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 = ext4_group_overhead_blocks(sb, group);
  83         ext4_fsblk_t metaend = start + overhead;
  84         struct buffer_head *bh = NULL;
  85         ext4_grpblk_t free_blocks_count, offset;
  86         int err = -EINVAL;
  87
  88         input->free_blocks_count = free_blocks_count =
  89                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  90
  91         if (test_opt(sb, DEBUG))
  92                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
  93                        "(%d free, %u reserved)\n",
  94                        ext4_bg_has_super(sb, input->group) ? "normal" :
  95                        "no-super", input->group, input->blocks_count,
  96                        free_blocks_count, input->reserved_blocks);
  97
  98         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
  99         if (group != sbi->s_groups_count)
 100                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 101                              input->group, sbi->s_groups_count);
 102         else if (offset != 0)
 103                         ext4_warning(sb, "Last group not full");
 104         else if (input->reserved_blocks > input->blocks_count / 5)
 105                 ext4_warning(sb, "Reserved blocks too high (%u)",
 106                              input->reserved_blocks);
 107         else if (free_blocks_count < 0)
 108                 ext4_warning(sb, "Bad blocks count %u",
 109                              input->blocks_count);
 110         else if (!(bh = sb_bread(sb, end - 1)))
 111                 ext4_warning(sb, "Cannot read last block (%llu)",
 112                              end - 1);
 113         else if (outside(input->block_bitmap, start, end))
 114                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 115                              (unsigned long long)input->block_bitmap);
 116         else if (outside(input->inode_bitmap, start, end))
 117                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 118                              (unsigned long long)input->inode_bitmap);
 119         else if (outside(input->inode_table, start, end) ||
 120                  outside(itend - 1, start, end))
 121                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 122                              (unsigned long long)input->inode_table, itend - 1);
 123         else if (input->inode_bitmap == input->block_bitmap)
 124                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 125                              (unsigned long long)input->block_bitmap);
 126         else if (inside(input->block_bitmap, input->inode_table, itend))
 127                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 128                              "(%llu-%llu)",
 129                              (unsigned long long)input->block_bitmap,
 130                              (unsigned long long)input->inode_table, itend - 1);
 131         else if (inside(input->inode_bitmap, input->inode_table, itend))
 132                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 133                              "(%llu-%llu)",
 134                              (unsigned long long)input->inode_bitmap,
 135                              (unsigned long long)input->inode_table, itend - 1);
 136         else if (inside(input->block_bitmap, start, metaend))
 137                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 138                              (unsigned long long)input->block_bitmap,
 139                              start, metaend - 1);
 140         else if (inside(input->inode_bitmap, start, metaend))
 141                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 142                              (unsigned long long)input->inode_bitmap,
 143                              start, metaend - 1);
 144         else if (inside(input->inode_table, start, metaend) ||
 145                  inside(itend - 1, start, metaend))
 146                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 147                              "(%llu-%llu)",
 148                              (unsigned long long)input->inode_table,
 149                              itend - 1, start, metaend - 1);
 150         else
 151                 err = 0;
 152         brelse(bh);
 153
 154         return err;
 155 }
 156
 157 /*
 158  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 159  * group each time.
 160  */
 161 struct ext4_new_flex_group_data {
 162         struct ext4_new_group_data *groups;     /* new_group_data for groups
 163                                                    in the flex group */
 164         __u16 *bg_flags;                        /* block group flags of groups
 165                                                    in @groups */
 166         ext4_group_t count;                     /* number of groups in @groups
 167                                                  */
 168 };
 169
 170 /*
 171  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 172  * @flexbg_size.
 173  *
 174  * Returns NULL on failure otherwise address of the allocated structure.
 175  */
 176 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 177 {
 178         struct ext4_new_flex_group_data *flex_gd;
 179
 180         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 181         if (flex_gd == NULL)
 182                 goto out3;
 183
 184         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
 185                 goto out2;
 186         flex_gd->count = flexbg_size;
 187
 188         flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
 189                                   flexbg_size, GFP_NOFS);
 190         if (flex_gd->groups == NULL)
 191                 goto out2;
 192
 193         flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
 194         if (flex_gd->bg_flags == NULL)
 195                 goto out1;
 196
 197         return flex_gd;
 198
 199 out1:
 200         kfree(flex_gd->groups);
 201 out2:
 202         kfree(flex_gd);
 203 out3:
 204         return NULL;
 205 }
 206
 207 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 208 {
 209         kfree(flex_gd->bg_flags);
 210         kfree(flex_gd->groups);
 211         kfree(flex_gd);
 212 }
 213
 214 /*
 215  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 216  * and inode tables for a flex group.
 217  *
 218  * This function is used by 64bit-resize.  Note that this function allocates
 219  * group tables from the 1st group of groups contained by @flexgd, which may
 220  * be a partial of a flex group.
 221  *
 222  * @sb: super block of fs to which the groups belongs
 223  *
 224  * Returns 0 on a successful allocation of the metadata blocks in the
 225  * block group.
 226  */
 227 static int ext4_alloc_group_tables(struct super_block *sb,
 228                                 struct ext4_new_flex_group_data *flex_gd,
 229                                 int flexbg_size)
 230 {
 231         struct ext4_new_group_data *group_data = flex_gd->groups;
 232         ext4_fsblk_t start_blk;
 233         ext4_fsblk_t last_blk;
 234         ext4_group_t src_group;
 235         ext4_group_t bb_index = 0;
 236         ext4_group_t ib_index = 0;
 237         ext4_group_t it_index = 0;
 238         ext4_group_t group;
 239         ext4_group_t last_group;
 240         unsigned overhead;
 241
 242         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 243
 244         src_group = group_data[0].group;
 245         last_group  = src_group + flex_gd->count - 1;
 246
 247         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 248                (last_group & ~(flexbg_size - 1))));
 249 next_group:
 250         group = group_data[0].group;
 251         if (src_group >= group_data[0].group + flex_gd->count)
 252                 return -ENOSPC;
 253         start_blk = ext4_group_first_block_no(sb, src_group);
 254         last_blk = start_blk + group_data[src_group - group].blocks_count;
 255
 256         overhead = ext4_group_overhead_blocks(sb, src_group);
 257
 258         start_blk += overhead;
 259
 260         /* We collect contiguous blocks as much as possible. */
 261         src_group++;
 262         for (; src_group <= last_group; src_group++) {
 263                 overhead = ext4_group_overhead_blocks(sb, src_group);
 264                 if (overhead != 0)
 265                         last_blk += group_data[src_group - group].blocks_count;
 266                 else
 267                         break;
 268         }
 269
 270         /* Allocate block bitmaps */
 271         for (; bb_index < flex_gd->count; bb_index++) {
 272                 if (start_blk >= last_blk)
 273                         goto next_group;
 274                 group_data[bb_index].block_bitmap = start_blk++;
 275                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 276                 group -= group_data[0].group;
 277                 group_data[group].free_blocks_count--;
 278                 if (flexbg_size > 1)
 279                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 280         }
 281
 282         /* Allocate inode bitmaps */
 283         for (; ib_index < flex_gd->count; ib_index++) {
 284                 if (start_blk >= last_blk)
 285                         goto next_group;
 286                 group_data[ib_index].inode_bitmap = start_blk++;
 287                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 288                 group -= group_data[0].group;
 289                 group_data[group].free_blocks_count--;
 290                 if (flexbg_size > 1)
 291                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 292         }
 293
 294         /* Allocate inode tables */
 295         for (; it_index < flex_gd->count; it_index++) {
 296                 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
 297                         goto next_group;
 298                 group_data[it_index].inode_table = start_blk;
 299                 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
 300                 group -= group_data[0].group;
 301                 group_data[group].free_blocks_count -=
 302                                         EXT4_SB(sb)->s_itb_per_group;
 303                 if (flexbg_size > 1)
 304                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 305
 306                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 307         }
 308
 309         if (test_opt(sb, DEBUG)) {
 310                 int i;
 311                 group = group_data[0].group;
 312
 313                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 314                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 315                        flexbg_size);
 316
 317                 for (i = 0; i < flex_gd->count; i++) {
 318                         printk(KERN_DEBUG "adding %s group %u: %u "
 319                                "blocks (%d free)\n",
 320                                ext4_bg_has_super(sb, group + i) ? "normal" :
 321                                "no-super", group + i,
 322                                group_data[i].blocks_count,
 323                                group_data[i].free_blocks_count);
 324                 }
 325         }
 326         return 0;
 327 }
 328
 329 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 330                                   ext4_fsblk_t blk)
 331 {
 332         struct buffer_head *bh;
 333         int err;
 334
 335         bh = sb_getblk(sb, blk);
 336         if (!bh)
 337                 return ERR_PTR(-EIO);
 338         if ((err = ext4_journal_get_write_access(handle, bh))) {
 339                 brelse(bh);
 340                 bh = ERR_PTR(err);
 341         } else {
 342                 memset(bh->b_data, 0, sb->s_blocksize);
 343                 set_buffer_uptodate(bh);
 344         }
 345
 346         return bh;
 347 }
 348
 349 /*
 350  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 351  * If that fails, restart the transaction & regain write access for the
 352  * buffer head which is used for block_bitmap modifications.
