linux/fs/ext4/resize.c
<<
>>
Prefs
   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
  19int 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         * If we are not using the primary superblock/GDT copy don't resize,
  28         * because the user tools have no way of handling this.  Probably a
  29         * bad time to do it anyways.
  30         */
  31        if (EXT4_SB(sb)->s_sbh->b_blocknr !=
  32            le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  33                ext4_warning(sb, "won't resize using backup superblock at %llu",
  34                        (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  35                return -EPERM;
  36        }
  37
  38        /*
  39         * We are not allowed to do online-resizing on a filesystem mounted
  40         * with error, because it can destroy the filesystem easily.
  41         */
  42        if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  43                ext4_warning(sb, "There are errors in the filesystem, "
  44                             "so online resizing is not allowed\n");
  45                return -EPERM;
  46        }
  47
  48        if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
  49                ret = -EBUSY;
  50
  51        return ret;
  52}
  53
  54void ext4_resize_end(struct super_block *sb)
  55{
  56        clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
  57        smp_mb__after_clear_bit();
  58}
  59
  60static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  61                                             ext4_group_t group) {
  62        return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  63               EXT4_DESC_PER_BLOCK_BITS(sb);
  64}
  65
  66static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
  67                                             ext4_group_t group) {
  68        group = ext4_meta_bg_first_group(sb, group);
  69        return ext4_group_first_block_no(sb, group);
  70}
  71
  72static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
  73                                                ext4_group_t group) {
  74        ext4_grpblk_t overhead;
  75        overhead = ext4_bg_num_gdb(sb, group);
  76        if (ext4_bg_has_super(sb, group))
  77                overhead += 1 +
  78                          le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
  79        return overhead;
  80}
  81
  82#define outside(b, first, last) ((b) < (first) || (b) >= (last))
  83#define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  84
  85static int verify_group_input(struct super_block *sb,
  86                              struct ext4_new_group_data *input)
  87{
  88        struct ext4_sb_info *sbi = EXT4_SB(sb);
  89        struct ext4_super_block *es = sbi->s_es;
  90        ext4_fsblk_t start = ext4_blocks_count(es);
  91        ext4_fsblk_t end = start + input->blocks_count;
  92        ext4_group_t group = input->group;
  93        ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  94        unsigned overhead;
  95        ext4_fsblk_t metaend;
  96        struct buffer_head *bh = NULL;
  97        ext4_grpblk_t free_blocks_count, offset;
  98        int err = -EINVAL;
  99
 100        if (group != sbi->s_groups_count) {
 101                ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 102                             input->group, sbi->s_groups_count);
 103                return -EINVAL;
 104        }
 105
 106        overhead = ext4_group_overhead_blocks(sb, group);
 107        metaend = start + overhead;
 108        input->free_blocks_count = free_blocks_count =
 109                input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 110
 111        if (test_opt(sb, DEBUG))
 112                printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 113                       "(%d free, %u reserved)\n",
 114                       ext4_bg_has_super(sb, input->group) ? "normal" :
 115                       "no-super", input->group, input->blocks_count,
 116                       free_blocks_count, input->reserved_blocks);
 117
 118        ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 119        if (offset != 0)
 120                        ext4_warning(sb, "Last group not full");
 121        else if (input->reserved_blocks > input->blocks_count / 5)
 122                ext4_warning(sb, "Reserved blocks too high (%u)",
 123                             input->reserved_blocks);
 124        else if (free_blocks_count < 0)
 125                ext4_warning(sb, "Bad blocks count %u",
 126                             input->blocks_count);
 127        else if (!(bh = sb_bread(sb, end - 1)))
 128                ext4_warning(sb, "Cannot read last block (%llu)",
 129                             end - 1);
 130        else if (outside(input->block_bitmap, start, end))
 131                ext4_warning(sb, "Block bitmap not in group (block %llu)",
 132                             (unsigned long long)input->block_bitmap);
 133        else if (outside(input->inode_bitmap, start, end))
 134                ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 135                             (unsigned long long)input->inode_bitmap);
 136        else if (outside(input->inode_table, start, end) ||
 137                 outside(itend - 1, start, end))
 138                ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 139                             (unsigned long long)input->inode_table, itend - 1);
 140        else if (input->inode_bitmap == input->block_bitmap)
 141                ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 142                             (unsigned long long)input->block_bitmap);
 143        else if (inside(input->block_bitmap, input->inode_table, itend))
 144                ext4_warning(sb, "Block bitmap (%llu) in inode table "
 145                             "(%llu-%llu)",
 146                             (unsigned long long)input->block_bitmap,
 147                             (unsigned long long)input->inode_table, itend - 1);
 148        else if (inside(input->inode_bitmap, input->inode_table, itend))
 149                ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 150                             "(%llu-%llu)",
 151                             (unsigned long long)input->inode_bitmap,
 152                             (unsigned long long)input->inode_table, itend - 1);
 153        else if (inside(input->block_bitmap, start, metaend))
 154                ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 155                             (unsigned long long)input->block_bitmap,
 156                             start, metaend - 1);
 157        else if (inside(input->inode_bitmap, start, metaend))
 158                ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 159                             (unsigned long long)input->inode_bitmap,
 160                             start, metaend - 1);
 161        else if (inside(input->inode_table, start, metaend) ||
 162                 inside(itend - 1, start, metaend))
 163                ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 164                             "(%llu-%llu)",
 165                             (unsigned long long)input->inode_table,
 166                             itend - 1, start, metaend - 1);
 167        else
 168                err = 0;
 169        brelse(bh);
 170
 171        return err;
 172}
 173
 174/*
 175 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 176 * group each time.
 177 */
 178struct ext4_new_flex_group_data {
 179        struct ext4_new_group_data *groups;     /* new_group_data for groups
 180                                                   in the flex group */
 181        __u16 *bg_flags;                        /* block group flags of groups
 182                                                   in @groups */
 183        ext4_group_t count;                     /* number of groups in @groups
 184                                                 */
 185};
 186
 187/*
 188 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 189 * @flexbg_size.
 190 *
 191 * Returns NULL on failure otherwise address of the allocated structure.
 192 */
 193static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 194{
 195        struct ext4_new_flex_group_data *flex_gd;
 196
 197        flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 198        if (flex_gd == NULL)
 199                goto out3;
 200
 201        if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
 202                goto out2;
 203        flex_gd->count = flexbg_size;
 204
 205        flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
 206                                  flexbg_size, GFP_NOFS);
 207        if (flex_gd->groups == NULL)
 208                goto out2;
 209
 210        flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
 211        if (flex_gd->bg_flags == NULL)
 212                goto out1;
 213
 214        return flex_gd;
 215
 216out1:
 217        kfree(flex_gd->groups);
 218out2:
 219        kfree(flex_gd);
 220out3:
 221        return NULL;
 222}
 223
 224static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 225{
 226        kfree(flex_gd->bg_flags);
 227        kfree(flex_gd->groups);
 228        kfree(flex_gd);
 229}
 230
 231/*
 232 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 233 * and inode tables for a flex group.
 234 *
 235 * This function is used by 64bit-resize.  Note that this function allocates
 236 * group tables from the 1st group of groups contained by @flexgd, which may
 237 * be a partial of a flex group.
 238 *
 239 * @sb: super block of fs to which the groups belongs
 240 *
 241 * Returns 0 on a successful allocation of the metadata blocks in the
 242 * block group.
