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                 lock_buffer(bh);
 151                 memset(bh->b_data, 0, sb->s_blocksize);
 152                 set_buffer_uptodate(bh);
 153                 unlock_buffer(bh);
 154         }
 155
 156         return bh;
 157 }
 158
 159 /*
 160  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 161  * If that fails, restart the transaction & regain write access for the
 162  * buffer head which is used for block_bitmap modifications.
 163  */
 164 static int extend_or_restart_transaction(handle_t *handle, int thresh,
 165                                          struct buffer_head *bh)
 166 {
 167         int err;
 168
 169         if (ext4_handle_has_enough_credits(handle, thresh))
 170                 return 0;
 171
 172         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 173         if (err < 0)
 174                 return err;
 175         if (err) {
 176                 if ((err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 177                         return err;
 178                 if ((err = ext4_journal_get_write_access(handle, bh)))
 179                         return err;
 180         }
 181
 182         return 0;
 183 }
 184
 185 /*
 186  * Set up the block and inode bitmaps, and the inode table for the new group.
 187  * This doesn't need to be part of the main transaction, since we are only
 188  * changing blocks outside the actual filesystem.  We still do journaling to
 189  * ensure the recovery is correct in case of a failure just after resize.
 190  * If any part of this fails, we simply abort the resize.
 191  */
 192 static int setup_new_group_blocks(struct super_block *sb,
 193                                   struct ext4_new_group_data *input)
 194 {
 195         struct ext4_sb_info *sbi = EXT4_SB(sb);
 196         ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
 197         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 198                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 199         unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
 200         struct buffer_head *bh;
 201         handle_t *handle;
 202         ext4_fsblk_t block;
 203         ext4_grpblk_t bit;
 204         int i;
 205         int err = 0, err2;
 206
 207         /* This transaction may be extended/restarted along the way */
 208         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 209
 210         if (IS_ERR(handle))
 211                 return PTR_ERR(handle);
 212
 213         BUG_ON(input->group != sbi->s_groups_count);
 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 (ext4_bg_has_super(sb, input->group)) {
 221                 ext4_debug("mark backup superblock %#04llx (+0)\n", start);
 222                 ext4_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                 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
 231
 232                 if ((err = extend_or_restart_transaction(handle, 1, bh)))
 233                         goto exit_bh;
 234
 235                 gdb = sb_getblk(sb, block);
 236                 if (!gdb) {
 237                         err = -EIO;
 238                         goto exit_bh;
 239                 }
 240                 if ((err = ext4_journal_get_write_access(handle, gdb))) {
 241                         brelse(gdb);
 242                         goto exit_bh;
 243                 }
 244                 lock_buffer(gdb);
 245                 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
 246                 set_buffer_uptodate(gdb);
 247                 unlock_buffer(gdb);
 248                 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 249                 if (unlikely(err)) {
 250                         brelse(gdb);
 251                         goto exit_bh;
 252                 }
 253                 ext4_set_bit(bit, bh->b_data);
 254                 brelse(gdb);
 255         }
 256
 257         /* Zero out all of the reserved backup group descriptor table blocks */
 258         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 259                         block, sbi->s_itb_per_group);
 260         err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
 261                                GFP_NOFS);
 262         if (err)
 263                 goto exit_bh;
 264         for (i = 0, bit = gdblocks + 1; i < reserved_gdb; i++, bit++)
 265                 ext4_set_bit(bit, bh->b_data);
 266
 267         ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
 268                    input->block_bitmap - start);
 269         ext4_set_bit(input->block_bitmap - start, bh->b_data);
 270         ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
 271                    input->inode_bitmap - start);
 272         ext4_set_bit(input->inode_bitmap - start, bh->b_data);
 273
 274         /* Zero out all of the inode table blocks */
 275         block = input->inode_table;
 276         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 277                         block, sbi->s_itb_per_group);