 353  */
 354 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 355 {
 356         int err;
 357
 358         if (ext4_handle_has_enough_credits(handle, thresh))
 359                 return 0;
 360
 361         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 362         if (err < 0)
 363                 return err;
 364         if (err) {
 365                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 366                 if (err)
 367                         return err;
 368         }
 369
 370         return 0;
 371 }
 372
 373 /*
 374  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 375  *
 376  * Helper function for ext4_setup_new_group_blocks() which set .
 377  *
 378  * @sb: super block
 379  * @handle: journal handle
 380  * @flex_gd: flex group data
 381  */
 382 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 383                         struct ext4_new_flex_group_data *flex_gd,
 384                         ext4_fsblk_t block, ext4_group_t count)
 385 {
 386         ext4_group_t count2;
 387
 388         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 389         for (count2 = count; count > 0; count -= count2, block += count2) {
 390                 ext4_fsblk_t start;
 391                 struct buffer_head *bh;
 392                 ext4_group_t group;
 393                 int err;
 394
 395                 ext4_get_group_no_and_offset(sb, block, &group, NULL);
 396                 start = ext4_group_first_block_no(sb, group);
 397                 group -= flex_gd->groups[0].group;
 398
 399                 count2 = sb->s_blocksize * 8 - (block - start);
 400                 if (count2 > count)
 401                         count2 = count;
 402
 403                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 404                         BUG_ON(flex_gd->count > 1);
 405                         continue;
 406                 }
 407
 408                 err = extend_or_restart_transaction(handle, 1);
 409                 if (err)
 410                         return err;
 411
 412                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 413                 if (!bh)
 414                         return -EIO;
 415
 416                 err = ext4_journal_get_write_access(handle, bh);
 417                 if (err)
 418                         return err;
 419                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 420                            block - start, count2);
 421                 ext4_set_bits(bh->b_data, block - start, count2);
 422
 423                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 424                 if (unlikely(err))
 425                         return err;
 426                 brelse(bh);
 427         }
 428
 429         return 0;
 430 }
 431
 432 /*
 433  * Set up the block and inode bitmaps, and the inode table for the new groups.
 434  * This doesn't need to be part of the main transaction, since we are only
 435  * changing blocks outside the actual filesystem.  We still do journaling to
 436  * ensure the recovery is correct in case of a failure just after resize.
 437  * If any part of this fails, we simply abort the resize.
 438  *
 439  * setup_new_flex_group_blocks handles a flex group as follow:
 440  *  1. copy super block and GDT, and initialize group tables if necessary.
 441  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 442  *     super block and GDT.
 443  *  2. allocate group tables in block bitmaps, that is, set bits in block
 444  *     bitmap for blocks taken by group tables.
 445  */
 446 static int setup_new_flex_group_blocks(struct super_block *sb,
 447                                 struct ext4_new_flex_group_data *flex_gd)
 448 {
 449         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 450         ext4_fsblk_t start;
 451         ext4_fsblk_t block;
 452         struct ext4_sb_info *sbi = EXT4_SB(sb);
 453         struct ext4_super_block *es = sbi->s_es;
 454         struct ext4_new_group_data *group_data = flex_gd->groups;
 455         __u16 *bg_flags = flex_gd->bg_flags;
 456         handle_t *handle;
 457         ext4_group_t group, count;
 458         struct buffer_head *bh = NULL;
 459         int reserved_gdb, i, j, err = 0, err2;
 460         int meta_bg;
 461
 462         BUG_ON(!flex_gd->count || !group_data ||
 463                group_data[0].group != sbi->s_groups_count);
 464
 465         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 466         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
 467
 468         /* This transaction may be extended/restarted along the way */
 469         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 470         if (IS_ERR(handle))
 471                 return PTR_ERR(handle);
 472
 473         group = group_data[0].group;
 474         for (i = 0; i < flex_gd->count; i++, group++) {
 475                 unsigned long gdblocks;
 476                 ext4_grpblk_t overhead;
 477
 478                 gdblocks = ext4_bg_num_gdb(sb, group);
 479                 start = ext4_group_first_block_no(sb, group);
 480
 481                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 482                         goto handle_itb;
 483
 484                 if (meta_bg == 1) {
 485                         ext4_group_t first_group;
 486                         first_group = ext4_meta_bg_first_group(sb, group);
 487                         if (first_group != group + 1 &&
 488                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 489                                 goto handle_itb;
 490                 }
 491
 492                 block = start + ext4_bg_has_super(sb, group);
 493                 /* Copy all of the GDT blocks into the backup in this group */
 494                 for (j = 0; j < gdblocks; j++, block++) {
 495                         struct buffer_head *gdb;
 496
 497                         ext4_debug("update backup group %#04llx\n", block);
 498                         err = extend_or_restart_transaction(handle, 1);
 499                         if (err)
 500                                 goto out;
 501
 502                         gdb = sb_getblk(sb, block);
 503                         if (!gdb) {
 504                                 err = -EIO;
 505                                 goto out;
 506                         }
 507
 508                         err = ext4_journal_get_write_access(handle, gdb);
 509                         if (err) {
 510                                 brelse(gdb);
 511                                 goto out;
 512                         }
 513                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 514                                gdb->b_size);
 515                         set_buffer_uptodate(gdb);
 516
 517                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 518                         if (unlikely(err)) {
 519                                 brelse(gdb);
 520                                 goto out;
 521                         }
 522                         brelse(gdb);
 523                 }
 524
 525                 /* Zero out all of the reserved backup group descriptor
 526                  * table blocks
 527                  */
 528                 if (ext4_bg_has_super(sb, group)) {
 529                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 530                                         reserved_gdb, GFP_NOFS);
 531                         if (err)
 532                                 goto out;
 533                 }
 534
 535 handle_itb:
 536                 /* Initialize group tables of the grop @group */
 537                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 538                         goto handle_bb;
 539
 540                 /* Zero out all of the inode table blocks */
 541                 block = group_data[i].inode_table;
 542                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 543                            block, sbi->s_itb_per_group);
 544                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 545                                        GFP_NOFS);
 546                 if (err)
 547                         goto out;
 548
 549 handle_bb:
 550                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 551                         goto handle_ib;
 552
 553                 /* Initialize block bitmap of the @group */
 554                 block = group_data[i].block_bitmap;
 555                 err = extend_or_restart_transaction(handle, 1);
 556                 if (err)
 557                         goto out;
 558
 559                 bh = bclean(handle, sb, block);
 560                 if (IS_ERR(bh)) {
 561                         err = PTR_ERR(bh);
 562                         goto out;
 563                 }
 564                 overhead = ext4_group_overhead_blocks(sb, group);
 565                 if (overhead != 0) {
 566                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 567                                    start);
 568                         ext4_set_bits(bh->b_data, 0, overhead);
 569                 }
 570                 ext4_mark_bitmap_end(group_data[i].blocks_count,
 571                                      sb->s_blocksize * 8, bh->b_data);
 572                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 573                 if (err)
 574                         goto out;