 243 */
 244static int ext4_alloc_group_tables(struct super_block *sb,
 245                                struct ext4_new_flex_group_data *flex_gd,
 246                                int flexbg_size)
 247{
 248        struct ext4_new_group_data *group_data = flex_gd->groups;
 249        ext4_fsblk_t start_blk;
 250        ext4_fsblk_t last_blk;
 251        ext4_group_t src_group;
 252        ext4_group_t bb_index = 0;
 253        ext4_group_t ib_index = 0;
 254        ext4_group_t it_index = 0;
 255        ext4_group_t group;
 256        ext4_group_t last_group;
 257        unsigned overhead;
 258        __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 259
 260        BUG_ON(flex_gd->count == 0 || group_data == NULL);
 261
 262        src_group = group_data[0].group;
 263        last_group  = src_group + flex_gd->count - 1;
 264
 265        BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 266               (last_group & ~(flexbg_size - 1))));
 267next_group:
 268        group = group_data[0].group;
 269        if (src_group >= group_data[0].group + flex_gd->count)
 270                return -ENOSPC;
 271        start_blk = ext4_group_first_block_no(sb, src_group);
 272        last_blk = start_blk + group_data[src_group - group].blocks_count;
 273
 274        overhead = ext4_group_overhead_blocks(sb, src_group);
 275
 276        start_blk += overhead;
 277
 278        /* We collect contiguous blocks as much as possible. */
 279        src_group++;
 280        for (; src_group <= last_group; src_group++) {
 281                overhead = ext4_group_overhead_blocks(sb, src_group);
 282                if (overhead == 0)
 283                        last_blk += group_data[src_group - group].blocks_count;
 284                else
 285                        break;
 286        }
 287
 288        /* Allocate block bitmaps */
 289        for (; bb_index < flex_gd->count; bb_index++) {
 290                if (start_blk >= last_blk)
 291                        goto next_group;
 292                group_data[bb_index].block_bitmap = start_blk++;
 293                group = ext4_get_group_number(sb, start_blk - 1);
 294                group -= group_data[0].group;
 295                group_data[group].free_blocks_count--;
 296                flex_gd->bg_flags[group] &= uninit_mask;
 297        }
 298
 299        /* Allocate inode bitmaps */
 300        for (; ib_index < flex_gd->count; ib_index++) {
 301                if (start_blk >= last_blk)
 302                        goto next_group;
 303                group_data[ib_index].inode_bitmap = 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                flex_gd->bg_flags[group] &= uninit_mask;
 308        }
 309
 310        /* Allocate inode tables */
 311        for (; it_index < flex_gd->count; it_index++) {
 312                unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 313                ext4_fsblk_t next_group_start;
 314
 315                if (start_blk + itb > last_blk)
 316                        goto next_group;
 317                group_data[it_index].inode_table = start_blk;
 318                group = ext4_get_group_number(sb, start_blk);
 319                next_group_start = ext4_group_first_block_no(sb, group + 1);
 320                group -= group_data[0].group;
 321
 322                if (start_blk + itb > next_group_start) {
 323                        flex_gd->bg_flags[group + 1] &= uninit_mask;
 324                        overhead = start_blk + itb - next_group_start;
 325                        group_data[group + 1].free_blocks_count -= overhead;
 326                        itb -= overhead;
 327                }
 328
 329                group_data[group].free_blocks_count -= itb;
 330                flex_gd->bg_flags[group] &= uninit_mask;
 331                start_blk += EXT4_SB(sb)->s_itb_per_group;
 332        }
 333
 334        if (test_opt(sb, DEBUG)) {
 335                int i;
 336                group = group_data[0].group;
 337
 338                printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 339                       "%d groups, flexbg size is %d:\n", flex_gd->count,
 340                       flexbg_size);
 341
 342                for (i = 0; i < flex_gd->count; i++) {
 343                        printk(KERN_DEBUG "adding %s group %u: %u "
 344                               "blocks (%d free)\n",
 345                               ext4_bg_has_super(sb, group + i) ? "normal" :
 346                               "no-super", group + i,
 347                               group_data[i].blocks_count,
 348                               group_data[i].free_blocks_count);
 349                }
 350        }
 351        return 0;
 352}
 353
 354static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 355                                  ext4_fsblk_t blk)
 356{
 357        struct buffer_head *bh;
 358        int err;
 359
 360        bh = sb_getblk(sb, blk);
 361        if (unlikely(!bh))
 362                return ERR_PTR(-ENOMEM);
 363        BUFFER_TRACE(bh, "get_write_access");
 364        if ((err = ext4_journal_get_write_access(handle, bh))) {
 365                brelse(bh);
 366                bh = ERR_PTR(err);
 367        } else {
 368                memset(bh->b_data, 0, sb->s_blocksize);
 369                set_buffer_uptodate(bh);
 370        }
 371
 372        return bh;
 373}
 374
 375/*
 376 * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 377 * If that fails, restart the transaction & regain write access for the
 378 * buffer head which is used for block_bitmap modifications.
 379 */
 380static int extend_or_restart_transaction(handle_t *handle, int thresh)
 381{
 382        int err;
 383
 384        if (ext4_handle_has_enough_credits(handle, thresh))
 385                return 0;
 386
 387        err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 388        if (err < 0)
 389                return err;
 390        if (err) {
 391                err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 392                if (err)
 393                        return err;
 394        }
 395
 396        return 0;
 397}
 398
 399/*
 400 * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 401 *
 402 * Helper function for ext4_setup_new_group_blocks() which set .
 403 *
 404 * @sb: super block
 405 * @handle: journal handle
 406 * @flex_gd: flex group data
 407 */
 408static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 409                        struct ext4_new_flex_group_data *flex_gd,
 410                        ext4_fsblk_t block, ext4_group_t count)
 411{
 412        ext4_group_t count2;
 413
 414        ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 415        for (count2 = count; count > 0; count -= count2, block += count2) {
 416                ext4_fsblk_t start;
 417                struct buffer_head *bh;
 418                ext4_group_t group;
 419                int err;
 420
 421                group = ext4_get_group_number(sb, block);
 422                start = ext4_group_first_block_no(sb, group);
 423                group -= flex_gd->groups[0].group;
 424
 425                count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
 426                if (count2 > count)
 427                        count2 = count;
 428
 429                if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 430                        BUG_ON(flex_gd->count > 1);
 431                        continue;
 432                }
 433
 434                err = extend_or_restart_transaction(handle, 1);
 435                if (err)
 436                        return err;
 437
 438                bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 439                if (unlikely(!bh))
 440                        return -ENOMEM;
 441
 442                BUFFER_TRACE(bh, "get_write_access");
 443                err = ext4_journal_get_write_access(handle, bh);
 444                if (err)
 445                        return err;
 446                ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 447                           block - start, count2);
 448                ext4_set_bits(bh->b_data, block - start, count2);
 449
 450                err = ext4_handle_dirty_metadata(handle, NULL, bh);
 451                if (unlikely(err))
 452                        return err;
 453                brelse(bh);
 454        }
 455
 456        return 0;
 457}
 458
 459/*
 460 * Set up the block and inode bitmaps, and the inode table for the new groups.
 461 * This doesn't need to be part of the main transaction, since we are only
 462 * changing blocks outside the actual filesystem.  We still do journaling to
 463 * ensure the recovery is correct in case of a failure just after resize.
 464 * If any part of this fails, we simply abort the resize.
 465 *
 466 * setup_new_flex_group_blocks handles a flex group as follow:
 467 *  1. copy super block and GDT, and initialize group tables if necessary.
 468 *     In this step, we only set bits in blocks bitmaps for blocks taken by
 469 *     super block and GDT.
 470 *  2. allocate group tables in block bitmaps, that is, set bits in block
 471 *     bitmap for blocks taken by group tables.
