fs/ext4/resize.c

   1 /*
   2  *  linux/fs/ext4/resize.c
   3  *
   4  * Support for resizing an ext4 filesystem while it is mounted.
   5  *
   6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   7  *
   8  * This could probably be made into a module, because it is not often in use.
   9  */
  10
  11
  12 #define EXT4FS_DEBUG
  13
  14 #include <linux/errno.h>
  15 #include <linux/slab.h>
  16
  17 #include "ext4_jbd2.h"
  18
  19 int ext4_resize_begin(struct super_block *sb)
  20 {
  21         int ret = 0;
  22
  23         if (!capable(CAP_SYS_RESOURCE))
  24                 return -EPERM;
  25
  26         /*
  27          * We are not allowed to do online-resizing on a filesystem mounted
  28          * with error, because it can destroy the filesystem easily.
  29          */
  30         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  31                 ext4_warning(sb, "There are errors in the filesystem, "
  32                              "so online resizing is not allowed\n");
  33                 return -EPERM;
  34         }
  35
  36         if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
  37                 ret = -EBUSY;
  38
  39         return ret;
  40 }
  41
  42 void ext4_resize_end(struct super_block *sb)
  43 {
  44         clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
  45         smp_mb__after_clear_bit();
  46 }
  47
  48 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  49 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  50
  51 static int verify_group_input(struct super_block *sb,
  52                               struct ext4_new_group_data *input)
  53 {
  54         struct ext4_sb_info *sbi = EXT4_SB(sb);
  55         struct ext4_super_block *es = sbi->s_es;
  56         ext4_fsblk_t start = ext4_blocks_count(es);
  57         ext4_fsblk_t end = start + input->blocks_count;
  58         ext4_group_t group = input->group;
  59         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  60         unsigned overhead = ext4_bg_has_super(sb, group) ?
  61                 (1 + ext4_bg_num_gdb(sb, group) +
  62                  le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
  63         ext4_fsblk_t metaend = start + overhead;
  64         struct buffer_head *bh = NULL;
  65         ext4_grpblk_t free_blocks_count, offset;
  66         int err = -EINVAL;
  67
  68         input->free_blocks_count = free_blocks_count =
  69                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  70
  71         if (test_opt(sb, DEBUG))
  72                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
  73                        "(%d free, %u reserved)\n",
  74                        ext4_bg_has_super(sb, input->group) ? "normal" :
  75                        "no-super", input->group, input->blocks_count,
  76                        free_blocks_count, input->reserved_blocks);
  77
  78         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
  79         if (group != sbi->s_groups_count)
  80                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
  81                              input->group, sbi->s_groups_count);
  82         else if (offset != 0)
  83                         ext4_warning(sb, "Last group not full");
  84         else if (input->reserved_blocks > input->blocks_count / 5)
  85                 ext4_warning(sb, "Reserved blocks too high (%u)",
  86                              input->reserved_blocks);
  87         else if (free_blocks_count < 0)
  88                 ext4_warning(sb, "Bad blocks count %u",
  89                              input->blocks_count);
  90         else if (!(bh = sb_bread(sb, end - 1)))
  91                 ext4_warning(sb, "Cannot read last block (%llu)",
  92                              end - 1);
  93         else if (outside(input->block_bitmap, start, end))
  94                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
  95                              (unsigned long long)input->block_bitmap);
  96         else if (outside(input->inode_bitmap, start, end))
  97                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
  98                              (unsigned long long)input->inode_bitmap);
  99         else if (outside(input->inode_table, start, end) ||
 100                  outside(itend - 1, start, end))
 101                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 102                              (unsigned long long)input->inode_table, itend - 1);
 103         else if (input->inode_bitmap == input->block_bitmap)
 104                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 105                              (unsigned long long)input->block_bitmap);
 106         else if (inside(input->block_bitmap, input->inode_table, itend))
 107                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 108                              "(%llu-%llu)",
 109                              (unsigned long long)input->block_bitmap,
 110                              (unsigned long long)input->inode_table, itend - 1);
 111         else if (inside(input->inode_bitmap, input->inode_table, itend))
 112                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 113                              "(%llu-%llu)",
 114                              (unsigned long long)input->inode_bitmap,
 115                              (unsigned long long)input->inode_table, itend - 1);
 116         else if (inside(input->block_bitmap, start, metaend))
 117                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 118                              (unsigned long long)input->block_bitmap,
 119                              start, metaend - 1);
 120         else if (inside(input->inode_bitmap, start, metaend))
 121                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 122                              (unsigned long long)input->inode_bitmap,
 123                              start, metaend - 1);
 124         else if (inside(input->inode_table, start, metaend) ||
 125                  inside(itend - 1, start, metaend))
 126                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 127                              "(%llu-%llu)",
 128                              (unsigned long long)input->inode_table,
 129                              itend - 1, start, metaend - 1);
 130         else
 131                 err = 0;
 132         brelse(bh);
 133
 134         return err;
 135 }
 136
 137 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 138                                   ext4_fsblk_t blk)
 139 {
 140         struct buffer_head *bh;
 141         int err;
 142
 143         bh = sb_getblk(sb, blk);
 144         if (!bh)
 145                 return ERR_PTR(-EIO);
 146         if ((err = ext4_journal_get_write_access(handle, bh))) {
 147                 brelse(bh);
 148                 bh = ERR_PTR(err);
 149         } else {
 150                 memset(bh->b_data, 0, sb->s_blocksize);
 151                 set_buffer_uptodate(bh);
 152         }
 153
 154         return bh;
 155 }
 156
 157 /*
 158  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 159  * If that fails, restart the transaction & regain write access for the
 160  * buffer head which is used for block_bitmap modifications.