 575                 brelse(bh);
 576
 577 handle_ib:
 578                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 579                         continue;
 580
 581                 /* Initialize inode bitmap of the @group */
 582                 block = group_data[i].inode_bitmap;
 583                 err = extend_or_restart_transaction(handle, 1);
 584                 if (err)
 585                         goto out;
 586                 /* Mark unused entries in inode bitmap used */
 587                 bh = bclean(handle, sb, block);
 588                 if (IS_ERR(bh)) {
 589                         err = PTR_ERR(bh);
 590                         goto out;
 591                 }
 592
 593                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 594                                      sb->s_blocksize * 8, bh->b_data);
 595                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 596                 if (err)
 597                         goto out;
 598                 brelse(bh);
 599         }
 600         bh = NULL;
 601
 602         /* Mark group tables in block bitmap */
 603         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 604                 count = group_table_count[j];
 605                 start = (&group_data[0].block_bitmap)[j];
 606                 block = start;
 607                 for (i = 1; i < flex_gd->count; i++) {
 608                         block += group_table_count[j];
 609                         if (block == (&group_data[i].block_bitmap)[j]) {
 610                                 count += group_table_count[j];
 611                                 continue;
 612                         }
 613                         err = set_flexbg_block_bitmap(sb, handle,
 614                                                 flex_gd, start, count);
 615                         if (err)
 616                                 goto out;
 617                         count = group_table_count[j];
 618                         start = group_data[i].block_bitmap;
 619                         block = start;
 620                 }
 621
 622                 if (count) {
 623                         err = set_flexbg_block_bitmap(sb, handle,
 624                                                 flex_gd, start, count);
 625                         if (err)
 626                                 goto out;
 627                 }
 628         }
 629
 630 out:
 631         brelse(bh);
 632         err2 = ext4_journal_stop(handle);
 633         if (err2 && !err)
 634                 err = err2;
 635
 636         return err;
 637 }
 638
 639 /*
 640  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 641  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 642  * calling this for the first time.  In a sparse filesystem it will be the
 643  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 644  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 645  */
 646 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 647                                   unsigned *five, unsigned *seven)
 648 {
 649         unsigned *min = three;
 650         int mult = 3;
 651         unsigned ret;
 652
 653         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 654                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 655                 ret = *min;
 656                 *min += 1;
 657                 return ret;
 658         }
 659
 660         if (*five < *min) {
 661                 min = five;
 662                 mult = 5;
 663         }
 664         if (*seven < *min) {
 665                 min = seven;
 666                 mult = 7;
 667         }
 668
 669         ret = *min;
 670         *min *= mult;
 671
 672         return ret;
 673 }
 674
 675 /*
 676  * Check that all of the backup GDT blocks are held in the primary GDT block.
 677  * It is assumed that they are stored in group order.  Returns the number of
 678  * groups in current filesystem that have BACKUPS, or -ve error code.
 679  */
 680 static int verify_reserved_gdb(struct super_block *sb,
 681                                ext4_group_t end,
 682                                struct buffer_head *primary)
 683 {
 684         const ext4_fsblk_t blk = primary->b_blocknr;
 685         unsigned three = 1;
 686         unsigned five = 5;
 687         unsigned seven = 7;
 688         unsigned grp;
 689         __le32 *p = (__le32 *)primary->b_data;
 690         int gdbackups = 0;
 691
 692         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 693                 if (le32_to_cpu(*p++) !=
 694                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 695                         ext4_warning(sb, "reserved GDT %llu"
 696                                      " missing grp %d (%llu)",
 697                                      blk, grp,
 698                                      grp *
 699                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 700                                      blk);
 701                         return -EINVAL;
 702                 }
 703                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 704                         return -EFBIG;
 705         }
 706
 707         return gdbackups;
 708 }
 709
 710 /*
 711  * Called when we need to bring a reserved group descriptor table block into
 712  * use from the resize inode.  The primary copy of the new GDT block currently
 713  * is an indirect block (under the double indirect block in the resize inode).
 714  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 715  * block, in group order.  Even though we know all the block numbers we need,
 716  * we check to ensure that the resize inode has actually reserved these blocks.
 717  *
 718  * Don't need to update the block bitmaps because the blocks are still in use.
 719  *
 720  * We get all of the error cases out of the way, so that we are sure to not
 721  * fail once we start modifying the data on disk, because JBD has no rollback.
 722  */
 723 static int add_new_gdb(handle_t *handle, struct inode *inode,
 724                        ext4_group_t group)
 725 {
 726         struct super_block *sb = inode->i_sb;
 727         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 728         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 729         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 730         struct buffer_head **o_group_desc, **n_group_desc;
 731         struct buffer_head *dind;
 732         struct buffer_head *gdb_bh;
 733         int gdbackups;
 734         struct ext4_iloc iloc;
 735         __le32 *data;
 736         int err;
 737
 738         if (test_opt(sb, DEBUG))
 739                 printk(KERN_DEBUG
 740                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 741                        gdb_num);
 742
 743         /*
 744          * If we are not using the primary superblock/GDT copy don't resize,
 745          * because the user tools have no way of handling this.  Probably a
 746          * bad time to do it anyways.
 747          */
 748         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 749             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 750                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 751                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 752                 return -EPERM;
 753         }
 754
 755         gdb_bh = sb_bread(sb, gdblock);
 756         if (!gdb_bh)
 757                 return -EIO;
 758
 759         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 760         if (gdbackups < 0) {
 761                 err = gdbackups;
 762                 goto exit_bh;
 763         }
 764
 765         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 766         dind = sb_bread(sb, le32_to_cpu(*data));
 767         if (!dind) {
 768                 err = -EIO;
 769                 goto exit_bh;
 770         }
 771
 772         data = (__le32 *)dind->b_data;
 773         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 774                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 775                              group, gdblock);
 776                 err = -EINVAL;
 777                 goto exit_dind;
 778         }
 779
 780         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 781         if (unlikely(err))
 782                 goto exit_dind;
 783
 784         err = ext4_journal_get_write_access(handle, gdb_bh);
 785         if (unlikely(err))
 786                 goto exit_dind;
 787
 788         err = ext4_journal_get_write_access(handle, dind);
 789         if (unlikely(err))
 790                 ext4_std_error(sb, err);
 791
 792         /* ext4_reserve_inode_write() gets a reference on the iloc */
 793         err = ext4_reserve_inode_write(handle, inode, &iloc);
 794         if (unlikely(err))
 795                 goto exit_dind;
 796
 797         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 798                                      sizeof(struct buffer_head *),
 799                                      GFP_NOFS);
 800         if (!n_group_desc) {
 801                 err = -ENOMEM;
 802                 ext4_warning(sb, "not enough memory for %lu groups",
 803                              gdb_num + 1);
 804                 goto exit_inode;
 805         }
 806
 807         /*
 808          * Finally, we have all of the possible failures behind us...
