fs/ext4/resize.c

   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 <[email protected]>
   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
  20 int ext4_resize_begin(struct super_block *sb)
  21 {
  22         int ret = 0;
  23
  24         if (!capable(CAP_SYS_RESOURCE))
  25                 return -EPERM;
  26
  27         /*
  28          * If we are not using the primary superblock/GDT copy don't resize,
  29          * because the user tools have no way of handling this.  Probably a
  30          * bad time to do it anyways.
  31          */
  32         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
  33             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  34                 ext4_warning(sb, "won't resize using backup superblock at %llu",
  35                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  36                 return -EPERM;
  37         }
  38
  39         /*
  40          * We are not allowed to do online-resizing on a filesystem mounted
  41          * with error, because it can destroy the filesystem easily.
  42          */
  43         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  44                 ext4_warning(sb, "There are errors in the filesystem, "
  45                              "so online resizing is not allowed");
  46                 return -EPERM;
  47         }
  48
  49         if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
  50                                   &EXT4_SB(sb)->s_ext4_flags))
  51                 ret = -EBUSY;
  52
  53         return ret;
  54 }
  55
  56 void ext4_resize_end(struct super_block *sb)
  57 {
  58         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
  59         smp_mb__after_atomic();
  60 }
  61
  62 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  63                                              ext4_group_t group) {
  64         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  65                EXT4_DESC_PER_BLOCK_BITS(sb);
  66 }
  67
  68 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
  69                                              ext4_group_t group) {
  70         group = ext4_meta_bg_first_group(sb, group);
  71         return ext4_group_first_block_no(sb, group);
  72 }
  73
  74 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
  75                                                 ext4_group_t group) {
  76         ext4_grpblk_t overhead;
  77         overhead = ext4_bg_num_gdb(sb, group);
  78         if (ext4_bg_has_super(sb, group))
  79                 overhead += 1 +
  80                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
  81         return overhead;
  82 }
  83
  84 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  85 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  86
  87 static int verify_group_input(struct super_block *sb,
  88                               struct ext4_new_group_data *input)
  89 {
  90         struct ext4_sb_info *sbi = EXT4_SB(sb);
  91         struct ext4_super_block *es = sbi->s_es;
  92         ext4_fsblk_t start = ext4_blocks_count(es);
  93         ext4_fsblk_t end = start + input->blocks_count;
  94         ext4_group_t group = input->group;
  95         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  96         unsigned overhead;
  97         ext4_fsblk_t metaend;
  98         struct buffer_head *bh = NULL;
  99         ext4_grpblk_t free_blocks_count, offset;
 100         int err = -EINVAL;
 101
 102         if (group != sbi->s_groups_count) {
 103                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 104                              input->group, sbi->s_groups_count);
 105                 return -EINVAL;
 106         }
 107
 108         overhead = ext4_group_overhead_blocks(sb, group);
 109         metaend = start + overhead;
 110         input->free_clusters_count = free_blocks_count =
 111                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 112
 113         if (test_opt(sb, DEBUG))
 114                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 115                        "(%d free, %u reserved)\n",
 116                        ext4_bg_has_super(sb, input->group) ? "normal" :
 117                        "no-super", input->group, input->blocks_count,
 118                        free_blocks_count, input->reserved_blocks);
 119
 120         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 121         if (offset != 0)
 122                         ext4_warning(sb, "Last group not full");
 123         else if (input->reserved_blocks > input->blocks_count / 5)
 124                 ext4_warning(sb, "Reserved blocks too high (%u)",
 125                              input->reserved_blocks);
 126         else if (free_blocks_count < 0)
 127                 ext4_warning(sb, "Bad blocks count %u",
 128                              input->blocks_count);
 129         else if (!(bh = sb_bread(sb, end - 1)))
 130                 ext4_warning(sb, "Cannot read last block (%llu)",
 131                              end - 1);
 132         else if (outside(input->block_bitmap, start, end))
 133                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 134                              (unsigned long long)input->block_bitmap);
 135         else if (outside(input->inode_bitmap, start, end))
 136                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 137                              (unsigned long long)input->inode_bitmap);
 138         else if (outside(input->inode_table, start, end) ||
 139                  outside(itend - 1, start, end))
 140                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 141                              (unsigned long long)input->inode_table, itend - 1);
 142         else if (input->inode_bitmap == input->block_bitmap)
 143                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 144                              (unsigned long long)input->block_bitmap);
 145         else if (inside(input->block_bitmap, input->inode_table, itend))
 146                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 147                              "(%llu-%llu)",
 148                              (unsigned long long)input->block_bitmap,
 149                              (unsigned long long)input->inode_table, itend - 1);
 150         else if (inside(input->inode_bitmap, input->inode_table, itend))
 151                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 152                              "(%llu-%llu)",
 153                              (unsigned long long)input->inode_bitmap,
 154                              (unsigned long long)input->inode_table, itend - 1);
 155         else if (inside(input->block_bitmap, start, metaend))
 156                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 157                              (unsigned long long)input->block_bitmap,
 158                              start, metaend - 1);
 159         else if (inside(input->inode_bitmap, start, metaend))
 160                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 161                              (unsigned long long)input->inode_bitmap,
 162                              start, metaend - 1);
 163         else if (inside(input->inode_table, start, metaend) ||
 164                  inside(itend - 1, start, metaend))
 165                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 166                              "(%llu-%llu)",
 167                              (unsigned long long)input->inode_table,
 168                              itend - 1, start, metaend - 1);
 169         else
 170                 err = 0;
 171         brelse(bh);
 172
 173         return err;
 174 }
 175
 176 /*
 177  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 178  * group each time.
