning(sb, __func__, "inodes_count overflow\n");
 815                return -EINVAL;
 816        }
 817
 818        if (reserved_gdb || gdb_off == 0) {
 819                if (!EXT3_HAS_COMPAT_FEATURE(sb,
 820                                             EXT3_FEATURE_COMPAT_RESIZE_INODE)
 821                    || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
 822                        ext3_warning(sb, __func__,
 823                                     "No reserved GDT blocks, can't resize");
 824                        return -EPERM;
 825                }
 826                inode = ext3_iget(sb, EXT3_RESIZE_INO);
 827                if (IS_ERR(inode)) {
 828                        ext3_warning(sb, __func__,
 829                                     "Error opening resize inode");
 830                        return PTR_ERR(inode);
 831                }
 832        }
 833
 834        if ((err = verify_group_input(sb, input)))
 835                goto exit_put;
 836
 837        if ((err = setup_new_group_blocks(sb, input)))
 838                goto exit_put;
 839
 840        /*
 841         * We will always be modifying at least the superblock and a GDT
 842         * block.  If we are adding a group past the last current GDT block,
 843         * we will also modify the inode and the dindirect block.  If we
 844         * are adding a group with superblock/GDT backups  we will also
 845         * modify each of the reserved GDT dindirect blocks.
 846         */
 847        handle = ext3_journal_start_sb(sb,
 848                                       ext3_bg_has_super(sb, input->group) ?
 849                                       3 + reserved_gdb : 4);
 850        if (IS_ERR(handle)) {
 851                err = PTR_ERR(handle);
 852                goto exit_put;
 853        }
 854
 855        mutex_lock(&sbi->s_resize_lock);
 856        if (input->group != sbi->s_groups_count) {
 857                ext3_warning(sb, __func__,
 858                             "multiple resizers run on filesystem!");
 859                err = -EBUSY;
 860                goto exit_journal;
 861        }
 862
 863        if ((err = ext3_journal_get_write_access(handle, sbi->s_sbh)))
 864                goto exit_journal;
 865
 866        /*
 867         * We will only either add reserved group blocks to a backup group
 868         * or remove reserved blocks for the first group in a new group block.
 869         * Doing both would be mean more complex code, and sane people don't
 870         * use non-sparse filesystems anymore.  This is already checked above.
 871         */
 872        if (gdb_off) {
 873                primary = sbi->s_group_desc[gdb_num];
 874                if ((err = ext3_journal_get_write_access(handle, primary)))
 875                        goto exit_journal;
 876
 877                if (reserved_gdb && ext3_bg_num_gdb(sb, input->group) &&
 878                    (err = reserve_backup_gdb(handle, inode, input)))
 879                        goto exit_journal;
 880        } else if ((err = add_new_gdb(handle, inode, input, &primary)))
 881                goto exit_journal;
 882
 883        /*
 884         * OK, now we've set up the new group.  Time to make it active.
 885         *
 886         * We do not lock all allocations via s_resize_lock
 887         * so we have to be safe wrt. concurrent accesses the group
 888         * data.  So we need to be careful to set all of the relevant
 889         * group descriptor data etc. *before* we enable the group.
 890         *
 891         * The key field here is sbi->s_groups_count: as long as
 892         * that retains its old value, nobody is going to access the new
 893         * group.
 894         *
 895         * So first we update all the descriptor metadata for the new
 896         * group; then we update the total disk blocks count; then we
 897         * update the groups count to enable the group; then finally we
 898         * update the free space counts so that the system can start
 899         * using the new disk blocks.
 900         */
 901
 902        /* Update group descriptor block for new group */
 903        gdp = (struct ext3_group_desc *)primary->b_data + gdb_off;
 904
 905        gdp->bg_block_bitmap = cpu_to_le32(input->block_bitmap);
 906        gdp->bg_inode_bitmap = cpu_to_le32(input->inode_bitmap);
 907        gdp->bg_inode_table = cpu_to_le32(input->inode_table);
 908        gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
 909        gdp->bg_free_inodes_count = cpu_to_le16(EXT3_INODES_PER_GROUP(sb));
 910
 911        /*
 912         * Make the new blocks and inodes valid next.  We do this before
 913         * increasing the group count so that once the group is enabled,
 914         * all of its blocks and inodes are already valid.