 809          *
 810          * Remove new GDT block from inode double-indirect block and clear out
 811          * the new GDT block for use (which also "frees" the backup GDT blocks
 812          * from the reserved inode).  We don't need to change the bitmaps for
 813          * these blocks, because they are marked as in-use from being in the
 814          * reserved inode, and will become GDT blocks (primary and backup).
 815          */
 816         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 817         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 818         if (unlikely(err)) {
 819                 ext4_std_error(sb, err);
 820                 goto exit_inode;
 821         }
 822         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 823         ext4_mark_iloc_dirty(handle, inode, &iloc);
 824         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 825         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 826         if (unlikely(err)) {
 827                 ext4_std_error(sb, err);
 828                 goto exit_inode;
 829         }
 830         brelse(dind);
 831
 832         o_group_desc = EXT4_SB(sb)->s_group_desc;
 833         memcpy(n_group_desc, o_group_desc,
 834                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 835         n_group_desc[gdb_num] = gdb_bh;
 836         EXT4_SB(sb)->s_group_desc = n_group_desc;
 837         EXT4_SB(sb)->s_gdb_count++;
 838         ext4_kvfree(o_group_desc);
 839
 840         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 841         err = ext4_handle_dirty_super(handle, sb);
 842         if (err)
 843                 ext4_std_error(sb, err);
 844
 845         return err;
 846
 847 exit_inode:
 848         ext4_kvfree(n_group_desc);
 849         brelse(iloc.bh);
 850 exit_dind:
 851         brelse(dind);
 852 exit_bh:
 853         brelse(gdb_bh);
 854
 855         ext4_debug("leaving with error %d\n", err);
 856         return err;
 857 }
 858
 859 /*
 860  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 861  */
 862 static int add_new_gdb_meta_bg(struct super_block *sb,
 863                                handle_t *handle, ext4_group_t group) {
 864         ext4_fsblk_t gdblock;
 865         struct buffer_head *gdb_bh;
 866         struct buffer_head **o_group_desc, **n_group_desc;
 867         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 868         int err;
 869
 870         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 871                    ext4_bg_has_super(sb, group);
 872         gdb_bh = sb_bread(sb, gdblock);
 873         if (!gdb_bh)
 874                 return -EIO;
 875         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 876                                      sizeof(struct buffer_head *),
 877                                      GFP_NOFS);
 878         if (!n_group_desc) {
 879                 err = -ENOMEM;
 880                 ext4_warning(sb, "not enough memory for %lu groups",
 881                              gdb_num + 1);
 882                 return err;
 883         }
 884
 885         o_group_desc = EXT4_SB(sb)->s_group_desc;
 886         memcpy(n_group_desc, o_group_desc,
 887                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 888         n_group_desc[gdb_num] = gdb_bh;
 889         EXT4_SB(sb)->s_group_desc = n_group_desc;
 890         EXT4_SB(sb)->s_gdb_count++;
 891         ext4_kvfree(o_group_desc);
 892         err = ext4_journal_get_write_access(handle, gdb_bh);
 893         if (unlikely(err))
 894                 brelse(gdb_bh);
 895         return err;
 896 }
 897
 898 /*
 899  * Called when we are adding a new group which has a backup copy of each of
 900  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 901  * We need to add these reserved backup GDT blocks to the resize inode, so
 902  * that they are kept for future resizing and not allocated to files.
 903  *
 904  * Each reserved backup GDT block will go into a different indirect block.
 905  * The indirect blocks are actually the primary reserved GDT blocks,
 906  * so we know in advance what their block numbers are.  We only get the
 907  * double-indirect block to verify it is pointing to the primary reserved
 908  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 909  * backup GDT blocks are stored in their reserved primary GDT block.
 910  */
 911 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 912                               ext4_group_t group)
 913 {
 914         struct super_block *sb = inode->i_sb;
 915         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 916         struct buffer_head **primary;
 917         struct buffer_head *dind;
 918         struct ext4_iloc iloc;
 919         ext4_fsblk_t blk;
 920         __le32 *data, *end;
 921         int gdbackups = 0;
 922         int res, i;
 923         int err;
 924
 925         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 926         if (!primary)
 927                 return -ENOMEM;
 928
 929         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 930         dind = sb_bread(sb, le32_to_cpu(*data));
 931         if (!dind) {
 932                 err = -EIO;
 933                 goto exit_free;
 934         }
 935
 936         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 937         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 938                                          EXT4_ADDR_PER_BLOCK(sb));
 939         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 940
 941         /* Get each reserved primary GDT block and verify it holds backups */
 942         for (res = 0; res < reserved_gdb; res++, blk++) {
 943                 if (le32_to_cpu(*data) != blk) {
 944                         ext4_warning(sb, "reserved block %llu"
 945                                      " not at offset %ld",
 946                                      blk,
 947                                      (long)(data - (__le32 *)dind->b_data));
 948                         err = -EINVAL;
 949                         goto exit_bh;
 950                 }
 951                 primary[res] = sb_bread(sb, blk);
 952                 if (!primary[res]) {
 953                         err = -EIO;
 954                         goto exit_bh;
 955                 }
 956                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 957                 if (gdbackups < 0) {
 958                         brelse(primary[res]);
 959                         err = gdbackups;
 960                         goto exit_bh;
 961                 }
 962                 if (++data >= end)
 963                         data = (__le32 *)dind->b_data;
 964         }
 965
 966         for (i = 0; i < reserved_gdb; i++) {
 967                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
 968                         goto exit_bh;
 969         }
 970
 971         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 972                 goto exit_bh;
 973
 974         /*
 975          * Finally we can add each of the reserved backup GDT blocks from
 976          * the new group to its reserved primary GDT block.
 977          */
 978         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
 979         for (i = 0; i < reserved_gdb; i++) {
 980                 int err2;
 981                 data = (__le32 *)primary[i]->b_data;
 982                 /* printk("reserving backup %lu[%u] = %lu\n",
 983                        primary[i]->b_blocknr, gdbackups,
 984                        blk + primary[i]->b_blocknr); */
 985                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 986                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 987                 if (!err)
 988                         err = err2;
 989         }
 990         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 991         ext4_mark_iloc_dirty(handle, inode, &iloc);
 992
 993 exit_bh:
 994         while (--res >= 0)
 995                 brelse(primary[res]);
 996         brelse(dind);
 997
 998 exit_free:
 999         kfree(primary);
1000
1001         return err;
1002 }
1003
1004 /*
1005  * Update the backup copies of the ext4 metadata.  These don't need to be part
1006  * of the main resize transaction, because e2fsck will re-write them if there
1007  * is a problem (basically only OOM will cause a problem).  However, we
1008  * _should_ update the backups if possible, in case the primary gets trashed
1009  * for some reason and we need to run e2fsck from a backup superblock.  The
1010  * important part is that the new block and inode counts are in the backup
1011  * superblocks, and the location of the new group metadata in the GDT backups.
1012  *
1013  * We do not need take the s_resize_lock for this, because these
1014  * blocks are not otherwise touched by the filesystem code when it is
1015  * mounted.  We don't need to worry about last changing from
1016  * sbi->s_groups_count, because the worst that can happen is that we
1017  * do not copy the full number of backups at this time.  The resize
1018  * which changed s_groups_count will backup again.