 179  */
 180 struct ext4_new_flex_group_data {
 181         struct ext4_new_group_data *groups;     /* new_group_data for groups
 182                                                    in the flex group */
 183         __u16 *bg_flags;                        /* block group flags of groups
 184                                                    in @groups */
 185         ext4_group_t count;                     /* number of groups in @groups
 186                                                  */
 187 };
 188
 189 /*
 190  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 191  * @flexbg_size.
 192  *
 193  * Returns NULL on failure otherwise address of the allocated structure.
 194  */
 195 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 196 {
 197         struct ext4_new_flex_group_data *flex_gd;
 198
 199         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 200         if (flex_gd == NULL)
 201                 goto out3;
 202
 203         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
 204                 goto out2;
 205         flex_gd->count = flexbg_size;
 206
 207         flex_gd->groups = kmalloc_array(flexbg_size,
 208                                         sizeof(struct ext4_new_group_data),
 209                                         GFP_NOFS);
 210         if (flex_gd->groups == NULL)
 211                 goto out2;
 212
 213         flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
 214                                           GFP_NOFS);
 215         if (flex_gd->bg_flags == NULL)
 216                 goto out1;
 217
 218         return flex_gd;
 219
 220 out1:
 221         kfree(flex_gd->groups);
 222 out2:
 223         kfree(flex_gd);
 224 out3:
 225         return NULL;
 226 }
 227
 228 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 229 {
 230         kfree(flex_gd->bg_flags);
 231         kfree(flex_gd->groups);
 232         kfree(flex_gd);
 233 }
 234
 235 /*
 236  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 237  * and inode tables for a flex group.
 238  *
 239  * This function is used by 64bit-resize.  Note that this function allocates
 240  * group tables from the 1st group of groups contained by @flexgd, which may
 241  * be a partial of a flex group.
 242  *
 243  * @sb: super block of fs to which the groups belongs
 244  *
 245  * Returns 0 on a successful allocation of the metadata blocks in the
 246  * block group.
 247  */
 248 static int ext4_alloc_group_tables(struct super_block *sb,
 249                                 struct ext4_new_flex_group_data *flex_gd,
 250                                 int flexbg_size)
 251 {
 252         struct ext4_new_group_data *group_data = flex_gd->groups;
 253         ext4_fsblk_t start_blk;
 254         ext4_fsblk_t last_blk;
 255         ext4_group_t src_group;
 256         ext4_group_t bb_index = 0;
 257         ext4_group_t ib_index = 0;
 258         ext4_group_t it_index = 0;
 259         ext4_group_t group;
 260         ext4_group_t last_group;
 261         unsigned overhead;
 262         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 263         int i;
 264
 265         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 266
 267         src_group = group_data[0].group;
 268         last_group  = src_group + flex_gd->count - 1;
 269
 270         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 271                (last_group & ~(flexbg_size - 1))));
 272 next_group:
 273         group = group_data[0].group;
 274         if (src_group >= group_data[0].group + flex_gd->count)
 275                 return -ENOSPC;
 276         start_blk = ext4_group_first_block_no(sb, src_group);
 277         last_blk = start_blk + group_data[src_group - group].blocks_count;
 278
 279         overhead = ext4_group_overhead_blocks(sb, src_group);
 280
 281         start_blk += overhead;
 282
 283         /* We collect contiguous blocks as much as possible. */
 284         src_group++;
 285         for (; src_group <= last_group; src_group++) {
 286                 overhead = ext4_group_overhead_blocks(sb, src_group);
 287                 if (overhead == 0)
 288                         last_blk += group_data[src_group - group].blocks_count;
 289                 else
 290                         break;
 291         }
 292
 293         /* Allocate block bitmaps */
 294         for (; bb_index < flex_gd->count; bb_index++) {
 295                 if (start_blk >= last_blk)
 296                         goto next_group;
 297                 group_data[bb_index].block_bitmap = start_blk++;
 298                 group = ext4_get_group_number(sb, start_blk - 1);
 299                 group -= group_data[0].group;
 300                 group_data[group].mdata_blocks++;
 301                 flex_gd->bg_flags[group] &= uninit_mask;
 302         }
 303
 304         /* Allocate inode bitmaps */
 305         for (; ib_index < flex_gd->count; ib_index++) {
 306                 if (start_blk >= last_blk)
 307                         goto next_group;
 308                 group_data[ib_index].inode_bitmap = start_blk++;
 309                 group = ext4_get_group_number(sb, start_blk - 1);
 310                 group -= group_data[0].group;
 311                 group_data[group].mdata_blocks++;
 312                 flex_gd->bg_flags[group] &= uninit_mask;
 313         }
 314
 315         /* Allocate inode tables */
 316         for (; it_index < flex_gd->count; it_index++) {
 317                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 318                 ext4_fsblk_t next_group_start;
 319
 320                 if (start_blk + itb > last_blk)
 321                         goto next_group;
 322                 group_data[it_index].inode_table = start_blk;
 323                 group = ext4_get_group_number(sb, start_blk);
 324                 next_group_start = ext4_group_first_block_no(sb, group + 1);
 325                 group -= group_data[0].group;
 326
 327                 if (start_blk + itb > next_group_start) {
 328                         flex_gd->bg_flags[group + 1] &= uninit_mask;
 329                         overhead = start_blk + itb - next_group_start;
 330                         group_data[group + 1].mdata_blocks += overhead;
 331                         itb -= overhead;
