linux/fs/ext4/resize.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/ext4/resize.c
   4 *
   5 * Support for resizing an ext4 filesystem while it is mounted.
   6 *
   7 * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   8 *
   9 * This could probably be made into a module, because it is not often in use.
  10 */
  11
  12
  13#define EXT4FS_DEBUG
  14
  15#include <linux/errno.h>
  16#include <linux/slab.h>
  17
  18#include "ext4_jbd2.h"
  19
  20struct ext4_rcu_ptr {
  21        struct rcu_head rcu;
  22        void *ptr;
  23};
  24
  25static void ext4_rcu_ptr_callback(struct rcu_head *head)
  26{
  27        struct ext4_rcu_ptr *ptr;
  28
  29        ptr = container_of(head, struct ext4_rcu_ptr, rcu);
  30        kvfree(ptr->ptr);
  31        kfree(ptr);
  32}
  33
  34void ext4_kvfree_array_rcu(void *to_free)
  35{
  36        struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
  37
  38        if (ptr) {
  39                ptr->ptr = to_free;
  40                call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
  41                return;
  42        }
  43        synchronize_rcu();
  44        kvfree(to_free);
  45}
  46
  47int ext4_resize_begin(struct super_block *sb)
  48{
  49        struct ext4_sb_info *sbi = EXT4_SB(sb);
  50        int ret = 0;
  51
  52        if (!capable(CAP_SYS_RESOURCE))
  53                return -EPERM;
  54
  55        /*
  56         * If we are not using the primary superblock/GDT copy don't resize,
  57         * because the user tools have no way of handling this.  Probably a
  58         * bad time to do it anyways.
  59         */
  60        if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
  61            le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  62                ext4_warning(sb, "won't resize using backup superblock at %llu",
  63                        (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  64                return -EPERM;
  65        }
  66
  67        /*
  68         * We are not allowed to do online-resizing on a filesystem mounted
  69         * with error, because it can destroy the filesystem easily.
  70         */
  71        if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  72                ext4_warning(sb, "There are errors in the filesystem, "
  73                             "so online resizing is not allowed");
  74                return -EPERM;
  75        }
  76
  77        if (ext4_has_feature_sparse_super2(sb)) {
  78                ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
  79                return -EOPNOTSUPP;
  80        }
  81
  82        if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
  83                                  &EXT4_SB(sb)->s_ext4_flags))
  84                ret = -EBUSY;
  85
  86        return ret;
  87}
  88
  89void ext4_resize_end(struct super_block *sb)
  90{
  91        clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
  92        smp_mb__after_atomic();
  93}
  94
  95static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  96                                             ext4_group_t group) {
  97        return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  98               EXT4_DESC_PER_BLOCK_BITS(sb);
  99}
 100
 101static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
 102                                             ext4_group_t group) {
 103        group = ext4_meta_bg_first_group(sb, group);
 104        return ext4_group_first_block_no(sb, group);
 105}
 106
 107static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
 108                                                ext4_group_t group) {
 109        ext4_grpblk_t overhead;
 110        overhead = ext4_bg_num_gdb(sb, group);
 111        if (ext4_bg_has_super(sb, group))
 112                overhead += 1 +
 113                          le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 114        return overhead;
 115}
 116
 117#define outside(b, first, last) ((b) < (first) || (b) >= (last))
 118#define inside(b, first, last)  ((b) >= (first) && (b) < (last))
 119
 120static int verify_group_input(struct super_block *sb,
 121                              struct ext4_new_group_data *input)
 122{
 123        struct ext4_sb_info *sbi = EXT4_SB(sb);
 124        struct ext4_super_block *es = sbi->s_es;
 125        ext4_fsblk_t start = ext4_blocks_count(es);
 126        ext4_fsblk_t end = start + input->blocks_count;
 127        ext4_group_t group = input->group;
 128        ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
 129        unsigned overhead;
 130        ext4_fsblk_t metaend;
 131        struct buffer_head *bh = NULL;
 132        ext4_grpblk_t free_blocks_count, offset;
 133        int err = -EINVAL;
 134
 135        if (group != sbi->s_groups_count) {
 136                ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 137                             input->group, sbi->s_groups_count);
 138                return -EINVAL;
 139        }
 140
 141        overhead = ext4_group_overhead_blocks(sb, group);
 142        metaend = start + overhead;
 143        input->free_clusters_count = free_blocks_count =
 144                input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 145
 146        if (test_opt(sb, DEBUG))
 147                printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 148                       "(%d free, %u reserved)\n",
 149                       ext4_bg_has_super(sb, input->group) ? "normal" :
 150                       "no-super", input->group, input->blocks_count,
 151                       free_blocks_count, input->reserved_blocks);
 152
 153        ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 154        if (offset != 0)
 155                        ext4_warning(sb, "Last group not full");
 156        else if (input->reserved_blocks > input->blocks_count / 5)
 157                ext4_warning(sb, "Reserved blocks too high (%u)",
 158                             input->reserved_blocks);
 159        else if (free_blocks_count < 0)
 160                ext4_warning(sb, "Bad blocks count %u",
 161                             input->blocks_count);
 162        else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
 163                err = PTR_ERR(bh);
 164                bh = NULL;
 165                ext4_warning(sb, "Cannot read last block (%llu)",
 166                             end - 1);
 167        } else if (outside(input->block_bitmap, start, end))
 168                ext4_warning(sb, "Block bitmap not in group (block %llu)",
 169                             (unsigned long long)input->block_bitmap);
 170        else if (outside(input->inode_bitmap, start, end))
 171                ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 172                             (unsigned long long)input->inode_bitmap);
 173        else if (outside(input->inode_table, start, end) ||
 174                 outside(itend - 1, start, end))
 175                ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 176                             (unsigned long long)input->inode_table, itend - 1);
 177        else if (input->inode_bitmap == input->block_bitmap)
 178                ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 179                             (unsigned long long)input->block_bitmap);
 180        else if (inside(input->block_bitmap, input->inode_table, itend))
 181                ext4_warning(sb, "Block bitmap (%llu) in inode table "
 182                             "(%llu-%llu)",
 183                             (unsigned long long)input->block_bitmap,
 184                             (unsigned long long)input->inode_table, itend - 1);
 185        else if (inside(input->inode_bitmap, input->inode_table, itend))
 186                ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 187                             "(%llu-%llu)",
 188                             (unsigned long long)input->inode_bitmap,
 189                             (unsigned long long)input->inode_table, itend - 1);
 190        else if (inside(input->block_bitmap, start, metaend))
 191                ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 192                             (unsigned long long)input->block_bitmap,
 193                             start, metaend - 1);
 194        else if (inside(input->inode_bitmap, start, metaend))
 195                ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 196                             (unsigned long long)input->inode_bitmap,
 197                             start, metaend - 1);
 198        else if (inside(input->inode_table, start, metaend) ||
 199                 inside(itend - 1, start, metaend))
 200                ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 201                             "(%llu-%llu)",
 202                             (unsigned long long)input->inode_table,
 203                             itend - 1, start, metaend - 1);
 204        else
 205                err = 0;
 206        brelse(bh);
 207
 208        return err;
 209}
 210
 211/*
 212 * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 213 * group each time.
 214 */
 215struct ext4_new_flex_group_data {
 216        struct ext4_new_group_data *groups;     /* new_group_data for groups
 217                                                   in the flex group */
 218        __u16 *bg_flags;                        /* block group flags of groups
 219                                                   in @groups */
 220        ext4_group_t count;                     /* number of groups in @groups
 221                                                 */
 222};
 223
 224/*
 225 * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 226 * @flexbg_size.
 227 *
 228 * Returns NULL on failure otherwise address of the allocated structure.
 229 */
 230static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 231{
 232        struct ext4_new_flex_group_data *flex_gd;
 233
 234        flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 235        if (flex_gd == NULL)
 236                goto out3;
 237
 238        if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
 239                goto out2;
 240        flex_gd->count = flexbg_size;
 241
 242        flex_gd->groups = kmalloc_array(flexbg_size,
 243                                        sizeof(struct ext4_new_group_data),
 244                                        GFP_NOFS);
 245        if (flex_gd->groups == NULL)
 246                goto out2;
 247
 248        flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
 249                                          GFP_NOFS);
 250        if (flex_gd->bg_flags == NULL)
 251                goto out1;
 252
 253        return flex_gd;
 254
 255out1:
 256        kfree(flex_gd->groups);
 257out2:
 258        kfree(flex_gd);
 259out3:
 260        return NULL;
 261}
 262
 263static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 264{
 265        kfree(flex_gd->bg_flags);
 266        kfree(flex_gd->groups);
 267        kfree(flex_gd);
 268}
 269
 270/*
 271 * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 272 * and inode tables for a flex group.