 278         err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
 279         if (err)
 280                 goto exit_bh;
 281         for (i = 0, bit = input->inode_table - start;
 282              i < sbi->s_itb_per_group; i++, bit++)
 283                 ext4_set_bit(bit, bh->b_data);
 284
 285         if ((err = extend_or_restart_transaction(handle, 2, bh)))
 286                 goto exit_bh;
 287
 288         ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
 289                              bh->b_data);
 290         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 291         if (unlikely(err)) {
 292                 ext4_std_error(sb, err);
 293                 goto exit_bh;
 294         }
 295         brelse(bh);
 296         /* Mark unused entries in inode bitmap used */
 297         ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
 298                    input->inode_bitmap, input->inode_bitmap - start);
 299         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 300                 err = PTR_ERR(bh);
 301                 goto exit_journal;
 302         }
 303
 304         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
 305                              bh->b_data);
 306         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 307         if (unlikely(err))
 308                 ext4_std_error(sb, err);
 309 exit_bh:
 310         brelse(bh);
 311
 312 exit_journal:
 313         if ((err2 = ext4_journal_stop(handle)) && !err)
 314                 err = err2;
 315
 316         return err;
 317 }
 318
 319 /*
 320  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 321  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 322  * calling this for the first time.  In a sparse filesystem it will be the
 323  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 324  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 325  */
 326 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 327                                   unsigned *five, unsigned *seven)
 328 {
 329         unsigned *min = three;
 330         int mult = 3;
 331         unsigned ret;
 332
 333         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 334                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 335                 ret = *min;
 336                 *min += 1;
 337                 return ret;
 338         }
 339
 340         if (*five < *min) {
 341                 min = five;
 342                 mult = 5;
 343         }
 344         if (*seven < *min) {
 345                 min = seven;
 346                 mult = 7;
 347         }
 348
 349         ret = *min;
 350         *min *= mult;
 351
 352         return ret;
 353 }
 354
 355 /*
 356  * Check that all of the backup GDT blocks are held in the primary GDT block.
 357  * It is assumed that they are stored in group order.  Returns the number of
 358  * groups in current filesystem that have BACKUPS, or -ve error code.
 359  */
 360 static int verify_reserved_gdb(struct super_block *sb,
 361                                struct buffer_head *primary)
 362 {
 363         const ext4_fsblk_t blk = primary->b_blocknr;
 364         const ext4_group_t end = EXT4_SB(sb)->s_groups_count;
 365         unsigned three = 1;
 366         unsigned five = 5;
 367         unsigned seven = 7;
 368         unsigned grp;
 369         __le32 *p = (__le32 *)primary->b_data;
 370         int gdbackups = 0;
 371
 372         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 373                 if (le32_to_cpu(*p++) !=
 374                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 375                         ext4_warning(sb, "reserved GDT %llu"
 376                                      " missing grp %d (%llu)",
 377                                      blk, grp,
 378                                      grp *
 379                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 380                                      blk);
 381                         return -EINVAL;
 382                 }
 383                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 384                         return -EFBIG;
 385         }
 386
 387         return gdbackups;
 388 }
 389
 390 /*
 391  * Called when we need to bring a reserved group descriptor table block into
 392  * use from the resize inode.  The primary copy of the new GDT block currently
 393  * is an indirect block (under the double indirect block in the resize inode).
 394  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 395  * block, in group order.  Even though we know all the block numbers we need,
 396  * we check to ensure that the resize inode has actually reserved these blocks.
 397  *
 398  * Don't need to update the block bitmaps because the blocks are still in use.
 399  *
 400  * We get all of the error cases out of the way, so that we are sure to not
 401  * fail once we start modifying the data on disk, because JBD has no rollback.