 332                 }
 333
 334                 group_data[group].mdata_blocks += itb;
 335                 flex_gd->bg_flags[group] &= uninit_mask;
 336                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 337         }
 338
 339         /* Update free clusters count to exclude metadata blocks */
 340         for (i = 0; i < flex_gd->count; i++) {
 341                 group_data[i].free_clusters_count -=
 342                                 EXT4_NUM_B2C(EXT4_SB(sb),
 343                                              group_data[i].mdata_blocks);
 344         }
 345
 346         if (test_opt(sb, DEBUG)) {
 347                 int i;
 348                 group = group_data[0].group;
 349
 350                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 351                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 352                        flexbg_size);
 353
 354                 for (i = 0; i < flex_gd->count; i++) {
 355                         ext4_debug(
 356                                "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
 357                                ext4_bg_has_super(sb, group + i) ? "normal" :
 358                                "no-super", group + i,
 359                                group_data[i].blocks_count,
 360                                group_data[i].free_clusters_count,
 361                                group_data[i].mdata_blocks);
 362                 }
 363         }
 364         return 0;
 365 }
 366
 367 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 368                                   ext4_fsblk_t blk)
 369 {
 370         struct buffer_head *bh;
 371         int err;
 372
 373         bh = sb_getblk(sb, blk);
 374         if (unlikely(!bh))
 375                 return ERR_PTR(-ENOMEM);
 376         BUFFER_TRACE(bh, "get_write_access");
 377         if ((err = ext4_journal_get_write_access(handle, bh))) {
 378                 brelse(bh);
 379                 bh = ERR_PTR(err);
 380         } else {
 381                 memset(bh->b_data, 0, sb->s_blocksize);
 382                 set_buffer_uptodate(bh);
 383         }
 384
 385         return bh;
 386 }
 387
 388 /*
 389  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 390  * If that fails, restart the transaction & regain write access for the
 391  * buffer head which is used for block_bitmap modifications.
 392  */
 393 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 394 {
 395         int err;
 396
 397         if (ext4_handle_has_enough_credits(handle, thresh))
 398                 return 0;
 399
 400         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 401         if (err < 0)
 402                 return err;
 403         if (err) {
 404                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 405                 if (err)
 406                         return err;
 407         }
 408
 409         return 0;
 410 }
 411
 412 /*
 413  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
 414  *
 415  * Helper function for ext4_setup_new_group_blocks() which set .
 416  *
 417  * @sb: super block
 418  * @handle: journal handle
 419  * @flex_gd: flex group data
 420  */
 421 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 422                         struct ext4_new_flex_group_data *flex_gd,
 423                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
 424 {
 425         struct ext4_sb_info *sbi = EXT4_SB(sb);
 426         ext4_group_t count = last_cluster - first_cluster + 1;
 427         ext4_group_t count2;
 428
 429         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
 430                    last_cluster);
 431         for (count2 = count; count > 0;
 432              count -= count2, first_cluster += count2) {
 433                 ext4_fsblk_t start;
 434                 struct buffer_head *bh;
 435                 ext4_group_t group;
 436                 int err;
 437
 438                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
 439                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
 440                 group -= flex_gd->groups[0].group;
 441
 442                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
 443                 if (count2 > count)
 444                         count2 = count;
 445
 446                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 447                         BUG_ON(flex_gd->count > 1);
 448                         continue;
 449                 }
 450
 451                 err = extend_or_restart_transaction(handle, 1);
 452                 if (err)
 453                         return err;
 454
 455                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 456                 if (unlikely(!bh))
 457                         return -ENOMEM;
 458
 459                 BUFFER_TRACE(bh, "get_write_access");
 460                 err = ext4_journal_get_write_access(handle, bh);
 461                 if (err)
 462                         return err;
 463                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
 464                            first_cluster, first_cluster - start, count2);
 465                 ext4_set_bits(bh->b_data, first_cluster - start, count2);
 466
 467                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 468                 if (unlikely(err))
 469                         return err;
 470                 brelse(bh);
 471         }
 472
 473         return 0;
 474 }
 475
 476 /*
 477  * Set up the block and inode bitmaps, and the inode table for the new groups.
 478  * This doesn't need to be part of the main transaction, since we are only
 479  * changing blocks outside the actual filesystem.  We still do journaling to
 480  * ensure the recovery is correct in case of a failure just after resize.
 481  * If any part of this fails, we simply abort the resize.
 482  *
 483  * setup_new_flex_group_blocks handles a flex group as follow:
 484  *  1. copy super block and GDT, and initialize group tables if necessary.
 485  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 486  *     super block and GDT.
 487  *  2. allocate group tables in block bitmaps, that is, set bits in block
 488  *     bitmap for blocks taken by group tables.