 273 *
 274 * This function is used by 64bit-resize.  Note that this function allocates
 275 * group tables from the 1st group of groups contained by @flexgd, which may
 276 * be a partial of a flex group.
 277 *
 278 * @sb: super block of fs to which the groups belongs
 279 *
 280 * Returns 0 on a successful allocation of the metadata blocks in the
 281 * block group.
 282 */
 283static int ext4_alloc_group_tables(struct super_block *sb,
 284                                struct ext4_new_flex_group_data *flex_gd,
 285                                int flexbg_size)
 286{
 287        struct ext4_new_group_data *group_data = flex_gd->groups;
 288        ext4_fsblk_t start_blk;
 289        ext4_fsblk_t last_blk;
 290        ext4_group_t src_group;
 291        ext4_group_t bb_index = 0;
 292        ext4_group_t ib_index = 0;
 293        ext4_group_t it_index = 0;
 294        ext4_group_t group;
 295        ext4_group_t last_group;
 296        unsigned overhead;
 297        __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 298        int i;
 299
 300        BUG_ON(flex_gd->count == 0 || group_data == NULL);
 301
 302        src_group = group_data[0].group;
 303        last_group  = src_group + flex_gd->count - 1;
 304
 305        BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 306               (last_group & ~(flexbg_size - 1))));
 307next_group:
 308        group = group_data[0].group;
 309        if (src_group >= group_data[0].group + flex_gd->count)
 310                return -ENOSPC;
 311        start_blk = ext4_group_first_block_no(sb, src_group);
 312        last_blk = start_blk + group_data[src_group - group].blocks_count;
 313
 314        overhead = ext4_group_overhead_blocks(sb, src_group);
 315
 316        start_blk += overhead;
 317
 318        /* We collect contiguous blocks as much as possible. */
 319        src_group++;
 320        for (; src_group <= last_group; src_group++) {
 321                overhead = ext4_group_overhead_blocks(sb, src_group);
 322                if (overhead == 0)
 323                        last_blk += group_data[src_group - group].blocks_count;
 324                else
 325                        break;
 326        }
 327
 328        /* Allocate block bitmaps */
 329        for (; bb_index < flex_gd->count; bb_index++) {
 330                if (start_blk >= last_blk)
 331                        goto next_group;
 332                group_data[bb_index].block_bitmap = start_blk++;
 333                group = ext4_get_group_number(sb, start_blk - 1);
 334                group -= group_data[0].group;
 335                group_data[group].mdata_blocks++;
 336                flex_gd->bg_flags[group] &= uninit_mask;
 337        }
 338
 339        /* Allocate inode bitmaps */
 340        for (; ib_index < flex_gd->count; ib_index++) {
 341                if (start_blk >= last_blk)
 342                        goto next_group;
 343                group_data[ib_index].inode_bitmap = start_blk++;
 344                group = ext4_get_group_number(sb, start_blk - 1);
 345                group -= group_data[0].group;
 346                group_data[group].mdata_blocks++;
 347                flex_gd->bg_flags[group] &= uninit_mask;
 348        }
 349
 350        /* Allocate inode tables */
 351        for (; it_index < flex_gd->count; it_index++) {
 352                unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 353                ext4_fsblk_t next_group_start;
 354
 355                if (start_blk + itb > last_blk)
 356                        goto next_group;
 357                group_data[it_index].inode_table = start_blk;
 358                group = ext4_get_group_number(sb, start_blk);
 359                next_group_start = ext4_group_first_block_no(sb, group + 1);
 360                group -= group_data[0].group;
 361
 362                if (start_blk + itb > next_group_start) {
 363                        flex_gd->bg_flags[group + 1] &= uninit_mask;
 364                        overhead = start_blk + itb - next_group_start;
 365                        group_data[group + 1].mdata_blocks += overhead;
 366                        itb -= overhead;
 367                }
 368
 369                group_data[group].mdata_blocks += itb;
 370                flex_gd->bg_flags[group] &= uninit_mask;
 371                start_blk += EXT4_SB(sb)->s_itb_per_group;
 372        }
 373
 374        /* Update free clusters count to exclude metadata blocks */
 375        for (i = 0; i < flex_gd->count; i++) {
 376                group_data[i].free_clusters_count -=
 377                                EXT4_NUM_B2C(EXT4_SB(sb),
 378                                             group_data[i].mdata_blocks);
 379        }
 380
 381        if (test_opt(sb, DEBUG)) {
 382                int i;
 383                group = group_data[0].group;
 384
 385                printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 386                       "%d groups, flexbg size is %d:\n", flex_gd->count,
 387                       flexbg_size);
 388
 389                for (i = 0; i < flex_gd->count; i++) {
 390                        ext4_debug(
 391                               "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
 392                               ext4_bg_has_super(sb, group + i) ? "normal" :
 393                               "no-super", group + i,
 394                               group_data[i].blocks_count,
 395                               group_data[i].free_clusters_count,
 396                               group_data[i].mdata_blocks);
 397                }
 398        }
 399        return 0;
 400}
 401
 402static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 403                                  ext4_fsblk_t blk)
 404{
 405        struct buffer_head *bh;
 406        int err;
 407
 408        bh = sb_getblk(sb, blk);
 409        if (unlikely(!bh))
 410                return ERR_PTR(-ENOMEM);
 411        BUFFER_TRACE(bh, "get_write_access");
 412        err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
 413        if (err) {
 414                brelse(bh);
 415                bh = ERR_PTR(err);
 416        } else {
 417                memset(bh->b_data, 0, sb->s_blocksize);
 418                set_buffer_uptodate(bh);
 419        }
 420
 421        return bh;
 422}
 423
 424static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
 425{
 426        return ext4_journal_ensure_credits_fn(handle, credits,
 427                EXT4_MAX_TRANS_DATA, 0, 0);
 428}
 429
 430/*
 431 * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
 432 *
 433 * Helper function for ext4_setup_new_group_blocks() which set .
 434 *
 435 * @sb: super block
 436 * @handle: journal handle
 437 * @flex_gd: flex group data
 438 */
 439static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 440                        struct ext4_new_flex_group_data *flex_gd,
 441                        ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
 442{
 443        struct ext4_sb_info *sbi = EXT4_SB(sb);
 444        ext4_group_t count = last_cluster - first_cluster + 1;
 445        ext4_group_t count2;
 446
 447        ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
 448                   last_cluster);
 449        for (count2 = count; count > 0;
 450             count -= count2, first_cluster += count2) {
 451                ext4_fsblk_t start;
 452                struct buffer_head *bh;
 453                ext4_group_t group;
 454                int err;
 455
 456                group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
 457                start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
 458                group -= flex_gd->groups[0].group;
 459
 460                count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
 461                if (count2 > count)
 462                        count2 = count;
 463
 464                if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 465                        BUG_ON(flex_gd->count > 1);
 466                        continue;
 467                }
 468
 469                err = ext4_resize_ensure_credits_batch(handle, 1);
 470                if (err < 0)
 471                        return err;
 472
 473                bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 474                if (unlikely(!bh))
 475                        return -ENOMEM;
 476
 477                BUFFER_TRACE(bh, "get_write_access");
 478                err = ext4_journal_get_write_access(handle, sb, bh,
 479                                                    EXT4_JTR_NONE);
 480                if (err) {
 481                        brelse(bh);
 482                        return err;
 483                }
 484                ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
 485                           first_cluster, first_cluster - start, count2);
 486                ext4_set_bits(bh->b_data, first_cluster - start, count2);
 487
 488                err = ext4_handle_dirty_metadata(handle, NULL, bh);
 489                brelse(bh);
 490                if (unlikely(err))
 491                        return err;
 492        }
 493
 494        return 0;
 495}
 496
 497/*
 498 * Set up the block and inode bitmaps, and the inode table for the new groups.
 499 * This doesn't need to be part of the main transaction, since we are only
 500 * changing blocks outside the actual filesystem.  We still do journaling to
 501 * ensure the recovery is correct in case of a failure just after resize.
 502 * If any part of this fails, we simply abort the resize.
 503 *
 504 * setup_new_flex_group_blocks handles a flex group as follow:
 505 *  1. copy super block and GDT, and initialize group tables if necessary.
 506 *     In this step, we only set bits in blocks bitmaps for blocks taken by
 507 *     super block and GDT.
 508 *  2. allocate group tables in block bitmaps, that is, set bits in block
 509 *     bitmap for blocks taken by group tables.
