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 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  49                                              ext4_group_t group) {
  50         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  51                EXT4_DESC_PER_BLOCK_BITS(sb);
  52 }
  53
  54 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
  55                                              ext4_group_t group) {
  56         group = ext4_meta_bg_first_group(sb, group);
  57         return ext4_group_first_block_no(sb, group);
  58 }
  59
  60 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
  61                                                 ext4_group_t group) {
  62         ext4_grpblk_t overhead;
  63         overhead = ext4_bg_num_gdb(sb, group);
  64         if (ext4_bg_has_super(sb, group))
  65                 overhead += 1 +
  66                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
  67         return overhead;
  68 }
  69
  70 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  71 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  72
  73 static int verify_group_input(struct super_block *sb,
  74                               struct ext4_new_group_data *input)
  75 {
  76         struct ext4_sb_info *sbi = EXT4_SB(sb);
  77         struct ext4_super_block *es = sbi->s_es;
  78         ext4_fsblk_t start = ext4_blocks_count(es);
  79         ext4_fsblk_t end = start + input->blocks_count;
  80         ext4_group_t group = input->group;
  81         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  82         unsigned overhead;
  83         ext4_fsblk_t metaend;
  84         struct buffer_head *bh = NULL;
  85         ext4_grpblk_t free_blocks_count, offset;
  86         int err = -EINVAL;
  87
  88         if (group != sbi->s_groups_count) {
  89                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
  90                              input->group, sbi->s_groups_count);
  91                 return -EINVAL;
  92         }
  93
  94         overhead = ext4_group_overhead_blocks(sb, group);
  95         metaend = start + overhead;
  96         input->free_blocks_count = free_blocks_count =
  97                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  98
  99         if (test_opt(sb, DEBUG))
 100                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 101                        "(%d free, %u reserved)\n",
 102                        ext4_bg_has_super(sb, input->group) ? "normal" :
 103                        "no-super", input->group, input->blocks_count,
 104                        free_blocks_count, input->reserved_blocks);
 105
 106         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 107         if (offset != 0)
 108                         ext4_warning(sb, "Last group not full");
 109         else if (input->reserved_blocks > input->blocks_count / 5)
 110                 ext4_warning(sb, "Reserved blocks too high (%u)",
 111                              input->reserved_blocks);
 112         else if (free_blocks_count < 0)
 113                 ext4_warning(sb, "Bad blocks count %u",
 114                              input->blocks_count);
 115         else if (!(bh = sb_bread(sb, end - 1)))
 116                 ext4_warning(sb, "Cannot read last block (%llu)",
 117                              end - 1);
 118         else if (outside(input->block_bitmap, start, end))
 119                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 120                              (unsigned long long)input->block_bitmap);
 121         else if (outside(input->inode_bitmap, start, end))
 122                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 123                              (unsigned long long)input->inode_bitmap);
 124         else if (outside(input->inode_table, start, end) ||
 125                  outside(itend - 1, start, end))
 126                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 127                              (unsigned long long)input->inode_table, itend - 1);
 128         else if (input->inode_bitmap == input->block_bitmap)
 129                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 130                              (unsigned long long)input->block_bitmap);
 131         else if (inside(input->block_bitmap, input->inode_table, itend))
 132                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 133                              "(%llu-%llu)",
 134                              (unsigned long long)input->block_bitmap,
 135                              (unsigned long long)input->inode_table, itend - 1);
 136         else if (inside(input->inode_bitmap, input->inode_table, itend))
 137                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 138                              "(%llu-%llu)",
 139                              (unsigned long long)input->inode_bitmap,
 140                              (unsigned long long)input->inode_table, itend - 1);
 141         else if (inside(input->block_bitmap, start, metaend))
 142                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 143                              (unsigned long long)input->block_bitmap,
 144                              start, metaend - 1);
 145         else if (inside(input->inode_bitmap, start, metaend))
 146                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 147                              (unsigned long long)input->inode_bitmap,
 148                              start, metaend - 1);
 149         else if (inside(input->inode_table, start, metaend) ||
 150                  inside(itend - 1, start, metaend))
 151                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 152                              "(%llu-%llu)",
 153                              (unsigned long long)input->inode_table,
 154                              itend - 1, start, metaend - 1);
 155         else
 156                 err = 0;
 157         brelse(bh);
 158
 159         return err;
 160 }
 161
 162 /*
 163  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 164  * group each time.
 165  */
 166 struct ext4_new_flex_group_data {
 167         struct ext4_new_group_data *groups;     /* new_group_data for groups
 168                                                    in the flex group */
 169         __u16 *bg_flags;                        /* block group flags of groups
 170                                                    in @groups */
 171         ext4_group_t count;                     /* number of groups in @groups
 172                                                  */
 173 };
 174
 175 /*
 176  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 177  * @flexbg_size.
 178  *
 179  * Returns NULL on failure otherwise address of the allocated structure.
 180  */
 181 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 182 {
 183         struct ext4_new_flex_group_data *flex_gd;
 184
 185         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 186         if (flex_gd == NULL)
 187                 goto out3;
 188
 189         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
 190                 goto out2;
 191         flex_gd->count = flexbg_size;
 192
 193         flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
 194                                   flexbg_size, GFP_NOFS);
 195         if (flex_gd->groups == NULL)
 196                 goto out2;
 197
 198         flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
 199         if (flex_gd->bg_flags == NULL)
 200                 goto out1;
 201
 202         return flex_gd;
 203
 204 out1:
 205         kfree(flex_gd->groups);
 206 out2:
 207         kfree(flex_gd);
 208 out3:
 209         return NULL;
 210 }
 211
 212 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 213 {
 214         kfree(flex_gd->bg_flags);
 215         kfree(flex_gd->groups);
 216         kfree(flex_gd);
 217 }
 218
 219 /*
 220  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 221  * and inode tables for a flex group.
 222  *
 223  * This function is used by 64bit-resize.  Note that this function allocates
 224  * group tables from the 1st group of groups contained by @flexgd, which may
 225  * be a partial of a flex group.
 226  *
 227  * @sb: super block of fs to which the groups belongs
 228  *
 229  * Returns 0 on a successful allocation of the metadata blocks in the
 230  * block group.