 489  */
 490 static int setup_new_flex_group_blocks(struct super_block *sb,
 491                                 struct ext4_new_flex_group_data *flex_gd)
 492 {
 493         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 494         ext4_fsblk_t start;
 495         ext4_fsblk_t block;
 496         struct ext4_sb_info *sbi = EXT4_SB(sb);
 497         struct ext4_super_block *es = sbi->s_es;
 498         struct ext4_new_group_data *group_data = flex_gd->groups;
 499         __u16 *bg_flags = flex_gd->bg_flags;
 500         handle_t *handle;
 501         ext4_group_t group, count;
 502         struct buffer_head *bh = NULL;
 503         int reserved_gdb, i, j, err = 0, err2;
 504         int meta_bg;
 505
 506         BUG_ON(!flex_gd->count || !group_data ||
 507                group_data[0].group != sbi->s_groups_count);
 508
 509         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 510         meta_bg = ext4_has_feature_meta_bg(sb);
 511
 512         /* This transaction may be extended/restarted along the way */
 513         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 514         if (IS_ERR(handle))
 515                 return PTR_ERR(handle);
 516
 517         group = group_data[0].group;
 518         for (i = 0; i < flex_gd->count; i++, group++) {
 519                 unsigned long gdblocks;
 520                 ext4_grpblk_t overhead;
 521
 522                 gdblocks = ext4_bg_num_gdb(sb, group);
 523                 start = ext4_group_first_block_no(sb, group);
 524
 525                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 526                         goto handle_itb;
 527
 528                 if (meta_bg == 1) {
 529                         ext4_group_t first_group;
 530                         first_group = ext4_meta_bg_first_group(sb, group);
 531                         if (first_group != group + 1 &&
 532                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 533                                 goto handle_itb;
 534                 }
 535
 536                 block = start + ext4_bg_has_super(sb, group);
 537                 /* Copy all of the GDT blocks into the backup in this group */
 538                 for (j = 0; j < gdblocks; j++, block++) {
 539                         struct buffer_head *gdb;
 540
 541                         ext4_debug("update backup group %#04llx\n", block);
 542                         err = extend_or_restart_transaction(handle, 1);
 543                         if (err)
 544                                 goto out;
 545
 546                         gdb = sb_getblk(sb, block);
 547                         if (unlikely(!gdb)) {
 548                                 err = -ENOMEM;
 549                                 goto out;
 550                         }
 551
 552                         BUFFER_TRACE(gdb, "get_write_access");
 553                         err = ext4_journal_get_write_access(handle, gdb);
 554                         if (err) {
 555                                 brelse(gdb);
 556                                 goto out;
 557                         }
 558                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 559                                gdb->b_size);
 560                         set_buffer_uptodate(gdb);
 561
 562                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 563                         if (unlikely(err)) {
 564                                 brelse(gdb);
 565                                 goto out;
 566                         }
 567                         brelse(gdb);
 568                 }
 569
 570                 /* Zero out all of the reserved backup group descriptor
 571                  * table blocks
 572                  */
 573                 if (ext4_bg_has_super(sb, group)) {
 574                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 575                                         reserved_gdb, GFP_NOFS);
 576                         if (err)
 577                                 goto out;
 578                 }
 579
 580 handle_itb:
 581                 /* Initialize group tables of the grop @group */
 582                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 583                         goto handle_bb;
 584
 585                 /* Zero out all of the inode table blocks */
 586                 block = group_data[i].inode_table;
 587                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 588                            block, sbi->s_itb_per_group);
 589                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 590                                        GFP_NOFS);
 591                 if (err)
 592                         goto out;
 593
 594 handle_bb:
 595                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 596                         goto handle_ib;
 597
 598                 /* Initialize block bitmap of the @group */
 599                 block = group_data[i].block_bitmap;
 600                 err = extend_or_restart_transaction(handle, 1);
 601                 if (err)
 602                         goto out;
 603
 604                 bh = bclean(handle, sb, block);
 605                 if (IS_ERR(bh)) {
 606                         err = PTR_ERR(bh);
 607                         bh = NULL;
 608                         goto out;
 609                 }
 610                 overhead = ext4_group_overhead_blocks(sb, group);
 611                 if (overhead != 0) {
 612                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 613                                    start);
 614                         ext4_set_bits(bh->b_data, 0,
 615                                       EXT4_NUM_B2C(sbi, overhead));
 616                 }
 617                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
 618                                      sb->s_blocksize * 8, bh->b_data);
 619                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 620                 if (err)
 621                         goto out;
 622                 brelse(bh);
 623
 624 handle_ib:
 625                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 626                         continue;
 627
 628                 /* Initialize inode bitmap of the @group */
 629                 block = group_data[i].inode_bitmap;
 630                 err = extend_or_restart_transaction(handle, 1);
 631                 if (err)
 632                         goto out;
 633                 /* Mark unused entries in inode bitmap used */
 634                 bh = bclean(handle, sb, block);
 635                 if (IS_ERR(bh)) {
 636                         err = PTR_ERR(bh);
 637                         bh = NULL;
 638                         goto out;
 639                 }
 640
 641                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 642                                      sb->s_blocksize * 8, bh->b_data);
 643                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 644                 if (err)
 645                         goto out;
 646                 brelse(bh);
 647         }
 648         bh = NULL;
 649
 650         /* Mark group tables in block bitmap */
 651         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 652                 count = group_table_count[j];
 653                 start = (&group_data[0].block_bitmap)[j];
 654                 block = start;
 655                 for (i = 1; i < flex_gd->count; i++) {
 656                         block += group_table_count[j];
 657                         if (block == (&group_data[i].block_bitmap)[j]) {
 658                                 count += group_table_count[j];
 659                                 continue;
 660                         }
 661                         err = set_flexbg_block_bitmap(sb, handle,