 915         *
 916         * We always allocate group-by-group, then block-by-block or
 917         * inode-by-inode within a group, so enabling these
 918         * blocks/inodes before the group is live won't actually let us
 919         * allocate the new space yet.
 920         */
 921        le32_add_cpu(&es->s_blocks_count, input->blocks_count);
 922        le32_add_cpu(&es->s_inodes_count, EXT3_INODES_PER_GROUP(sb));
 923
 924        /*
 925         * We need to protect s_groups_count against other CPUs seeing
 926         * inconsistent state in the superblock.
 927         *
 928         * The precise rules we use are:
 929         *
 930         * * Writers of s_groups_count *must* hold s_resize_lock
 931         * AND
 932         * * Writers must perform a smp_wmb() after updating all dependent
 933         *   data and before modifying the groups count
 934         *
 935         * * Readers must hold s_resize_lock over the access
 936         * OR
 937         * * Readers must perform an smp_rmb() after reading the groups count
 938         *   and before reading any dependent data.
 939         *
 940         * NB. These rules can be relaxed when checking the group count
 941         * while freeing data, as we can only allocate from a block
 942         * group after serialising against the group count, and we can
 943         * only then free after serialising in turn against that
 944         * allocation.
 945         */
 946        smp_wmb();
 947
 948        /* Update the global fs size fields */
 949        sbi->s_groups_count++;
 950
 951        err = ext3_journal_dirty_metadata(handle, primary);
 952        if (err)
 953                goto exit_journal;
 954
 955        /* Update the reserved block counts only once the new group is
 956         * active. */
 957        le32_add_cpu(&es->s_r_blocks_count, input->reserved_blocks);
 958
 959        /* Update the free space counts */
 960        percpu_counter_add(&sbi->s_freeblocks_counter,
 961                           input->free_blocks_count);
 962        percpu_counter_add(&sbi->s_freeinodes_counter,
 963                           EXT3_INODES_PER_GROUP(sb));
 964
 965        err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
 966
 967exit_journal:
 968        mutex_unlock(&sbi->s_resize_lock);
 969        if ((err2 = ext3_journal_stop(handle)) && !err)
 970                err = err2;
 971        if (!err) {
 972                update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
 973                               sizeof(struct ext3_super_block));
 974                update_backups(sb, primary->b_blocknr, primary->b_data,
 975                               primary->b_size);
 976        }
 977exit_put:
 978        iput(inode);
 979        return err;
 980} /* ext3_group_add */
 981
 982/* Extend the filesystem to the new number of blocks specified.  This entry
 983 * point is only used to extend the current filesystem to the end of the last
 984 * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
 985 * for emergencies (because it has no dependencies on reserved blocks).
 986 *
 987 * If we _really_ wanted, we could use default values to call ext3_group_add()
 988 * allow the "remount" trick to work for arbitrary resizing, assuming enough
 989 * GDT blocks are reserved to grow to the desired size.