 402  */
 403 static int add_new_gdb(handle_t *handle, struct inode *inode,
 404                        struct ext4_new_group_data *input,
 405                        struct buffer_head **primary)
 406 {
 407         struct super_block *sb = inode->i_sb;
 408         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 409         unsigned long gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 410         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 411         struct buffer_head **o_group_desc, **n_group_desc;
 412         struct buffer_head *dind;
 413         int gdbackups;
 414         struct ext4_iloc iloc;
 415         __le32 *data;
 416         int err;
 417
 418         if (test_opt(sb, DEBUG))
 419                 printk(KERN_DEBUG
 420                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 421                        gdb_num);
 422
 423         /*
 424          * If we are not using the primary superblock/GDT copy don't resize,
 425          * because the user tools have no way of handling this.  Probably a
 426          * bad time to do it anyways.
 427          */
 428         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 429             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 430                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 431                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 432                 return -EPERM;
 433         }
 434
 435         *primary = sb_bread(sb, gdblock);
 436         if (!*primary)
 437                 return -EIO;
 438
 439         if ((gdbackups = verify_reserved_gdb(sb, *primary)) < 0) {
 440                 err = gdbackups;
 441                 goto exit_bh;
 442         }
 443
 444         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 445         dind = sb_bread(sb, le32_to_cpu(*data));
 446         if (!dind) {
 447                 err = -EIO;
 448                 goto exit_bh;
 449         }
 450
 451         data = (__le32 *)dind->b_data;
 452         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 453                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 454                              input->group, gdblock);
 455                 err = -EINVAL;
 456                 goto exit_dind;
 457         }
 458
 459         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 460         if (unlikely(err))
 461                 goto exit_dind;
 462
 463         err = ext4_journal_get_write_access(handle, *primary);
 464         if (unlikely(err))
 465                 goto exit_sbh;
 466
 467         err = ext4_journal_get_write_access(handle, dind);
 468         if (unlikely(err))
 469                 ext4_std_error(sb, err);
 470
 471         /* ext4_reserve_inode_write() gets a reference on the iloc */
 472         err = ext4_reserve_inode_write(handle, inode, &iloc);
 473         if (unlikely(err))
 474                 goto exit_dindj;
 475
 476         n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 477                         GFP_NOFS);
 478         if (!n_group_desc) {
 479                 err = -ENOMEM;
 480                 ext4_warning(sb,
 481                               "not enough memory for %lu groups", gdb_num + 1);
 482                 goto exit_inode;
 483         }
 484
 485         /*
 486          * Finally, we have all of the possible failures behind us...
 487          *
 488          * Remove new GDT block from inode double-indirect block and clear out
 489          * the new GDT block for use (which also "frees" the backup GDT blocks
 490          * from the reserved inode).  We don't need to change the bitmaps for
 491          * these blocks, because they are marked as in-use from being in the
 492          * reserved inode, and will become GDT blocks (primary and backup).
 493          */
 494         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 495         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 496         if (unlikely(err)) {
 497                 ext4_std_error(sb, err);
 498                 goto exit_inode;
 499         }
 500         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 501         ext4_mark_iloc_dirty(handle, inode, &iloc);
 502         memset((*primary)->b_data, 0, sb->s_blocksize);
 503         err = ext4_handle_dirty_metadata(handle, NULL, *primary);
 504         if (unlikely(err)) {
 505                 ext4_std_error(sb, err);
 506                 goto exit_inode;
 507         }
 508         brelse(dind);
 509
 510         o_group_desc = EXT4_SB(sb)->s_group_desc;
 511         memcpy(n_group_desc, o_group_desc,
 512                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 513         n_group_desc[gdb_num] = *primary;
 514         EXT4_SB(sb)->s_group_desc = n_group_desc;
 515         EXT4_SB(sb)->s_gdb_count++;
 516         kfree(o_group_desc);
 517
 518         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 519         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 520         if (err)
 521                 ext4_std_error(sb, err);
 522
 523         return err;
 524
 525 exit_inode:
 526         /* ext4_handle_release_buffer(handle, iloc.bh); */
 527         brelse(iloc.bh);
 528 exit_dindj:
 529         /* ext4_handle_release_buffer(handle, dind); */
 530 exit_sbh:
 531         /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
 532 exit_dind:
 533         brelse(dind);
 534 exit_bh:
 535         brelse(*primary);
 536
 537         ext4_debug("leaving with error %d\n", err);
 538         return err;
 539 }
 540
 541 /*
 542  * Called when we are adding a new group which has a backup copy of each of
 543  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 544  * We need to add these reserved backup GDT blocks to the resize inode, so
 545  * that they are kept for future resizing and not allocated to files.