 662                                                       flex_gd,
 663                                                       EXT4_B2C(sbi, start),
 664                                                       EXT4_B2C(sbi,
 665                                                                start + count
 666                                                                - 1));
 667                         if (err)
 668                                 goto out;
 669                         count = group_table_count[j];
 670                         start = (&group_data[i].block_bitmap)[j];
 671                         block = start;
 672                 }
 673
 674                 if (count) {
 675                         err = set_flexbg_block_bitmap(sb, handle,
 676                                                       flex_gd,
 677                                                       EXT4_B2C(sbi, start),
 678                                                       EXT4_B2C(sbi,
 679                                                                start + count
 680                                                                - 1));
 681                         if (err)
 682                                 goto out;
 683                 }
 684         }
 685
 686 out:
 687         brelse(bh);
 688         err2 = ext4_journal_stop(handle);
 689         if (err2 && !err)
 690                 err = err2;
 691
 692         return err;
 693 }
 694
 695 /*
 696  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 697  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 698  * calling this for the first time.  In a sparse filesystem it will be the
 699  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 700  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 701  */
 702 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 703                                   unsigned *five, unsigned *seven)
 704 {
 705         unsigned *min = three;
 706         int mult = 3;
 707         unsigned ret;
 708
 709         if (!ext4_has_feature_sparse_super(sb)) {
 710                 ret = *min;
 711                 *min += 1;
 712                 return ret;
 713         }
 714
 715         if (*five < *min) {
 716                 min = five;
 717                 mult = 5;
 718         }
 719         if (*seven < *min) {
 720                 min = seven;
 721                 mult = 7;
 722         }
 723
 724         ret = *min;
 725         *min *= mult;
 726
 727         return ret;
 728 }
 729
 730 /*
 731  * Check that all of the backup GDT blocks are held in the primary GDT block.
 732  * It is assumed that they are stored in group order.  Returns the number of
 733  * groups in current filesystem that have BACKUPS, or -ve error code.
 734  */
 735 static int verify_reserved_gdb(struct super_block *sb,
 736                                ext4_group_t end,
 737                                struct buffer_head *primary)
 738 {
 739         const ext4_fsblk_t blk = primary->b_blocknr;
 740         unsigned three = 1;
 741         unsigned five = 5;
 742         unsigned seven = 7;
 743         unsigned grp;
 744         __le32 *p = (__le32 *)primary->b_data;
 745         int gdbackups = 0;
 746
 747         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 748                 if (le32_to_cpu(*p++) !=
 749                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 750                         ext4_warning(sb, "reserved GDT %llu"
 751                                      " missing grp %d (%llu)",
 752                                      blk, grp,
 753                                      grp *
 754                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 755                                      blk);
 756                         return -EINVAL;
 757                 }
 758                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 759                         return -EFBIG;
 760         }
 761
 762         return gdbackups;
 763 }
 764
 765 /*
 766  * Called when we need to bring a reserved group descriptor table block into
 767  * use from the resize inode.  The primary copy of the new GDT block currently
 768  * is an indirect block (under the double indirect block in the resize inode).
 769  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 770  * block, in group order.  Even though we know all the block numbers we need,
 771  * we check to ensure that the resize inode has actually reserved these blocks.
 772  *
 773  * Don't need to update the block bitmaps because the blocks are still in use.
 774  *
 775  * We get all of the error cases out of the way, so that we are sure to not
 776  * fail once we start modifying the data on disk, because JBD has no rollback.
 777  */
 778 static int add_new_gdb(handle_t *handle, struct inode *inode,
 779                        ext4_group_t group)
 780 {
 781         struct super_block *sb = inode->i_sb;
 782         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 783         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 784         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 785         struct buffer_head **o_group_desc, **n_group_desc;
 786         struct buffer_head *dind;
 787         struct buffer_head *gdb_bh;
 788         int gdbackups;
 789         struct ext4_iloc iloc;
 790         __le32 *data;
 791         int err;
 792
 793         if (test_opt(sb, DEBUG))
 794                 printk(KERN_DEBUG
 795                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 796                        gdb_num);
 797
 798         gdb_bh = sb_bread(sb, gdblock);
 799         if (!gdb_bh)
 800                 return -EIO;
 801
 802         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 803         if (gdbackups < 0) {
 804                 err = gdbackups;
 805                 goto exit_bh;
 806         }
 807
 808         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 809         dind = sb_bread(sb, le32_to_cpu(*data));
 810         if (!dind) {
 811                 err = -EIO;
 812                 goto exit_bh;
 813         }
 814
 815         data = (__le32 *)dind->b_data;
 816         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 817                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 818                              group, gdblock);
 819                 err = -EINVAL;
 820                 goto exit_dind;
 821         }
 822
 823         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 824         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 825         if (unlikely(err))
 826                 goto exit_dind;
 827
 828         BUFFER_TRACE(gdb_bh, "get_write_access");
 829         err = ext4_journal_get_write_access(handle, gdb_bh);
 830         if (unlikely(err))
 831                 goto exit_dind;
 832
 833         BUFFER_TRACE(dind, "get_write_access");
 834         err = ext4_journal_get_write_access(handle, dind);
 835         if (unlikely(err))
 836                 ext4_std_error(sb, err);
 837
 838         /* ext4_reserve_inode_write() gets a reference on the iloc */
 839         err = ext4_reserve_inode_write(handle, inode, &iloc);
 840         if (unlikely(err))
 841                 goto exit_dind;
 842
 843         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 844                                      sizeof(struct buffer_head *),
 845                                      GFP_NOFS);
 846         if (!n_group_desc) {
 847                 err = -ENOMEM;
 848                 ext4_warning(sb, "not enough memory for %lu groups",
 849                              gdb_num + 1);
 850                 goto exit_inode;
 851         }
 852
 853         /*
 854          * Finally, we have all of the possible failures behind us...