 231  */
 232 static int ext4_alloc_group_tables(struct super_block *sb,
 233                                 struct ext4_new_flex_group_data *flex_gd,
 234                                 int flexbg_size)
 235 {
 236         struct ext4_new_group_data *group_data = flex_gd->groups;
 237         ext4_fsblk_t start_blk;
 238         ext4_fsblk_t last_blk;
 239         ext4_group_t src_group;
 240         ext4_group_t bb_index = 0;
 241         ext4_group_t ib_index = 0;
 242         ext4_group_t it_index = 0;
 243         ext4_group_t group;
 244         ext4_group_t last_group;
 245         unsigned overhead;
 246
 247         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 248
 249         src_group = group_data[0].group;
 250         last_group  = src_group + flex_gd->count - 1;
 251
 252         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 253                (last_group & ~(flexbg_size - 1))));
 254 next_group:
 255         group = group_data[0].group;
 256         if (src_group >= group_data[0].group + flex_gd->count)
 257                 return -ENOSPC;
 258         start_blk = ext4_group_first_block_no(sb, src_group);
 259         last_blk = start_blk + group_data[src_group - group].blocks_count;
 260
 261         overhead = ext4_group_overhead_blocks(sb, src_group);
 262
 263         start_blk += overhead;
 264
 265         /* We collect contiguous blocks as much as possible. */
 266         src_group++;
 267         for (; src_group <= last_group; src_group++) {
 268                 overhead = ext4_group_overhead_blocks(sb, src_group);
 269                 if (overhead != 0)
 270                         last_blk += group_data[src_group - group].blocks_count;
 271                 else
 272                         break;
 273         }
 274
 275         /* Allocate block bitmaps */
 276         for (; bb_index < flex_gd->count; bb_index++) {
 277                 if (start_blk >= last_blk)
 278                         goto next_group;
 279                 group_data[bb_index].block_bitmap = start_blk++;
 280                 group = ext4_get_group_number(sb, start_blk - 1);
 281                 group -= group_data[0].group;
 282                 group_data[group].free_blocks_count--;
 283                 if (flexbg_size > 1)
 284                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 285         }
 286
 287         /* Allocate inode bitmaps */
 288         for (; ib_index < flex_gd->count; ib_index++) {
 289                 if (start_blk >= last_blk)
 290                         goto next_group;
 291                 group_data[ib_index].inode_bitmap = start_blk++;
 292                 group = ext4_get_group_number(sb, start_blk - 1);
 293                 group -= group_data[0].group;
 294                 group_data[group].free_blocks_count--;
 295                 if (flexbg_size > 1)
 296                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 297         }
 298
 299         /* Allocate inode tables */
 300         for (; it_index < flex_gd->count; it_index++) {
 301                 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
 302                         goto next_group;
 303                 group_data[it_index].inode_table = start_blk;
 304                 group = ext4_get_group_number(sb, start_blk - 1);
 305                 group -= group_data[0].group;
 306                 group_data[group].free_blocks_count -=
 307                                         EXT4_SB(sb)->s_itb_per_group;
 308                 if (flexbg_size > 1)
 309                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 310
 311                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 312         }
 313
 314         if (test_opt(sb, DEBUG)) {
 315                 int i;
 316                 group = group_data[0].group;
 317
 318                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 319                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 320                        flexbg_size);
 321
 322                 for (i = 0; i < flex_gd->count; i++) {
 323                         printk(KERN_DEBUG "adding %s group %u: %u "
 324                                "blocks (%d free)\n",
 325                                ext4_bg_has_super(sb, group + i) ? "normal" :
 326                                "no-super", group + i,
 327                                group_data[i].blocks_count,
 328                                group_data[i].free_blocks_count);
 329                 }
 330         }
 331         return 0;
 332 }
 333
 334 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 335                                   ext4_fsblk_t blk)
 336 {
 337         struct buffer_head *bh;
 338         int err;
 339
 340         bh = sb_getblk(sb, blk);
 341         if (unlikely(!bh))
 342                 return ERR_PTR(-ENOMEM);
 343         if ((err = ext4_journal_get_write_access(handle, bh))) {
 344                 brelse(bh);
 345                 bh = ERR_PTR(err);
 346         } else {
 347                 memset(bh->b_data, 0, sb->s_blocksize);
 348                 set_buffer_uptodate(bh);
 349         }
 350
 351         return bh;
 352 }
 353
 354 /*
 355  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 356  * If that fails, restart the transaction & regain write access for the
 357  * buffer head which is used for block_bitmap modifications.
 358  */
 359 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 360 {
 361         int err;
 362
 363         if (ext4_handle_has_enough_credits(handle, thresh))
 364                 return 0;
 365
 366         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 367         if (err < 0)
 368                 return err;
 369         if (err) {
 370                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 371                 if (err)
 372                         return err;
 373         }
 374
 375         return 0;
 376 }
 377
 378 /*
 379  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 380  *
 381  * Helper function for ext4_setup_new_group_blocks() which set .
 382  *
 383  * @sb: super block
 384  * @handle: journal handle
 385  * @flex_gd: flex group data
 386  */
 387 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 388                         struct ext4_new_flex_group_data *flex_gd,
 389                         ext4_fsblk_t block, ext4_group_t count)
 390 {
 391         ext4_group_t count2;
 392
 393         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 394         for (count2 = count; count > 0; count -= count2, block += count2) {
 395                 ext4_fsblk_t start;
 396                 struct buffer_head *bh;
 397                 ext4_group_t group;
 398                 int err;
 399
 400                 group = ext4_get_group_number(sb, block);
 401                 start = ext4_group_first_block_no(sb, group);
 402                 group -= flex_gd->groups[0].group;
 403
 404                 count2 = sb->s_blocksize * 8 - (block - start);
 405                 if (count2 > count)
 406                         count2 = count;
 407
 408                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 409                         BUG_ON(flex_gd->count > 1);
 410                         continue;
 411                 }
 412
 413                 err = extend_or_restart_transaction(handle, 1);
 414                 if (err)
 415                         return err;
 416
 417                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 418                 if (unlikely(!bh))
 419                         return -ENOMEM;
 420
 421                 err = ext4_journal_get_write_access(handle, bh);
 422                 if (err)
 423                         return err;
 424                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 425                            block - start, count2);
 426                 ext4_set_bits(bh->b_data, block - start, count2);
 427
 428                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 429                 if (unlikely(err))
 430                         return err;
 431                 brelse(bh);
 432         }
 433
 434         return 0;
 435 }
 436
 437 /*
 438  * Set up the block and inode bitmaps, and the inode table for the new groups.
 439  * This doesn't need to be part of the main transaction, since we are only
 440  * changing blocks outside the actual filesystem.  We still do journaling to
 441  * ensure the recovery is correct in case of a failure just after resize.
 442  * If any part of this fails, we simply abort the resize.
 443  *
 444  * setup_new_flex_group_blocks handles a flex group as follow:
 445  *  1. copy super block and GDT, and initialize group tables if necessary.
 446  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 447  *     super block and GDT.
 448  *  2. allocate group tables in block bitmaps, that is, set bits in block
 449  *     bitmap for blocks taken by group tables.