 990 */
 991int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
 992                      ext3_fsblk_t n_blocks_count)
 993{
 994        ext3_fsblk_t o_blocks_count;
 995        ext3_grpblk_t last;
 996        ext3_grpblk_t add;
 997        struct buffer_head * bh;
 998        handle_t *handle;
 999        int err;
1000        unsigned long freed_blocks;
1001
1002        /* We don't need to worry about locking wrt other resizers just
1003         * yet: we're going to revalidate es->s_blocks_count after
1004         * taking the s_resize_lock below. */
1005        o_blocks_count = le32_to_cpu(es->s_blocks_count);
1006
1007        if (test_opt(sb, DEBUG))
1008                printk(KERN_DEBUG "EXT3-fs: extending last group from "E3FSBLK
1009                       " up to "E3FSBLK" blocks\n",
1010                       o_blocks_count, n_blocks_count);
1011
1012        if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1013                return 0;
1014
1015        if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1016                printk(KERN_ERR "EXT3-fs: filesystem on %s:"
1017                        " too large to resize to "E3FSBLK" blocks safely\n",
1018                        sb->s_id, n_blocks_count);
1019                if (sizeof(sector_t) < 8)
1020                        ext3_warning(sb, __func__,
1021                        "CONFIG_LBDAF not enabled\n");
1022                return -EINVAL;
1023        }
1024
1025        if (n_blocks_count < o_blocks_count) {
1026                ext3_warning(sb, __func__,
1027                             "can't shrink FS - resize aborted");
1028                return -EBUSY;
1029        }
1030
1031        /* Handle the remaining blocks in the last group only. */
1032        last = (o_blocks_count - le32_to_cpu(es->s_first_data_block)) %
1033                EXT3_BLOCKS_PER_GROUP(sb);
1034
1035        if (last == 0) {
1036                ext3_warning(sb, __func__,
1037                             "need to use ext2online to resize further");
1038                return -EPERM;
1039        }
1040
1041        add = EXT3_BLOCKS_PER_GROUP(sb) - last;
1042
1043        if (o_blocks_count + add < o_blocks_count) {
1044                ext3_warning(sb, __func__, "blocks_count overflow");
1045                return -EINVAL;
1046        }
1047
1048        if (o_blocks_count + add > n_blocks_count)
1049                add = n_blocks_count - o_blocks_count;
1050
1051        if (o_blocks_count + add < n_blocks_count)
1052                ext3_warning(sb, __func__,
1053                             "will only finish group ("E3FSBLK
1054                             " blocks, %u new)",
1055                             o_blocks_count + add, add);
1056
1057        /* See if the device is actually as big as what was requested */
1058        bh = sb_bread(sb, o_blocks_count + add -1);
1059        if (!bh) {
1060                ext3_warning(sb, __func__,
1061                             "can't read last block, resize aborted");
1062                return -ENOSPC;
1063        }
1064        brelse(bh);
1065
1066        /* We will update the superblock, one block bitmap, and
1067         * one group descriptor via ext3_free_blocks().
1068         */
1069        handle = ext3_journal_start_sb(sb, 3);
1070        if (IS_ERR(handle)) {
1071                err = PTR_ERR(handle);
1072                ext3_warning(sb, __func__, "error %d on journal start",err);
1073                goto exit_put;
1074        }
1075
1076        mutex_lock(&EXT3_SB(sb)->s_resize_lock);
1077        if (o_blocks_count != le32_to_cpu(es->s_blocks_count)) {
1078                ext3_warning(sb, __func__,
1079                             "multiple resizers run on filesystem!");
1080                mutex_unlock(&EXT3_SB(sb)->s_resize_lock);
1081                ext3_journal_stop(handle);
1082                err = -EBUSY;
1083                goto exit_put;
1084        }
1085
1086        if ((err = ext3_journal_get_write_access(handle,
1087                                                 EXT3_SB(sb)->s_sbh))) {
1088                ext3_warning(sb, __func__,
1089                             "error %d on journal write access", err);
1090                mutex_unlock(&EXT3_SB(sb)->s_resize_lock);
1091                ext3_journal_stop(handle);
1092                goto exit_put;
1093        }
1094        es->s_blocks_count = cpu_to_le32(o_blocks_count + add);
1095        err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1096        mutex_unlock(&EXT3_SB(sb)->s_resize_lock);
1097        if (err) {
1098                ext3_warning(sb, __func__,
1099                             "error %d on journal dirty metadata", err);
1100                ext3_journal_stop(handle);
1101                goto exit_put;
1102        }
1103        ext3_debug("freeing blocks "E3FSBLK" through "E3FSBLK"\n",
1104                   o_blocks_count, o_blocks_count + add);
1105        ext3_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
1106        ext3_debug("freed blocks "E3FSBLK" through "E3FSBLK"\n",
1107                   o_blocks_count, o_blocks_count + add);
1108        if ((err = ext3_journal_stop(handle)))
1109                goto exit_put;
1110        if (test_opt(sb, DEBUG))
1111                printk(KERN_DEBUG "EXT3-fs: extended group to %u blocks\n",
1112                       le32_to_cpu(es->s_blocks_count));
1113        update_backups(sb, EXT3_SB(sb)->s_sbh->b_blocknr, (char *)es,
1114                       sizeof(struct ext3_super_block));
1115exit_put:
1116        return err;
1117} /* ext3_group_extend */
1118
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.