 510 */
 511static int setup_new_flex_group_blocks(struct super_block *sb,
 512                                struct ext4_new_flex_group_data *flex_gd)
 513{
 514        int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 515        ext4_fsblk_t start;
 516        ext4_fsblk_t block;
 517        struct ext4_sb_info *sbi = EXT4_SB(sb);
 518        struct ext4_super_block *es = sbi->s_es;
 519        struct ext4_new_group_data *group_data = flex_gd->groups;
 520        __u16 *bg_flags = flex_gd->bg_flags;
 521        handle_t *handle;
 522        ext4_group_t group, count;
 523        struct buffer_head *bh = NULL;
 524        int reserved_gdb, i, j, err = 0, err2;
 525        int meta_bg;
 526
 527        BUG_ON(!flex_gd->count || !group_data ||
 528               group_data[0].group != sbi->s_groups_count);
 529
 530        reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 531        meta_bg = ext4_has_feature_meta_bg(sb);
 532
 533        /* This transaction may be extended/restarted along the way */
 534        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 535        if (IS_ERR(handle))
 536                return PTR_ERR(handle);
 537
 538        group = group_data[0].group;
 539        for (i = 0; i < flex_gd->count; i++, group++) {
 540                unsigned long gdblocks;
 541                ext4_grpblk_t overhead;
 542
 543                gdblocks = ext4_bg_num_gdb(sb, group);
 544                start = ext4_group_first_block_no(sb, group);
 545
 546                if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 547                        goto handle_itb;
 548
 549                if (meta_bg == 1) {
 550                        ext4_group_t first_group;
 551                        first_group = ext4_meta_bg_first_group(sb, group);
 552                        if (first_group != group + 1 &&
 553                            first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 554                                goto handle_itb;
 555                }
 556
 557                block = start + ext4_bg_has_super(sb, group);
 558                /* Copy all of the GDT blocks into the backup in this group */
 559                for (j = 0; j < gdblocks; j++, block++) {
 560                        struct buffer_head *gdb;
 561
 562                        ext4_debug("update backup group %#04llx\n", block);
 563                        err = ext4_resize_ensure_credits_batch(handle, 1);
 564                        if (err < 0)
 565                                goto out;
 566
 567                        gdb = sb_getblk(sb, block);
 568                        if (unlikely(!gdb)) {
 569                                err = -ENOMEM;
 570                                goto out;
 571                        }
 572
 573                        BUFFER_TRACE(gdb, "get_write_access");
 574                        err = ext4_journal_get_write_access(handle, sb, gdb,
 575                                                            EXT4_JTR_NONE);
 576                        if (err) {
 577                                brelse(gdb);
 578                                goto out;
 579                        }
 580                        memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
 581                                s_group_desc, j)->b_data, gdb->b_size);
 582                        set_buffer_uptodate(gdb);
 583
 584                        err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 585                        if (unlikely(err)) {
 586                                brelse(gdb);
 587                                goto out;
 588                        }
 589                        brelse(gdb);
 590                }
 591
 592                /* Zero out all of the reserved backup group descriptor
 593                 * table blocks
 594                 */
 595                if (ext4_bg_has_super(sb, group)) {
 596                        err = sb_issue_zeroout(sb, gdblocks + start + 1,
 597                                        reserved_gdb, GFP_NOFS);
 598                        if (err)
 599                                goto out;
 600                }
 601
 602handle_itb:
 603                /* Initialize group tables of the grop @group */
 604                if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 605                        goto handle_bb;
 606
 607                /* Zero out all of the inode table blocks */
 608                block = group_data[i].inode_table;
 609                ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 610                           block, sbi->s_itb_per_group);
 611                err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 612                                       GFP_NOFS);
 613                if (err)
 614                        goto out;
 615
 616handle_bb:
 617                if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 618                        goto handle_ib;
 619
 620                /* Initialize block bitmap of the @group */
 621                block = group_data[i].block_bitmap;
 622                err = ext4_resize_ensure_credits_batch(handle, 1);
 623                if (err < 0)
 624                        goto out;
 625
 626                bh = bclean(handle, sb, block);
 627                if (IS_ERR(bh)) {
 628                        err = PTR_ERR(bh);
 629                        goto out;
 630                }
 631                overhead = ext4_group_overhead_blocks(sb, group);
 632                if (overhead != 0) {
 633                        ext4_debug("mark backup superblock %#04llx (+0)\n",
 634                                   start);
 635                        ext4_set_bits(bh->b_data, 0,
 636                                      EXT4_NUM_B2C(sbi, overhead));
 637                }
 638                ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
 639                                     sb->s_blocksize * 8, bh->b_data);
 640                err = ext4_handle_dirty_metadata(handle, NULL, bh);
 641                brelse(bh);
 642                if (err)
 643                        goto out;
 644
 645handle_ib:
 646                if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 647                        continue;
 648
 649                /* Initialize inode bitmap of the @group */
 650                block = group_data[i].inode_bitmap;
 651                err = ext4_resize_ensure_credits_batch(handle, 1);
 652                if (err < 0)
 653                        goto out;
 654                /* Mark unused entries in inode bitmap used */
 655                bh = bclean(handle, sb, block);
 656                if (IS_ERR(bh)) {
 657                        err = PTR_ERR(bh);
 658                        goto out;
 659                }
 660
 661                ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 662                                     sb->s_blocksize * 8, bh->b_data);
 663                err = ext4_handle_dirty_metadata(handle, NULL, bh);
 664                brelse(bh);
 665                if (err)
 666                        goto out;
 667        }
 668
 669        /* Mark group tables in block bitmap */
 670        for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 671                count = group_table_count[j];
 672                start = (&group_data[0].block_bitmap)[j];
 673                block = start;
 674                for (i = 1; i < flex_gd->count; i++) {
 675                        block += group_table_count[j];
 676                        if (block == (&group_data[i].block_bitmap)[j]) {
 677                                count += group_table_count[j];
 678                                continue;
 679                        }
 680                        err = set_flexbg_block_bitmap(sb, handle,
 681                                                      flex_gd,
 682                                                      EXT4_B2C(sbi, start),
 683                                                      EXT4_B2C(sbi,
 684                                                               start + count
 685                                                               - 1));
 686                        if (err)
 687                                goto out;
 688                        count = group_table_count[j];
 689                        start = (&group_data[i].block_bitmap)[j];
 690                        block = start;
 691                }
 692
 693                if (count) {
 694                        err = set_flexbg_block_bitmap(sb, handle,
 695                                                      flex_gd,
 696                                                      EXT4_B2C(sbi, start),
 697                                                      EXT4_B2C(sbi,
 698                                                               start + count
 699                                                               - 1));
 700                        if (err)
 701                                goto out;
 702                }
 703        }
 704
 705out:
 706        err2 = ext4_journal_stop(handle);
 707        if (err2 && !err)
 708                err = err2;
 709
 710        return err;
 711}
 712
 713/*
 714 * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 715 * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 716 * calling this for the first time.  In a sparse filesystem it will be the
 717 * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 718 * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 719 */
 720static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 721                                  unsigned *five, unsigned *seven)
 722{
 723        unsigned *min = three;
 724        int mult = 3;
 725        unsigned ret;
 726
 727        if (!ext4_has_feature_sparse_super(sb)) {
 728                ret = *min;
 729                *min += 1;
 730                return ret;
 731        }
 732
 733        if (*five < *min) {
 734                min = five;
 735                mult = 5;
 736        }
 737        if (*seven < *min) {
 738                min = seven;
 739                mult = 7;
 740        }
 741
 742        ret = *min;
 743        *min *= mult;
 744
 745        return ret;
 746}
 747
 748/*
 749 * Check that all of the backup GDT blocks are held in the primary GDT block.
 750 * It is assumed that they are stored in group order.  Returns the number of
 751 * groups in current filesystem that have BACKUPS, or -ve error code.
 752 */
 753static int verify_reserved_gdb(struct super_block *sb,
 754                               ext4_group_t end,
 755                               struct buffer_head *primary)
 756{
 757        const ext4_fsblk_t blk = primary->b_blocknr;
 758        unsigned three = 1;
 759        unsigned five = 5;
 760        unsigned seven = 7;
 761        unsigned grp;
 762        __le32 *p = (__le32 *)primary->b_data;
 763        int gdbackups = 0;
 764
 765        while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 766                if (le32_to_cpu(*p++) !=
 767                    grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 768                        ext4_warning(sb, "reserved GDT %llu"
 769                                     " missing grp %d (%llu)",
 770                                     blk, grp,
 771                                     grp *
 772                                     (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 773                                     blk);
 774                        return -EINVAL;
 775                }
 776                if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 777                        return -EFBIG;
 778        }
 779
 780        return gdbackups;
 781}
 782
 783/*
 784 * Called when we need to bring a reserved group descriptor table block into
 785 * use from the resize inode.  The primary copy of the new GDT block currently
 786 * is an indirect block (under the double indirect block in the resize inode).
 787 * The new backup GDT blocks will be stored as leaf blocks in this indirect
 788 * block, in group order.  Even though we know all the block numbers we need,
 789 * we check to ensure that the resize inode has actually reserved these blocks.
 790 *
 791 * Don't need to update the block bitmaps because the blocks are still in use.