 450  */
 451 static int setup_new_flex_group_blocks(struct super_block *sb,
 452                                 struct ext4_new_flex_group_data *flex_gd)
 453 {
 454         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 455         ext4_fsblk_t start;
 456         ext4_fsblk_t block;
 457         struct ext4_sb_info *sbi = EXT4_SB(sb);
 458         struct ext4_super_block *es = sbi->s_es;
 459         struct ext4_new_group_data *group_data = flex_gd->groups;
 460         __u16 *bg_flags = flex_gd->bg_flags;
 461         handle_t *handle;
 462         ext4_group_t group, count;
 463         struct buffer_head *bh = NULL;
 464         int reserved_gdb, i, j, err = 0, err2;
 465         int meta_bg;
 466
 467         BUG_ON(!flex_gd->count || !group_data ||
 468                group_data[0].group != sbi->s_groups_count);
 469
 470         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 471         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
 472
 473         /* This transaction may be extended/restarted along the way */
 474         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 475         if (IS_ERR(handle))
 476                 return PTR_ERR(handle);
 477
 478         group = group_data[0].group;
 479         for (i = 0; i < flex_gd->count; i++, group++) {
 480                 unsigned long gdblocks;
 481                 ext4_grpblk_t overhead;
 482
 483                 gdblocks = ext4_bg_num_gdb(sb, group);
 484                 start = ext4_group_first_block_no(sb, group);
 485
 486                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 487                         goto handle_itb;
 488
 489                 if (meta_bg == 1) {
 490                         ext4_group_t first_group;
 491                         first_group = ext4_meta_bg_first_group(sb, group);
 492                         if (first_group != group + 1 &&
 493                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 494                                 goto handle_itb;
 495                 }
 496
 497                 block = start + ext4_bg_has_super(sb, group);
 498                 /* Copy all of the GDT blocks into the backup in this group */
 499                 for (j = 0; j < gdblocks; j++, block++) {
 500                         struct buffer_head *gdb;
 501
 502                         ext4_debug("update backup group %#04llx\n", block);
 503                         err = extend_or_restart_transaction(handle, 1);
 504                         if (err)
 505                                 goto out;
 506
 507                         gdb = sb_getblk(sb, block);
 508                         if (unlikely(!gdb)) {
 509                                 err = -ENOMEM;
 510                                 goto out;
 511                         }
 512
 513                         err = ext4_journal_get_write_access(handle, gdb);
 514                         if (err) {
 515                                 brelse(gdb);
 516                                 goto out;
 517                         }
 518                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 519                                gdb->b_size);
 520                         set_buffer_uptodate(gdb);
 521
 522                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 523                         if (unlikely(err)) {
 524                                 brelse(gdb);
 525                                 goto out;
 526                         }
 527                         brelse(gdb);
 528                 }
 529
 530                 /* Zero out all of the reserved backup group descriptor
 531                  * table blocks
 532                  */
 533                 if (ext4_bg_has_super(sb, group)) {
 534                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 535                                         reserved_gdb, GFP_NOFS);
 536                         if (err)
 537                                 goto out;
 538                 }
 539
 540 handle_itb:
 541                 /* Initialize group tables of the grop @group */
 542                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 543                         goto handle_bb;
 544
 545                 /* Zero out all of the inode table blocks */
 546                 block = group_data[i].inode_table;
 547                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 548                            block, sbi->s_itb_per_group);
 549                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 550                                        GFP_NOFS);
 551                 if (err)
 552                         goto out;
 553
 554 handle_bb:
 555                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 556                         goto handle_ib;
 557
 558                 /* Initialize block bitmap of the @group */
 559                 block = group_data[i].block_bitmap;
 560                 err = extend_or_restart_transaction(handle, 1);
 561                 if (err)
 562                         goto out;
 563
 564                 bh = bclean(handle, sb, block);
 565                 if (IS_ERR(bh)) {
 566                         err = PTR_ERR(bh);
 567                         goto out;
 568                 }
 569                 overhead = ext4_group_overhead_blocks(sb, group);
 570                 if (overhead != 0) {
 571                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 572                                    start);
 573                         ext4_set_bits(bh->b_data, 0, overhead);
 574                 }
 575                 ext4_mark_bitmap_end(group_data[i].blocks_count,
 576                                      sb->s_blocksize * 8, bh->b_data);
 577                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 578                 if (err)
 579                         goto out;
 580                 brelse(bh);
 581
 582 handle_ib:
 583                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 584                         continue;
 585
 586                 /* Initialize inode bitmap of the @group */
 587                 block = group_data[i].inode_bitmap;
 588                 err = extend_or_restart_transaction(handle, 1);
 589                 if (err)
 590                         goto out;
 591                 /* Mark unused entries in inode bitmap used */
 592                 bh = bclean(handle, sb, block);
 593                 if (IS_ERR(bh)) {
 594                         err = PTR_ERR(bh);
 595                         goto out;
 596                 }
 597
 598                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 599                                      sb->s_blocksize * 8, bh->b_data);
 600                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 601                 if (err)
 602                         goto out;
 603                 brelse(bh);
 604         }
 605         bh = NULL;
 606
 607         /* Mark group tables in block bitmap */
 608         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 609                 count = group_table_count[j];
 610                 start = (&group_data[0].block_bitmap)[j];
 611                 block = start;
 612                 for (i = 1; i < flex_gd->count; i++) {
 613                         block += group_table_count[j];
 614                         if (block == (&group_data[i].block_bitmap)[j]) {
 615                                 count += group_table_count[j];
 616                                 continue;
 617                         }
 618                         err = set_flexbg_block_bitmap(sb, handle,
 619                                                 flex_gd, start, count);
 620                         if (err)
 621                                 goto out;
 622                         count = group_table_count[j];
 623                         start = group_data[i].block_bitmap;
 624                         block = start;
 625                 }
 626
 627                 if (count) {
 628                         err = set_flexbg_block_bitmap(sb, handle,
 629                                                 flex_gd, start, count);
 630                         if (err)
 631                                 goto out;
 632                 }
 633         }
 634
 635 out:
 636         brelse(bh);
 637         err2 = ext4_journal_stop(handle);
 638         if (err2 && !err)
 639                 err = err2;
 640
 641         return err;
 642 }
 643
 644 /*
 645  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 646  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 647  * calling this for the first time.  In a sparse filesystem it will be the
 648  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 649  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 650  */
 651 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 652                                   unsigned *five, unsigned *seven)
 653 {
 654         unsigned *min = three;
 655         int mult = 3;
 656         unsigned ret;
 657
 658         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 659                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 660                 ret = *min;
 661                 *min += 1;
 662                 return ret;
 663         }
 664
 665         if (*five < *min) {
 666                 min = five;
 667                 mult = 5;
 668         }
 669         if (*seven < *min) {
 670                 min = seven;
 671                 mult = 7;
 672         }
 673
 674         ret = *min;
 675         *min *= mult;
 676
 677         return ret;
 678 }
 679
 680 /*
 681  * Check that all of the backup GDT blocks are held in the primary GDT block.
 682  * It is assumed that they are stored in group order.  Returns the number of
 683  * groups in current filesystem that have BACKUPS, or -ve error code.
 684  */
 685 static int verify_reserved_gdb(struct super_block *sb,
 686                                ext4_group_t end,
 687                                struct buffer_head *primary)
 688 {
 689         const ext4_fsblk_t blk = primary->b_blocknr;
 690         unsigned three = 1;
 691         unsigned five = 5;
 692         unsigned seven = 7;
 693         unsigned grp;
 694         __le32 *p = (__le32 *)primary->b_data;
 695         int gdbackups = 0;
 696
 697         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 698                 if (le32_to_cpu(*p++) !=
 699                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 700                         ext4_warning(sb, "reserved GDT %llu"
 701                                      " missing grp %d (%llu)",
 702                                      blk, grp,
 703                                      grp *
 704                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 705                                      blk);
 706                         return -EINVAL;
 707                 }
 708                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 709                         return -EFBIG;
 710         }
 711
 712         return gdbackups;
 713 }
 714
 715 /*
 716  * Called when we need to bring a reserved group descriptor table block into
 717  * use from the resize inode.  The primary copy of the new GDT block currently
 718  * is an indirect block (under the double indirect block in the resize inode).
 719  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 720  * block, in group order.  Even though we know all the block numbers we need,
 721  * we check to ensure that the resize inode has actually reserved these blocks.
 722  *
 723  * Don't need to update the block bitmaps because the blocks are still in use.
 724  *
 725  * We get all of the error cases out of the way, so that we are sure to not
 726  * fail once we start modifying the data on disk, because JBD has no rollback.
 727  */
 728 static int add_new_gdb(handle_t *handle, struct inode *inode,
 729                        ext4_group_t group)
 730 {
 731         struct super_block *sb = inode->i_sb;
 732         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 733         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 734         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 735         struct buffer_head **o_group_desc, **n_group_desc;
 736         struct buffer_head *dind;
 737         struct buffer_head *gdb_bh;
 738         int gdbackups;
 739         struct ext4_iloc iloc;
 740         __le32 *data;
 741         int err;
 742
 743         if (test_opt(sb, DEBUG))
 744                 printk(KERN_DEBUG
 745                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 746                        gdb_num);
 747
 748         /*
 749          * If we are not using the primary superblock/GDT copy don't resize,
 750          * because the user tools have no way of handling this.  Probably a
 751          * bad time to do it anyways.