 161  */
 162 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 163 {
 164         int err;
 165
 166         if (ext4_handle_has_enough_credits(handle, thresh))
 167                 return 0;
 168
 169         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 170         if (err < 0)
 171                 return err;
 172         if (err) {
 173                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 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  */
 188 static int setup_new_group_blocks(struct super_block *sb,
 189                                   struct ext4_new_group_data *input)
 190 {
 191         struct ext4_sb_info *sbi = EXT4_SB(sb);
 192         ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
 193         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 194                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 195         unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
 196         struct buffer_head *bh;
 197         handle_t *handle;
 198         ext4_fsblk_t block;
 199         ext4_grpblk_t bit;
 200         int i;
 201         int err = 0, err2;
 202
 203         /* This transaction may be extended/restarted along the way */
 204         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 205
 206         if (IS_ERR(handle))
 207                 return PTR_ERR(handle);
 208
 209         BUG_ON(input->group != sbi->s_groups_count);
 210
 211         /* Copy all of the GDT blocks into the backup in this group */
 212         for (i = 0, bit = 1, block = start + 1;
 213              i < gdblocks; i++, block++, bit++) {
 214                 struct buffer_head *gdb;
 215
 216                 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
 217                 err = extend_or_restart_transaction(handle, 1);
 218                 if (err)
 219                         goto exit_journal;
 220
 221                 gdb = sb_getblk(sb, block);
 222                 if (!gdb) {
 223                         err = -EIO;
 224                         goto exit_journal;
 225                 }
 226                 if ((err = ext4_journal_get_write_access(handle, gdb))) {
 227                         brelse(gdb);
 228                         goto exit_journal;
 229                 }
 230                 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
 231                 set_buffer_uptodate(gdb);
 232                 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 233                 if (unlikely(err)) {
 234                         brelse(gdb);
 235                         goto exit_journal;
 236                 }
 237                 brelse(gdb);
 238         }
 239
 240         /* Zero out all of the reserved backup group descriptor table blocks */
 241         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 242                         block, sbi->s_itb_per_group);
 243         err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
 244                                GFP_NOFS);
 245         if (err)
 246                 goto exit_journal;
 247
 248         err = extend_or_restart_transaction(handle, 2);
 249         if (err)
 250                 goto exit_journal;
 251
 252         bh = bclean(handle, sb, input->block_bitmap);
 253         if (IS_ERR(bh)) {
 254                 err = PTR_ERR(bh);
 255                 goto exit_journal;
 256         }
 257
 258         if (ext4_bg_has_super(sb, input->group)) {
 259                 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
 260                 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
 261         }
 262
 263         ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
 264                    input->block_bitmap - start);
 265         ext4_set_bit(input->block_bitmap - start, bh->b_data);
 266         ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
 267                    input->inode_bitmap - start);
 268         ext4_set_bit(input->inode_bitmap - start, bh->b_data);
 269
 270         /* Zero out all of the inode table blocks */
 271         block = input->inode_table;
 272         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 273                         block, sbi->s_itb_per_group);
 274         err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
 275         if (err)
 276                 goto exit_bh;
 277         ext4_set_bits(bh->b_data, input->inode_table - start,
 278                       sbi->s_itb_per_group);
 279
 280
 281         ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
 282                              bh->b_data);
 283         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 284         if (unlikely(err)) {
 285                 ext4_std_error(sb, err);
 286                 goto exit_bh;
 287         }
 288         brelse(bh);
 289         /* Mark unused entries in inode bitmap used */
 290         ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
 291                    input->inode_bitmap, input->inode_bitmap - start);