 546  *
 547  * Each reserved backup GDT block will go into a different indirect block.
 548  * The indirect blocks are actually the primary reserved GDT blocks,
 549  * so we know in advance what their block numbers are.  We only get the
 550  * double-indirect block to verify it is pointing to the primary reserved
 551  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 552  * backup GDT blocks are stored in their reserved primary GDT block.
 553  */
 554 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 555                               struct ext4_new_group_data *input)
 556 {
 557         struct super_block *sb = inode->i_sb;
 558         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 559         struct buffer_head **primary;
 560         struct buffer_head *dind;
 561         struct ext4_iloc iloc;
 562         ext4_fsblk_t blk;
 563         __le32 *data, *end;
 564         int gdbackups = 0;
 565         int res, i;
 566         int err;
 567
 568         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 569         if (!primary)
 570                 return -ENOMEM;
 571
 572         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 573         dind = sb_bread(sb, le32_to_cpu(*data));
 574         if (!dind) {
 575                 err = -EIO;
 576                 goto exit_free;
 577         }
 578
 579         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 580         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 581                                          EXT4_ADDR_PER_BLOCK(sb));
 582         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 583
 584         /* Get each reserved primary GDT block and verify it holds backups */
 585         for (res = 0; res < reserved_gdb; res++, blk++) {
 586                 if (le32_to_cpu(*data) != blk) {
 587                         ext4_warning(sb, "reserved block %llu"
 588                                      " not at offset %ld",
 589                                      blk,
 590                                      (long)(data - (__le32 *)dind->b_data));
 591                         err = -EINVAL;
 592                         goto exit_bh;
 593                 }
 594                 primary[res] = sb_bread(sb, blk);
 595                 if (!primary[res]) {
 596                         err = -EIO;
 597                         goto exit_bh;
 598                 }
 599                 if ((gdbackups = verify_reserved_gdb(sb, primary[res])) < 0) {
 600                         brelse(primary[res]);
 601                         err = gdbackups;
 602                         goto exit_bh;
 603                 }
 604                 if (++data >= end)
 605                         data = (__le32 *)dind->b_data;
 606         }
 607
 608         for (i = 0; i < reserved_gdb; i++) {
 609                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
 610                         /*
 611                         int j;
 612                         for (j = 0; j < i; j++)
 613                                 ext4_handle_release_buffer(handle, primary[j]);
 614                          */
 615                         goto exit_bh;
 616                 }
 617         }
 618
 619         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 620                 goto exit_bh;
 621
 622         /*
 623          * Finally we can add each of the reserved backup GDT blocks from
 624          * the new group to its reserved primary GDT block.
 625          */
 626         blk = input->group * EXT4_BLOCKS_PER_GROUP(sb);
 627         for (i = 0; i < reserved_gdb; i++) {
 628                 int err2;
 629                 data = (__le32 *)primary[i]->b_data;
 630                 /* printk("reserving backup %lu[%u] = %lu\n",
 631                        primary[i]->b_blocknr, gdbackups,
 632                        blk + primary[i]->b_blocknr); */
 633                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 634                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 635                 if (!err)
 636                         err = err2;
 637         }
 638         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 639         ext4_mark_iloc_dirty(handle, inode, &iloc);
 640
 641 exit_bh:
 642         while (--res >= 0)
 643                 brelse(primary[res]);
 644         brelse(dind);
 645
 646 exit_free:
 647         kfree(primary);
 648
 649         return err;
 650 }
 651
 652 /*
 653  * Update the backup copies of the ext4 metadata.  These don't need to be part
 654  * of the main resize transaction, because e2fsck will re-write them if there
 655  * is a problem (basically only OOM will cause a problem).  However, we
 656  * _should_ update the backups if possible, in case the primary gets trashed
 657  * for some reason and we need to run e2fsck from a backup superblock.  The
 658  * important part is that the new block and inode counts are in the backup
 659  * superblocks, and the location of the new group metadata in the GDT backups.