 752          */
 753         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 754             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 755                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 756                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 757                 return -EPERM;
 758         }
 759
 760         gdb_bh = sb_bread(sb, gdblock);
 761         if (!gdb_bh)
 762                 return -EIO;
 763
 764         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 765         if (gdbackups < 0) {
 766                 err = gdbackups;
 767                 goto exit_bh;
 768         }
 769
 770         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 771         dind = sb_bread(sb, le32_to_cpu(*data));
 772         if (!dind) {
 773                 err = -EIO;
 774                 goto exit_bh;
 775         }
 776
 777         data = (__le32 *)dind->b_data;
 778         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 779                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 780                              group, gdblock);
 781                 err = -EINVAL;
 782                 goto exit_dind;
 783         }
 784
 785         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 786         if (unlikely(err))
 787                 goto exit_dind;
 788
 789         err = ext4_journal_get_write_access(handle, gdb_bh);
 790         if (unlikely(err))
 791                 goto exit_dind;
 792
 793         err = ext4_journal_get_write_access(handle, dind);
 794         if (unlikely(err))
 795                 ext4_std_error(sb, err);
 796
 797         /* ext4_reserve_inode_write() gets a reference on the iloc */
 798         err = ext4_reserve_inode_write(handle, inode, &iloc);
 799         if (unlikely(err))
 800                 goto exit_dind;
 801
 802         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 803                                      sizeof(struct buffer_head *),
 804                                      GFP_NOFS);
 805         if (!n_group_desc) {
 806                 err = -ENOMEM;
 807                 ext4_warning(sb, "not enough memory for %lu groups",
 808                              gdb_num + 1);
 809                 goto exit_inode;
 810         }
 811
 812         /*
 813          * Finally, we have all of the possible failures behind us...
 814          *
 815          * Remove new GDT block from inode double-indirect block and clear out
 816          * the new GDT block for use (which also "frees" the backup GDT blocks
 817          * from the reserved inode).  We don't need to change the bitmaps for
 818          * these blocks, because they are marked as in-use from being in the
 819          * reserved inode, and will become GDT blocks (primary and backup).
 820          */
 821         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 822         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 823         if (unlikely(err)) {
 824                 ext4_std_error(sb, err);
 825                 goto exit_inode;
 826         }
 827         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 828         ext4_mark_iloc_dirty(handle, inode, &iloc);
 829         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 830         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 831         if (unlikely(err)) {
 832                 ext4_std_error(sb, err);
 833                 goto exit_inode;
 834         }
 835         brelse(dind);
 836
 837         o_group_desc = EXT4_SB(sb)->s_group_desc;
 838         memcpy(n_group_desc, o_group_desc,
 839                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 840         n_group_desc[gdb_num] = gdb_bh;
 841         EXT4_SB(sb)->s_group_desc = n_group_desc;
 842         EXT4_SB(sb)->s_gdb_count++;
 843         ext4_kvfree(o_group_desc);
 844
 845         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 846         err = ext4_handle_dirty_super(handle, sb);
 847         if (err)
 848                 ext4_std_error(sb, err);
 849
 850         return err;
 851
 852 exit_inode:
 853         ext4_kvfree(n_group_desc);
 854         brelse(iloc.bh);
 855 exit_dind:
 856         brelse(dind);
 857 exit_bh:
 858         brelse(gdb_bh);
 859
 860         ext4_debug("leaving with error %d\n", err);
 861         return err;
 862 }
 863
 864 /*
 865  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 866  */
 867 static int add_new_gdb_meta_bg(struct super_block *sb,
 868                                handle_t *handle, ext4_group_t group) {
 869         ext4_fsblk_t gdblock;
 870         struct buffer_head *gdb_bh;
 871         struct buffer_head **o_group_desc, **n_group_desc;
 872         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 873         int err;
 874
 875         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 876                    ext4_bg_has_super(sb, group);
 877         gdb_bh = sb_bread(sb, gdblock);
 878         if (!gdb_bh)
 879                 return -EIO;
 880         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 881                                      sizeof(struct buffer_head *),
 882                                      GFP_NOFS);
 883         if (!n_group_desc) {
 884                 err = -ENOMEM;
 885                 ext4_warning(sb, "not enough memory for %lu groups",
 886                              gdb_num + 1);
 887                 return err;
 888         }
 889
 890         o_group_desc = EXT4_SB(sb)->s_group_desc;
 891         memcpy(n_group_desc, o_group_desc,
 892                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 893         n_group_desc[gdb_num] = gdb_bh;
 894         EXT4_SB(sb)->s_group_desc = n_group_desc;
 895         EXT4_SB(sb)->s_gdb_count++;
 896         ext4_kvfree(o_group_desc);
 897         err = ext4_journal_get_write_access(handle, gdb_bh);
 898         if (unlikely(err))
 899                 brelse(gdb_bh);
 900         return err;
 901 }
 902
 903 /*
 904  * Called when we are adding a new group which has a backup copy of each of
 905  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 906  * We need to add these reserved backup GDT blocks to the resize inode, so
 907  * that they are kept for future resizing and not allocated to files.
 908  *
 909  * Each reserved backup GDT block will go into a different indirect block.
 910  * The indirect blocks are actually the primary reserved GDT blocks,
 911  * so we know in advance what their block numbers are.  We only get the
 912  * double-indirect block to verify it is pointing to the primary reserved
 913  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 914  * backup GDT blocks are stored in their reserved primary GDT block.
 915  */
 916 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 917                               ext4_group_t group)
 918 {
 919         struct super_block *sb = inode->i_sb;
 920         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 921         struct buffer_head **primary;
 922         struct buffer_head *dind;
 923         struct ext4_iloc iloc;
 924         ext4_fsblk_t blk;
 925         __le32 *data, *end;
 926         int gdbackups = 0;
 927         int res, i;
 928         int err;
 929
 930         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 931         if (!primary)
 932                 return -ENOMEM;
 933
 934         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 935         dind = sb_bread(sb, le32_to_cpu(*data));
 936         if (!dind) {
 937                 err = -EIO;
 938                 goto exit_free;
 939         }
 940
 941         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 942         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 943                                          EXT4_ADDR_PER_BLOCK(sb));
 944         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 945
 946         /* Get each reserved primary GDT block and verify it holds backups */
 947         for (res = 0; res < reserved_gdb; res++, blk++) {
 948                 if (le32_to_cpu(*data) != blk) {
 949                         ext4_warning(sb, "reserved block %llu"
 950                                      " not at offset %ld",
 951                                      blk,
 952                                      (long)(data - (__le32 *)dind->b_data));
 953                         err = -EINVAL;
 954                         goto exit_bh;
 955                 }
 956                 primary[res] = sb_bread(sb, blk);
 957                 if (!primary[res]) {
 958                         err = -EIO;
 959                         goto exit_bh;
 960                 }
 961                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 962                 if (gdbackups < 0) {
 963                         brelse(primary[res]);
 964                         err = gdbackups;
 965                         goto exit_bh;
 966                 }
 967                 if (++data >= end)
 968                         data = (__le32 *)dind->b_data;
 969         }
 970
 971         for (i = 0; i < reserved_gdb; i++) {
 972                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
 973                         goto exit_bh;
 974         }
 975
 976         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 977                 goto exit_bh;
 978
 979         /*
 980          * Finally we can add each of the reserved backup GDT blocks from
 981          * the new group to its reserved primary GDT block.
 982          */
 983         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
 984         for (i = 0; i < reserved_gdb; i++) {
 985                 int err2;
 986                 data = (__le32 *)primary[i]->b_data;
 987                 /* printk("reserving backup %lu[%u] = %lu\n",
 988                        primary[i]->b_blocknr, gdbackups,
 989                        blk + primary[i]->b_blocknr); */
 990                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 991                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 992                 if (!err)
 993                         err = err2;
 994         }
 995         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 996         ext4_mark_iloc_dirty(handle, inode, &iloc);
 997
 998 exit_bh:
 999         while (--res >= 0)
1000                 brelse(primary[res]);
1001         brelse(dind);
1002
1003 exit_free:
1004         kfree(primary);
1005
1006         return err;
1007 }
1008
1009 /*
1010  * Update the backup copies of the ext4 metadata.  These don't need to be part
1011  * of the main resize transaction, because e2fsck will re-write them if there
1012  * is a problem (basically only OOM will cause a problem).  However, we
1013  * _should_ update the backups if possible, in case the primary gets trashed
1014  * for some reason and we need to run e2fsck from a backup superblock.  The
1015  * important part is that the new block and inode counts are in the backup
1016  * superblocks, and the location of the new group metadata in the GDT backups.