 292         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 293                 err = PTR_ERR(bh);
 294                 goto exit_journal;
 295         }
 296
 297         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
 298                              bh->b_data);
 299         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 300         if (unlikely(err))
 301                 ext4_std_error(sb, err);
 302 exit_bh:
 303         brelse(bh);
 304
 305 exit_journal:
 306         if ((err2 = ext4_journal_stop(handle)) && !err)
 307                 err = err2;
 308
 309         return err;
 310 }
 311
 312 /*
 313  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 314  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 315  * calling this for the first time.  In a sparse filesystem it will be the
 316  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 317  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 318  */
 319 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 320                                   unsigned *five, unsigned *seven)
 321 {
 322         unsigned *min = three;
 323         int mult = 3;
 324         unsigned ret;
 325
 326         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 327                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 328                 ret = *min;
 329                 *min += 1;
 330                 return ret;
 331         }
 332
 333         if (*five < *min) {
 334                 min = five;
 335                 mult = 5;
 336         }
 337         if (*seven < *min) {
 338                 min = seven;
 339                 mult = 7;
 340         }
 341
 342         ret = *min;
 343         *min *= mult;
 344
 345         return ret;
 346 }
 347
 348 /*
 349  * Check that all of the backup GDT blocks are held in the primary GDT block.
 350  * It is assumed that they are stored in group order.  Returns the number of
 351  * groups in current filesystem that have BACKUPS, or -ve error code.
 352  */
 353 static int verify_reserved_gdb(struct super_block *sb,
 354                                struct buffer_head *primary)
 355 {
 356         const ext4_fsblk_t blk = primary->b_blocknr;
 357         const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
 358         unsigned three = 1;
 359         unsigned five = 5;
 360         unsigned seven = 7;
 361         unsigned grp;
 362         __le32 *p = (__le32 *)primary->b_data;
 363         int gdbackups = 0;
 364
 365         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 366                 if (le32_to_cpu(*p++) !=
 367                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 368                         ext4_warning(sb, "reserved GDT %llu"
 369                                      " missing grp %d (%llu)",
 370                                      blk, grp,
 371                                      grp *
 372                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 373                                      blk);
 374                         return -EINVAL;
 375                 }
 376                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 377                         return -EFBIG;
 378         }
 379
 380         return gdbackups;
 381 }
 382
 383 /*
 384  * Called when we need to bring a reserved group descriptor table block into
 385  * use from the resize inode.  The primary copy of the new GDT block currently
 386  * is an indirect block (under the double indirect block in the resize inode).
 387  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 388  * block, in group order.  Even though we know all the block numbers we need,
 389  * we check to ensure that the resize inode has actually reserved these blocks.
 390  *
 391  * Don't need to update the block bitmaps because the blocks are still in use.
 392  *
 393  * We get all of the error cases out of the way, so that we are sure to not
 394  * fail once we start modifying the data on disk, because JBD has no rollback.
 395  */
 396 static int add_new_gdb(handle_t *handle, struct inode *inode,
 397                        ext4_group_t group)
 398 {
 399         struct super_block *sb = inode->i_sb;
 400         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 401         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 402         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 403         struct buffer_head **o_group_desc, **n_group_desc;
 404         struct buffer_head *dind;
 405         struct buffer_head *gdb_bh;
 406         int gdbackups;
 407         struct ext4_iloc iloc;
 408         __le32 *data;
 409         int err;
 410
 411         if (test_opt(sb, DEBUG))
 412                 printk(KERN_DEBUG
 413                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 414                        gdb_num);
 415
 416         /*
 417          * If we are not using the primary superblock/GDT copy don't resize,
 418          * because the user tools have no way of handling this.  Probably a
 419          * bad time to do it anyways.