 792 *
 793 * We get all of the error cases out of the way, so that we are sure to not
 794 * fail once we start modifying the data on disk, because JBD has no rollback.
 795 */
 796static int add_new_gdb(handle_t *handle, struct inode *inode,
 797                       ext4_group_t group)
 798{
 799        struct super_block *sb = inode->i_sb;
 800        struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 801        unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 802        ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 803        struct buffer_head **o_group_desc, **n_group_desc = NULL;
 804        struct buffer_head *dind = NULL;
 805        struct buffer_head *gdb_bh = NULL;
 806        int gdbackups;
 807        struct ext4_iloc iloc = { .bh = NULL };
 808        __le32 *data;
 809        int err;
 810
 811        if (test_opt(sb, DEBUG))
 812                printk(KERN_DEBUG
 813                       "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 814                       gdb_num);
 815
 816        gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 817        if (IS_ERR(gdb_bh))
 818                return PTR_ERR(gdb_bh);
 819
 820        gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 821        if (gdbackups < 0) {
 822                err = gdbackups;
 823                goto errout;
 824        }
 825
 826        data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 827        dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
 828        if (IS_ERR(dind)) {
 829                err = PTR_ERR(dind);
 830                dind = NULL;
 831                goto errout;
 832        }
 833
 834        data = (__le32 *)dind->b_data;
 835        if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 836                ext4_warning(sb, "new group %u GDT block %llu not reserved",
 837                             group, gdblock);
 838                err = -EINVAL;
 839                goto errout;
 840        }
 841
 842        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 843        err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
 844                                            EXT4_JTR_NONE);
 845        if (unlikely(err))
 846                goto errout;
 847
 848        BUFFER_TRACE(gdb_bh, "get_write_access");
 849        err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
 850        if (unlikely(err))
 851                goto errout;
 852
 853        BUFFER_TRACE(dind, "get_write_access");
 854        err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
 855        if (unlikely(err)) {
 856                ext4_std_error(sb, err);
 857                goto errout;
 858        }
 859
 860        /* ext4_reserve_inode_write() gets a reference on the iloc */
 861        err = ext4_reserve_inode_write(handle, inode, &iloc);
 862        if (unlikely(err))
 863                goto errout;
 864
 865        n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 866                                GFP_KERNEL);
 867        if (!n_group_desc) {
 868                err = -ENOMEM;
 869                ext4_warning(sb, "not enough memory for %lu groups",
 870                             gdb_num + 1);
 871                goto errout;
 872        }
 873
 874        /*
 875         * Finally, we have all of the possible failures behind us...
 876         *
 877         * Remove new GDT block from inode double-indirect block and clear out
 878         * the new GDT block for use (which also "frees" the backup GDT blocks
 879         * from the reserved inode).  We don't need to change the bitmaps for
 880         * these blocks, because they are marked as in-use from being in the
 881         * reserved inode, and will become GDT blocks (primary and backup).
 882         */
 883        data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 884        err = ext4_handle_dirty_metadata(handle, NULL, dind);
 885        if (unlikely(err)) {
 886                ext4_std_error(sb, err);
 887                goto errout;
 888        }
 889        inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
 890                           (9 - EXT4_SB(sb)->s_cluster_bits);
 891        ext4_mark_iloc_dirty(handle, inode, &iloc);
 892        memset(gdb_bh->b_data, 0, sb->s_blocksize);
 893        err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 894        if (unlikely(err)) {
 895                ext4_std_error(sb, err);
 896                iloc.bh = NULL;
 897                goto errout;
 898        }
 899        brelse(dind);
 900
 901        rcu_read_lock();
 902        o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 903        memcpy(n_group_desc, o_group_desc,
 904               EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 905        rcu_read_unlock();
 906        n_group_desc[gdb_num] = gdb_bh;
 907        rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 908        EXT4_SB(sb)->s_gdb_count++;
 909        ext4_kvfree_array_rcu(o_group_desc);
 910
 911        lock_buffer(EXT4_SB(sb)->s_sbh);
 912        le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 913        ext4_superblock_csum_set(sb);
 914        unlock_buffer(EXT4_SB(sb)->s_sbh);
 915        err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 916        if (err)
 917                ext4_std_error(sb, err);
 918        return err;
 919errout:
 920        kvfree(n_group_desc);
 921        brelse(iloc.bh);
 922        brelse(dind);
 923        brelse(gdb_bh);
 924
 925        ext4_debug("leaving with error %d\n", err);
 926        return err;
 927}
 928
 929/*
 930 * add_new_gdb_meta_bg is the sister of add_new_gdb.
 931 */
 932static int add_new_gdb_meta_bg(struct super_block *sb,
 933                               handle_t *handle, ext4_group_t group) {
 934        ext4_fsblk_t gdblock;
 935        struct buffer_head *gdb_bh;
 936        struct buffer_head **o_group_desc, **n_group_desc;
 937        unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 938        int err;
 939
 940        gdblock = ext4_meta_bg_first_block_no(sb, group) +
 941                   ext4_bg_has_super(sb, group);
 942        gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 943        if (IS_ERR(gdb_bh))
 944                return PTR_ERR(gdb_bh);
 945        n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 946                                GFP_KERNEL);
 947        if (!n_group_desc) {
 948                brelse(gdb_bh);
 949                err = -ENOMEM;
 950                ext4_warning(sb, "not enough memory for %lu groups",
 951                             gdb_num + 1);
 952                return err;
 953        }
 954
 955        rcu_read_lock();
 956        o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 957        memcpy(n_group_desc, o_group_desc,
 958               EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 959        rcu_read_unlock();
 960        n_group_desc[gdb_num] = gdb_bh;
 961
 962        BUFFER_TRACE(gdb_bh, "get_write_access");
 963        err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
 964        if (err) {
 965                kvfree(n_group_desc);
 966                brelse(gdb_bh);
 967                return err;
 968        }
 969
 970        rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 971        EXT4_SB(sb)->s_gdb_count++;
 972        ext4_kvfree_array_rcu(o_group_desc);
 973        return err;
 974}
 975
 976/*
 977 * Called when we are adding a new group which has a backup copy of each of
 978 * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 979 * We need to add these reserved backup GDT blocks to the resize inode, so
 980 * that they are kept for future resizing and not allocated to files.
 981 *
 982 * Each reserved backup GDT block will go into a different indirect block.
 983 * The indirect blocks are actually the primary reserved GDT blocks,
 984 * so we know in advance what their block numbers are.  We only get the
 985 * double-indirect block to verify it is pointing to the primary reserved
 986 * GDT blocks so we don't overwrite a data block by accident.  The reserved
 987 * backup GDT blocks are stored in their reserved primary GDT block.
 988 */
 989static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 990                              ext4_group_t group)
 991{
 992        struct super_block *sb = inode->i_sb;
 993        int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 994        int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
 995        struct buffer_head **primary;
 996        struct buffer_head *dind;
 997        struct ext4_iloc iloc;
 998        ext4_fsblk_t blk;
 999        __le32 *data, *end;
1000        int gdbackups = 0;
1001        int res, i;
1002        int err;
1003
1004        primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1005        if (!primary)
1006                return -ENOMEM;
1007
1008        data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1009        dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1010        if (IS_ERR(dind)) {
1011                err = PTR_ERR(dind);
1012                dind = NULL;
1013                goto exit_free;
1014        }
1015
1016        blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1017        data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1018                                         EXT4_ADDR_PER_BLOCK(sb));
1019        end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1020
1021        /* Get each reserved primary GDT block and verify it holds backups */
1022        for (res = 0; res < reserved_gdb; res++, blk++) {
1023                if (le32_to_cpu(*data) != blk) {
1024                        ext4_warning(sb, "reserved block %llu"
1025                                     " not at offset %ld",
1026                                     blk,
1027                                     (long)(data - (__le32 *)dind->b_data));
1028                        err = -EINVAL;
1029                        goto exit_bh;
1030                }
1031                primary[res] = ext4_sb_bread(sb, blk, 0);
1032                if (IS_ERR(primary[res])) {
1033                        err = PTR_ERR(primary[res]);
1034                        primary[res] = NULL;
1035                        goto exit_bh;
1036                }
1037                gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1038                if (gdbackups < 0) {
1039                        brelse(primary[res]);
1040                        err = gdbackups;
1041                        goto exit_bh;
1042                }
1043                if (++data >= end)
1044                        data = (__le32 *)dind->b_data;
1045        }
1046
1047        for (i = 0; i < reserved_gdb; i++) {
1048                BUFFER_TRACE(primary[i], "get_write_access");
1049                if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1050                                                         EXT4_JTR_NONE)))
1051                        goto exit_bh;
1052        }
1053
1054        if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1055                goto exit_bh;
1056
1057        /*
1058         * Finally we can add each of the reserved backup GDT blocks from
1059         * the new group to its reserved primary GDT block.