 660  *
 661  * We do not need take the s_resize_lock for this, because these
 662  * blocks are not otherwise touched by the filesystem code when it is
 663  * mounted.  We don't need to worry about last changing from
 664  * sbi->s_groups_count, because the worst that can happen is that we
 665  * do not copy the full number of backups at this time.  The resize
 666  * which changed s_groups_count will backup again.
 667  */
 668 static void update_backups(struct super_block *sb,
 669                            int blk_off, char *data, int size)
 670 {
 671         struct ext4_sb_info *sbi = EXT4_SB(sb);
 672         const ext4_group_t last = sbi->s_groups_count;
 673         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
 674         unsigned three = 1;
 675         unsigned five = 5;
 676         unsigned seven = 7;
 677         ext4_group_t group;
 678         int rest = sb->s_blocksize - size;
 679         handle_t *handle;
 680         int err = 0, err2;
 681
 682         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 683         if (IS_ERR(handle)) {
 684                 group = 1;
 685                 err = PTR_ERR(handle);
 686                 goto exit_err;
 687         }
 688
 689         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
 690                 struct buffer_head *bh;
 691
 692                 /* Out of journal space, and can't get more - abort - so sad */
 693                 if (ext4_handle_valid(handle) &&
 694                     handle->h_buffer_credits == 0 &&
 695                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
 696                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 697                         break;
 698
 699                 bh = sb_getblk(sb, group * bpg + blk_off);
 700                 if (!bh) {
 701                         err = -EIO;
 702                         break;
 703                 }
 704                 ext4_debug("update metadata backup %#04lx\n",
 705                           (unsigned long)bh->b_blocknr);
 706                 if ((err = ext4_journal_get_write_access(handle, bh)))
 707                         break;
 708                 lock_buffer(bh);
 709                 memcpy(bh->b_data, data, size);
 710                 if (rest)
 711                         memset(bh->b_data + size, 0, rest);
 712                 set_buffer_uptodate(bh);
 713                 unlock_buffer(bh);
 714                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 715                 if (unlikely(err))
 716                         ext4_std_error(sb, err);
 717                 brelse(bh);
 718         }
 719         if ((err2 = ext4_journal_stop(handle)) && !err)
 720                 err = err2;
 721
 722         /*
 723          * Ugh! Need to have e2fsck write the backup copies.  It is too
 724          * late to revert the resize, we shouldn't fail just because of
 725          * the backup copies (they are only needed in case of corruption).
 726          *
 727          * However, if we got here we have a journal problem too, so we
 728          * can't really start a transaction to mark the superblock.
 729          * Chicken out and just set the flag on the hope it will be written
 730          * to disk, and if not - we will simply wait until next fsck.
 731          */
 732 exit_err:
 733         if (err) {
 734                 ext4_warning(sb, "can't update backup for group %u (err %d), "
 735                              "forcing fsck on next reboot", group, err);
 736                 sbi->s_mount_state &= ~EXT4_VALID_FS;
 737                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
 738                 mark_buffer_dirty(sbi->s_sbh);
 739         }
 740 }
 741
 742 /* Add group descriptor data to an existing or new group descriptor block.
 743  * Ensure we handle all possible error conditions _before_ we start modifying
 744  * the filesystem, because we cannot abort the transaction and not have it
 745  * write the data to disk.