 472 */
 473static int setup_new_flex_group_blocks(struct super_block *sb,
 474                                struct ext4_new_flex_group_data *flex_gd)
 475{
 476        int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 477        ext4_fsblk_t start;
 478        ext4_fsblk_t block;
 479        struct ext4_sb_info *sbi = EXT4_SB(sb);
 480        struct ext4_super_block *es = sbi->s_es;
 481        struct ext4_new_group_data *group_data = flex_gd->groups;
 482        __u16 *bg_flags = flex_gd->bg_flags;
 483        handle_t *handle;
 484        ext4_group_t group, count;
 485        struct buffer_head *bh = NULL;
 486        int reserved_gdb, i, j, err = 0, err2;
 487        int meta_bg;
 488
 489        BUG_ON(!flex_gd->count || !group_data ||
 490               group_data[0].group != sbi->s_groups_count);
 491
 492        reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 493        meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
 494
 495        /* This transaction may be extended/restarted along the way */
 496        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 497        if (IS_ERR(handle))
 498                return PTR_ERR(handle);
 499
 500        group = group_data[0].group;
 501        for (i = 0; i < flex_gd->count; i++, group++) {
 502                unsigned long gdblocks;
 503                ext4_grpblk_t overhead;
 504
 505                gdblocks = ext4_bg_num_gdb(sb, group);
 506                start = ext4_group_first_block_no(sb, group);
 507
 508                if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 509                        goto handle_itb;
 510
 511                if (meta_bg == 1) {
 512                        ext4_group_t first_group;
 513                        first_group = ext4_meta_bg_first_group(sb, group);
 514                        if (first_group != group + 1 &&
 515                            first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 516                                goto handle_itb;
 517                }
 518
 519                block = start + ext4_bg_has_super(sb, group);
 520                /* Copy all of the GDT blocks into the backup in this group */
 521                for (j = 0; j < gdblocks; j++, block++) {
 522                        struct buffer_head *gdb;
 523
 524                        ext4_debug("update backup group %#04llx\n", block);
 525                        err = extend_or_restart_transaction(handle, 1);
 526                        if (err)
 527                                goto out;
 528
 529                        gdb = sb_getblk(sb, block);
 530                        if (unlikely(!gdb)) {
 531                                err = -ENOMEM;
 532                                goto out;
 533                        }
 534
 535                        BUFFER_TRACE(gdb, "get_write_access");
 536                        err = ext4_journal_get_write_access(handle, gdb);
 537                        if (err) {
 538                                brelse(gdb);
 539                                goto out;
 540                        }
 541                        memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 542                               gdb->b_size);
 543                        set_buffer_uptodate(gdb);
 544
 545                        err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 546                        if (unlikely(err)) {
 547                                brelse(gdb);
 548                                goto out;
 549                        }
 550                        brelse(gdb);
 551                }
 552
 553                /* Zero out all of the reserved backup group descriptor
 554                 * table blocks
 555                 */
 556                if (ext4_bg_has_super(sb, group)) {
 557                        err = sb_issue_zeroout(sb, gdblocks + start + 1,
 558                                        reserved_gdb, GFP_NOFS);
 559                        if (err)
 560                                goto out;
 561                }
 562
 563handle_itb:
 564                /* Initialize group tables of the grop @group */
 565                if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 566                        goto handle_bb;
 567
 568                /* Zero out all of the inode table blocks */
 569                block = group_data[i].inode_table;
 570                ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 571                           block, sbi->s_itb_per_group);
 572                err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 573                                       GFP_NOFS);
 574                if (err)
 575                        goto out;
 576
 577handle_bb:
 578                if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 579                        goto handle_ib;
 580
 581                /* Initialize block bitmap of the @group */
 582                block = group_data[i].block_bitmap;
 583                err = extend_or_restart_transaction(handle, 1);
 584                if (err)
 585                        goto out;
 586
 587                bh = bclean(handle, sb, block);
 588                if (IS_ERR(bh)) {
 589                        err = PTR_ERR(bh);
 590                        bh = NULL;
 591                        goto out;
 592                }
 593                overhead = ext4_group_overhead_blocks(sb, group);
 594                if (overhead != 0) {
 595                        ext4_debug("mark backup superblock %#04llx (+0)\n",
 596                                   start);
 597                        ext4_set_bits(bh->b_data, 0, overhead);
 598                }
 599                ext4_mark_bitmap_end(group_data[i].blocks_count,
 600                                     sb->s_blocksize * 8, bh->b_data);
 601                err = ext4_handle_dirty_metadata(handle, NULL, bh);
 602                if (err)
 603                        goto out;
 604                brelse(bh);
 605
 606handle_ib:
 607                if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 608                        continue;
 609
 610                /* Initialize inode bitmap of the @group */
 611                block = group_data[i].inode_bitmap;
 612                err = extend_or_restart_transaction(handle, 1);
 613                if (err)
 614                        goto out;
 615                /* Mark unused entries in inode bitmap used */
 616                bh = bclean(handle, sb, block);
 617                if (IS_ERR(bh)) {
 618                        err = PTR_ERR(bh);
 619                        bh = NULL;
 620                        goto out;
 621                }
 622
 623                ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 624                                     sb->s_blocksize * 8, bh->b_data);
 625                err = ext4_handle_dirty_metadata(handle, NULL, bh);
 626                if (err)
 627                        goto out;
 628                brelse(bh);
 629        }
 630        bh = NULL;
 631
 632        /* Mark group tables in block bitmap */
 633        for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 634                count = group_table_count[j];
 635                start = (&group_data[0].block_bitmap)[j];
 636                block = start;
 637                for (i = 1; i < flex_gd->count; i++) {
 638                        block += group_table_count[j];
 639                        if (block == (&group_data[i].block_bitmap)[j]) {
 640                                count += group_table_count[j];
 641                                continue;
 642                        }
 643                        err = set_flexbg_block_bitmap(sb, handle,
 644                                                flex_gd, start, count);
 645                        if (err)
 646                                goto out;
 647                        count = group_table_count[j];
 648                        start = (&group_data[i].block_bitmap)[j];
 649                        block = start;
 650                }
 651
 652                if (count) {
 653                        err = set_flexbg_block_bitmap(sb, handle,
 654                                                flex_gd, start, count);
 655                        if (err)
 656                                goto out;
 657                }
 658        }
 659
 660out:
 661        brelse(bh);
 662        err2 = ext4_journal_stop(handle);
 663        if (err2 && !err)
 664                err = err2;
 665
 666        return err;
 667}
 668
 669/*
 670 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 671 * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 672 * calling this for the first time.  In a sparse filesystem it will be the
 673 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 674 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 675 */
 676static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 677                                  unsigned *five, unsigned *seven)
 678{
 679        unsigned *min = three;
 680        int mult = 3;
 681        unsigned ret;
 682
 683        if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 684                                        EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 685                ret = *min;
 686                *min += 1;
 687                return ret;
 688        }
 689
 690        if (*five < *min) {
 691                min = five;
 692                mult = 5;
 693        }
 694        if (*seven < *min) {
 695                min = seven;
 696                mult = 7;
 697        }
 698
 699        ret = *min;
 700        *min *= mult;
 701
 702        return ret;
 703}
 704
 705/*
 706 * Check that all of the backup GDT blocks are held in the primary GDT block.
 707 * It is assumed that they are stored in group order.  Returns the number of
 708 * groups in current filesystem that have BACKUPS, or -ve error code.
 709 */
 710static int verify_reserved_gdb(struct super_block *sb,
 711                               ext4_group_t end,
 712                               struct buffer_head *primary)
 713{
 714        const ext4_fsblk_t blk = primary->b_blocknr;
 715        unsigned three = 1;
 716        unsigned five = 5;
 717        unsigned seven = 7;
 718        unsigned grp;
 719        __le32 *p = (__le32 *)primary->b_data;
 720        int gdbackups = 0;
 721
 722        while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 723                if (le32_to_cpu(*p++) !=
 724                    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 725                        ext4_warning(sb, "reserved GDT %llu"
 726                                     " missing grp %d (%llu)",
 727                                     blk, grp,
 728                                     grp *
 729                                     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 730                                     blk);
 731                        return -EINVAL;
 732                }
 733                if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 734                        return -EFBIG;
 735        }
 736
 737        return gdbackups;
 738}
 739
 740/*
 741 * Called when we need to bring a reserved group descriptor table block into
 742 * use from the resize inode.  The primary copy of the new GDT block currently
 743 * is an indirect block (under the double indirect block in the resize inode).
 744 * The new backup GDT blocks will be stored as leaf blocks in this indirect
 745 * block, in group order.  Even though we know all the block numbers we need,
 746 * we check to ensure that the resize inode has actually reserved these blocks.
 747 *
 748 * Don't need to update the block bitmaps because the blocks are still in use.
 749 *
 750 * We get all of the error cases out of the way, so that we are sure to not
 751 * fail once we start modifying the data on disk, because JBD has no rollback.