1060         */
1061        blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1062        for (i = 0; i < reserved_gdb; i++) {
1063                int err2;
1064                data = (__le32 *)primary[i]->b_data;
1065                /* printk("reserving backup %lu[%u] = %lu\n",
1066                       primary[i]->b_blocknr, gdbackups,
1067                       blk + primary[i]->b_blocknr); */
1068                data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1069                err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1070                if (!err)
1071                        err = err2;
1072        }
1073
1074        inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1075        ext4_mark_iloc_dirty(handle, inode, &iloc);
1076
1077exit_bh:
1078        while (--res >= 0)
1079                brelse(primary[res]);
1080        brelse(dind);
1081
1082exit_free:
1083        kfree(primary);
1084
1085        return err;
1086}
1087
1088/*
1089 * Update the backup copies of the ext4 metadata.  These don't need to be part
1090 * of the main resize transaction, because e2fsck will re-write them if there
1091 * is a problem (basically only OOM will cause a problem).  However, we
1092 * _should_ update the backups if possible, in case the primary gets trashed
1093 * for some reason and we need to run e2fsck from a backup superblock.  The
1094 * important part is that the new block and inode counts are in the backup
1095 * superblocks, and the location of the new group metadata in the GDT backups.
1096 *
1097 * We do not need take the s_resize_lock for this, because these
1098 * blocks are not otherwise touched by the filesystem code when it is
1099 * mounted.  We don't need to worry about last changing from
1100 * sbi->s_groups_count, because the worst that can happen is that we
1101 * do not copy the full number of backups at this time.  The resize
1102 * which changed s_groups_count will backup again.
1103 */
1104static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1105                           int size, int meta_bg)
1106{
1107        struct ext4_sb_info *sbi = EXT4_SB(sb);
1108        ext4_group_t last;
1109        const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1110        unsigned three = 1;
1111        unsigned five = 5;
1112        unsigned seven = 7;
1113        ext4_group_t group = 0;
1114        int rest = sb->s_blocksize - size;
1115        handle_t *handle;
1116        int err = 0, err2;
1117
1118        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1119        if (IS_ERR(handle)) {
1120                group = 1;
1121                err = PTR_ERR(handle);
1122                goto exit_err;
1123        }
1124
1125        if (meta_bg == 0) {
1126                group = ext4_list_backups(sb, &three, &five, &seven);
1127                last = sbi->s_groups_count;
1128        } else {
1129                group = ext4_get_group_number(sb, blk_off) + 1;
1130                last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1131        }
1132
1133        while (group < sbi->s_groups_count) {
1134                struct buffer_head *bh;
1135                ext4_fsblk_t backup_block;
1136
1137                /* Out of journal space, and can't get more - abort - so sad */
1138                err = ext4_resize_ensure_credits_batch(handle, 1);
1139                if (err < 0)
1140                        break;
1141
1142                if (meta_bg == 0)
1143                        backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1144                else
1145                        backup_block = (ext4_group_first_block_no(sb, group) +
1146                                        ext4_bg_has_super(sb, group));
1147
1148                bh = sb_getblk(sb, backup_block);
1149                if (unlikely(!bh)) {
1150                        err = -ENOMEM;
1151                        break;
1152                }
1153                ext4_debug("update metadata backup %llu(+%llu)\n",
1154                           backup_block, backup_block -
1155                           ext4_group_first_block_no(sb, group));
1156                BUFFER_TRACE(bh, "get_write_access");
1157                if ((err = ext4_journal_get_write_access(handle, sb, bh,
1158                                                         EXT4_JTR_NONE)))
1159                        break;
1160                lock_buffer(bh);
1161                memcpy(bh->b_data, data, size);
1162                if (rest)
1163                        memset(bh->b_data + size, 0, rest);
1164                set_buffer_uptodate(bh);
1165                unlock_buffer(bh);
1166                err = ext4_handle_dirty_metadata(handle, NULL, bh);
1167                if (unlikely(err))
1168                        ext4_std_error(sb, err);
1169                brelse(bh);
1170
1171                if (meta_bg == 0)
1172                        group = ext4_list_backups(sb, &three, &five, &seven);
1173                else if (group == last)
1174                        break;
1175                else
1176                        group = last;
1177        }
1178        if ((err2 = ext4_journal_stop(handle)) && !err)
1179                err = err2;
1180
1181        /*
1182         * Ugh! Need to have e2fsck write the backup copies.  It is too
1183         * late to revert the resize, we shouldn't fail just because of
1184         * the backup copies (they are only needed in case of corruption).
1185         *
1186         * However, if we got here we have a journal problem too, so we
1187         * can't really start a transaction to mark the superblock.
1188         * Chicken out and just set the flag on the hope it will be written
1189         * to disk, and if not - we will simply wait until next fsck.
1190         */
1191exit_err:
1192        if (err) {
1193                ext4_warning(sb, "can't update backup for group %u (err %d), "
1194                             "forcing fsck on next reboot", group, err);
1195                sbi->s_mount_state &= ~EXT4_VALID_FS;
1196                sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1197                mark_buffer_dirty(sbi->s_sbh);
1198        }
1199}
1200
1201/*
1202 * ext4_add_new_descs() adds @count group descriptor of groups
1203 * starting at @group
1204 *
1205 * @handle: journal handle
1206 * @sb: super block
1207 * @group: the group no. of the first group desc to be added
1208 * @resize_inode: the resize inode
1209 * @count: number of group descriptors to be added
1210 */
1211static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1212                              ext4_group_t group, struct inode *resize_inode,
1213                              ext4_group_t count)
1214{
1215        struct ext4_sb_info *sbi = EXT4_SB(sb);
1216        struct ext4_super_block *es = sbi->s_es;
1217        struct buffer_head *gdb_bh;
1218        int i, gdb_off, gdb_num, err = 0;
1219        int meta_bg;
1220
1221        meta_bg = ext4_has_feature_meta_bg(sb);
1222        for (i = 0; i < count; i++, group++) {
1223                int reserved_gdb = ext4_bg_has_super(sb, group) ?
1224                        le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1225
1226                gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1227                gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1228
1229                /*
1230                 * We will only either add reserved group blocks to a backup group
1231                 * or remove reserved blocks for the first group in a new group block.
1232                 * Doing both would be mean more complex code, and sane people don't
1233                 * use non-sparse filesystems anymore.  This is already checked above.
1234                 */
1235                if (gdb_off) {
1236                        gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1237                                                     gdb_num);
1238                        BUFFER_TRACE(gdb_bh, "get_write_access");
1239                        err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1240                                                            EXT4_JTR_NONE);
1241
1242                        if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1243                                err = reserve_backup_gdb(handle, resize_inode, group);
1244                } else if (meta_bg != 0) {
1245                        err = add_new_gdb_meta_bg(sb, handle, group);
1246                } else {
1247                        err = add_new_gdb(handle, resize_inode, group);
1248                }
1249                if (err)
1250                        break;
1251        }
1252        return err;
1253}
1254
1255static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1256{
1257        struct buffer_head *bh = sb_getblk(sb, block);
1258        if (unlikely(!bh))
1259                return NULL;
1260        if (!bh_uptodate_or_lock(bh)) {
1261                if (ext4_read_bh(bh, 0, NULL) < 0) {
1262                        brelse(bh);
1263                        return NULL;
1264                }
1265        }
1266
1267        return bh;
1268}
1269
1270static int ext4_set_bitmap_checksums(struct super_block *sb,
1271                                     ext4_group_t group,
1272                                     struct ext4_group_desc *gdp,
1273                                     struct ext4_new_group_data *group_data)
1274{
1275        struct buffer_head *bh;
1276
1277        if (!ext4_has_metadata_csum(sb))
1278                return 0;
1279
1280        bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1281        if (!bh)
1282                return -EIO;
1283        ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1284                                   EXT4_INODES_PER_GROUP(sb) / 8);
1285        brelse(bh);
1286
1287        bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1288        if (!bh)
1289                return -EIO;
1290        ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1291        brelse(bh);
1292
1293        return 0;
1294}
1295
1296/*
1297 * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1298 */
1299static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1300                                struct ext4_new_flex_group_data *flex_gd)
1301{
1302        struct ext4_new_group_data      *group_data = flex_gd->groups;
1303        struct ext4_group_desc          *gdp;
1304        struct ext4_sb_info             *sbi = EXT4_SB(sb);
1305        struct buffer_head              *gdb_bh;
1306        ext4_group_t                    group;
1307        __u16                           *bg_flags = flex_gd->bg_flags;
1308        int                             i, gdb_off, gdb_num, err = 0;
1309
1310
1311        for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1312                group = group_data->group;
1313
1314                gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1315                gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1316
1317                /*
1318                 * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1319                 */
1320                gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1321                /* Update group descriptor block for new group */
1322                gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1323                                                 gdb_off * EXT4_DESC_SIZE(sb));
1324
1325                memset(gdp, 0, EXT4_DESC_SIZE(sb));
1326                ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1327                ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1328                err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1329                if (err) {
1330                        ext4_std_error(sb, err);
1331                        break;
1332                }
1333
1334                ext4_inode_table_set(sb, gdp, group_data->inode_table);
1335                ext4_free_group_clusters_set(sb, gdp,
1336                                             group_data->free_clusters_count);
1337                ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1338                if (ext4_has_group_desc_csum(sb))
1339                        ext4_itable_unused_set(sb, gdp,
1340                                               EXT4_INODES_PER_GROUP(sb));
1341                gdp->bg_flags = cpu_to_le16(*bg_flags);
1342                ext4_group_desc_csum_set(sb, group, gdp);
1343
1344                err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1345                if (unlikely(err)) {
1346                        ext4_std_error(sb, err);
1347                        break;
1348                }
1349
1350                /*
1351                 * We can allocate memory for mb_alloc based on the new group
1352                 * descriptor
1353                 */
1354                err = ext4_mb_add_groupinfo(sb, group, gdp);
1355                if (err)
1356                        break;
1357        }
1358        return err;
1359}
1360
1361/*
1362 * ext4_update_super() updates the super block so that the newly added
1363 * groups can be seen by the filesystem.