1017  *
1018  * We do not need take the s_resize_lock for this, because these
1019  * blocks are not otherwise touched by the filesystem code when it is
1020  * mounted.  We don't need to worry about last changing from
1021  * sbi->s_groups_count, because the worst that can happen is that we
1022  * do not copy the full number of backups at this time.  The resize
1023  * which changed s_groups_count will backup again.
1024  */
1025 static void update_backups(struct super_block *sb, int blk_off, char *data,
1026                            int size, int meta_bg)
1027 {
1028         struct ext4_sb_info *sbi = EXT4_SB(sb);
1029         ext4_group_t last;
1030         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1031         unsigned three = 1;
1032         unsigned five = 5;
1033         unsigned seven = 7;
1034         ext4_group_t group = 0;
1035         int rest = sb->s_blocksize - size;
1036         handle_t *handle;
1037         int err = 0, err2;
1038
1039         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1040         if (IS_ERR(handle)) {
1041                 group = 1;
1042                 err = PTR_ERR(handle);
1043                 goto exit_err;
1044         }
1045
1046         if (meta_bg == 0) {
1047                 group = ext4_list_backups(sb, &three, &five, &seven);
1048                 last = sbi->s_groups_count;
1049         } else {
1050                 group = ext4_meta_bg_first_group(sb, group) + 1;
1051                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1052         }
1053
1054         while (group < sbi->s_groups_count) {
1055                 struct buffer_head *bh;
1056                 ext4_fsblk_t backup_block;
1057
1058                 /* Out of journal space, and can't get more - abort - so sad */
1059                 if (ext4_handle_valid(handle) &&
1060                     handle->h_buffer_credits == 0 &&
1061                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1062                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1063                         break;
1064
1065                 if (meta_bg == 0)
1066                         backup_block = group * bpg + blk_off;
1067                 else
1068                         backup_block = (ext4_group_first_block_no(sb, group) +
1069                                         ext4_bg_has_super(sb, group));
1070
1071                 bh = sb_getblk(sb, backup_block);
1072                 if (unlikely(!bh)) {
1073                         err = -ENOMEM;
1074                         break;
1075                 }
1076                 ext4_debug("update metadata backup %llu(+%llu)\n",
1077                            backup_block, backup_block -
1078                            ext4_group_first_block_no(sb, group));
1079                 if ((err = ext4_journal_get_write_access(handle, bh)))
1080                         break;
1081                 lock_buffer(bh);
1082                 memcpy(bh->b_data, data, size);
1083                 if (rest)
1084                         memset(bh->b_data + size, 0, rest);
1085                 set_buffer_uptodate(bh);
1086                 unlock_buffer(bh);
1087                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1088                 if (unlikely(err))
1089                         ext4_std_error(sb, err);
1090                 brelse(bh);
1091
1092                 if (meta_bg == 0)
1093                         group = ext4_list_backups(sb, &three, &five, &seven);
1094                 else if (group == last)
1095                         break;
1096                 else
1097                         group = last;
1098         }
1099         if ((err2 = ext4_journal_stop(handle)) && !err)
1100                 err = err2;
1101
1102         /*
1103          * Ugh! Need to have e2fsck write the backup copies.  It is too
1104          * late to revert the resize, we shouldn't fail just because of
1105          * the backup copies (they are only needed in case of corruption).
1106          *
1107          * However, if we got here we have a journal problem too, so we
1108          * can't really start a transaction to mark the superblock.
1109          * Chicken out and just set the flag on the hope it will be written
1110          * to disk, and if not - we will simply wait until next fsck.
1111          */
1112 exit_err:
1113         if (err) {
1114                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1115                              "forcing fsck on next reboot", group, err);
1116                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1117                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1118                 mark_buffer_dirty(sbi->s_sbh);
1119         }
1120 }
1121
1122 /*
1123  * ext4_add_new_descs() adds @count group descriptor of groups
1124  * starting at @group
1125  *
1126  * @handle: journal handle
1127  * @sb: super block
1128  * @group: the group no. of the first group desc to be added
1129  * @resize_inode: the resize inode
1130  * @count: number of group descriptors to be added
1131  */
1132 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1133                               ext4_group_t group, struct inode *resize_inode,
1134                               ext4_group_t count)
1135 {
1136         struct ext4_sb_info *sbi = EXT4_SB(sb);
1137         struct ext4_super_block *es = sbi->s_es;
1138         struct buffer_head *gdb_bh;
1139         int i, gdb_off, gdb_num, err = 0;
1140         int meta_bg;
1141
1142         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1143         for (i = 0; i < count; i++, group++) {
1144                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1145                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1146
1147                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1148                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1149
1150                 /*
1151                  * We will only either add reserved group blocks to a backup group
1152                  * or remove reserved blocks for the first group in a new group block.
1153                  * Doing both would be mean more complex code, and sane people don't
1154                  * use non-sparse filesystems anymore.  This is already checked above.
1155                  */
1156                 if (gdb_off) {
1157                         gdb_bh = sbi->s_group_desc[gdb_num];
1158                         err = ext4_journal_get_write_access(handle, gdb_bh);
1159
1160                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1161                                 err = reserve_backup_gdb(handle, resize_inode, group);
1162                 } else if (meta_bg != 0) {
1163                         err = add_new_gdb_meta_bg(sb, handle, group);
1164                 } else {
1165                         err = add_new_gdb(handle, resize_inode, group);
1166                 }
1167                 if (err)
1168                         break;
1169         }
1170         return err;
1171 }
1172
1173 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1174 {
1175         struct buffer_head *bh = sb_getblk(sb, block);
1176         if (unlikely(!bh))
1177                 return NULL;
1178         if (!bh_uptodate_or_lock(bh)) {
1179                 if (bh_submit_read(bh) < 0) {
1180                         brelse(bh);
1181                         return NULL;
1182                 }
1183         }
1184
1185         return bh;
1186 }
1187
1188 static int ext4_set_bitmap_checksums(struct super_block *sb,
1189                                      ext4_group_t group,
1190                                      struct ext4_group_desc *gdp,
1191                                      struct ext4_new_group_data *group_data)
1192 {
1193         struct buffer_head *bh;
1194
1195         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1196                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1197                 return 0;
1198
1199         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1200         if (!bh)
1201                 return -EIO;
1202         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1203                                    EXT4_INODES_PER_GROUP(sb) / 8);
1204         brelse(bh);
1205
1206         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1207         if (!bh)
1208                 return -EIO;
1209         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1210         brelse(bh);
1211
1212         return 0;
1213 }
1214
1215 /*
1216  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1217  */
1218 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1219                                 struct ext4_new_flex_group_data *flex_gd)
1220 {
1221         struct ext4_new_group_data      *group_data = flex_gd->groups;
1222         struct ext4_group_desc          *gdp;
1223         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1224         struct buffer_head              *gdb_bh;
1225         ext4_group_t                    group;
1226         __u16                           *bg_flags = flex_gd->bg_flags;
1227         int                             i, gdb_off, gdb_num, err = 0;
1228
1229
1230         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1231                 group = group_data->group;
1232
1233                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1234                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1235
1236                 /*
1237                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1238                  */
1239                 gdb_bh = sbi->s_group_desc[gdb_num];
1240                 /* Update group descriptor block for new group */
1241                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1242                                                  gdb_off * EXT4_DESC_SIZE(sb));
1243
1244                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1245                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1246                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1247                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1248                 if (err) {
1249                         ext4_std_error(sb, err);
1250                         break;
1251                 }
1252
1253                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1254                 ext4_free_group_clusters_set(sb, gdp,
1255                         EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1256                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1257                 if (ext4_has_group_desc_csum(sb))
1258                         ext4_itable_unused_set(sb, gdp,
1259                                                EXT4_INODES_PER_GROUP(sb));
1260                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1261                 ext4_group_desc_csum_set(sb, group, gdp);
1262
1263                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1264                 if (unlikely(err)) {
1265                         ext4_std_error(sb, err);
1266                         break;
1267                 }
1268
1269                 /*
1270                  * We can allocate memory for mb_alloc based on the new group
1271                  * descriptor
1272                  */
1273                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1274                 if (err)
1275                         break;
1276         }
1277         return err;
1278 }
1279
1280 /*
1281  * ext4_update_super() updates the super block so that the newly added
1282  * groups can be seen by the filesystem.