1019  */
1020 static void update_backups(struct super_block *sb, int blk_off, char *data,
1021                            int size, int meta_bg)
1022 {
1023         struct ext4_sb_info *sbi = EXT4_SB(sb);
1024         ext4_group_t last;
1025         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1026         unsigned three = 1;
1027         unsigned five = 5;
1028         unsigned seven = 7;
1029         ext4_group_t group = 0;
1030         int rest = sb->s_blocksize - size;
1031         handle_t *handle;
1032         int err = 0, err2;
1033
1034         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
1035         if (IS_ERR(handle)) {
1036                 group = 1;
1037                 err = PTR_ERR(handle);
1038                 goto exit_err;
1039         }
1040
1041         if (meta_bg == 0) {
1042                 group = ext4_list_backups(sb, &three, &five, &seven);
1043                 last = sbi->s_groups_count;
1044         } else {
1045                 group = ext4_meta_bg_first_group(sb, group) + 1;
1046                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1047         }
1048
1049         while (group < sbi->s_groups_count) {
1050                 struct buffer_head *bh;
1051                 ext4_fsblk_t backup_block;
1052
1053                 /* Out of journal space, and can't get more - abort - so sad */
1054                 if (ext4_handle_valid(handle) &&
1055                     handle->h_buffer_credits == 0 &&
1056                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1057                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1058                         break;
1059
1060                 if (meta_bg == 0)
1061                         backup_block = group * bpg + blk_off;
1062                 else
1063                         backup_block = (ext4_group_first_block_no(sb, group) +
1064                                         ext4_bg_has_super(sb, group));
1065
1066                 bh = sb_getblk(sb, backup_block);
1067                 if (!bh) {
1068                         err = -EIO;
1069                         break;
1070                 }
1071                 ext4_debug("update metadata backup %llu(+%llu)\n",
1072                            backup_block, backup_block -
1073                            ext4_group_first_block_no(sb, group));
1074                 if ((err = ext4_journal_get_write_access(handle, bh)))
1075                         break;
1076                 lock_buffer(bh);
1077                 memcpy(bh->b_data, data, size);
1078                 if (rest)
1079                         memset(bh->b_data + size, 0, rest);
1080                 set_buffer_uptodate(bh);
1081                 unlock_buffer(bh);
1082                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1083                 if (unlikely(err))
1084                         ext4_std_error(sb, err);
1085                 brelse(bh);
1086
1087                 if (meta_bg == 0)
1088                         group = ext4_list_backups(sb, &three, &five, &seven);
1089                 else if (group == last)
1090                         break;
1091                 else
1092                         group = last;
1093         }
1094         if ((err2 = ext4_journal_stop(handle)) && !err)
1095                 err = err2;
1096
1097         /*
1098          * Ugh! Need to have e2fsck write the backup copies.  It is too
1099          * late to revert the resize, we shouldn't fail just because of
1100          * the backup copies (they are only needed in case of corruption).
1101          *
1102          * However, if we got here we have a journal problem too, so we
1103          * can't really start a transaction to mark the superblock.
1104          * Chicken out and just set the flag on the hope it will be written
1105          * to disk, and if not - we will simply wait until next fsck.
1106          */
1107 exit_err:
1108         if (err) {
1109                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1110                              "forcing fsck on next reboot", group, err);
1111                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1112                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1113                 mark_buffer_dirty(sbi->s_sbh);
1114         }
1115 }
1116
1117 /*
1118  * ext4_add_new_descs() adds @count group descriptor of groups
1119  * starting at @group
1120  *
1121  * @handle: journal handle
1122  * @sb: super block
1123  * @group: the group no. of the first group desc to be added
1124  * @resize_inode: the resize inode
1125  * @count: number of group descriptors to be added
1126  */
1127 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1128                               ext4_group_t group, struct inode *resize_inode,
1129                               ext4_group_t count)
1130 {
1131         struct ext4_sb_info *sbi = EXT4_SB(sb);
1132         struct ext4_super_block *es = sbi->s_es;
1133         struct buffer_head *gdb_bh;
1134         int i, gdb_off, gdb_num, err = 0;
1135         int meta_bg;
1136
1137         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1138         for (i = 0; i < count; i++, group++) {
1139                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1140                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1141
1142                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1143                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1144
1145                 /*
1146                  * We will only either add reserved group blocks to a backup group
1147                  * or remove reserved blocks for the first group in a new group block.
1148                  * Doing both would be mean more complex code, and sane people don't
1149                  * use non-sparse filesystems anymore.  This is already checked above.
1150                  */
1151                 if (gdb_off) {
1152                         gdb_bh = sbi->s_group_desc[gdb_num];
1153                         err = ext4_journal_get_write_access(handle, gdb_bh);
1154
1155                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1156                                 err = reserve_backup_gdb(handle, resize_inode, group);
1157                 } else if (meta_bg != 0) {
1158                         err = add_new_gdb_meta_bg(sb, handle, group);
1159                 } else {
1160                         err = add_new_gdb(handle, resize_inode, group);
1161                 }
1162                 if (err)
1163                         break;
1164         }
1165         return err;
1166 }
1167
1168 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1169 {
1170         struct buffer_head *bh = sb_getblk(sb, block);
1171         if (!bh)
1172                 return NULL;
1173         if (!bh_uptodate_or_lock(bh)) {
1174                 if (bh_submit_read(bh) < 0) {
1175                         brelse(bh);
1176                         return NULL;
1177                 }
1178         }
1179
1180         return bh;
1181 }
1182
1183 static int ext4_set_bitmap_checksums(struct super_block *sb,
1184                                      ext4_group_t group,
1185                                      struct ext4_group_desc *gdp,
1186                                      struct ext4_new_group_data *group_data)
1187 {
1188         struct buffer_head *bh;
1189
1190         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1191                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1192                 return 0;
1193
1194         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1195         if (!bh)
1196                 return -EIO;
1197         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1198                                    EXT4_INODES_PER_GROUP(sb) / 8);
1199         brelse(bh);
1200
1201         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1202         if (!bh)
1203                 return -EIO;
1204         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1205         brelse(bh);
1206
1207         return 0;
1208 }
1209
1210 /*
1211  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1212  */
1213 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1214                                 struct ext4_new_flex_group_data *flex_gd)
1215 {
1216         struct ext4_new_group_data      *group_data = flex_gd->groups;
1217         struct ext4_group_desc          *gdp;
1218         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1219         struct buffer_head              *gdb_bh;
1220         ext4_group_t                    group;
1221         __u16                           *bg_flags = flex_gd->bg_flags;
1222         int                             i, gdb_off, gdb_num, err = 0;
1223
1224
1225         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1226                 group = group_data->group;
1227
1228                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1229                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1230
1231                 /*
1232                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1233                  */
1234                 gdb_bh = sbi->s_group_desc[gdb_num];
1235                 /* Update group descriptor block for new group */
1236                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1237                                                  gdb_off * EXT4_DESC_SIZE(sb));
1238
1239                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1240                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1241                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1242                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1243                 if (err) {
1244                         ext4_std_error(sb, err);
1245                         break;
1246                 }
1247
1248                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1249                 ext4_free_group_clusters_set(sb, gdp,
1250                                              EXT4_B2C(sbi, group_data->free_blocks_count));
1251                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1252                 if (ext4_has_group_desc_csum(sb))
1253                         ext4_itable_unused_set(sb, gdp,
1254                                                EXT4_INODES_PER_GROUP(sb));
1255                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1256                 ext4_group_desc_csum_set(sb, group, gdp);
1257
1258                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1259                 if (unlikely(err)) {
1260                         ext4_std_error(sb, err);
1261                         break;
1262                 }
1263
1264                 /*
1265                  * We can allocate memory for mb_alloc based on the new group
1266                  * descriptor
1267                  */
1268                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1269                 if (err)
1270                         break;
1271         }
1272         return err;
1273 }
1274
1275 /*
1276  * ext4_update_super() updates the super block so that the newly added
1277  * groups can be seen by the filesystem.