1364 *
1365 * @sb: super block
1366 * @flex_gd: new added groups
1367 */
1368static void ext4_update_super(struct super_block *sb,
1369                             struct ext4_new_flex_group_data *flex_gd)
1370{
1371        ext4_fsblk_t blocks_count = 0;
1372        ext4_fsblk_t free_blocks = 0;
1373        ext4_fsblk_t reserved_blocks = 0;
1374        struct ext4_new_group_data *group_data = flex_gd->groups;
1375        struct ext4_sb_info *sbi = EXT4_SB(sb);
1376        struct ext4_super_block *es = sbi->s_es;
1377        int i;
1378
1379        BUG_ON(flex_gd->count == 0 || group_data == NULL);
1380        /*
1381         * Make the new blocks and inodes valid next.  We do this before
1382         * increasing the group count so that once the group is enabled,
1383         * all of its blocks and inodes are already valid.
1384         *
1385         * We always allocate group-by-group, then block-by-block or
1386         * inode-by-inode within a group, so enabling these
1387         * blocks/inodes before the group is live won't actually let us
1388         * allocate the new space yet.
1389         */
1390        for (i = 0; i < flex_gd->count; i++) {
1391                blocks_count += group_data[i].blocks_count;
1392                free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1393        }
1394
1395        reserved_blocks = ext4_r_blocks_count(es) * 100;
1396        reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1397        reserved_blocks *= blocks_count;
1398        do_div(reserved_blocks, 100);
1399
1400        lock_buffer(sbi->s_sbh);
1401        ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1402        ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1403        le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1404                     flex_gd->count);
1405        le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1406                     flex_gd->count);
1407
1408        ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1409        /*
1410         * We need to protect s_groups_count against other CPUs seeing
1411         * inconsistent state in the superblock.
1412         *
1413         * The precise rules we use are:
1414         *
1415         * * Writers must perform a smp_wmb() after updating all
1416         *   dependent data and before modifying the groups count
1417         *
1418         * * Readers must perform an smp_rmb() after reading the groups
1419         *   count and before reading any dependent data.
1420         *
1421         * NB. These rules can be relaxed when checking the group count
1422         * while freeing data, as we can only allocate from a block
1423         * group after serialising against the group count, and we can
1424         * only then free after serialising in turn against that
1425         * allocation.
1426         */
1427        smp_wmb();
1428
1429        /* Update the global fs size fields */
1430        sbi->s_groups_count += flex_gd->count;
1431        sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1432                        (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1433
1434        /* Update the reserved block counts only once the new group is
1435         * active. */
1436        ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1437                                reserved_blocks);
1438        ext4_superblock_csum_set(sb);
1439        unlock_buffer(sbi->s_sbh);
1440
1441        /* Update the free space counts */
1442        percpu_counter_add(&sbi->s_freeclusters_counter,
1443                           EXT4_NUM_B2C(sbi, free_blocks));
1444        percpu_counter_add(&sbi->s_freeinodes_counter,
1445                           EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1446
1447        ext4_debug("free blocks count %llu",
1448                   percpu_counter_read(&sbi->s_freeclusters_counter));
1449        if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1450                ext4_group_t flex_group;
1451                struct flex_groups *fg;
1452
1453                flex_group = ext4_flex_group(sbi, group_data[0].group);
1454                fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1455                atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1456                             &fg->free_clusters);
1457                atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1458                           &fg->free_inodes);
1459        }
1460
1461        /*
1462         * Update the fs overhead information
1463         */
1464        ext4_calculate_overhead(sb);
1465
1466        if (test_opt(sb, DEBUG))
1467                printk(KERN_DEBUG "EXT4-fs: added group %u:"
1468                       "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1469                       blocks_count, free_blocks, reserved_blocks);
1470}
1471
1472/* Add a flex group to an fs. Ensure we handle all possible error conditions
1473 * _before_ we start modifying the filesystem, because we cannot abort the
1474 * transaction and not have it write the data to disk.
1475 */
1476static int ext4_flex_group_add(struct super_block *sb,
1477                               struct inode *resize_inode,
1478                               struct ext4_new_flex_group_data *flex_gd)
1479{
1480        struct ext4_sb_info *sbi = EXT4_SB(sb);
1481        struct ext4_super_block *es = sbi->s_es;
1482        ext4_fsblk_t o_blocks_count;
1483        ext4_grpblk_t last;
1484        ext4_group_t group;
1485        handle_t *handle;
1486        unsigned reserved_gdb;
1487        int err = 0, err2 = 0, credit;
1488
1489        BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1490
1491        reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1492        o_blocks_count = ext4_blocks_count(es);
1493        ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1494        BUG_ON(last);
1495
1496        err = setup_new_flex_group_blocks(sb, flex_gd);
1497        if (err)
1498                goto exit;
1499        /*
1500         * We will always be modifying at least the superblock and  GDT
1501         * blocks.  If we are adding a group past the last current GDT block,
1502         * we will also modify the inode and the dindirect block.  If we
1503         * are adding a group with superblock/GDT backups  we will also
1504         * modify each of the reserved GDT dindirect blocks.
1505         */
1506        credit = 3;     /* sb, resize inode, resize inode dindirect */
1507        /* GDT blocks */
1508        credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1509        credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1510        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1511        if (IS_ERR(handle)) {
1512                err = PTR_ERR(handle);
1513                goto exit;
1514        }
1515
1516        BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1517        err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1518                                            EXT4_JTR_NONE);
1519        if (err)
1520                goto exit_journal;
1521
1522        group = flex_gd->groups[0].group;
1523        BUG_ON(group != sbi->s_groups_count);
1524        err = ext4_add_new_descs(handle, sb, group,
1525                                resize_inode, flex_gd->count);
1526        if (err)
1527                goto exit_journal;
1528
1529        err = ext4_setup_new_descs(handle, sb, flex_gd);
1530        if (err)
1531                goto exit_journal;
1532
1533        ext4_update_super(sb, flex_gd);
1534
1535        err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1536
1537exit_journal:
1538        err2 = ext4_journal_stop(handle);
1539        if (!err)
1540                err = err2;
1541
1542        if (!err) {
1543                int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1544                int gdb_num_end = ((group + flex_gd->count - 1) /
1545                                   EXT4_DESC_PER_BLOCK(sb));
1546                int meta_bg = ext4_has_feature_meta_bg(sb);
1547                sector_t old_gdb = 0;
1548
1549                update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1550                               sizeof(struct ext4_super_block), 0);
1551                for (; gdb_num <= gdb_num_end; gdb_num++) {
1552                        struct buffer_head *gdb_bh;
1553
1554                        gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1555                                                     gdb_num);
1556                        if (old_gdb == gdb_bh->b_blocknr)
1557                                continue;
1558                        update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1559                                       gdb_bh->b_size, meta_bg);
1560                        old_gdb = gdb_bh->b_blocknr;
1561                }
1562        }
1563exit:
1564        return err;
1565}
1566
1567static int ext4_setup_next_flex_gd(struct super_block *sb,
1568                                    struct ext4_new_flex_group_data *flex_gd,
1569                                    ext4_fsblk_t n_blocks_count,
1570                                    unsigned long flexbg_size)
1571{
1572        struct ext4_sb_info *sbi = EXT4_SB(sb);
1573        struct ext4_super_block *es = sbi->s_es;
1574        struct ext4_new_group_data *group_data = flex_gd->groups;
1575        ext4_fsblk_t o_blocks_count;
1576        ext4_group_t n_group;
1577        ext4_group_t group;
1578        ext4_group_t last_group;
1579        ext4_grpblk_t last;
1580        ext4_grpblk_t clusters_per_group;
1581        unsigned long i;
1582
1583        clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1584
1585        o_blocks_count = ext4_blocks_count(es);
1586
1587        if (o_blocks_count == n_blocks_count)
1588                return 0;
1589
1590        ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1591        BUG_ON(last);
1592        ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1593
1594        last_group = group | (flexbg_size - 1);
1595        if (last_group > n_group)
1596                last_group = n_group;
1597
1598        flex_gd->count = last_group - group + 1;
1599
1600        for (i = 0; i < flex_gd->count; i++) {
1601                int overhead;
1602
1603                group_data[i].group = group + i;
1604                group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1605                overhead = ext4_group_overhead_blocks(sb, group + i);
1606                group_data[i].mdata_blocks = overhead;
1607                group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1608                if (ext4_has_group_desc_csum(sb)) {
1609                        flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1610                                               EXT4_BG_INODE_UNINIT;
1611                        if (!test_opt(sb, INIT_INODE_TABLE))
1612                                flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1613                } else
1614                        flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1615        }
1616
1617        if (last_group == n_group && ext4_has_group_desc_csum(sb))
1618                /* We need to initialize block bitmap of last group. */
1619                flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1620
1621        if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1622                group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1623                group_data[i - 1].free_clusters_count -= clusters_per_group -
1624                                                       last - 1;
1625        }
1626
1627        return 1;
1628}
1629
1630/* Add group descriptor data to an existing or new group descriptor block.