 420          */
 421         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 422             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 423                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 424                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 425                 return -EPERM;
 426         }
 427
 428         gdb_bh = sb_bread(sb, gdblock);
 429         if (!gdb_bh)
 430                 return -EIO;
 431
 432         gdbackups = verify_reserved_gdb(sb, gdb_bh);
 433         if (gdbackups < 0) {
 434                 err = gdbackups;
 435                 goto exit_bh;
 436         }
 437
 438         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 439         dind = sb_bread(sb, le32_to_cpu(*data));
 440         if (!dind) {
 441                 err = -EIO;
 442                 goto exit_bh;
 443         }
 444
 445         data = (__le32 *)dind->b_data;
 446         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 447                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 448                              group, gdblock);
 449                 err = -EINVAL;
 450                 goto exit_dind;
 451         }
 452
 453         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 454         if (unlikely(err))
 455                 goto exit_dind;
 456
 457         err = ext4_journal_get_write_access(handle, gdb_bh);
 458         if (unlikely(err))
 459                 goto exit_sbh;
 460
 461         err = ext4_journal_get_write_access(handle, dind);
 462         if (unlikely(err))
 463                 ext4_std_error(sb, err);
 464
 465         /* ext4_reserve_inode_write() gets a reference on the iloc */
 466         err = ext4_reserve_inode_write(handle, inode, &iloc);
 467         if (unlikely(err))
 468                 goto exit_dindj;
 469
 470         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 471                                      sizeof(struct buffer_head *),
 472                                      GFP_NOFS);
 473         if (!n_group_desc) {
 474                 err = -ENOMEM;
 475                 ext4_warning(sb, "not enough memory for %lu groups",
 476                              gdb_num + 1);
 477                 goto exit_inode;
 478         }
 479
 480         /*
 481          * Finally, we have all of the possible failures behind us...
 482          *
 483          * Remove new GDT block from inode double-indirect block and clear out
 484          * the new GDT block for use (which also "frees" the backup GDT blocks
 485          * from the reserved inode).  We don't need to change the bitmaps for
 486          * these blocks, because they are marked as in-use from being in the
 487          * reserved inode, and will become GDT blocks (primary and backup).
 488          */
 489         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 490         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 491         if (unlikely(err)) {
 492                 ext4_std_error(sb, err);
 493                 goto exit_inode;
 494         }
 495         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 496         ext4_mark_iloc_dirty(handle, inode, &iloc);
 497         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 498         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 499         if (unlikely(err)) {
 500                 ext4_std_error(sb, err);
 501                 goto exit_inode;
 502         }
 503         brelse(dind);
 504
 505         o_group_desc = EXT4_SB(sb)->s_group_desc;
 506         memcpy(n_group_desc, o_group_desc,
 507                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 508         n_group_desc[gdb_num] = gdb_bh;
 509         EXT4_SB(sb)->s_group_desc = n_group_desc;
 510         EXT4_SB(sb)->s_gdb_count++;
 511         ext4_kvfree(o_group_desc);
 512
 513         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 514         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 515         if (err)
 516                 ext4_std_error(sb, err);
 517
 518         return err;
 519
 520 exit_inode:
 521         ext4_kvfree(n_group_desc);
 522         /* ext4_handle_release_buffer(handle, iloc.bh); */
 523         brelse(iloc.bh);
 524 exit_dindj:
 525         /* ext4_handle_release_buffer(handle, dind); */
 526 exit_sbh:
 527         /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
 528 exit_dind:
 529         brelse(dind);
 530 exit_bh:
 531         brelse(gdb_bh);
 532
 533         ext4_debug("leaving with error %d\n", err);
 534         return err;
 535 }
 536
 537 /*
 538  * Called when we are adding a new group which has a backup copy of each of
 539  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 540  * We need to add these reserved backup GDT blocks to the resize inode, so
 541  * that they are kept for future resizing and not allocated to files.
 542  *
 543  * Each reserved backup GDT block will go into a different indirect block.
 544  * The indirect blocks are actually the primary reserved GDT blocks,
 545  * so we know in advance what their block numbers are.  We only get the
 546  * double-indirect block to verify it is pointing to the primary reserved
 547  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 548  * backup GDT blocks are stored in their reserved primary GDT block.