 855          *
 856          * Remove new GDT block from inode double-indirect block and clear out
 857          * the new GDT block for use (which also "frees" the backup GDT blocks
 858          * from the reserved inode).  We don't need to change the bitmaps for
 859          * these blocks, because they are marked as in-use from being in the
 860          * reserved inode, and will become GDT blocks (primary and backup).
 861          */
 862         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 863         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 864         if (unlikely(err)) {
 865                 ext4_std_error(sb, err);
 866                 goto exit_inode;
 867         }
 868         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
 869                            (9 - EXT4_SB(sb)->s_cluster_bits);
 870         ext4_mark_iloc_dirty(handle, inode, &iloc);
 871         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 872         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 873         if (unlikely(err)) {
 874                 ext4_std_error(sb, err);
 875                 goto exit_inode;
 876         }
 877         brelse(dind);
 878
 879         o_group_desc = EXT4_SB(sb)->s_group_desc;
 880         memcpy(n_group_desc, o_group_desc,
 881                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 882         n_group_desc[gdb_num] = gdb_bh;
 883         EXT4_SB(sb)->s_group_desc = n_group_desc;
 884         EXT4_SB(sb)->s_gdb_count++;
 885         kvfree(o_group_desc);
 886
 887         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 888         err = ext4_handle_dirty_super(handle, sb);
 889         if (err)
 890                 ext4_std_error(sb, err);
 891
 892         return err;
 893
 894 exit_inode:
 895         kvfree(n_group_desc);
 896         brelse(iloc.bh);
 897 exit_dind:
 898         brelse(dind);
 899 exit_bh:
 900         brelse(gdb_bh);
 901
 902         ext4_debug("leaving with error %d\n", err);
 903         return err;
 904 }
 905
 906 /*
 907  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 908  */
 909 static int add_new_gdb_meta_bg(struct super_block *sb,
 910                                handle_t *handle, ext4_group_t group) {
 911         ext4_fsblk_t gdblock;
 912         struct buffer_head *gdb_bh;
 913         struct buffer_head **o_group_desc, **n_group_desc;
 914         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 915         int err;
 916
 917         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 918                    ext4_bg_has_super(sb, group);
 919         gdb_bh = sb_bread(sb, gdblock);
 920         if (!gdb_bh)
 921                 return -EIO;
 922         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 923                                      sizeof(struct buffer_head *),
 924                                      GFP_NOFS);
 925         if (!n_group_desc) {
 926                 err = -ENOMEM;
 927                 ext4_warning(sb, "not enough memory for %lu groups",
 928                              gdb_num + 1);
 929                 return err;
 930         }
 931
 932         o_group_desc = EXT4_SB(sb)->s_group_desc;
 933         memcpy(n_group_desc, o_group_desc,
 934                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 935         n_group_desc[gdb_num] = gdb_bh;
 936         EXT4_SB(sb)->s_group_desc = n_group_desc;
 937         EXT4_SB(sb)->s_gdb_count++;
 938         kvfree(o_group_desc);
 939         BUFFER_TRACE(gdb_bh, "get_write_access");
 940         err = ext4_journal_get_write_access(handle, gdb_bh);
 941         if (unlikely(err))
 942                 brelse(gdb_bh);
 943         return err;
 944 }
 945
 946 /*
 947  * Called when we are adding a new group which has a backup copy of each of
 948  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 949  * We need to add these reserved backup GDT blocks to the resize inode, so
 950  * that they are kept for future resizing and not allocated to files.
 951  *
 952  * Each reserved backup GDT block will go into a different indirect block.
 953  * The indirect blocks are actually the primary reserved GDT blocks,
 954  * so we know in advance what their block numbers are.  We only get the
 955  * double-indirect block to verify it is pointing to the primary reserved
 956  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 957  * backup GDT blocks are stored in their reserved primary GDT block.