1631 * Ensure we handle all possible error conditions _before_ we start modifying
1632 * the filesystem, because we cannot abort the transaction and not have it
1633 * write the data to disk.
1634 *
1635 * If we are on a GDT block boundary, we need to get the reserved GDT block.
1636 * Otherwise, we may need to add backup GDT blocks for a sparse group.
1637 *
1638 * We only need to hold the superblock lock while we are actually adding
1639 * in the new group's counts to the superblock.  Prior to that we have
1640 * not really "added" the group at all.  We re-check that we are still
1641 * adding in the last group in case things have changed since verifying.
1642 */
1643int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1644{
1645        struct ext4_new_flex_group_data flex_gd;
1646        struct ext4_sb_info *sbi = EXT4_SB(sb);
1647        struct ext4_super_block *es = sbi->s_es;
1648        int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1649                le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1650        struct inode *inode = NULL;
1651        int gdb_off;
1652        int err;
1653        __u16 bg_flags = 0;
1654
1655        gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1656
1657        if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1658                ext4_warning(sb, "Can't resize non-sparse filesystem further");
1659                return -EPERM;
1660        }
1661
1662        if (ext4_blocks_count(es) + input->blocks_count <
1663            ext4_blocks_count(es)) {
1664                ext4_warning(sb, "blocks_count overflow");
1665                return -EINVAL;
1666        }
1667
1668        if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1669            le32_to_cpu(es->s_inodes_count)) {
1670                ext4_warning(sb, "inodes_count overflow");
1671                return -EINVAL;
1672        }
1673
1674        if (reserved_gdb || gdb_off == 0) {
1675                if (!ext4_has_feature_resize_inode(sb) ||
1676                    !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1677                        ext4_warning(sb,
1678                                     "No reserved GDT blocks, can't resize");
1679                        return -EPERM;
1680                }
1681                inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1682                if (IS_ERR(inode)) {
1683                        ext4_warning(sb, "Error opening resize inode");
1684                        return PTR_ERR(inode);
1685                }
1686        }
1687
1688
1689        err = verify_group_input(sb, input);
1690        if (err)
1691                goto out;
1692
1693        err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1694        if (err)
1695                goto out;
1696
1697        err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1698        if (err)
1699                goto out;
1700
1701        flex_gd.count = 1;
1702        flex_gd.groups = input;
1703        flex_gd.bg_flags = &bg_flags;
1704        err = ext4_flex_group_add(sb, inode, &flex_gd);
1705out:
1706        iput(inode);
1707        return err;
1708} /* ext4_group_add */
1709
1710/*
1711 * extend a group without checking assuming that checking has been done.
1712 */
1713static int ext4_group_extend_no_check(struct super_block *sb,
1714                                      ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1715{
1716        struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1717        handle_t *handle;
1718        int err = 0, err2;
1719
1720        /* We will update the superblock, one block bitmap, and
1721         * one group descriptor via ext4_group_add_blocks().
1722         */
1723        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1724        if (IS_ERR(handle)) {
1725                err = PTR_ERR(handle);
1726                ext4_warning(sb, "error %d on journal start", err);
1727                return err;
1728        }
1729
1730        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1731        err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1732                                            EXT4_JTR_NONE);
1733        if (err) {
1734                ext4_warning(sb, "error %d on journal write access", err);
1735                goto errout;
1736        }
1737
1738        lock_buffer(EXT4_SB(sb)->s_sbh);
1739        ext4_blocks_count_set(es, o_blocks_count + add);
1740        ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1741        ext4_superblock_csum_set(sb);
1742        unlock_buffer(EXT4_SB(sb)->s_sbh);
1743        ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1744                   o_blocks_count + add);
1745        /* We add the blocks to the bitmap and set the group need init bit */
1746        err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1747        if (err)
1748                goto errout;
1749        ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1750        ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1751                   o_blocks_count + add);
1752errout:
1753        err2 = ext4_journal_stop(handle);
1754        if (err2 && !err)
1755                err = err2;
1756
1757        if (!err) {
1758                if (test_opt(sb, DEBUG))
1759                        printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1760                               "blocks\n", ext4_blocks_count(es));
1761                update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1762                               (char *)es, sizeof(struct ext4_super_block), 0);
1763        }
1764        return err;
1765}
1766
1767/*
1768 * Extend the filesystem to the new number of blocks specified.  This entry
1769 * point is only used to extend the current filesystem to the end of the last
1770 * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1771 * for emergencies (because it has no dependencies on reserved blocks).
1772 *
1773 * If we _really_ wanted, we could use default values to call ext4_group_add()
1774 * allow the "remount" trick to work for arbitrary resizing, assuming enough
1775 * GDT blocks are reserved to grow to the desired size.