1278  *
1279  * @sb: super block
1280  * @flex_gd: new added groups
1281  */
1282 static void ext4_update_super(struct super_block *sb,
1283                              struct ext4_new_flex_group_data *flex_gd)
1284 {
1285         ext4_fsblk_t blocks_count = 0;
1286         ext4_fsblk_t free_blocks = 0;
1287         ext4_fsblk_t reserved_blocks = 0;
1288         struct ext4_new_group_data *group_data = flex_gd->groups;
1289         struct ext4_sb_info *sbi = EXT4_SB(sb);
1290         struct ext4_super_block *es = sbi->s_es;
1291         int i;
1292
1293         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1294         /*
1295          * Make the new blocks and inodes valid next.  We do this before
1296          * increasing the group count so that once the group is enabled,
1297          * all of its blocks and inodes are already valid.
1298          *
1299          * We always allocate group-by-group, then block-by-block or
1300          * inode-by-inode within a group, so enabling these
1301          * blocks/inodes before the group is live won't actually let us
1302          * allocate the new space yet.
1303          */
1304         for (i = 0; i < flex_gd->count; i++) {
1305                 blocks_count += group_data[i].blocks_count;
1306                 free_blocks += group_data[i].free_blocks_count;
1307         }
1308
1309         reserved_blocks = ext4_r_blocks_count(es) * 100;
1310         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1311         reserved_blocks *= blocks_count;
1312         do_div(reserved_blocks, 100);
1313
1314         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1315         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1316         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1317                      flex_gd->count);
1318         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1319                      flex_gd->count);
1320
1321         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1322         /*
1323          * We need to protect s_groups_count against other CPUs seeing
1324          * inconsistent state in the superblock.
1325          *
1326          * The precise rules we use are:
1327          *
1328          * * Writers must perform a smp_wmb() after updating all
1329          *   dependent data and before modifying the groups count
1330          *
1331          * * Readers must perform an smp_rmb() after reading the groups
1332          *   count and before reading any dependent data.
1333          *
1334          * NB. These rules can be relaxed when checking the group count
1335          * while freeing data, as we can only allocate from a block
1336          * group after serialising against the group count, and we can
1337          * only then free after serialising in turn against that
1338          * allocation.
1339          */
1340         smp_wmb();
1341
1342         /* Update the global fs size fields */
1343         sbi->s_groups_count += flex_gd->count;
1344
1345         /* Update the reserved block counts only once the new group is
1346          * active. */
1347         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1348                                 reserved_blocks);
1349
1350         /* Update the free space counts */
1351         percpu_counter_add(&sbi->s_freeclusters_counter,
1352                            EXT4_B2C(sbi, free_blocks));
1353         percpu_counter_add(&sbi->s_freeinodes_counter,
1354                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1355
1356         ext4_debug("free blocks count %llu",
1357                    percpu_counter_read(&sbi->s_freeclusters_counter));
1358         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1359                                       EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1360             sbi->s_log_groups_per_flex) {
1361                 ext4_group_t flex_group;
1362                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1363                 atomic_add(EXT4_B2C(sbi, free_blocks),
1364                            &sbi->s_flex_groups[flex_group].free_clusters);
1365                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1366                            &sbi->s_flex_groups[flex_group].free_inodes);
1367         }
1368
1369         /*
1370          * Update the fs overhead information
1371          */
1372         ext4_calculate_overhead(sb);
1373
1374         if (test_opt(sb, DEBUG))
1375                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1376                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1377                        blocks_count, free_blocks, reserved_blocks);
1378 }
1379
1380 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1381  * _before_ we start modifying the filesystem, because we cannot abort the
1382  * transaction and not have it write the data to disk.
1383  */
1384 static int ext4_flex_group_add(struct super_block *sb,
1385                                struct inode *resize_inode,
1386                                struct ext4_new_flex_group_data *flex_gd)
1387 {
1388         struct ext4_sb_info *sbi = EXT4_SB(sb);
1389         struct ext4_super_block *es = sbi->s_es;
1390         ext4_fsblk_t o_blocks_count;
1391         ext4_grpblk_t last;
1392         ext4_group_t group;
1393         handle_t *handle;
1394         unsigned reserved_gdb;
1395         int err = 0, err2 = 0, credit;
1396
1397         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1398
1399         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1400         o_blocks_count = ext4_blocks_count(es);
1401         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1402         BUG_ON(last);
1403
1404         err = setup_new_flex_group_blocks(sb, flex_gd);
1405         if (err)
1406                 goto exit;
1407         /*
1408          * We will always be modifying at least the superblock and  GDT
1409          * block.  If we are adding a group past the last current GDT block,
1410          * we will also modify the inode and the dindirect block.  If we
1411          * are adding a group with superblock/GDT backups  we will also
1412          * modify each of the reserved GDT dindirect blocks.
1413          */
1414         credit = flex_gd->count * 4 + reserved_gdb;
1415         handle = ext4_journal_start_sb(sb, credit);
1416         if (IS_ERR(handle)) {
1417                 err = PTR_ERR(handle);
1418                 goto exit;
1419         }
1420
1421         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1422         if (err)
1423                 goto exit_journal;
1424
1425         group = flex_gd->groups[0].group;
1426         BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1427         err = ext4_add_new_descs(handle, sb, group,
1428                                 resize_inode, flex_gd->count);
1429         if (err)
1430                 goto exit_journal;
1431
1432         err = ext4_setup_new_descs(handle, sb, flex_gd);
1433         if (err)
1434                 goto exit_journal;
1435
1436         ext4_update_super(sb, flex_gd);
1437
1438         err = ext4_handle_dirty_super(handle, sb);
1439
1440 exit_journal:
1441         err2 = ext4_journal_stop(handle);
1442         if (!err)
1443                 err = err2;
1444
1445         if (!err) {
1446                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1447                 int gdb_num_end = ((group + flex_gd->count - 1) /
1448                                    EXT4_DESC_PER_BLOCK(sb));
1449                 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1450                                 EXT4_FEATURE_INCOMPAT_META_BG);
1451                 sector_t old_gdb = 0;
1452
1453                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1454                                sizeof(struct ext4_super_block), 0);
1455                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1456                         struct buffer_head *gdb_bh;
1457
1458                         gdb_bh = sbi->s_group_desc[gdb_num];
1459                         if (old_gdb == gdb_bh->b_blocknr)
1460                                 continue;
1461                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1462                                        gdb_bh->b_size, meta_bg);
1463                         old_gdb = gdb_bh->b_blocknr;
1464                 }
1465         }
1466 exit:
1467         return err;
1468 }
1469
1470 static int ext4_setup_next_flex_gd(struct super_block *sb,
1471                                     struct ext4_new_flex_group_data *flex_gd,
1472                                     ext4_fsblk_t n_blocks_count,
1473                                     unsigned long flexbg_size)
1474 {
1475         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1476         struct ext4_new_group_data *group_data = flex_gd->groups;
1477         ext4_fsblk_t o_blocks_count;
1478         ext4_group_t n_group;
1479         ext4_group_t group;
1480         ext4_group_t last_group;
1481         ext4_grpblk_t last;
1482         ext4_grpblk_t blocks_per_group;
1483         unsigned long i;
1484
1485         blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1486
1487         o_blocks_count = ext4_blocks_count(es);
1488
1489         if (o_blocks_count == n_blocks_count)
1490                 return 0;
1491
1492         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1493         BUG_ON(last);
1494         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1495
1496         last_group = group | (flexbg_size - 1);
1497         if (last_group > n_group)
1498                 last_group = n_group;
1499
1500         flex_gd->count = last_group - group + 1;
1501
1502         for (i = 0; i < flex_gd->count; i++) {
1503                 int overhead;
1504
1505                 group_data[i].group = group + i;
1506                 group_data[i].blocks_count = blocks_per_group;
1507                 overhead = ext4_group_overhead_blocks(sb, group + i);
1508                 group_data[i].free_blocks_count = blocks_per_group - overhead;
1509                 if (ext4_has_group_desc_csum(sb))
1510                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1511                                                EXT4_BG_INODE_UNINIT;
1512                 else
1513                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1514         }
1515
1516         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1517                 /* We need to initialize block bitmap of last group. */
1518                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1519
1520         if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1521                 group_data[i - 1].blocks_count = last + 1;
1522                 group_data[i - 1].free_blocks_count -= blocks_per_group-
1523                                         last - 1;
1524         }
1525
1526         return 1;
1527 }
1528
1529 /* Add group descriptor data to an existing or new group descriptor block.