 746  *
 747  * If we are on a GDT block boundary, we need to get the reserved GDT block.
 748  * Otherwise, we may need to add backup GDT blocks for a sparse group.
 749  *
 750  * We only need to hold the superblock lock while we are actually adding
 751  * in the new group's counts to the superblock.  Prior to that we have
 752  * not really "added" the group at all.  We re-check that we are still
 753  * adding in the last group in case things have changed since verifying.
 754  */
 755 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
 756 {
 757         struct ext4_sb_info *sbi = EXT4_SB(sb);
 758         struct ext4_super_block *es = sbi->s_es;
 759         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 760                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
 761         struct buffer_head *primary = NULL;
 762         struct ext4_group_desc *gdp;
 763         struct inode *inode = NULL;
 764         handle_t *handle;
 765         int gdb_off, gdb_num;
 766         int err, err2;
 767
 768         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
 769         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
 770
 771         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
 772                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 773                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
 774                 return -EPERM;
 775         }
 776
 777         if (ext4_blocks_count(es) + input->blocks_count <
 778             ext4_blocks_count(es)) {
 779                 ext4_warning(sb, "blocks_count overflow");
 780                 return -EINVAL;
 781         }
 782
 783         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
 784             le32_to_cpu(es->s_inodes_count)) {
 785                 ext4_warning(sb, "inodes_count overflow");
 786                 return -EINVAL;
 787         }
 788
 789         if (reserved_gdb || gdb_off == 0) {
 790                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
 791                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
 792                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
 793                         ext4_warning(sb,
 794                                      "No reserved GDT blocks, can't resize");
 795                         return -EPERM;
 796                 }
 797                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
 798                 if (IS_ERR(inode)) {
 799                         ext4_warning(sb, "Error opening resize inode");
 800                         return PTR_ERR(inode);
 801                 }
 802         }
 803
 804
 805         if ((err = verify_group_input(sb, input)))
 806                 goto exit_put;
 807
 808         if ((err = setup_new_group_blocks(sb, input)))
 809                 goto exit_put;
 810
 811         /*
 812          * We will always be modifying at least the superblock and a GDT
 813          * block.  If we are adding a group past the last current GDT block,
 814          * we will also modify the inode and the dindirect block.  If we
 815          * are adding a group with superblock/GDT backups  we will also
 816          * modify each of the reserved GDT dindirect blocks.
 817          */
 818         handle = ext4_journal_start_sb(sb,
 819                                        ext4_bg_has_super(sb, input->group) ?
 820                                        3 + reserved_gdb : 4);
 821         if (IS_ERR(handle)) {
 822                 err = PTR_ERR(handle);
 823                 goto exit_put;
 824         }
 825
 826         if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
 827                 goto exit_journal;
 828
 829         /*
 830          * We will only either add reserved group blocks to a backup group
 831          * or remove reserved blocks for the first group in a new group block.
 832          * Doing both would be mean more complex code, and sane people don't
 833          * use non-sparse filesystems anymore.  This is already checked above.
 834          */
 835         if (gdb_off) {
 836                 primary = sbi->s_group_desc[gdb_num];
 837                 if ((err = ext4_journal_get_write_access(handle, primary)))
 838                         goto exit_journal;
 839
 840                 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group) &&
 841                     (err = reserve_backup_gdb(handle, inode, input)))
 842                         goto exit_journal;
 843         } else if ((err = add_new_gdb(handle, inode, input, &primary)))
 844                 goto exit_journal;
 845
 846         /*
 847          * OK, now we've set up the new group.  Time to make it active.
 848          *
 849          * so we have to be safe wrt. concurrent accesses the group
 850          * data.  So we need to be careful to set all of the relevant
 851          * group descriptor data etc. *before* we enable the group.
 852          *
 853          * The key field here is sbi->s_groups_count: as long as
 854          * that retains its old value, nobody is going to access the new
 855          * group.