1776 */
1777int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1778                      ext4_fsblk_t n_blocks_count)
1779{
1780        ext4_fsblk_t o_blocks_count;
1781        ext4_grpblk_t last;
1782        ext4_grpblk_t add;
1783        struct buffer_head *bh;
1784        int err;
1785        ext4_group_t group;
1786
1787        o_blocks_count = ext4_blocks_count(es);
1788
1789        if (test_opt(sb, DEBUG))
1790                ext4_msg(sb, KERN_DEBUG,
1791                         "extending last group from %llu to %llu blocks",
1792                         o_blocks_count, n_blocks_count);
1793
1794        if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1795                return 0;
1796
1797        if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1798                ext4_msg(sb, KERN_ERR,
1799                         "filesystem too large to resize to %llu blocks safely",
1800                         n_blocks_count);
1801                return -EINVAL;
1802        }
1803
1804        if (n_blocks_count < o_blocks_count) {
1805                ext4_warning(sb, "can't shrink FS - resize aborted");
1806                return -EINVAL;
1807        }
1808
1809        /* Handle the remaining blocks in the last group only. */
1810        ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1811
1812        if (last == 0) {
1813                ext4_warning(sb, "need to use ext2online to resize further");
1814                return -EPERM;
1815        }
1816
1817        add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1818
1819        if (o_blocks_count + add < o_blocks_count) {
1820                ext4_warning(sb, "blocks_count overflow");
1821                return -EINVAL;
1822        }
1823
1824        if (o_blocks_count + add > n_blocks_count)
1825                add = n_blocks_count - o_blocks_count;
1826
1827        if (o_blocks_count + add < n_blocks_count)
1828                ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1829                             o_blocks_count + add, add);
1830
1831        /* See if the device is actually as big as what was requested */
1832        bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1833        if (IS_ERR(bh)) {
1834                ext4_warning(sb, "can't read last block, resize aborted");
1835                return -ENOSPC;
1836        }
1837        brelse(bh);
1838
1839        err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1840        return err;
1841} /* ext4_group_extend */
1842
1843
1844static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1845{
1846        return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1847}
1848
1849/*
1850 * Release the resize inode and drop the resize_inode feature if there
1851 * are no more reserved gdt blocks, and then convert the file system
1852 * to enable meta_bg
1853 */
1854static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1855{
1856        handle_t *handle;
1857        struct ext4_sb_info *sbi = EXT4_SB(sb);
1858        struct ext4_super_block *es = sbi->s_es;
1859        struct ext4_inode_info *ei = EXT4_I(inode);
1860        ext4_fsblk_t nr;
1861        int i, ret, err = 0;
1862        int credits = 1;
1863
1864        ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1865        if (inode) {
1866                if (es->s_reserved_gdt_blocks) {
1867                        ext4_error(sb, "Unexpected non-zero "
1868                                   "s_reserved_gdt_blocks");
1869                        return -EPERM;
1870                }
1871
1872                /* Do a quick sanity check of the resize inode */
1873                if (inode->i_blocks != 1 << (inode->i_blkbits -
1874                                             (9 - sbi->s_cluster_bits)))
1875                        goto invalid_resize_inode;
1876                for (i = 0; i < EXT4_N_BLOCKS; i++) {
1877                        if (i == EXT4_DIND_BLOCK) {
1878                                if (ei->i_data[i])
1879                                        continue;
1880                                else
1881                                        goto invalid_resize_inode;
1882                        }
1883                        if (ei->i_data[i])
1884                                goto invalid_resize_inode;
1885                }
1886                credits += 3;   /* block bitmap, bg descriptor, resize inode */
1887        }
1888
1889        handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1890        if (IS_ERR(handle))
1891                return PTR_ERR(handle);
1892
1893        BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1894        err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1895                                            EXT4_JTR_NONE);
1896        if (err)
1897                goto errout;
1898
1899        lock_buffer(sbi->s_sbh);
1900        ext4_clear_feature_resize_inode(sb);
1901        ext4_set_feature_meta_bg(sb);
1902        sbi->s_es->s_first_meta_bg =
1903                cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1904        ext4_superblock_csum_set(sb);
1905        unlock_buffer(sbi->s_sbh);
1906
1907        err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1908        if (err) {
1909                ext4_std_error(sb, err);
1910                goto errout;
1911        }
1912
1913        if (inode) {
1914                nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1915                ext4_free_blocks(handle, inode, NULL, nr, 1,
1916                                 EXT4_FREE_BLOCKS_METADATA |
1917                                 EXT4_FREE_BLOCKS_FORGET);
1918                ei->i_data[EXT4_DIND_BLOCK] = 0;
1919                inode->i_blocks = 0;
1920
1921                err = ext4_mark_inode_dirty(handle, inode);
1922                if (err)
1923                        ext4_std_error(sb, err);
1924        }
1925
1926errout:
1927        ret = ext4_journal_stop(handle);
1928        if (!err)
1929                err = ret;
1930        return ret;
1931
1932invalid_resize_inode:
1933        ext4_error(sb, "corrupted/inconsistent resize inode");
1934        return -EINVAL;
1935}
1936
1937/*
1938 * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1939 *
1940 * @sb: super block of the fs to be resized
1941 * @n_blocks_count: the number of blocks resides in the resized fs
1942 */
1943int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1944{
1945        struct ext4_new_flex_group_data *flex_gd = NULL;
1946        struct ext4_sb_info *sbi = EXT4_SB(sb);
1947        struct ext4_super_block *es = sbi->s_es;
1948        struct buffer_head *bh;
1949        struct inode *resize_inode = NULL;
1950        ext4_grpblk_t add, offset;
1951        unsigned long n_desc_blocks;
1952        unsigned long o_desc_blocks;
1953        ext4_group_t o_group;
1954        ext4_group_t n_group;
1955        ext4_fsblk_t o_blocks_count;
1956        ext4_fsblk_t n_blocks_count_retry = 0;
1957        unsigned long last_update_time = 0;
1958        int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1959        int meta_bg;
1960
1961        /* See if the device is actually as big as what was requested */
1962        bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1963        if (IS_ERR(bh)) {
1964                ext4_warning(sb, "can't read last block, resize aborted");
1965                return -ENOSPC;
1966        }
1967        brelse(bh);
1968
1969retry:
1970        o_blocks_count = ext4_blocks_count(es);
1971
1972        ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1973                 "to %llu blocks", o_blocks_count, n_blocks_count);
1974
1975        if (n_blocks_count < o_blocks_count) {
1976                /* On-line shrinking not supported */
1977                ext4_warning(sb, "can't shrink FS - resize aborted");
1978                return -EINVAL;
1979        }
1980
1981        if (n_blocks_count == o_blocks_count)
1982                /* Nothing need to do */
1983                return 0;
1984
1985        n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1986        if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1987                ext4_warning(sb, "resize would cause inodes_count overflow");
1988                return -EINVAL;
1989        }
1990        ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1991
1992        n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1993        o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1994
1995        meta_bg = ext4_has_feature_meta_bg(sb);
1996
1997        if (ext4_has_feature_resize_inode(sb)) {
1998                if (meta_bg) {
1999                        ext4_error(sb, "resize_inode and meta_bg enabled "
2000                                   "simultaneously");
2001                        return -EINVAL;
2002                }
2003                if (n_desc_blocks > o_desc_blocks +
2004                    le16_to_cpu(es->s_reserved_gdt_blocks)) {
2005                        n_blocks_count_retry = n_blocks_count;
2006                        n_desc_blocks = o_desc_blocks +
2007                                le16_to_cpu(es->s_reserved_gdt_blocks);
2008                        n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2009                        n_blocks_count = (ext4_fsblk_t)n_group *
2010                                EXT4_BLOCKS_PER_GROUP(sb) +
2011                                le32_to_cpu(es->s_first_data_block);
2012                        n_group--; /* set to last group number */
2013                }
2014
2015                if (!resize_inode)
2016                        resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2017                                                 EXT4_IGET_SPECIAL);
2018                if (IS_ERR(resize_inode)) {
2019                        ext4_warning(sb, "Error opening resize inode");
2020                        return PTR_ERR(resize_inode);
2021                }
2022        }
2023
2024        if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2025                err = ext4_convert_meta_bg(sb, resize_inode);
2026                if (err)
2027                        goto out;
2028                if (resize_inode) {
2029                        iput(resize_inode);
2030                        resize_inode = NULL;
2031                }
2032                if (n_blocks_count_retry) {
2033                        n_blocks_count = n_blocks_count_retry;
2034                        n_blocks_count_retry = 0;
2035                        goto retry;
2036                }
2037        }
2038
2039        /*
2040         * Make sure the last group has enough space so that it's
2041         * guaranteed to have enough space for all metadata blocks
2042         * that it might need to hold.  (We might not need to store
2043         * the inode table blocks in the last block group, but there
2044         * will be cases where this might be needed.)
2045         */
2046        if ((ext4_group_first_block_no(sb, n_group) +
2047             ext4_group_overhead_blocks(sb, n_group) + 2 +
2048             sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2049                n_blocks_count = ext4_group_first_block_no(sb, n_group);
2050                n_group--;
2051                n_blocks_count_retry = 0;
2052                if (resize_inode) {
2053                        iput(resize_inode);
2054                        resize_inode = NULL;
2055                }
2056                goto retry;
2057        }
2058
2059        /* extend the last group */
2060        if (n_group == o_group)
2061                add = n_blocks_count - o_blocks_count;
2062        else
2063                add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2064        if (add > 0) {
2065                err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2066                if (err)
2067                        goto out;
2068        }
2069
2070        if (ext4_blocks_count(es) == n_blocks_count)
2071                goto out;
2072
2073        err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2074        if (err)
2075                goto out;
2076
2077        err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2078        if (err)
2079                goto out;
2080
2081        flex_gd = alloc_flex_gd(flexbg_size);
2082        if (flex_gd == NULL) {
2083                err = -ENOMEM;
2084                goto out;
2085        }
2086
2087        /* Add flex groups. Note that a regular group is a
2088         * flex group with 1 group.
2089         */
2090        while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2091                                              flexbg_size)) {
2092                if (jiffies - last_update_time > HZ * 10) {
2093                        if (last_update_time)
2094                                ext4_msg(sb, KERN_INFO,
2095                                         "resized to %llu blocks",
2096                                         ext4_blocks_count(es));
2097                        last_update_time = jiffies;
2098                }
2099                if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2100                        break;
2101                err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2102                if (unlikely(err))
2103                        break;
2104        }
2105
2106        if (!err && n_blocks_count_retry) {
2107                n_blocks_count = n_blocks_count_retry;
2108                n_blocks_count_retry = 0;
2109                free_flex_gd(flex_gd);
2110                flex_gd = NULL;
2111                if (resize_inode) {
2112                        iput(resize_inode);
2113                        resize_inode = NULL;
2114                }
2115                goto retry;
2116        }
2117
2118out:
2119        if (flex_gd)
2120                free_flex_gd(flex_gd);
2121        if (resize_inode != NULL)
2122                iput(resize_inode);
2123        if (err)
2124                ext4_warning(sb, "error (%d) occurred during "
2125                             "file system resize", err);
2126        ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2127                 ext4_blocks_count(es));
2128        return err;
2129}
2130