1530  * Ensure we handle all possible error conditions _before_ we start modifying
1531  * the filesystem, because we cannot abort the transaction and not have it
1532  * write the data to disk.
1533  *
1534  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1535  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1536  *
1537  * We only need to hold the superblock lock while we are actually adding
1538  * in the new group's counts to the superblock.  Prior to that we have
1539  * not really "added" the group at all.  We re-check that we are still
1540  * adding in the last group in case things have changed since verifying.
1541  */
1542 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1543 {
1544         struct ext4_new_flex_group_data flex_gd;
1545         struct ext4_sb_info *sbi = EXT4_SB(sb);
1546         struct ext4_super_block *es = sbi->s_es;
1547         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1548                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1549         struct inode *inode = NULL;
1550         int gdb_off, gdb_num;
1551         int err;
1552         __u16 bg_flags = 0;
1553
1554         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1555         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1556
1557         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1558                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1559                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1560                 return -EPERM;
1561         }
1562
1563         if (ext4_blocks_count(es) + input->blocks_count <
1564             ext4_blocks_count(es)) {
1565                 ext4_warning(sb, "blocks_count overflow");
1566                 return -EINVAL;
1567         }
1568
1569         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1570             le32_to_cpu(es->s_inodes_count)) {
1571                 ext4_warning(sb, "inodes_count overflow");
1572                 return -EINVAL;
1573         }
1574
1575         if (reserved_gdb || gdb_off == 0) {
1576                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1577                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
1578                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1579                         ext4_warning(sb,
1580                                      "No reserved GDT blocks, can't resize");
1581                         return -EPERM;
1582                 }
1583                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1584                 if (IS_ERR(inode)) {
1585                         ext4_warning(sb, "Error opening resize inode");
1586                         return PTR_ERR(inode);
1587                 }
1588         }
1589
1590
1591         err = verify_group_input(sb, input);
1592         if (err)
1593                 goto out;
1594
1595         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1596         if (err)
1597                 return err;
1598
1599         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1600         if (err)
1601                 goto out;
1602
1603         flex_gd.count = 1;
1604         flex_gd.groups = input;
1605         flex_gd.bg_flags = &bg_flags;
1606         err = ext4_flex_group_add(sb, inode, &flex_gd);
1607 out:
1608         iput(inode);
1609         return err;
1610 } /* ext4_group_add */
1611
1612 /*
1613  * extend a group without checking assuming that checking has been done.
1614  */
1615 static int ext4_group_extend_no_check(struct super_block *sb,
1616                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1617 {
1618         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1619         handle_t *handle;
1620         int err = 0, err2;
1621
1622         /* We will update the superblock, one block bitmap, and
1623          * one group descriptor via ext4_group_add_blocks().
1624          */
1625         handle = ext4_journal_start_sb(sb, 3);
1626         if (IS_ERR(handle)) {
1627                 err = PTR_ERR(handle);
1628                 ext4_warning(sb, "error %d on journal start", err);
1629                 return err;
1630         }
1631
1632         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1633         if (err) {
1634                 ext4_warning(sb, "error %d on journal write access", err);
1635                 goto errout;
1636         }
1637
1638         ext4_blocks_count_set(es, o_blocks_count + add);
1639         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1640         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1641                    o_blocks_count + add);
1642         /* We add the blocks to the bitmap and set the group need init bit */
1643         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1644         if (err)
1645                 goto errout;
1646         ext4_handle_dirty_super(handle, sb);
1647         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1648                    o_blocks_count + add);
1649 errout:
1650         err2 = ext4_journal_stop(handle);
1651         if (err2 && !err)
1652                 err = err2;
1653
1654         if (!err) {
1655                 ext4_fsblk_t first_block;
1656                 first_block = ext4_group_first_block_no(sb, 0);
1657                 if (test_opt(sb, DEBUG))
1658                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1659                                "blocks\n", ext4_blocks_count(es));
1660                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr - first_block,
1661                                (char *)es, sizeof(struct ext4_super_block), 0);
1662         }
1663         return err;
1664 }
1665
1666 /*
1667  * Extend the filesystem to the new number of blocks specified.  This entry
1668  * point is only used to extend the current filesystem to the end of the last
1669  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1670  * for emergencies (because it has no dependencies on reserved blocks).
1671  *
1672  * If we _really_ wanted, we could use default values to call ext4_group_add()
1673  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1674  * GDT blocks are reserved to grow to the desired size.