 549  */
 550 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 551                               ext4_group_t group)
 552 {
 553         struct super_block *sb = inode->i_sb;
 554         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 555         struct buffer_head **primary;
 556         struct buffer_head *dind;
 557         struct ext4_iloc iloc;
 558         ext4_fsblk_t blk;
 559         __le32 *data, *end;
 560         int gdbackups = 0;
 561         int res, i;
 562         int err;
 563
 564         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 565         if (!primary)
 566                 return -ENOMEM;
 567
 568         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 569         dind = sb_bread(sb, le32_to_cpu(*data));
 570         if (!dind) {
 571                 err = -EIO;
 572                 goto exit_free;
 573         }
 574
 575         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 576         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 577                                          EXT4_ADDR_PER_BLOCK(sb));
 578         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 579
 580         /* Get each reserved primary GDT block and verify it holds backups */
 581         for (res = 0; res < reserved_gdb; res++, blk++) {
 582                 if (le32_to_cpu(*data) != blk) {
 583                         ext4_warning(sb, "reserved block %llu"
 584                                      " not at offset %ld",
 585                                      blk,
 586                                      (long)(data - (__le32 *)dind->b_data));
 587                         err = -EINVAL;
 588                         goto exit_bh;
 589                 }
 590                 primary[res] = sb_bread(sb, blk);
 591                 if (!primary[res]) {
 592                         err = -EIO;
 593                         goto exit_bh;
 594                 }
 595                 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
 596                         brelse(primary[res]);
 597                         err = gdbackups;
 598                         goto exit_bh;
 599                 }
 600                 if (++data >= end)
 601                         data = (__le32 *)dind->b_data;
 602         }
 603
 604         for (i = 0; i < reserved_gdb; i++) {
 605                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
 606                         /*
 607                         int j;
 608                         for (j = 0; j < i; j++)
 609                                 ext4_handle_release_buffer(handle, primary[j]);
 610                          */
 611                         goto exit_bh;
 612                 }
 613         }
 614
 615         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 616                 goto exit_bh;
 617
 618         /*
 619          * Finally we can add each of the reserved backup GDT blocks from
 620          * the new group to its reserved primary GDT block.
 621          */
 622         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
 623         for (i = 0; i < reserved_gdb; i++) {
 624                 int err2;
 625                 data = (__le32 *)primary[i]->b_data;
 626                 /* printk("reserving backup %lu[%u] = %lu\n",
 627                        primary[i]->b_blocknr, gdbackups,
 628                        blk + primary[i]->b_blocknr); */
 629                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 630                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 631                 if (!err)
 632                         err = err2;
 633         }
 634         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 635         ext4_mark_iloc_dirty(handle, inode, &iloc);
 636
 637 exit_bh:
 638         while (--res >= 0)
 639                 brelse(primary[res]);
 640         brelse(dind);
 641
 642 exit_free:
 643         kfree(primary);
 644
 645         return err;
 646 }
 647
 648 /*
 649  * Update the backup copies of the ext4 metadata.  These don't need to be part
 650  * of the main resize transaction, because e2fsck will re-write them if there
 651  * is a problem (basically only OOM will cause a problem).  However, we
 652  * _should_ update the backups if possible, in case the primary gets trashed
 653  * for some reason and we need to run e2fsck from a backup superblock.  The
 654  * important part is that the new block and inode counts are in the backup
 655  * superblocks, and the location of the new group metadata in the GDT backups.
 656  *
 657  * We do not need take the s_resize_lock for this, because these
 658  * blocks are not otherwise touched by the filesystem code when it is
 659  * mounted.  We don't need to worry about last changing from
 660  * sbi->s_groups_count, because the worst that can happen is that we
 661  * do not copy the full number of backups at this time.  The resize
 662  * which changed s_groups_count will backup again.
 663  */
 664 static void update_backups(struct super_block *sb,
 665                            int blk_off, char *data, int size)
 666 {
 667         struct ext4_sb_info *sbi = EXT4_SB(sb);
 668         const ext4_group_t last = sbi->s_groups_count;
 669         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
 670         unsigned three = 1;
 671         unsigned five = 5;
 672         unsigned seven = 7;
 673         ext4_group_t group;
 674         int rest = sb->s_blocksize - size;
 675         handle_t *handle;
 676         int err = 0, err2;
 677
 678         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 679         if (IS_ERR(handle)) {
 680                 group = 1;
 681                 err = PTR_ERR(handle);
 682                 goto exit_err;
 683         }
 684
 685         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
 686                 struct buffer_head *bh;
 687
 688                 /* Out of journal space, and can't get more - abort - so sad */
 689                 if (ext4_handle_valid(handle) &&
 690                     handle->h_buffer_credits == 0 &&
 691                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
 692                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 693                         break;
 694
 695                 bh = sb_getblk(sb, group * bpg + blk_off);
 696                 if (!bh) {
 697                         err = -EIO;
 698                         break;
 699                 }
 700                 ext4_debug("update metadata backup %#04lx\n",
 701                           (unsigned long)bh->b_blocknr);
 702                 if ((err = ext4_journal_get_write_access(handle, bh)))
 703                         break;
 704                 lock_buffer(bh);
 705                 memcpy(bh->b_data, data, size);
 706                 if (rest)
 707                         memset(bh->b_data + size, 0, rest);
 708                 set_buffer_uptodate(bh);
 709                 unlock_buffer(bh);
 710                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 711                 if (unlikely(err))
 712                         ext4_std_error(sb, err);
 713                 brelse(bh);
 714         }
 715         if ((err2 = ext4_journal_stop(handle)) && !err)
 716                 err = err2;
 717
 718         /*
 719          * Ugh! Need to have e2fsck write the backup copies.  It is too
 720          * late to revert the resize, we shouldn't fail just because of
 721          * the backup copies (they are only needed in case of corruption).