 752 */
 753static int add_new_gdb(handle_t *handle, struct inode *inode,
 754                       ext4_group_t group)
 755{
 756        struct super_block *sb = inode->i_sb;
 757        struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 758        unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 759        ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 760        struct buffer_head **o_group_desc, **n_group_desc;
 761        struct buffer_head *dind;
 762        struct buffer_head *gdb_bh;
 763        int gdbackups;
 764        struct ext4_iloc iloc;
 765        __le32 *data;
 766        int err;
 767
 768        if (test_opt(sb, DEBUG))
 769                printk(KERN_DEBUG
 770                       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 771                       gdb_num);
 772
 773        gdb_bh = sb_bread(sb, gdblock);
 774        if (!gdb_bh)
 775                return -EIO;
 776
 777        gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 778        if (gdbackups < 0) {
 779                err = gdbackups;
 780                goto exit_bh;
 781        }
 782
 783        data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 784        dind = sb_bread(sb, le32_to_cpu(*data));
 785        if (!dind) {
 786                err = -EIO;
 787                goto exit_bh;
 788        }
 789
 790        data = (__le32 *)dind->b_data;
 791        if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 792                ext4_warning(sb, "new group %u GDT block %llu not reserved",
 793                             group, gdblock);
 794                err = -EINVAL;
 795                goto exit_dind;
 796        }
 797
 798        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 799        err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 800        if (unlikely(err))
 801                goto exit_dind;
 802
 803        BUFFER_TRACE(gdb_bh, "get_write_access");
 804        err = ext4_journal_get_write_access(handle, gdb_bh);
 805        if (unlikely(err))
 806                goto exit_dind;
 807
 808        BUFFER_TRACE(dind, "get_write_access");
 809        err = ext4_journal_get_write_access(handle, dind);
 810        if (unlikely(err))
 811                ext4_std_error(sb, err);
 812
 813        /* ext4_reserve_inode_write() gets a reference on the iloc */
 814        err = ext4_reserve_inode_write(handle, inode, &iloc);
 815        if (unlikely(err))
 816                goto exit_dind;
 817
 818        n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 819                                     sizeof(struct buffer_head *),
 820                                     GFP_NOFS);
 821        if (!n_group_desc) {
 822                err = -ENOMEM;
 823                ext4_warning(sb, "not enough memory for %lu groups",
 824                             gdb_num + 1);
 825                goto exit_inode;
 826        }
 827
 828        /*
 829         * Finally, we have all of the possible failures behind us...
 830         *
 831         * Remove new GDT block from inode double-indirect block and clear out
 832         * the new GDT block for use (which also "frees" the backup GDT blocks
 833         * from the reserved inode).  We don't need to change the bitmaps for
 834         * these blocks, because they are marked as in-use from being in the
 835         * reserved inode, and will become GDT blocks (primary and backup).
 836         */
 837        data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 838        err = ext4_handle_dirty_metadata(handle, NULL, dind);
 839        if (unlikely(err)) {
 840                ext4_std_error(sb, err);
 841                goto exit_inode;
 842        }
 843        inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 844        ext4_mark_iloc_dirty(handle, inode, &iloc);
 845        memset(gdb_bh->b_data, 0, sb->s_blocksize);
 846        err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 847        if (unlikely(err)) {
 848                ext4_std_error(sb, err);
 849                goto exit_inode;
 850        }
 851        brelse(dind);
 852
 853        o_group_desc = EXT4_SB(sb)->s_group_desc;
 854        memcpy(n_group_desc, o_group_desc,
 855               EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 856        n_group_desc[gdb_num] = gdb_bh;
 857        EXT4_SB(sb)->s_group_desc = n_group_desc;
 858        EXT4_SB(sb)->s_gdb_count++;
 859        ext4_kvfree(o_group_desc);
 860
 861        le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 862        err = ext4_handle_dirty_super(handle, sb);
 863        if (err)
 864                ext4_std_error(sb, err);
 865
 866        return err;
 867
 868exit_inode:
 869        ext4_kvfree(n_group_desc);
 870        brelse(iloc.bh);
 871exit_dind:
 872        brelse(dind);
 873exit_bh:
 874        brelse(gdb_bh);
 875
 876        ext4_debug("leaving with error %d\n", err);
 877        return err;
 878}
 879
 880/*
 881 * add_new_gdb_meta_bg is the sister of add_new_gdb.
 882 */
 883static int add_new_gdb_meta_bg(struct super_block *sb,
 884                               handle_t *handle, ext4_group_t group) {
 885        ext4_fsblk_t gdblock;
 886        struct buffer_head *gdb_bh;
 887        struct buffer_head **o_group_desc, **n_group_desc;
 888        unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 889        int err;
 890
 891        gdblock = ext4_meta_bg_first_block_no(sb, group) +
 892                   ext4_bg_has_super(sb, group);
 893        gdb_bh = sb_bread(sb, gdblock);
 894        if (!gdb_bh)
 895                return -EIO;
 896        n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 897                                     sizeof(struct buffer_head *),
 898                                     GFP_NOFS);
 899        if (!n_group_desc) {
 900                err = -ENOMEM;
 901                ext4_warning(sb, "not enough memory for %lu groups",
 902                             gdb_num + 1);
 903                return err;
 904        }
 905
 906        o_group_desc = EXT4_SB(sb)->s_group_desc;
 907        memcpy(n_group_desc, o_group_desc,
 908               EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 909        n_group_desc[gdb_num] = gdb_bh;
 910        EXT4_SB(sb)->s_group_desc = n_group_desc;
 911        EXT4_SB(sb)->s_gdb_count++;
 912        ext4_kvfree(o_group_desc);
 913        BUFFER_TRACE(gdb_bh, "get_write_access");
 914        err = ext4_journal_get_write_access(handle, gdb_bh);
 915        if (unlikely(err))
 916                brelse(gdb_bh);
 917        return err;
 918}
 919
 920/*
 921 * Called when we are adding a new group which has a backup copy of each of
 922 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 923 * We need to add these reserved backup GDT blocks to the resize inode, so
 924 * that they are kept for future resizing and not allocated to files.
 925 *
 926 * Each reserved backup GDT block will go into a different indirect block.
 927 * The indirect blocks are actually the primary reserved GDT blocks,
 928 * so we know in advance what their block numbers are.  We only get the
 929 * double-indirect block to verify it is pointing to the primary reserved
 930 * GDT blocks so we don't overwrite a data block by accident.  The reserved
 931 * backup GDT blocks are stored in their reserved primary GDT block.
 932 */
 933static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 934                              ext4_group_t group)
 935{
 936        struct super_block *sb = inode->i_sb;
 937        int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 938        struct buffer_head **primary;
 939        struct buffer_head *dind;
 940        struct ext4_iloc iloc;
 941        ext4_fsblk_t blk;
 942        __le32 *data, *end;
 943        int gdbackups = 0;
 944        int res, i;
 945        int err;
 946
 947        primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 948        if (!primary)
 949                return -ENOMEM;
 950
 951        data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 952        dind = sb_bread(sb, le32_to_cpu(*data));
 953        if (!dind) {
 954                err = -EIO;
 955                goto exit_free;
 956        }
 957
 958        blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 959        data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 960                                         EXT4_ADDR_PER_BLOCK(sb));
 961        end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 962
 963        /* Get each reserved primary GDT block and verify it holds backups */
 964        for (res = 0; res < reserved_gdb; res++, blk++) {
 965                if (le32_to_cpu(*data) != blk) {
 966                        ext4_warning(sb, "reserved block %llu"
 967                                     " not at offset %ld",
 968                                     blk,
 969                                     (long)(data - (__le32 *)dind->b_data));
 970                        err = -EINVAL;
 971                        goto exit_bh;
 972                }
 973                primary[res] = sb_bread(sb, blk);
 974                if (!primary[res]) {
 975                        err = -EIO;
 976                        goto exit_bh;
 977                }
 978                gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 979                if (gdbackups < 0) {
 980                        brelse(primary[res]);
 981                        err = gdbackups;
 982                        goto exit_bh;
 983                }
 984                if (++data >= end)
 985                        data = (__le32 *)dind->b_data;
 986        }
 987
 988        for (i = 0; i < reserved_gdb; i++) {
 989                BUFFER_TRACE(primary[i], "get_write_access");
 990                if ((err = ext4_journal_get_write_access(handle, primary[i])))
 991                        goto exit_bh;
 992        }
 993
 994        if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 995                goto exit_bh;
 996
 997        /*
 998         * Finally we can add each of the reserved backup GDT blocks from
 999         * the new group to its reserved primary GDT block.