 958  */
 959 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 960                               ext4_group_t group)
 961 {
 962         struct super_block *sb = inode->i_sb;
 963         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 964         int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
 965         struct buffer_head **primary;
 966         struct buffer_head *dind;
 967         struct ext4_iloc iloc;
 968         ext4_fsblk_t blk;
 969         __le32 *data, *end;
 970         int gdbackups = 0;
 971         int res, i;
 972         int err;
 973
 974         primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
 975         if (!primary)
 976                 return -ENOMEM;
 977
 978         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 979         dind = sb_bread(sb, le32_to_cpu(*data));
 980         if (!dind) {
 981                 err = -EIO;
 982                 goto exit_free;
 983         }
 984
 985         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 986         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 987                                          EXT4_ADDR_PER_BLOCK(sb));
 988         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 989
 990         /* Get each reserved primary GDT block and verify it holds backups */
 991         for (res = 0; res < reserved_gdb; res++, blk++) {
 992                 if (le32_to_cpu(*data) != blk) {
 993                         ext4_warning(sb, "reserved block %llu"
 994                                      " not at offset %ld",
 995                                      blk,
 996                                      (long)(data - (__le32 *)dind->b_data));
 997                         err = -EINVAL;
 998                         goto exit_bh;
 999                 }
1000                 primary[res] = sb_bread(sb, blk);
1001                 if (!primary[res]) {
1002                         err = -EIO;
1003                         goto exit_bh;
1004                 }
1005                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1006                 if (gdbackups < 0) {
1007                         brelse(primary[res]);
1008                         err = gdbackups;
1009                         goto exit_bh;
1010                 }
1011                 if (++data >= end)
1012                         data = (__le32 *)dind->b_data;
1013         }
1014
1015         for (i = 0; i < reserved_gdb; i++) {
1016                 BUFFER_TRACE(primary[i], "get_write_access");
1017                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1018                         goto exit_bh;
1019         }
1020
1021         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1022                 goto exit_bh;
1023
1024         /*
1025          * Finally we can add each of the reserved backup GDT blocks from
1026          * the new group to its reserved primary GDT block.
1027          */
1028         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1029         for (i = 0; i < reserved_gdb; i++) {
1030                 int err2;
1031                 data = (__le32 *)primary[i]->b_data;
1032                 /* printk("reserving backup %lu[%u] = %lu\n",
1033                        primary[i]->b_blocknr, gdbackups,
1034                        blk + primary[i]->b_blocknr); */
1035                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1036                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1037                 if (!err)
1038                         err = err2;
1039         }
1040
1041         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1042         ext4_mark_iloc_dirty(handle, inode, &iloc);
1043
1044 exit_bh:
1045         while (--res >= 0)
1046                 brelse(primary[res]);
1047         brelse(dind);
1048
1049 exit_free:
1050         kfree(primary);
1051
1052         return err;
1053 }
1054
1055 /*
1056  * Update the backup copies of the ext4 metadata.  These don't need to be part
1057  * of the main resize transaction, because e2fsck will re-write them if there
1058  * is a problem (basically only OOM will cause a problem).  However, we
1059  * _should_ update the backups if possible, in case the primary gets trashed
1060  * for some reason and we need to run e2fsck from a backup superblock.  The
1061  * important part is that the new block and inode counts are in the backup
1062  * superblocks, and the location of the new group metadata in the GDT backups.
1063  *
1064  * We do not need take the s_resize_lock for this, because these
1065  * blocks are not otherwise touched by the filesystem code when it is
1066  * mounted.  We don't need to worry about last changing from
1067  * sbi->s_groups_count, because the worst that can happen is that we
1068  * do not copy the full number of backups at this time.  The resize
1069  * which changed s_groups_count will backup again.
1070  */
1071 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1072                            int size, int meta_bg)
1073 {
1074         struct ext4_sb_info *sbi = EXT4_SB(sb);
1075         ext4_group_t last;
1076         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1077         unsigned three = 1;
1078         unsigned five = 5;
1079         unsigned seven = 7;
1080         ext4_group_t group = 0;
1081         int rest = sb->s_blocksize - size;
1082         handle_t *handle;
1083         int err = 0, err2;
1084
1085         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1086         if (IS_ERR(handle)) {
1087                 group = 1;
1088                 err = PTR_ERR(handle);
1089                 goto exit_err;
1090         }
1091
1092         if (meta_bg == 0) {
1093                 group = ext4_list_backups(sb, &three, &five, &seven);
1094                 last = sbi->s_groups_count;
1095         } else {
1096                 group = ext4_get_group_number(sb, blk_off) + 1;
1097                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1098         }
1099
1100         while (group < sbi->s_groups_count) {
1101                 struct buffer_head *bh;
1102                 ext4_fsblk_t backup_block;
1103
1104                 /* Out of journal space, and can't get more - abort - so sad */
1105                 if (ext4_handle_valid(handle) &&
1106                     handle->h_buffer_credits == 0 &&
1107                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1108                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1109                         break;
1110
1111                 if (meta_bg == 0)
1112                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1113                 else
1114                         backup_block = (ext4_group_first_block_no(sb, group) +
1115                                         ext4_bg_has_super(sb, group));
1116
1117                 bh = sb_getblk(sb, backup_block);
1118                 if (unlikely(!bh)) {
1119                         err = -ENOMEM;
1120                         break;
1121                 }
1122                 ext4_debug("update metadata backup %llu(+%llu)\n",
1123                            backup_block, backup_block -
1124                            ext4_group_first_block_no(sb, group));
1125                 BUFFER_TRACE(bh, "get_write_access");
1126                 if ((err = ext4_journal_get_write_access(handle, bh)))
1127                         break;
1128                 lock_buffer(bh);
1129                 memcpy(bh->b_data, data, size);
1130                 if (rest)
1131                         memset(bh->b_data + size, 0, rest);
1132                 set_buffer_uptodate(bh);
1133                 unlock_buffer(bh);
1134                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1135                 if (unlikely(err))
1136                         ext4_std_error(sb, err);
1137                 brelse(bh);
1138
1139                 if (meta_bg == 0)
1140                         group = ext4_list_backups(sb, &three, &five, &seven);
1141                 else if (group == last)
1142                         break;
1143                 else
1144                         group = last;
1145         }
1146         if ((err2 = ext4_journal_stop(handle)) && !err)
1147                 err = err2;
1148
1149         /*
1150          * Ugh! Need to have e2fsck write the backup copies.  It is too
1151          * late to revert the resize, we shouldn't fail just because of
1152          * the backup copies (they are only needed in case of corruption).