 722          *
 723          * However, if we got here we have a journal problem too, so we
 724          * can't really start a transaction to mark the superblock.
 725          * Chicken out and just set the flag on the hope it will be written
 726          * to disk, and if not - we will simply wait until next fsck.
 727          */
 728 exit_err:
 729         if (err) {
 730                 ext4_warning(sb, "can't update backup for group %u (err %d), "
 731                              "forcing fsck on next reboot", group, err);
 732                 sbi->s_mount_state &= ~EXT4_VALID_FS;
 733                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
 734                 mark_buffer_dirty(sbi->s_sbh);
 735         }
 736 }
 737
 738 /* Add group descriptor data to an existing or new group descriptor block.
 739  * Ensure we handle all possible error conditions _before_ we start modifying
 740  * the filesystem, because we cannot abort the transaction and not have it
 741  * write the data to disk.
 742  *
 743  * If we are on a GDT block boundary, we need to get the reserved GDT block.
 744  * Otherwise, we may need to add backup GDT blocks for a sparse group.
 745  *
 746  * We only need to hold the superblock lock while we are actually adding
 747  * in the new group's counts to the superblock.  Prior to that we have
 748  * not really "added" the group at all.  We re-check that we are still
 749  * adding in the last group in case things have changed since verifying.
 750  */
 751 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
 752 {
 753         struct ext4_sb_info *sbi = EXT4_SB(sb);
 754         struct ext4_super_block *es = sbi->s_es;
 755         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 756                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
 757         struct buffer_head *primary = NULL;
 758         struct ext4_group_desc *gdp;
 759         struct inode *inode = NULL;
 760         handle_t *handle;
 761         int gdb_off, gdb_num;
 762         int err, err2;
 763
 764         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 765         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
 766
 767         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
 768                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 769                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
 770                 return -EPERM;
 771         }
 772
 773         if (ext4_blocks_count(es) + input->blocks_count <
 774             ext4_blocks_count(es)) {
 775                 ext4_warning(sb, "blocks_count overflow");
 776                 return -EINVAL;
 777         }
 778
 779         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
 780             le32_to_cpu(es->s_inodes_count)) {
 781                 ext4_warning(sb, "inodes_count overflow");
 782                 return -EINVAL;
 783         }
 784
 785         if (reserved_gdb || gdb_off == 0) {
 786                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
 787                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
 788                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
 789                         ext4_warning(sb,
 790                                      "No reserved GDT blocks, can't resize");
 791                         return -EPERM;
 792                 }
 793                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
 794                 if (IS_ERR(inode)) {
 795                         ext4_warning(sb, "Error opening resize inode");
 796                         return PTR_ERR(inode);
 797                 }
 798         }
 799
 800
 801         if ((err = verify_group_input(sb, input)))
 802                 goto exit_put;
 803
 804         if ((err = setup_new_group_blocks(sb, input)))
 805                 goto exit_put;
 806
 807         /*
 808          * We will always be modifying at least the superblock and a GDT
 809          * block.  If we are adding a group past the last current GDT block,
 810          * we will also modify the inode and the dindirect block.  If we
 811          * are adding a group with superblock/GDT backups  we will also
 812          * modify each of the reserved GDT dindirect blocks.
 813          */
 814         handle = ext4_journal_start_sb(sb,
 815                                        ext4_bg_has_super(sb, input->group) ?
 816                                        3 + reserved_gdb : 4);
 817         if (IS_ERR(handle)) {
 818                 err = PTR_ERR(handle);
 819                 goto exit_put;
 820         }
 821
 822         if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
 823                 goto exit_journal;
 824
 825         /*
 826          * We will only either add reserved group blocks to a backup group
 827          * or remove reserved blocks for the first group in a new group block.
 828          * Doing both would be mean more complex code, and sane people don't
 829          * use non-sparse filesystems anymore.  This is already checked above.