1000         */
1001        blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1002        for (i = 0; i < reserved_gdb; i++) {
1003                int err2;
1004                data = (__le32 *)primary[i]->b_data;
1005                /* printk("reserving backup %lu[%u] = %lu\n",
1006                       primary[i]->b_blocknr, gdbackups,
1007                       blk + primary[i]->b_blocknr); */
1008                data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1009                err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1010                if (!err)
1011                        err = err2;
1012        }
1013        inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1014        ext4_mark_iloc_dirty(handle, inode, &iloc);
1015
1016exit_bh:
1017        while (--res >= 0)
1018                brelse(primary[res]);
1019        brelse(dind);
1020
1021exit_free:
1022        kfree(primary);
1023
1024        return err;
1025}
1026
1027/*
1028 * Update the backup copies of the ext4 metadata.  These don't need to be part
1029 * of the main resize transaction, because e2fsck will re-write them if there
1030 * is a problem (basically only OOM will cause a problem).  However, we
1031 * _should_ update the backups if possible, in case the primary gets trashed
1032 * for some reason and we need to run e2fsck from a backup superblock.  The
1033 * important part is that the new block and inode counts are in the backup
1034 * superblocks, and the location of the new group metadata in the GDT backups.
1035 *
1036 * We do not need take the s_resize_lock for this, because these
1037 * blocks are not otherwise touched by the filesystem code when it is
1038 * mounted.  We don't need to worry about last changing from
1039 * sbi->s_groups_count, because the worst that can happen is that we
1040 * do not copy the full number of backups at this time.  The resize
1041 * which changed s_groups_count will backup again.
1042 */
1043static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1044                           int size, int meta_bg)
1045{
1046        struct ext4_sb_info *sbi = EXT4_SB(sb);
1047        ext4_group_t last;
1048        const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1049        unsigned three = 1;
1050        unsigned five = 5;
1051        unsigned seven = 7;
1052        ext4_group_t group = 0;
1053        int rest = sb->s_blocksize - size;
1054        handle_t *handle;
1055        int err = 0, err2;
1056
1057        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1058        if (IS_ERR(handle)) {
1059                group = 1;
1060                err = PTR_ERR(handle);
1061                goto exit_err;
1062        }
1063
1064        if (meta_bg == 0) {
1065                group = ext4_list_backups(sb, &three, &five, &seven);
1066                last = sbi->s_groups_count;
1067        } else {
1068                group = ext4_get_group_number(sb, blk_off) + 1;
1069                last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1070        }
1071
1072        while (group < sbi->s_groups_count) {
1073                struct buffer_head *bh;
1074                ext4_fsblk_t backup_block;
1075
1076                /* Out of journal space, and can't get more - abort - so sad */
1077                if (ext4_handle_valid(handle) &&
1078                    handle->h_buffer_credits == 0 &&
1079                    ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1080                    (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1081                        break;
1082
1083                if (meta_bg == 0)
1084                        backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1085                else
1086                        backup_block = (ext4_group_first_block_no(sb, group) +
1087                                        ext4_bg_has_super(sb, group));
1088
1089                bh = sb_getblk(sb, backup_block);
1090                if (unlikely(!bh)) {
1091                        err = -ENOMEM;
1092                        break;
1093                }
1094                ext4_debug("update metadata backup %llu(+%llu)\n",
1095                           backup_block, backup_block -
1096                           ext4_group_first_block_no(sb, group));
1097                BUFFER_TRACE(bh, "get_write_access");
1098                if ((err = ext4_journal_get_write_access(handle, bh)))
1099                        break;
1100                lock_buffer(bh);
1101                memcpy(bh->b_data, data, size);
1102                if (rest)
1103                        memset(bh->b_data + size, 0, rest);
1104                set_buffer_uptodate(bh);
1105                unlock_buffer(bh);
1106                err = ext4_handle_dirty_metadata(handle, NULL, bh);
1107                if (unlikely(err))
1108                        ext4_std_error(sb, err);
1109                brelse(bh);
1110
1111                if (meta_bg == 0)
1112                        group = ext4_list_backups(sb, &three, &five, &seven);
1113                else if (group == last)
1114                        break;
1115                else
1116                        group = last;
1117        }
1118        if ((err2 = ext4_journal_stop(handle)) && !err)
1119                err = err2;
1120
1121        /*
1122         * Ugh! Need to have e2fsck write the backup copies.  It is too
1123         * late to revert the resize, we shouldn't fail just because of
1124         * the backup copies (they are only needed in case of corruption).
1125         *
1126         * However, if we got here we have a journal problem too, so we
1127         * can't really start a transaction to mark the superblock.
1128         * Chicken out and just set the flag on the hope it will be written
1129         * to disk, and if not - we will simply wait until next fsck.
1130         */
1131exit_err:
1132        if (err) {
1133                ext4_warning(sb, "can't update backup for group %u (err %d), "
1134                             "forcing fsck on next reboot", group, err);
1135                sbi->s_mount_state &= ~EXT4_VALID_FS;
1136                sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1137                mark_buffer_dirty(sbi->s_sbh);
1138        }
1139}
1140
1141/*
1142 * ext4_add_new_descs() adds @count group descriptor of groups
1143 * starting at @group
1144 *
1145 * @handle: journal handle
1146 * @sb: super block
1147 * @group: the group no. of the first group desc to be added
1148 * @resize_inode: the resize inode
1149 * @count: number of group descriptors to be added
1150 */
1151static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1152                              ext4_group_t group, struct inode *resize_inode,
1153                              ext4_group_t count)
1154{
1155        struct ext4_sb_info *sbi = EXT4_SB(sb);
1156        struct ext4_super_block *es = sbi->s_es;
1157        struct buffer_head *gdb_bh;
1158        int i, gdb_off, gdb_num, err = 0;
1159        int meta_bg;
1160
1161        meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1162        for (i = 0; i < count; i++, group++) {
1163                int reserved_gdb = ext4_bg_has_super(sb, group) ?
1164                        le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1165
1166                gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1167                gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1168
1169                /*
1170                 * We will only either add reserved group blocks to a backup group
1171                 * or remove reserved blocks for the first group in a new group block.
1172                 * Doing both would be mean more complex code, and sane people don't
1173                 * use non-sparse filesystems anymore.  This is already checked above.
1174                 */
1175                if (gdb_off) {
1176                        gdb_bh = sbi->s_group_desc[gdb_num];
1177                        BUFFER_TRACE(gdb_bh, "get_write_access");
1178                        err = ext4_journal_get_write_access(handle, gdb_bh);
1179
1180                        if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1181                                err = reserve_backup_gdb(handle, resize_inode, group);
1182                } else if (meta_bg != 0) {
1183                        err = add_new_gdb_meta_bg(sb, handle, group);
1184                } else {
1185                        err = add_new_gdb(handle, resize_inode, group);
1186                }
1187                if (err)
1188                        break;
1189        }
1190        return err;
1191}
1192
1193static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1194{
1195        struct buffer_head *bh = sb_getblk(sb, block);
1196        if (unlikely(!bh))
1197                return NULL;
1198        if (!bh_uptodate_or_lock(bh)) {
1199                if (bh_submit_read(bh) < 0) {
1200                        brelse(bh);
1201                        return NULL;
1202                }
1203        }
1204
1205        return bh;
1206}
1207
1208static int ext4_set_bitmap_checksums(struct super_block *sb,
1209                                     ext4_group_t group,
1210                                     struct ext4_group_desc *gdp,
1211                                     struct ext4_new_group_data *group_data)
1212{
1213        struct buffer_head *bh;
1214
1215        if (!ext4_has_metadata_csum(sb))
1216                return 0;
1217
1218        bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1219        if (!bh)
1220                return -EIO;
1221        ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1222                                   EXT4_INODES_PER_GROUP(sb) / 8);
1223        brelse(bh);
1224
1225        bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1226        if (!bh)
1227                return -EIO;
1228        ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1229        brelse(bh);
1230
1231        return 0;
1232}
1233
1234/*
1235 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1236 */
1237static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1238                                struct ext4_new_flex_group_data *flex_gd)
1239{
1240        struct ext4_new_group_data      *group_data = flex_gd->groups;
1241        struct ext4_group_desc          *gdp;
1242        struct ext4_sb_info             *sbi = EXT4_SB(sb);
1243        struct buffer_head              *gdb_bh;
1244        ext4_group_t                    group;
1245        __u16                           *bg_flags = flex_gd->bg_flags;
1246        int                             i, gdb_off, gdb_num, err = 0;
1247        
1248
1249        for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1250                group = group_data->group;
1251
1252                gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1253                gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1254
1255                /*
1256                 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1257                 */
1258                gdb_bh = sbi->s_group_desc[gdb_num];
1259                /* Update group descriptor block for new group */
1260                gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1261                                                 gdb_off * EXT4_DESC_SIZE(sb));
1262
1263                memset(gdp, 0, EXT4_DESC_SIZE(sb));
1264                ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1265                ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1266                err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1267                if (err) {
1268                        ext4_std_error(sb, err);
1269                        break;
1270                }
1271
1272                ext4_inode_table_set(sb, gdp, group_data->inode_table);
1273                ext4_free_group_clusters_set(sb, gdp,
1274                        EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1275                ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1276                if (ext4_has_group_desc_csum(sb))
1277                        ext4_itable_unused_set(sb, gdp,
1278                                               EXT4_INODES_PER_GROUP(sb));
1279                gdp->bg_flags = cpu_to_le16(*bg_flags);
1280                ext4_group_desc_csum_set(sb, group, gdp);
1281
1282                err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1283                if (unlikely(err)) {
1284                        ext4_std_error(sb, err);
1285                        break;
1286                }
1287
1288                /*
1289                 * We can allocate memory for mb_alloc based on the new group
1290                 * descriptor
1291                 */
1292                err = ext4_mb_add_groupinfo(sb, group, gdp);
1293                if (err)
1294                        break;
1295        }
1296        return err;
1297}
1298
1299/*
1300 * ext4_update_super() updates the super block so that the newly added
1301 * groups can be seen by the filesystem.