1283  *
1284  * @sb: super block
1285  * @flex_gd: new added groups
1286  */
1287 static void ext4_update_super(struct super_block *sb,
1288                              struct ext4_new_flex_group_data *flex_gd)
1289 {
1290         ext4_fsblk_t blocks_count = 0;
1291         ext4_fsblk_t free_blocks = 0;
1292         ext4_fsblk_t reserved_blocks = 0;
1293         struct ext4_new_group_data *group_data = flex_gd->groups;
1294         struct ext4_sb_info *sbi = EXT4_SB(sb);
1295         struct ext4_super_block *es = sbi->s_es;
1296         int i;
1297
1298         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1299         /*
1300          * Make the new blocks and inodes valid next.  We do this before
1301          * increasing the group count so that once the group is enabled,
1302          * all of its blocks and inodes are already valid.
1303          *
1304          * We always allocate group-by-group, then block-by-block or
1305          * inode-by-inode within a group, so enabling these
1306          * blocks/inodes before the group is live won't actually let us
1307          * allocate the new space yet.
1308          */
1309         for (i = 0; i < flex_gd->count; i++) {
1310                 blocks_count += group_data[i].blocks_count;
1311                 free_blocks += group_data[i].free_blocks_count;
1312         }
1313
1314         reserved_blocks = ext4_r_blocks_count(es) * 100;
1315         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1316         reserved_blocks *= blocks_count;
1317         do_div(reserved_blocks, 100);
1318
1319         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1320         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1321         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1322                      flex_gd->count);
1323         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1324                      flex_gd->count);
1325
1326         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1327         /*
1328          * We need to protect s_groups_count against other CPUs seeing
1329          * inconsistent state in the superblock.
1330          *
1331          * The precise rules we use are:
1332          *
1333          * * Writers must perform a smp_wmb() after updating all
1334          *   dependent data and before modifying the groups count
1335          *
1336          * * Readers must perform an smp_rmb() after reading the groups
1337          *   count and before reading any dependent data.
1338          *
1339          * NB. These rules can be relaxed when checking the group count
1340          * while freeing data, as we can only allocate from a block
1341          * group after serialising against the group count, and we can
1342          * only then free after serialising in turn against that
1343          * allocation.
1344          */
1345         smp_wmb();
1346
1347         /* Update the global fs size fields */
1348         sbi->s_groups_count += flex_gd->count;
1349         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1350                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1351
1352         /* Update the reserved block counts only once the new group is
1353          * active. */
1354         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1355                                 reserved_blocks);
1356
1357         /* Update the free space counts */
1358         percpu_counter_add(&sbi->s_freeclusters_counter,
1359                            EXT4_NUM_B2C(sbi, free_blocks));
1360         percpu_counter_add(&sbi->s_freeinodes_counter,
1361                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1362
1363         ext4_debug("free blocks count %llu",
1364                    percpu_counter_read(&sbi->s_freeclusters_counter));
1365         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1366                                       EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1367             sbi->s_log_groups_per_flex) {
1368                 ext4_group_t flex_group;
1369                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1370                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1371                              &sbi->s_flex_groups[flex_group].free_clusters);
1372                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1373                            &sbi->s_flex_groups[flex_group].free_inodes);
1374         }
1375
1376         /*
1377          * Update the fs overhead information
1378          */
1379         ext4_calculate_overhead(sb);
1380
1381         if (test_opt(sb, DEBUG))
1382                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1383                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1384                        blocks_count, free_blocks, reserved_blocks);
1385 }
1386
1387 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1388  * _before_ we start modifying the filesystem, because we cannot abort the
1389  * transaction and not have it write the data to disk.
1390  */
1391 static int ext4_flex_group_add(struct super_block *sb,
1392                                struct inode *resize_inode,
1393                                struct ext4_new_flex_group_data *flex_gd)
1394 {
1395         struct ext4_sb_info *sbi = EXT4_SB(sb);
1396         struct ext4_super_block *es = sbi->s_es;
1397         ext4_fsblk_t o_blocks_count;
1398         ext4_grpblk_t last;
1399         ext4_group_t group;
1400         handle_t *handle;
1401         unsigned reserved_gdb;
1402         int err = 0, err2 = 0, credit;
1403
1404         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1405
1406         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1407         o_blocks_count = ext4_blocks_count(es);
1408         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1409         BUG_ON(last);
1410
1411         err = setup_new_flex_group_blocks(sb, flex_gd);
1412         if (err)
1413                 goto exit;
1414         /*
1415          * We will always be modifying at least the superblock and  GDT
1416          * block.  If we are adding a group past the last current GDT block,
1417          * we will also modify the inode and the dindirect block.  If we
1418          * are adding a group with superblock/GDT backups  we will also
1419          * modify each of the reserved GDT dindirect blocks.
1420          */
1421         credit = flex_gd->count * 4 + reserved_gdb;
1422         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1423         if (IS_ERR(handle)) {
1424                 err = PTR_ERR(handle);
1425                 goto exit;
1426         }
1427
1428         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1429         if (err)
1430                 goto exit_journal;
1431
1432         group = flex_gd->groups[0].group;
1433         BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1434         err = ext4_add_new_descs(handle, sb, group,
1435                                 resize_inode, flex_gd->count);
1436         if (err)
1437                 goto exit_journal;
1438
1439         err = ext4_setup_new_descs(handle, sb, flex_gd);
1440         if (err)
1441                 goto exit_journal;
1442
1443         ext4_update_super(sb, flex_gd);
1444
1445         err = ext4_handle_dirty_super(handle, sb);
1446
1447 exit_journal:
1448         err2 = ext4_journal_stop(handle);
1449         if (!err)
1450                 err = err2;
1451
1452         if (!err) {
1453                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1454                 int gdb_num_end = ((group + flex_gd->count - 1) /
1455                                    EXT4_DESC_PER_BLOCK(sb));
1456                 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1457                                 EXT4_FEATURE_INCOMPAT_META_BG);
1458                 sector_t old_gdb = 0;
1459
1460                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1461                                sizeof(struct ext4_super_block), 0);
1462                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1463                         struct buffer_head *gdb_bh;
1464
1465                         gdb_bh = sbi->s_group_desc[gdb_num];
1466                         if (old_gdb == gdb_bh->b_blocknr)
1467                                 continue;
1468                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1469                                        gdb_bh->b_size, meta_bg);
1470                         old_gdb = gdb_bh->b_blocknr;
1471                 }
1472         }
1473 exit:
1474         return err;
1475 }
1476
1477 static int ext4_setup_next_flex_gd(struct super_block *sb,
1478                                     struct ext4_new_flex_group_data *flex_gd,
1479                                     ext4_fsblk_t n_blocks_count,
1480                                     unsigned long flexbg_size)
1481 {
1482         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1483         struct ext4_new_group_data *group_data = flex_gd->groups;
1484         ext4_fsblk_t o_blocks_count;
1485         ext4_group_t n_group;
1486         ext4_group_t group;
1487         ext4_group_t last_group;
1488         ext4_grpblk_t last;
1489         ext4_grpblk_t blocks_per_group;
1490         unsigned long i;
1491
1492         blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1493
1494         o_blocks_count = ext4_blocks_count(es);
1495
1496         if (o_blocks_count == n_blocks_count)
1497                 return 0;
1498
1499         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1500         BUG_ON(last);
1501         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1502
1503         last_group = group | (flexbg_size - 1);
1504         if (last_group > n_group)
1505                 last_group = n_group;
1506
1507         flex_gd->count = last_group - group + 1;
1508
1509         for (i = 0; i < flex_gd->count; i++) {
1510                 int overhead;
1511
1512                 group_data[i].group = group + i;
1513                 group_data[i].blocks_count = blocks_per_group;
1514                 overhead = ext4_group_overhead_blocks(sb, group + i);
1515                 group_data[i].free_blocks_count = blocks_per_group - overhead;
1516                 if (ext4_has_group_desc_csum(sb)) {
1517                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1518                                                EXT4_BG_INODE_UNINIT;
1519                         if (!test_opt(sb, INIT_INODE_TABLE))
1520                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1521                 } else
1522                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1523         }
1524
1525         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1526                 /* We need to initialize block bitmap of last group. */
1527                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1528
1529         if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1530                 group_data[i - 1].blocks_count = last + 1;
1531                 group_data[i - 1].free_blocks_count -= blocks_per_group-
1532                                         last - 1;
1533         }
1534
1535         return 1;
1536 }
1537
1538 /* Add group descriptor data to an existing or new group descriptor block.