 830          */
 831         if (gdb_off) {
 832                 primary = sbi->s_group_desc[gdb_num];
 833                 if ((err = ext4_journal_get_write_access(handle, primary)))
 834                         goto exit_journal;
 835
 836                 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
 837                         err = reserve_backup_gdb(handle, inode, input->group);
 838                         if (err)
 839                                 goto exit_journal;
 840                 }
 841         } else {
 842                 /*
 843                  * Note that we can access new group descriptor block safely
 844                  * only if add_new_gdb() succeeds.
 845                  */
 846                 err = add_new_gdb(handle, inode, input->group);
 847                 if (err)
 848                         goto exit_journal;
 849                 primary = sbi->s_group_desc[gdb_num];
 850         }
 851
 852         /*
 853          * OK, now we've set up the new group.  Time to make it active.
 854          *
 855          * so we have to be safe wrt. concurrent accesses the group
 856          * data.  So we need to be careful to set all of the relevant
 857          * group descriptor data etc. *before* we enable the group.
 858          *
 859          * The key field here is sbi->s_groups_count: as long as
 860          * that retains its old value, nobody is going to access the new
 861          * group.
 862          *
 863          * So first we update all the descriptor metadata for the new
 864          * group; then we update the total disk blocks count; then we
 865          * update the groups count to enable the group; then finally we
 866          * update the free space counts so that the system can start
 867          * using the new disk blocks.
 868          */
 869
 870         /* Update group descriptor block for new group */
 871         gdp = (struct ext4_group_desc *)((char *)primary->b_data +
 872                                          gdb_off * EXT4_DESC_SIZE(sb));
 873
 874         memset(gdp, 0, EXT4_DESC_SIZE(sb));
 875         ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
 876         ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
 877         ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
 878         ext4_free_blks_set(sb, gdp, input->free_blocks_count);
 879         ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
 880         gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
 881         gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
 882
 883         /*
 884          * We can allocate memory for mb_alloc based on the new group
 885          * descriptor
 886          */
 887         err = ext4_mb_add_groupinfo(sb, input->group, gdp);
 888         if (err)
 889                 goto exit_journal;
 890
 891         /*
 892          * Make the new blocks and inodes valid next.  We do this before
 893          * increasing the group count so that once the group is enabled,
 894          * all of its blocks and inodes are already valid.
 895          *
 896          * We always allocate group-by-group, then block-by-block or
 897          * inode-by-inode within a group, so enabling these
 898          * blocks/inodes before the group is live won't actually let us
 899          * allocate the new space yet.
 900          */
 901         ext4_blocks_count_set(es, ext4_blocks_count(es) +
 902                 input->blocks_count);
 903         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
 904
 905         /*
 906          * We need to protect s_groups_count against other CPUs seeing
 907          * inconsistent state in the superblock.
 908          *
 909          * The precise rules we use are:
 910          *
 911          * * Writers must perform a smp_wmb() after updating all dependent
 912          *   data and before modifying the groups count
 913          *
 914          * * Readers must perform an smp_rmb() after reading the groups count
 915          *   and before reading any dependent data.
 916          *
 917          * NB. These rules can be relaxed when checking the group count
 918          * while freeing data, as we can only allocate from a block
 919          * group after serialising against the group count, and we can
 920          * only then free after serialising in turn against that
 921          * allocation.
 922          */
 923         smp_wmb();
 924
 925         /* Update the global fs size fields */
 926         sbi->s_groups_count++;
 927
 928         err = ext4_handle_dirty_metadata(handle, NULL, primary);
 929         if (unlikely(err)) {
 930                 ext4_std_error(sb, err);
 931                 goto exit_journal;
 932         }
 933
 934         /* Update the reserved block counts only once the new group is
 935          * active. */
 936         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
 937                 input->reserved_blocks);
 938
 939         /* Update the free space counts */
 940         percpu_counter_add(&sbi->s_freeblocks_counter,
 941                            input->free_blocks_count);
 942         percpu_counter_add(&sbi->s_freeinodes_counter,
 943                            EXT4_INODES_PER_GROUP(sb));
 944
 945         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
 946             sbi->s_log_groups_per_flex) {
 947                 ext4_group_t flex_group;
 948                 flex_group = ext4_flex_group(sbi, input->group);
 949                 atomic_add(input->free_blocks_count,
 950                            &sbi->s_flex_groups[flex_group].free_blocks);
 951                 atomic_add(EXT4_INODES_PER_GROUP(sb),
 952                            &sbi->s_flex_groups[flex_group].free_inodes);
 953         }
 954
 955         ext4_handle_dirty_super(handle, sb);
 956
 957 exit_journal:
 958         if ((err2 = ext4_journal_stop(handle)) && !err)
 959                 err = err2;
 960         if (!err && primary) {
 961                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
 962                                sizeof(struct ext4_super_block));
 963                 update_backups(sb, primary->b_blocknr, primary->b_data,
 964                                primary->b_size);
 965         }
 966 exit_put:
 967         iput(inode);
 968         return err;
 969 } /* ext4_group_add */
 970
 971 /*
 972  * Extend the filesystem to the new number of blocks specified.  This entry
 973  * point is only used to extend the current filesystem to the end of the last
 974  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
 975  * for emergencies (because it has no dependencies on reserved blocks).