1302 *
1303 * @sb: super block
1304 * @flex_gd: new added groups
1305 */
1306static void ext4_update_super(struct super_block *sb,
1307                             struct ext4_new_flex_group_data *flex_gd)
1308{
1309        ext4_fsblk_t blocks_count = 0;
1310        ext4_fsblk_t free_blocks = 0;
1311        ext4_fsblk_t reserved_blocks = 0;
1312        struct ext4_new_group_data *group_data = flex_gd->groups;
1313        struct ext4_sb_info *sbi = EXT4_SB(sb);
1314        struct ext4_super_block *es = sbi->s_es;
1315        int i;
1316
1317        BUG_ON(flex_gd->count == 0 || group_data == NULL);
1318        /*
1319         * Make the new blocks and inodes valid next.  We do this before
1320         * increasing the group count so that once the group is enabled,
1321         * all of its blocks and inodes are already valid.
1322         *
1323         * We always allocate group-by-group, then block-by-block or
1324         * inode-by-inode within a group, so enabling these
1325         * blocks/inodes before the group is live won't actually let us
1326         * allocate the new space yet.
1327         */
1328        for (i = 0; i < flex_gd->count; i++) {
1329                blocks_count += group_data[i].blocks_count;
1330                free_blocks += group_data[i].free_blocks_count;
1331        }
1332
1333        reserved_blocks = ext4_r_blocks_count(es) * 100;
1334        reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1335        reserved_blocks *= blocks_count;
1336        do_div(reserved_blocks, 100);
1337
1338        ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1339        ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1340        le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1341                     flex_gd->count);
1342        le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1343                     flex_gd->count);
1344
1345        ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1346        /*
1347         * We need to protect s_groups_count against other CPUs seeing
1348         * inconsistent state in the superblock.
1349         *
1350         * The precise rules we use are:
1351         *
1352         * * Writers must perform a smp_wmb() after updating all
1353         *   dependent data and before modifying the groups count
1354         *
1355         * * Readers must perform an smp_rmb() after reading the groups
1356         *   count and before reading any dependent data.
1357         *
1358         * NB. These rules can be relaxed when checking the group count
1359         * while freeing data, as we can only allocate from a block
1360         * group after serialising against the group count, and we can
1361         * only then free after serialising in turn against that
1362         * allocation.
1363         */
1364        smp_wmb();
1365
1366        /* Update the global fs size fields */
1367        sbi->s_groups_count += flex_gd->count;
1368        sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1369                        (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1370
1371        /* Update the reserved block counts only once the new group is
1372         * active. */
1373        ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1374                                reserved_blocks);
1375
1376        /* Update the free space counts */
1377        percpu_counter_add(&sbi->s_freeclusters_counter,
1378                           EXT4_NUM_B2C(sbi, free_blocks));
1379        percpu_counter_add(&sbi->s_freeinodes_counter,
1380                           EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1381
1382        ext4_debug("free blocks count %llu",
1383                   percpu_counter_read(&sbi->s_freeclusters_counter));
1384        if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1385                                      EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1386            sbi->s_log_groups_per_flex) {
1387                ext4_group_t flex_group;
1388                flex_group = ext4_flex_group(sbi, group_data[0].group);
1389                atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1390                             &sbi->s_flex_groups[flex_group].free_clusters);
1391                atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1392                           &sbi->s_flex_groups[flex_group].free_inodes);
1393        }
1394
1395        /*
1396         * Update the fs overhead information
1397         */
1398        ext4_calculate_overhead(sb);
1399
1400        if (test_opt(sb, DEBUG))
1401                printk(KERN_DEBUG "EXT4-fs: added group %u:"
1402                       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1403                       blocks_count, free_blocks, reserved_blocks);
1404}
1405
1406/* Add a flex group to an fs. Ensure we handle all possible error conditions
1407 * _before_ we start modifying the filesystem, because we cannot abort the
1408 * transaction and not have it write the data to disk.
1409 */
1410static int ext4_flex_group_add(struct super_block *sb,
1411                               struct inode *resize_inode,
1412                               struct ext4_new_flex_group_data *flex_gd)
1413{
1414        struct ext4_sb_info *sbi = EXT4_SB(sb);
1415        struct ext4_super_block *es = sbi->s_es;
1416        ext4_fsblk_t o_blocks_count;
1417        ext4_grpblk_t last;
1418        ext4_group_t group;
1419        handle_t *handle;
1420        unsigned reserved_gdb;
1421        int err = 0, err2 = 0, credit;
1422
1423        BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1424
1425        reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1426        o_blocks_count = ext4_blocks_count(es);
1427        ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1428        BUG_ON(last);
1429
1430        err = setup_new_flex_group_blocks(sb, flex_gd);
1431        if (err)
1432                goto exit;
1433        /*
1434         * We will always be modifying at least the superblock and  GDT
1435         * blocks.  If we are adding a group past the last current GDT block,
1436         * we will also modify the inode and the dindirect block.  If we
1437         * are adding a group with superblock/GDT backups  we will also
1438         * modify each of the reserved GDT dindirect blocks.