 856          *
 857          * So first we update all the descriptor metadata for the new
 858          * group; then we update the total disk blocks count; then we
 859          * update the groups count to enable the group; then finally we
 860          * update the free space counts so that the system can start
 861          * using the new disk blocks.
 862          */
 863
 864         /* Update group descriptor block for new group */
 865         gdp = (struct ext4_group_desc *)((char *)primary->b_data +
 866                                          gdb_off * EXT4_DESC_SIZE(sb));
 867
 868         memset(gdp, 0, EXT4_DESC_SIZE(sb));
 869         ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
 870         ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
 871         ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
 872         ext4_free_blks_set(sb, gdp, input->free_blocks_count);
 873         ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
 874         gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
 875         gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
 876
 877         /*
 878          * We can allocate memory for mb_alloc based on the new group
 879          * descriptor
 880          */
 881         err = ext4_mb_add_groupinfo(sb, input->group, gdp);
 882         if (err)
 883                 goto exit_journal;
 884
 885         /*
 886          * Make the new blocks and inodes valid next.  We do this before
 887          * increasing the group count so that once the group is enabled,
 888          * all of its blocks and inodes are already valid.
 889          *
 890          * We always allocate group-by-group, then block-by-block or
 891          * inode-by-inode within a group, so enabling these
 892          * blocks/inodes before the group is live won't actually let us
 893          * allocate the new space yet.
 894          */
 895         ext4_blocks_count_set(es, ext4_blocks_count(es) +
 896                 input->blocks_count);
 897         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
 898
 899         /*
 900          * We need to protect s_groups_count against other CPUs seeing
 901          * inconsistent state in the superblock.
 902          *
 903          * The precise rules we use are:
 904          *
 905          * * Writers must perform a smp_wmb() after updating all dependent
 906          *   data and before modifying the groups count
 907          *
 908          * * Readers must perform an smp_rmb() after reading the groups count
 909          *   and before reading any dependent data.
 910          *
 911          * NB. These rules can be relaxed when checking the group count
 912          * while freeing data, as we can only allocate from a block
 913          * group after serialising against the group count, and we can
 914          * only then free after serialising in turn against that
 915          * allocation.
 916          */
 917         smp_wmb();
 918
 919         /* Update the global fs size fields */
 920         sbi->s_groups_count++;
 921
 922         err = ext4_handle_dirty_metadata(handle, NULL, primary);
 923         if (unlikely(err)) {
 924                 ext4_std_error(sb, err);
 925                 goto exit_journal;
 926         }
 927
 928         /* Update the reserved block counts only once the new group is
 929          * active. */
 930         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
 931                 input->reserved_blocks);
 932
 933         /* Update the free space counts */
 934         percpu_counter_add(&sbi->s_freeblocks_counter,
 935                            input->free_blocks_count);
 936         percpu_counter_add(&sbi->s_freeinodes_counter,
 937                            EXT4_INODES_PER_GROUP(sb));
 938
 939         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
 940             sbi->s_log_groups_per_flex) {
 941                 ext4_group_t flex_group;
 942                 flex_group = ext4_flex_group(sbi, input->group);
 943                 atomic_add(input->free_blocks_count,
 944                            &sbi->s_flex_groups[flex_group].free_blocks);
 945                 atomic_add(EXT4_INODES_PER_GROUP(sb),
 946                            &sbi->s_flex_groups[flex_group].free_inodes);
 947         }
 948
 949         ext4_handle_dirty_super(handle, sb);
 950
 951 exit_journal:
 952         if ((err2 = ext4_journal_stop(handle)) && !err)
 953                 err = err2;
 954         if (!err) {
 955                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
 956                                sizeof(struct ext4_super_block));
 957                 update_backups(sb, primary->b_blocknr, primary->b_data,
 958                                primary->b_size);
 959         }
 960 exit_put:
 961         iput(inode);
 962         return err;
 963 } /* ext4_group_add */
 964
 965 /*
 966  * Extend the filesystem to the new number of blocks specified.  This entry
 967  * point is only used to extend the current filesystem to the end of the last
 968  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
 969  * for emergencies (because it has no dependencies on reserved blocks).