 976  *
 977  * If we _really_ wanted, we could use default values to call ext4_group_add()
 978  * allow the "remount" trick to work for arbitrary resizing, assuming enough
 979  * GDT blocks are reserved to grow to the desired size.
 980  */
 981 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
 982                       ext4_fsblk_t n_blocks_count)
 983 {
 984         ext4_fsblk_t o_blocks_count;
 985         ext4_grpblk_t last;
 986         ext4_grpblk_t add;
 987         struct buffer_head *bh;
 988         handle_t *handle;
 989         int err, err2;
 990         ext4_group_t group;
 991
 992         o_blocks_count = ext4_blocks_count(es);
 993
 994         if (test_opt(sb, DEBUG))
 995                 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
 996                        o_blocks_count, n_blocks_count);
 997
 998         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
 999                 return 0;
1000
1001         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1002                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1003                         " too large to resize to %llu blocks safely\n",
1004                         sb->s_id, n_blocks_count);
1005                 if (sizeof(sector_t) < 8)
1006                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1007                 return -EINVAL;
1008         }
1009
1010         if (n_blocks_count < o_blocks_count) {
1011                 ext4_warning(sb, "can't shrink FS - resize aborted");
1012                 return -EINVAL;
1013         }
1014
1015         /* Handle the remaining blocks in the last group only. */
1016         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1017
1018         if (last == 0) {
1019                 ext4_warning(sb, "need to use ext2online to resize further");
1020                 return -EPERM;
1021         }
1022
1023         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1024
1025         if (o_blocks_count + add < o_blocks_count) {
1026                 ext4_warning(sb, "blocks_count overflow");
1027                 return -EINVAL;
1028         }
1029
1030         if (o_blocks_count + add > n_blocks_count)
1031                 add = n_blocks_count - o_blocks_count;
1032
1033         if (o_blocks_count + add < n_blocks_count)
1034                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1035                              o_blocks_count + add, add);
1036
1037         /* See if the device is actually as big as what was requested */
1038         bh = sb_bread(sb, o_blocks_count + add - 1);
1039         if (!bh) {
1040                 ext4_warning(sb, "can't read last block, resize aborted");
1041                 return -ENOSPC;
1042         }
1043         brelse(bh);
1044
1045         /* We will update the superblock, one block bitmap, and
1046          * one group descriptor via ext4_free_blocks().
1047          */
1048         handle = ext4_journal_start_sb(sb, 3);
1049         if (IS_ERR(handle)) {
1050                 err = PTR_ERR(handle);
1051                 ext4_warning(sb, "error %d on journal start", err);
1052                 goto exit_put;
1053         }
1054
1055         if ((err = ext4_journal_get_write_access(handle,
1056                                                  EXT4_SB(sb)->s_sbh))) {
1057                 ext4_warning(sb, "error %d on journal write access", err);
1058                 ext4_journal_stop(handle);
1059                 goto exit_put;
1060         }
1061         ext4_blocks_count_set(es, o_blocks_count + add);
1062         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1063                    o_blocks_count + add);
1064         /* We add the blocks to the bitmap and set the group need init bit */
1065         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1066         ext4_handle_dirty_super(handle, sb);
1067         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1068                    o_blocks_count + add);
1069         err2 = ext4_journal_stop(handle);
1070         if (!err && err2)
1071                 err = err2;
1072
1073         if (err)
1074                 goto exit_put;
1075
1076         if (test_opt(sb, DEBUG))
1077                 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1078                        ext4_blocks_count(es));
1079         update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1080                        sizeof(struct ext4_super_block));
1081 exit_put:
1082         return err;
1083 } /* ext4_group_extend */