1439         */
1440        credit = 3;     /* sb, resize inode, resize inode dindirect */
1441        /* GDT blocks */
1442        credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1443        credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1444        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1445        if (IS_ERR(handle)) {
1446                err = PTR_ERR(handle);
1447                goto exit;
1448        }
1449
1450        BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1451        err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1452        if (err)
1453                goto exit_journal;
1454
1455        group = flex_gd->groups[0].group;
1456        BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1457        err = ext4_add_new_descs(handle, sb, group,
1458                                resize_inode, flex_gd->count);
1459        if (err)
1460                goto exit_journal;
1461
1462        err = ext4_setup_new_descs(handle, sb, flex_gd);
1463        if (err)
1464                goto exit_journal;
1465
1466        ext4_update_super(sb, flex_gd);
1467
1468        err = ext4_handle_dirty_super(handle, sb);
1469
1470exit_journal:
1471        err2 = ext4_journal_stop(handle);
1472        if (!err)
1473                err = err2;
1474
1475        if (!err) {
1476                int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1477                int gdb_num_end = ((group + flex_gd->count - 1) /
1478                                   EXT4_DESC_PER_BLOCK(sb));
1479                int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1480                                EXT4_FEATURE_INCOMPAT_META_BG);
1481                sector_t old_gdb = 0;
1482
1483                update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1484                               sizeof(struct ext4_super_block), 0);
1485                for (; gdb_num <= gdb_num_end; gdb_num++) {
1486                        struct buffer_head *gdb_bh;
1487
1488                        gdb_bh = sbi->s_group_desc[gdb_num];
1489                        if (old_gdb == gdb_bh->b_blocknr)
1490                                continue;
1491                        update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1492                                       gdb_bh->b_size, meta_bg);
1493                        old_gdb = gdb_bh->b_blocknr;
1494                }
1495        }
1496exit:
1497        return err;
1498}
1499
1500static int ext4_setup_next_flex_gd(struct super_block *sb,
1501                                    struct ext4_new_flex_group_data *flex_gd,
1502                                    ext4_fsblk_t n_blocks_count,
1503                                    unsigned long flexbg_size)
1504{
1505        struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1506        struct ext4_new_group_data *group_data = flex_gd->groups;
1507        ext4_fsblk_t o_blocks_count;
1508        ext4_group_t n_group;
1509        ext4_group_t group;
1510        ext4_group_t last_group;
1511        ext4_grpblk_t last;
1512        ext4_grpblk_t blocks_per_group;
1513        unsigned long i;
1514
1515        blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1516
1517        o_blocks_count = ext4_blocks_count(es);
1518
1519        if (o_blocks_count == n_blocks_count)
1520                return 0;
1521
1522        ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1523        BUG_ON(last);
1524        ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1525
1526        last_group = group | (flexbg_size - 1);
1527        if (last_group > n_group)
1528                last_group = n_group;
1529
1530        flex_gd->count = last_group - group + 1;
1531
1532        for (i = 0; i < flex_gd->count; i++) {
1533                int overhead;
1534
1535                group_data[i].group = group + i;
1536                group_data[i].blocks_count = blocks_per_group;
1537                overhead = ext4_group_overhead_blocks(sb, group + i);
1538                group_data[i].free_blocks_count = blocks_per_group - overhead;
1539                if (ext4_has_group_desc_csum(sb)) {
1540                        flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1541                                               EXT4_BG_INODE_UNINIT;
1542                        if (!test_opt(sb, INIT_INODE_TABLE))
1543                                flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1544                } else
1545                        flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1546        }
1547
1548        if (last_group == n_group && ext4_has_group_desc_csum(sb))
1549                /* We need to initialize block bitmap of last group. */
1550                flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1551
1552        if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1553                group_data[i - 1].blocks_count = last + 1;
1554                group_data[i - 1].free_blocks_count -= blocks_per_group-
1555                                        last - 1;
1556        }
1557
1558        return 1;
1559}
1560
1561/* Add group descriptor data to an existing or new group descriptor block.
1562 * Ensure we handle all possible error conditions _before_ we start modifying
1563 * the filesystem, because we cannot abort the transaction and not have it
1564 * write the data to disk.
1565 *
1566 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1567 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1568 *
1569 * We only need to hold the superblock lock while we are actually adding
1570 * in the new group's counts to the superblock.  Prior to that we have
1571 * not really "added" the group at all.  We re-check that we are still
1572 * adding in the last group in case things have changed since verifying.
1573 */
1574int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1575{
1576        struct ext4_new_flex_group_data flex_gd;
1577        struct ext4_sb_info *sbi = EXT4_SB(sb);
1578        struct ext4_super_block *es = sbi->s_es;
1579        int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1580                le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1581        struct inode *inode = NULL;
1582        int gdb_off;
1583        int err;
1584        __u16 bg_flags = 0;
1585
1586        gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1587
1588        if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1589                                        EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1590                ext4_warning(sb, "Can't resize non-sparse filesystem further");
1591                return -EPERM;
1592        }
1593
1594        if (ext4_blocks_count(es) + input->blocks_count <
1595            ext4_blocks_count(es)) {
1596                ext4_warning(sb, "blocks_count overflow");
1597                return -EINVAL;
1598        }
1599
1600        if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1601            le32_to_cpu(es->s_inodes_count)) {
1602                ext4_warning(sb, "inodes_count overflow");
1603                return -EINVAL;
1604        }
1605
1606        if (reserved_gdb || gdb_off == 0) {
1607                if (!EXT4_HAS_COMPAT_FEATURE(sb,
1608                                             EXT4_FEATURE_COMPAT_RESIZE_INODE)
1609                    || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1610                        ext4_warning(sb,
1611                                     "No reserved GDT blocks, can't resize");
1612                        return -EPERM;
1613                }
1614                inode = ext4_iget(sb, EXT4_RESIZE_INO);
1615                if (IS_ERR(inode)) {
1616                        ext4_warning(sb, "Error opening resize inode");
1617                        return PTR_ERR(inode);
1618                }
1619        }
1620
1621
1622        err = verify_group_input(sb, input);
1623        if (err)
1624                goto out;
1625
1626        err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1627        if (err)
1628                goto out;
1629
1630        err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1631        if (err)
1632                goto out;
1633
1634        flex_gd.count = 1;
1635        flex_gd.groups = input;
1636        flex_gd.bg_flags = &bg_flags;
1637        err = ext4_flex_group_add(sb, inode, &flex_gd);
1638out:
1639        iput(inode);
1640        return err;
1641} /* ext4_group_add */
1642
1643/*
1644 * extend a group without checking assuming that checking has been done.
1645 */
1646static int ext4_group_extend_no_check(struct super_block *sb,
1647                                      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1648{
1649        struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1650        handle_t *handle;
1651        int err = 0, err2;
1652
1653        /* We will update the superblock, one block bitmap, and
1654         * one group descriptor via ext4_group_add_blocks().
1655         */
1656        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1657        if (IS_ERR(handle)) {
1658                err = PTR_ERR(handle);
1659                ext4_warning(sb, "error %d on journal start", err);
1660                return err;
1661        }
1662
1663        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1664        err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1665        if (err) {
1666                ext4_warning(sb, "error %d on journal write access", err);
1667                goto errout;
1668        }
1669
1670        ext4_blocks_count_set(es, o_blocks_count + add);
1671        ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1672        ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1673                   o_blocks_count + add);
1674        /* We add the blocks to the bitmap and set the group need init bit */
1675        err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1676        if (err)
1677                goto errout;
1678        ext4_handle_dirty_super(handle, sb);
1679        ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1680                   o_blocks_count + add);
1681errout:
1682        err2 = ext4_journal_stop(handle);
1683        if (err2 && !err)
1684                err = err2;
1685
1686        if (!err) {
1687                if (test_opt(sb, DEBUG))
1688                        printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1689                               "blocks\n", ext4_blocks_count(es));
1690                update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1691                               (char *)es, sizeof(struct ext4_super_block), 0);
1692        }
1693        return err;
1694}
1695
1696/*
1697 * Extend the filesystem to the new number of blocks specified.  This entry
1698 * point is only used to extend the current filesystem to the end of the last
1699 * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1700 * for emergencies (because it has no dependencies on reserved blocks).
1701 *
1702 * If we _really_ wanted, we could use default values to call ext4_group_add()
1703 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1704 * GDT blocks are reserved to grow to the desired size.