1539  * Ensure we handle all possible error conditions _before_ we start modifying
1540  * the filesystem, because we cannot abort the transaction and not have it
1541  * write the data to disk.
1542  *
1543  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1544  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1545  *
1546  * We only need to hold the superblock lock while we are actually adding
1547  * in the new group's counts to the superblock.  Prior to that we have
1548  * not really "added" the group at all.  We re-check that we are still
1549  * adding in the last group in case things have changed since verifying.
1550  */
1551 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1552 {
1553         struct ext4_new_flex_group_data flex_gd;
1554         struct ext4_sb_info *sbi = EXT4_SB(sb);
1555         struct ext4_super_block *es = sbi->s_es;
1556         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1557                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1558         struct inode *inode = NULL;
1559         int gdb_off;
1560         int err;
1561         __u16 bg_flags = 0;
1562
1563         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1564
1565         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1566                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1567                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1568                 return -EPERM;
1569         }
1570
1571         if (ext4_blocks_count(es) + input->blocks_count <
1572             ext4_blocks_count(es)) {
1573                 ext4_warning(sb, "blocks_count overflow");
1574                 return -EINVAL;
1575         }
1576
1577         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1578             le32_to_cpu(es->s_inodes_count)) {
1579                 ext4_warning(sb, "inodes_count overflow");
1580                 return -EINVAL;
1581         }
1582
1583         if (reserved_gdb || gdb_off == 0) {
1584                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1585                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
1586                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1587                         ext4_warning(sb,
1588                                      "No reserved GDT blocks, can't resize");
1589                         return -EPERM;
1590                 }
1591                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1592                 if (IS_ERR(inode)) {
1593                         ext4_warning(sb, "Error opening resize inode");
1594                         return PTR_ERR(inode);
1595                 }
1596         }
1597
1598
1599         err = verify_group_input(sb, input);
1600         if (err)
1601                 goto out;
1602
1603         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1604         if (err)
1605                 goto out;
1606
1607         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1608         if (err)
1609                 goto out;
1610
1611         flex_gd.count = 1;
1612         flex_gd.groups = input;
1613         flex_gd.bg_flags = &bg_flags;
1614         err = ext4_flex_group_add(sb, inode, &flex_gd);
1615 out:
1616         iput(inode);
1617         return err;
1618 } /* ext4_group_add */
1619
1620 /*
1621  * extend a group without checking assuming that checking has been done.
1622  */
1623 static int ext4_group_extend_no_check(struct super_block *sb,
1624                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1625 {
1626         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1627         handle_t *handle;
1628         int err = 0, err2;
1629
1630         /* We will update the superblock, one block bitmap, and
1631          * one group descriptor via ext4_group_add_blocks().
1632          */
1633         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1634         if (IS_ERR(handle)) {
1635                 err = PTR_ERR(handle);
1636                 ext4_warning(sb, "error %d on journal start", err);
1637                 return err;
1638         }
1639
1640         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1641         if (err) {
1642                 ext4_warning(sb, "error %d on journal write access", err);
1643                 goto errout;
1644         }
1645
1646         ext4_blocks_count_set(es, o_blocks_count + add);
1647         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1648         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1649                    o_blocks_count + add);
1650         /* We add the blocks to the bitmap and set the group need init bit */
1651         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1652         if (err)
1653                 goto errout;
1654         ext4_handle_dirty_super(handle, sb);
1655         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1656                    o_blocks_count + add);
1657 errout:
1658         err2 = ext4_journal_stop(handle);
1659         if (err2 && !err)
1660                 err = err2;
1661
1662         if (!err) {
1663                 if (test_opt(sb, DEBUG))
1664                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1665                                "blocks\n", ext4_blocks_count(es));
1666                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1667                                (char *)es, sizeof(struct ext4_super_block), 0);
1668         }
1669         return err;
1670 }
1671
1672 /*
1673  * Extend the filesystem to the new number of blocks specified.  This entry
1674  * point is only used to extend the current filesystem to the end of the last
1675  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1676  * for emergencies (because it has no dependencies on reserved blocks).
1677  *
1678  * If we _really_ wanted, we could use default values to call ext4_group_add()
1679  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1680  * GDT blocks are reserved to grow to the desired size.