 970  *
 971  * If we _really_ wanted, we could use default values to call ext4_group_add()
 972  * allow the "remount" trick to work for arbitrary resizing, assuming enough
 973  * GDT blocks are reserved to grow to the desired size.
 974  */
 975 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
 976                       ext4_fsblk_t n_blocks_count)
 977 {
 978         ext4_fsblk_t o_blocks_count;
 979         ext4_grpblk_t last;
 980         ext4_grpblk_t add;
 981         struct buffer_head *bh;
 982         handle_t *handle;
 983         int err, err2;
 984         ext4_group_t group;
 985
 986         o_blocks_count = ext4_blocks_count(es);
 987
 988         if (test_opt(sb, DEBUG))
 989                 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu uto %llu blocks\n",
 990                        o_blocks_count, n_blocks_count);
 991
 992         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
 993                 return 0;
 994
 995         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
 996                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
 997                         " too large to resize to %llu blocks safely\n",
 998                         sb->s_id, n_blocks_count);
 999                 if (sizeof(sector_t) < 8)
1000                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1001                 return -EINVAL;
1002         }
1003
1004         if (n_blocks_count < o_blocks_count) {
1005                 ext4_warning(sb, "can't shrink FS - resize aborted");
1006                 return -EINVAL;
1007         }
1008
1009         /* Handle the remaining blocks in the last group only. */
1010         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1011
1012         if (last == 0) {
1013                 ext4_warning(sb, "need to use ext2online to resize further");
1014                 return -EPERM;
1015         }
1016
1017         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1018
1019         if (o_blocks_count + add < o_blocks_count) {
1020                 ext4_warning(sb, "blocks_count overflow");
1021                 return -EINVAL;
1022         }
1023
1024         if (o_blocks_count + add > n_blocks_count)
1025                 add = n_blocks_count - o_blocks_count;
1026
1027         if (o_blocks_count + add < n_blocks_count)
1028                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1029                              o_blocks_count + add, add);
1030
1031         /* See if the device is actually as big as what was requested */
1032         bh = sb_bread(sb, o_blocks_count + add - 1);
1033         if (!bh) {
1034                 ext4_warning(sb, "can't read last block, resize aborted");
1035                 return -ENOSPC;
1036         }
1037         brelse(bh);
1038
1039         /* We will update the superblock, one block bitmap, and
1040          * one group descriptor via ext4_free_blocks().
1041          */
1042         handle = ext4_journal_start_sb(sb, 3);
1043         if (IS_ERR(handle)) {
1044                 err = PTR_ERR(handle);
1045                 ext4_warning(sb, "error %d on journal start", err);
1046                 goto exit_put;
1047         }
1048
1049         if ((err = ext4_journal_get_write_access(handle,
1050                                                  EXT4_SB(sb)->s_sbh))) {
1051                 ext4_warning(sb, "error %d on journal write access", err);
1052                 ext4_journal_stop(handle);
1053                 goto exit_put;
1054         }
1055         ext4_blocks_count_set(es, o_blocks_count + add);
1056         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1057                    o_blocks_count + add);
1058         /* We add the blocks to the bitmap and set the group need init bit */
1059         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1060         ext4_handle_dirty_super(handle, sb);
1061         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1062                    o_blocks_count + add);
1063         err2 = ext4_journal_stop(handle);
1064         if (!err && err2)
1065                 err = err2;
1066
1067         if (err)
1068                 goto exit_put;
1069
1070         if (test_opt(sb, DEBUG))
1071                 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1072                        ext4_blocks_count(es));
1073         update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1074                        sizeof(struct ext4_super_block));
1075 exit_put:
1076         return err;
1077 } /* ext4_group_extend */