1705 */
1706int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1707                      ext4_fsblk_t n_blocks_count)
1708{
1709        ext4_fsblk_t o_blocks_count;
1710        ext4_grpblk_t last;
1711        ext4_grpblk_t add;
1712        struct buffer_head *bh;
1713        int err;
1714        ext4_group_t group;
1715
1716        o_blocks_count = ext4_blocks_count(es);
1717
1718        if (test_opt(sb, DEBUG))
1719                ext4_msg(sb, KERN_DEBUG,
1720                         "extending last group from %llu to %llu blocks",
1721                         o_blocks_count, n_blocks_count);
1722
1723        if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1724                return 0;
1725
1726        if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1727                ext4_msg(sb, KERN_ERR,
1728                         "filesystem too large to resize to %llu blocks safely",
1729                         n_blocks_count);
1730                if (sizeof(sector_t) < 8)
1731                        ext4_warning(sb, "CONFIG_LBDAF not enabled");
1732                return -EINVAL;
1733        }
1734
1735        if (n_blocks_count < o_blocks_count) {
1736                ext4_warning(sb, "can't shrink FS - resize aborted");
1737                return -EINVAL;
1738        }
1739
1740        /* Handle the remaining blocks in the last group only. */
1741        ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1742
1743        if (last == 0) {
1744                ext4_warning(sb, "need to use ext2online to resize further");
1745                return -EPERM;
1746        }
1747
1748        add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1749
1750        if (o_blocks_count + add < o_blocks_count) {
1751                ext4_warning(sb, "blocks_count overflow");
1752                return -EINVAL;
1753        }
1754
1755        if (o_blocks_count + add > n_blocks_count)
1756                add = n_blocks_count - o_blocks_count;
1757
1758        if (o_blocks_count + add < n_blocks_count)
1759                ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1760                             o_blocks_count + add, add);
1761
1762        /* See if the device is actually as big as what was requested */
1763        bh = sb_bread(sb, o_blocks_count + add - 1);
1764        if (!bh) {
1765                ext4_warning(sb, "can't read last block, resize aborted");
1766                return -ENOSPC;
1767        }
1768        brelse(bh);
1769
1770        err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1771        return err;
1772} /* ext4_group_extend */
1773
1774
1775static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1776{
1777        return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1778}
1779
1780/*
1781 * Release the resize inode and drop the resize_inode feature if there
1782 * are no more reserved gdt blocks, and then convert the file system
1783 * to enable meta_bg
1784 */
1785static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1786{
1787        handle_t *handle;
1788        struct ext4_sb_info *sbi = EXT4_SB(sb);
1789        struct ext4_super_block *es = sbi->s_es;
1790        struct ext4_inode_info *ei = EXT4_I(inode);
1791        ext4_fsblk_t nr;
1792        int i, ret, err = 0;
1793        int credits = 1;
1794
1795        ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1796        if (inode) {
1797                if (es->s_reserved_gdt_blocks) {
1798                        ext4_error(sb, "Unexpected non-zero "
1799                                   "s_reserved_gdt_blocks");
1800                        return -EPERM;
1801                }
1802
1803                /* Do a quick sanity check of the resize inode */
1804                if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1805                        goto invalid_resize_inode;
1806                for (i = 0; i < EXT4_N_BLOCKS; i++) {
1807                        if (i == EXT4_DIND_BLOCK) {
1808                                if (ei->i_data[i])
1809                                        continue;
1810                                else
1811                                        goto invalid_resize_inode;
1812                        }
1813                        if (ei->i_data[i])
1814                                goto invalid_resize_inode;
1815                }
1816                credits += 3;   /* block bitmap, bg descriptor, resize inode */
1817        }
1818
1819        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1820        if (IS_ERR(handle))
1821                return PTR_ERR(handle);
1822
1823        BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1824        err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1825        if (err)
1826                goto errout;
1827
1828        EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1829        EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1830        sbi->s_es->s_first_meta_bg =
1831                cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1832
1833        err = ext4_handle_dirty_super(handle, sb);
1834        if (err) {
1835                ext4_std_error(sb, err);
1836                goto errout;
1837        }
1838
1839        if (inode) {
1840                nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1841                ext4_free_blocks(handle, inode, NULL, nr, 1,
1842                                 EXT4_FREE_BLOCKS_METADATA |
1843                                 EXT4_FREE_BLOCKS_FORGET);
1844                ei->i_data[EXT4_DIND_BLOCK] = 0;
1845                inode->i_blocks = 0;
1846
1847                err = ext4_mark_inode_dirty(handle, inode);
1848                if (err)
1849                        ext4_std_error(sb, err);
1850        }
1851
1852errout:
1853        ret = ext4_journal_stop(handle);
1854        if (!err)
1855                err = ret;
1856        return ret;
1857
1858invalid_resize_inode:
1859        ext4_error(sb, "corrupted/inconsistent resize inode");
1860        return -EINVAL;
1861}
1862
1863/*
1864 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1865 *
1866 * @sb: super block of the fs to be resized
1867 * @n_blocks_count: the number of blocks resides in the resized fs
1868 */
1869int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1870{
1871        struct ext4_new_flex_group_data *flex_gd = NULL;
1872        struct ext4_sb_info *sbi = EXT4_SB(sb);
1873        struct ext4_super_block *es = sbi->s_es;
1874        struct buffer_head *bh;
1875        struct inode *resize_inode = NULL;
1876        ext4_grpblk_t add, offset;
1877        unsigned long n_desc_blocks;
1878        unsigned long o_desc_blocks;
1879        ext4_group_t o_group;
1880        ext4_group_t n_group;
1881        ext4_fsblk_t o_blocks_count;
1882        ext4_fsblk_t n_blocks_count_retry = 0;
1883        unsigned long last_update_time = 0;
1884        int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1885        int meta_bg;
1886
1887        /* See if the device is actually as big as what was requested */
1888        bh = sb_bread(sb, n_blocks_count - 1);
1889        if (!bh) {
1890                ext4_warning(sb, "can't read last block, resize aborted");
1891                return -ENOSPC;
1892        }
1893        brelse(bh);
1894
1895retry:
1896        o_blocks_count = ext4_blocks_count(es);
1897
1898        ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1899                 "to %llu blocks", o_blocks_count, n_blocks_count);
1900
1901        if (n_blocks_count < o_blocks_count) {
1902                /* On-line shrinking not supported */
1903                ext4_warning(sb, "can't shrink FS - resize aborted");
1904                return -EINVAL;
1905        }
1906
1907        if (n_blocks_count == o_blocks_count)
1908                /* Nothing need to do */
1909                return 0;
1910
1911        n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1912        if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1913                ext4_warning(sb, "resize would cause inodes_count overflow");
1914                return -EINVAL;
1915        }
1916        ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1917
1918        n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1919        o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1920
1921        meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1922
1923        if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1924                if (meta_bg) {
1925                        ext4_error(sb, "resize_inode and meta_bg enabled "
1926                                   "simultaneously");
1927                        return -EINVAL;
1928                }
1929                if (n_desc_blocks > o_desc_blocks +
1930                    le16_to_cpu(es->s_reserved_gdt_blocks)) {
1931                        n_blocks_count_retry = n_blocks_count;
1932                        n_desc_blocks = o_desc_blocks +
1933                                le16_to_cpu(es->s_reserved_gdt_blocks);
1934                        n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1935                        n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1936                        n_group--; /* set to last group number */
1937                }
1938
1939                if (!resize_inode)
1940                        resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1941                if (IS_ERR(resize_inode)) {
1942                        ext4_warning(sb, "Error opening resize inode");
1943                        return PTR_ERR(resize_inode);
1944                }
1945        }
1946
1947        if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1948                err = ext4_convert_meta_bg(sb, resize_inode);
1949                if (err)
1950                        goto out;
1951                if (resize_inode) {
1952                        iput(resize_inode);
1953                        resize_inode = NULL;
1954                }
1955                if (n_blocks_count_retry) {
1956                        n_blocks_count = n_blocks_count_retry;
1957                        n_blocks_count_retry = 0;
1958                        goto retry;
1959                }
1960        }
1961
1962        /* extend the last group */
1963        if (n_group == o_group)
1964                add = n_blocks_count - o_blocks_count;
1965        else
1966                add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1967        if (add > 0) {
1968                err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1969                if (err)
1970                        goto out;
1971        }
1972
1973        if (ext4_blocks_count(es) == n_blocks_count)
1974                goto out;
1975
1976        err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1977        if (err)
1978                return err;
1979
1980        err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1981        if (err)
1982                goto out;
1983
1984        flex_gd = alloc_flex_gd(flexbg_size);
1985        if (flex_gd == NULL) {
1986                err = -ENOMEM;
1987                goto out;
1988        }
1989
1990        /* Add flex groups. Note that a regular group is a
1991         * flex group with 1 group.
1992         */
1993        while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1994                                              flexbg_size)) {
1995                if (jiffies - last_update_time > HZ * 10) {
1996                        if (last_update_time)
1997                                ext4_msg(sb, KERN_INFO,
1998                                         "resized to %llu blocks",
1999                                         ext4_blocks_count(es));
2000                        last_update_time = jiffies;
2001                }
2002                if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2003                        break;
2004                err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2005                if (unlikely(err))
2006                        break;
2007        }
2008
2009        if (!err && n_blocks_count_retry) {
2010                n_blocks_count = n_blocks_count_retry;
2011                n_blocks_count_retry = 0;
2012                free_flex_gd(flex_gd);
2013                flex_gd = NULL;
2014                goto retry;
2015        }
2016
2017out:
2018        if (flex_gd)
2019                free_flex_gd(flex_gd);
2020        if (resize_inode != NULL)
2021                iput(resize_inode);
2022        ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2023        return err;
2024}
2025