1675  */
1676 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1677                       ext4_fsblk_t n_blocks_count)
1678 {
1679         ext4_fsblk_t o_blocks_count;
1680         ext4_grpblk_t last;
1681         ext4_grpblk_t add;
1682         struct buffer_head *bh;
1683         int err;
1684         ext4_group_t group;
1685
1686         o_blocks_count = ext4_blocks_count(es);
1687
1688         if (test_opt(sb, DEBUG))
1689                 ext4_msg(sb, KERN_DEBUG,
1690                          "extending last group from %llu to %llu blocks",
1691                          o_blocks_count, n_blocks_count);
1692
1693         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1694                 return 0;
1695
1696         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1697                 ext4_msg(sb, KERN_ERR,
1698                          "filesystem too large to resize to %llu blocks safely",
1699                          n_blocks_count);
1700                 if (sizeof(sector_t) < 8)
1701                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1702                 return -EINVAL;
1703         }
1704
1705         if (n_blocks_count < o_blocks_count) {
1706                 ext4_warning(sb, "can't shrink FS - resize aborted");
1707                 return -EINVAL;
1708         }
1709
1710         /* Handle the remaining blocks in the last group only. */
1711         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1712
1713         if (last == 0) {
1714                 ext4_warning(sb, "need to use ext2online to resize further");
1715                 return -EPERM;
1716         }
1717
1718         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1719
1720         if (o_blocks_count + add < o_blocks_count) {
1721                 ext4_warning(sb, "blocks_count overflow");
1722                 return -EINVAL;
1723         }
1724
1725         if (o_blocks_count + add > n_blocks_count)
1726                 add = n_blocks_count - o_blocks_count;
1727
1728         if (o_blocks_count + add < n_blocks_count)
1729                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1730                              o_blocks_count + add, add);
1731
1732         /* See if the device is actually as big as what was requested */
1733         bh = sb_bread(sb, o_blocks_count + add - 1);
1734         if (!bh) {
1735                 ext4_warning(sb, "can't read last block, resize aborted");
1736                 return -ENOSPC;
1737         }
1738         brelse(bh);
1739
1740         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1741         return err;
1742 } /* ext4_group_extend */
1743
1744
1745 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1746 {
1747         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1748 }
1749
1750 /*
1751  * Release the resize inode and drop the resize_inode feature if there
1752  * are no more reserved gdt blocks, and then convert the file system
1753  * to enable meta_bg
1754  */
1755 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1756 {
1757         handle_t *handle;
1758         struct ext4_sb_info *sbi = EXT4_SB(sb);
1759         struct ext4_super_block *es = sbi->s_es;
1760         struct ext4_inode_info *ei = EXT4_I(inode);
1761         ext4_fsblk_t nr;
1762         int i, ret, err = 0;
1763         int credits = 1;
1764
1765         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1766         if (inode) {
1767                 if (es->s_reserved_gdt_blocks) {
1768                         ext4_error(sb, "Unexpected non-zero "
1769                                    "s_reserved_gdt_blocks");
1770                         return -EPERM;
1771                 }
1772
1773                 /* Do a quick sanity check of the resize inode */
1774                 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1775                         goto invalid_resize_inode;
1776                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1777                         if (i == EXT4_DIND_BLOCK) {
1778                                 if (ei->i_data[i])
1779                                         continue;
1780                                 else
1781                                         goto invalid_resize_inode;
1782                         }
1783                         if (ei->i_data[i])
1784                                 goto invalid_resize_inode;
1785                 }
1786                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1787         }
1788
1789         handle = ext4_journal_start_sb(sb, credits);
1790         if (IS_ERR(handle))
1791                 return PTR_ERR(handle);
1792
1793         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1794         if (err)
1795                 goto errout;
1796
1797         EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1798         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1799         sbi->s_es->s_first_meta_bg =
1800                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1801
1802         err = ext4_handle_dirty_super(handle, sb);
1803         if (err) {
1804                 ext4_std_error(sb, err);
1805                 goto errout;
1806         }
1807
1808         if (inode) {
1809                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1810                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1811                                  EXT4_FREE_BLOCKS_METADATA |
1812                                  EXT4_FREE_BLOCKS_FORGET);
1813                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1814                 inode->i_blocks = 0;
1815
1816                 err = ext4_mark_inode_dirty(handle, inode);
1817                 if (err)
1818                         ext4_std_error(sb, err);
1819         }
1820
1821 errout:
1822         ret = ext4_journal_stop(handle);
1823         if (!err)
1824                 err = ret;
1825         return ret;
1826
1827 invalid_resize_inode:
1828         ext4_error(sb, "corrupted/inconsistent resize inode");
1829         return -EINVAL;
1830 }
1831
1832 /*
1833  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1834  *
1835  * @sb: super block of the fs to be resized
1836  * @n_blocks_count: the number of blocks resides in the resized fs
1837  */
1838 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1839 {
1840         struct ext4_new_flex_group_data *flex_gd = NULL;
1841         struct ext4_sb_info *sbi = EXT4_SB(sb);
1842         struct ext4_super_block *es = sbi->s_es;
1843         struct buffer_head *bh;
1844         struct inode *resize_inode = NULL;
1845         ext4_grpblk_t add, offset;
1846         unsigned long n_desc_blocks;
1847         unsigned long o_desc_blocks;
1848         ext4_group_t o_group;
1849         ext4_group_t n_group;
1850         ext4_fsblk_t o_blocks_count;
1851         ext4_fsblk_t n_blocks_count_retry = 0;
1852         unsigned long last_update_time = 0;
1853         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1854         int meta_bg;
1855
1856         /* See if the device is actually as big as what was requested */
1857         bh = sb_bread(sb, n_blocks_count - 1);
1858         if (!bh) {
1859                 ext4_warning(sb, "can't read last block, resize aborted");
1860                 return -ENOSPC;
1861         }
1862         brelse(bh);
1863
1864 retry:
1865         o_blocks_count = ext4_blocks_count(es);
1866
1867         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1868                  "to %llu blocks", o_blocks_count, n_blocks_count);
1869
1870         if (n_blocks_count < o_blocks_count) {
1871                 /* On-line shrinking not supported */
1872                 ext4_warning(sb, "can't shrink FS - resize aborted");
1873                 return -EINVAL;
1874         }
1875
1876         if (n_blocks_count == o_blocks_count)
1877                 /* Nothing need to do */
1878                 return 0;
1879
1880         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &offset);
1881         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1882
1883         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1884         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1885
1886         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1887
1888         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1889                 if (meta_bg) {
1890                         ext4_error(sb, "resize_inode and meta_bg enabled "
1891                                    "simultaneously");
1892                         return -EINVAL;
1893                 }
1894                 if (n_desc_blocks > o_desc_blocks +
1895                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
1896                         n_blocks_count_retry = n_blocks_count;
1897                         n_desc_blocks = o_desc_blocks +
1898                                 le16_to_cpu(es->s_reserved_gdt_blocks);
1899                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1900                         n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1901                         n_group--; /* set to last group number */
1902                 }
1903
1904                 if (!resize_inode)
1905                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1906                 if (IS_ERR(resize_inode)) {
1907                         ext4_warning(sb, "Error opening resize inode");
1908                         return PTR_ERR(resize_inode);
1909                 }
1910         }
1911
1912         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1913                 err = ext4_convert_meta_bg(sb, resize_inode);
1914                 if (err)
1915                         goto out;
1916                 if (resize_inode) {
1917                         iput(resize_inode);
1918                         resize_inode = NULL;
1919                 }
1920                 if (n_blocks_count_retry) {
1921                         n_blocks_count = n_blocks_count_retry;
1922                         n_blocks_count_retry = 0;
1923                         goto retry;
1924                 }
1925         }
1926
1927         /* extend the last group */
1928         if (n_group == o_group)
1929                 add = n_blocks_count - o_blocks_count;
1930         else
1931                 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1932         if (add > 0) {
1933                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1934                 if (err)
1935                         goto out;
1936         }
1937
1938         if (ext4_blocks_count(es) == n_blocks_count)
1939                 goto out;
1940
1941         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1942         if (err)
1943                 return err;
1944
1945         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1946         if (err)
1947                 goto out;
1948
1949         flex_gd = alloc_flex_gd(flexbg_size);
1950         if (flex_gd == NULL) {
1951                 err = -ENOMEM;
1952                 goto out;
1953         }
1954
1955         /* Add flex groups. Note that a regular group is a
1956          * flex group with 1 group.
1957          */
1958         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1959                                               flexbg_size)) {
1960                 if (jiffies - last_update_time > HZ * 10) {
1961                         if (last_update_time)
1962                                 ext4_msg(sb, KERN_INFO,
1963                                          "resized to %llu blocks",
1964                                          ext4_blocks_count(es));
1965                         last_update_time = jiffies;
1966                 }
1967                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1968                         break;
1969                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1970                 if (unlikely(err))
1971                         break;
1972         }
1973
1974         if (!err && n_blocks_count_retry) {
1975                 n_blocks_count = n_blocks_count_retry;
1976                 n_blocks_count_retry = 0;
1977                 free_flex_gd(flex_gd);
1978                 flex_gd = NULL;
1979                 goto retry;
1980         }
1981
1982 out:
1983         if (flex_gd)
1984                 free_flex_gd(flex_gd);
1985         if (resize_inode != NULL)
1986                 iput(resize_inode);
1987         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
1988         return err;
1989 }