1681  */
1682 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1683                       ext4_fsblk_t n_blocks_count)
1684 {
1685         ext4_fsblk_t o_blocks_count;
1686         ext4_grpblk_t last;
1687         ext4_grpblk_t add;
1688         struct buffer_head *bh;
1689         int err;
1690         ext4_group_t group;
1691
1692         o_blocks_count = ext4_blocks_count(es);
1693
1694         if (test_opt(sb, DEBUG))
1695                 ext4_msg(sb, KERN_DEBUG,
1696                          "extending last group from %llu to %llu blocks",
1697                          o_blocks_count, n_blocks_count);
1698
1699         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1700                 return 0;
1701
1702         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1703                 ext4_msg(sb, KERN_ERR,
1704                          "filesystem too large to resize to %llu blocks safely",
1705                          n_blocks_count);
1706                 if (sizeof(sector_t) < 8)
1707                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1708                 return -EINVAL;
1709         }
1710
1711         if (n_blocks_count < o_blocks_count) {
1712                 ext4_warning(sb, "can't shrink FS - resize aborted");
1713                 return -EINVAL;
1714         }
1715
1716         /* Handle the remaining blocks in the last group only. */
1717         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1718
1719         if (last == 0) {
1720                 ext4_warning(sb, "need to use ext2online to resize further");
1721                 return -EPERM;
1722         }
1723
1724         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1725
1726         if (o_blocks_count + add < o_blocks_count) {
1727                 ext4_warning(sb, "blocks_count overflow");
1728                 return -EINVAL;
1729         }
1730
1731         if (o_blocks_count + add > n_blocks_count)
1732                 add = n_blocks_count - o_blocks_count;
1733
1734         if (o_blocks_count + add < n_blocks_count)
1735                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1736                              o_blocks_count + add, add);
1737
1738         /* See if the device is actually as big as what was requested */
1739         bh = sb_bread(sb, o_blocks_count + add - 1);
1740         if (!bh) {
1741                 ext4_warning(sb, "can't read last block, resize aborted");
1742                 return -ENOSPC;
1743         }
1744         brelse(bh);
1745
1746         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1747         return err;
1748 } /* ext4_group_extend */
1749
1750
1751 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1752 {
1753         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1754 }
1755
1756 /*
1757  * Release the resize inode and drop the resize_inode feature if there
1758  * are no more reserved gdt blocks, and then convert the file system
1759  * to enable meta_bg
1760  */
1761 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1762 {
1763         handle_t *handle;
1764         struct ext4_sb_info *sbi = EXT4_SB(sb);
1765         struct ext4_super_block *es = sbi->s_es;
1766         struct ext4_inode_info *ei = EXT4_I(inode);
1767         ext4_fsblk_t nr;
1768         int i, ret, err = 0;
1769         int credits = 1;
1770
1771         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1772         if (inode) {
1773                 if (es->s_reserved_gdt_blocks) {
1774                         ext4_error(sb, "Unexpected non-zero "
1775                                    "s_reserved_gdt_blocks");
1776                         return -EPERM;
1777                 }
1778
1779                 /* Do a quick sanity check of the resize inode */
1780                 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1781                         goto invalid_resize_inode;
1782                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1783                         if (i == EXT4_DIND_BLOCK) {
1784                                 if (ei->i_data[i])
1785                                         continue;
1786                                 else
1787                                         goto invalid_resize_inode;
1788                         }
1789                         if (ei->i_data[i])
1790                                 goto invalid_resize_inode;
1791                 }
1792                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1793         }
1794
1795         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1796         if (IS_ERR(handle))
1797                 return PTR_ERR(handle);
1798
1799         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1800         if (err)
1801                 goto errout;
1802
1803         EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1804         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1805         sbi->s_es->s_first_meta_bg =
1806                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1807
1808         err = ext4_handle_dirty_super(handle, sb);
1809         if (err) {
1810                 ext4_std_error(sb, err);
1811                 goto errout;
1812         }
1813
1814         if (inode) {
1815                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1816                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1817                                  EXT4_FREE_BLOCKS_METADATA |
1818                                  EXT4_FREE_BLOCKS_FORGET);
1819                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1820                 inode->i_blocks = 0;
1821
1822                 err = ext4_mark_inode_dirty(handle, inode);
1823                 if (err)
1824                         ext4_std_error(sb, err);
1825         }
1826
1827 errout:
1828         ret = ext4_journal_stop(handle);
1829         if (!err)
1830                 err = ret;
1831         return ret;
1832
1833 invalid_resize_inode:
1834         ext4_error(sb, "corrupted/inconsistent resize inode");
1835         return -EINVAL;
1836 }
1837
1838 /*
1839  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1840  *
1841  * @sb: super block of the fs to be resized
1842  * @n_blocks_count: the number of blocks resides in the resized fs
1843  */
1844 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1845 {
1846         struct ext4_new_flex_group_data *flex_gd = NULL;
1847         struct ext4_sb_info *sbi = EXT4_SB(sb);
1848         struct ext4_super_block *es = sbi->s_es;
1849         struct buffer_head *bh;
1850         struct inode *resize_inode = NULL;
1851         ext4_grpblk_t add, offset;
1852         unsigned long n_desc_blocks;
1853         unsigned long o_desc_blocks;
1854         ext4_group_t o_group;
1855         ext4_group_t n_group;
1856         ext4_fsblk_t o_blocks_count;
1857         ext4_fsblk_t n_blocks_count_retry = 0;
1858         unsigned long last_update_time = 0;
1859         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1860         int meta_bg;
1861
1862         /* See if the device is actually as big as what was requested */
1863         bh = sb_bread(sb, n_blocks_count - 1);
1864         if (!bh) {
1865                 ext4_warning(sb, "can't read last block, resize aborted");
1866                 return -ENOSPC;
1867         }
1868         brelse(bh);
1869
1870 retry:
1871         o_blocks_count = ext4_blocks_count(es);
1872
1873         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1874                  "to %llu blocks", o_blocks_count, n_blocks_count);
1875
1876         if (n_blocks_count < o_blocks_count) {
1877                 /* On-line shrinking not supported */
1878                 ext4_warning(sb, "can't shrink FS - resize aborted");
1879                 return -EINVAL;
1880         }
1881
1882         if (n_blocks_count == o_blocks_count)
1883                 /* Nothing need to do */
1884                 return 0;
1885
1886         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1887         if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1888                 ext4_warning(sb, "resize would cause inodes_count overflow");
1889                 return -EINVAL;
1890         }
1891         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1892
1893         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1894         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1895
1896         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1897
1898         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1899                 if (meta_bg) {
1900                         ext4_error(sb, "resize_inode and meta_bg enabled "
1901                                    "simultaneously");
1902                         return -EINVAL;
1903                 }
1904                 if (n_desc_blocks > o_desc_blocks +
1905                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
1906                         n_blocks_count_retry = n_blocks_count;
1907                         n_desc_blocks = o_desc_blocks +
1908                                 le16_to_cpu(es->s_reserved_gdt_blocks);
1909                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1910                         n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1911                         n_group--; /* set to last group number */
1912                 }
1913
1914                 if (!resize_inode)
1915                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1916                 if (IS_ERR(resize_inode)) {
1917                         ext4_warning(sb, "Error opening resize inode");
1918                         return PTR_ERR(resize_inode);
1919                 }
1920         }
1921
1922         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1923                 err = ext4_convert_meta_bg(sb, resize_inode);
1924                 if (err)
1925                         goto out;
1926                 if (resize_inode) {
1927                         iput(resize_inode);
1928                         resize_inode = NULL;
1929                 }
1930                 if (n_blocks_count_retry) {
1931                         n_blocks_count = n_blocks_count_retry;
1932                         n_blocks_count_retry = 0;
1933                         goto retry;
1934                 }
1935         }
1936
1937         /* extend the last group */
1938         if (n_group == o_group)
1939                 add = n_blocks_count - o_blocks_count;
1940         else
1941                 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1942         if (add > 0) {
1943                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1944                 if (err)
1945                         goto out;
1946         }
1947
1948         if (ext4_blocks_count(es) == n_blocks_count)
1949                 goto out;
1950
1951         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1952         if (err)
1953                 return err;
1954
1955         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1956         if (err)
1957                 goto out;
1958
1959         flex_gd = alloc_flex_gd(flexbg_size);
1960         if (flex_gd == NULL) {
1961                 err = -ENOMEM;
1962                 goto out;
1963         }
1964
1965         /* Add flex groups. Note that a regular group is a
1966          * flex group with 1 group.
1967          */
1968         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1969                                               flexbg_size)) {
1970                 if (jiffies - last_update_time > HZ * 10) {
1971                         if (last_update_time)
1972                                 ext4_msg(sb, KERN_INFO,
1973                                          "resized to %llu blocks",
1974                                          ext4_blocks_count(es));
1975                         last_update_time = jiffies;
1976                 }
1977                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1978                         break;
1979                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1980                 if (unlikely(err))
1981                         break;
1982         }
1983
1984         if (!err && n_blocks_count_retry) {
1985                 n_blocks_count = n_blocks_count_retry;
1986                 n_blocks_count_retry = 0;
1987                 free_flex_gd(flex_gd);
1988                 flex_gd = NULL;
1989                 goto retry;
1990         }
1991
1992 out:
1993         if (flex_gd)
1994                 free_flex_gd(flex_gd);
1995         if (resize_inode != NULL)
1996                 iput(resize_inode);
1997         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
1998         return err;
1999 }