1153          *
1154          * However, if we got here we have a journal problem too, so we
1155          * can't really start a transaction to mark the superblock.
1156          * Chicken out and just set the flag on the hope it will be written
1157          * to disk, and if not - we will simply wait until next fsck.
1158          */
1159 exit_err:
1160         if (err) {
1161                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1162                              "forcing fsck on next reboot", group, err);
1163                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1164                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1165                 mark_buffer_dirty(sbi->s_sbh);
1166         }
1167 }
1168
1169 /*
1170  * ext4_add_new_descs() adds @count group descriptor of groups
1171  * starting at @group
1172  *
1173  * @handle: journal handle
1174  * @sb: super block
1175  * @group: the group no. of the first group desc to be added
1176  * @resize_inode: the resize inode
1177  * @count: number of group descriptors to be added
1178  */
1179 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1180                               ext4_group_t group, struct inode *resize_inode,
1181                               ext4_group_t count)
1182 {
1183         struct ext4_sb_info *sbi = EXT4_SB(sb);
1184         struct ext4_super_block *es = sbi->s_es;
1185         struct buffer_head *gdb_bh;
1186         int i, gdb_off, gdb_num, err = 0;
1187         int meta_bg;
1188
1189         meta_bg = ext4_has_feature_meta_bg(sb);
1190         for (i = 0; i < count; i++, group++) {
1191                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1192                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1193
1194                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1195                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1196
1197                 /*
1198                  * We will only either add reserved group blocks to a backup group
1199                  * or remove reserved blocks for the first group in a new group block.
1200                  * Doing both would be mean more complex code, and sane people don't
1201                  * use non-sparse filesystems anymore.  This is already checked above.
1202                  */
1203                 if (gdb_off) {
1204                         gdb_bh = sbi->s_group_desc[gdb_num];
1205                         BUFFER_TRACE(gdb_bh, "get_write_access");
1206                         err = ext4_journal_get_write_access(handle, gdb_bh);
1207
1208                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1209                                 err = reserve_backup_gdb(handle, resize_inode, group);
1210                 } else if (meta_bg != 0) {
1211                         err = add_new_gdb_meta_bg(sb, handle, group);
1212                 } else {
1213                         err = add_new_gdb(handle, resize_inode, group);
1214                 }
1215                 if (err)
1216                         break;
1217         }
1218         return err;
1219 }
1220
1221 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1222 {
1223         struct buffer_head *bh = sb_getblk(sb, block);
1224         if (unlikely(!bh))
1225                 return NULL;
1226         if (!bh_uptodate_or_lock(bh)) {
1227                 if (bh_submit_read(bh) < 0) {
1228                         brelse(bh);
1229                         return NULL;
1230                 }
1231         }
1232
1233         return bh;
1234 }
1235
1236 static int ext4_set_bitmap_checksums(struct super_block *sb,
1237                                      ext4_group_t group,
1238                                      struct ext4_group_desc *gdp,
1239                                      struct ext4_new_group_data *group_data)
1240 {
1241         struct buffer_head *bh;
1242
1243         if (!ext4_has_metadata_csum(sb))
1244                 return 0;
1245
1246         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1247         if (!bh)
1248                 return -EIO;
1249         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1250                                    EXT4_INODES_PER_GROUP(sb) / 8);
1251         brelse(bh);
1252
1253         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1254         if (!bh)
1255                 return -EIO;
1256         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1257         brelse(bh);
1258
1259         return 0;
1260 }
1261
1262 /*
1263  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1264  */
1265 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1266                                 struct ext4_new_flex_group_data *flex_gd)
1267 {
1268         struct ext4_new_group_data      *group_data = flex_gd->groups;
1269         struct ext4_group_desc          *gdp;
1270         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1271         struct buffer_head              *gdb_bh;
1272         ext4_group_t                    group;
1273         __u16                           *bg_flags = flex_gd->bg_flags;
1274         int                             i, gdb_off, gdb_num, err = 0;
1275
1276
1277         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1278                 group = group_data->group;
1279
1280                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1281                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1282
1283                 /*
1284                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1285                  */
1286                 gdb_bh = sbi->s_group_desc[gdb_num];
1287                 /* Update group descriptor block for new group */
1288                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1289                                                  gdb_off * EXT4_DESC_SIZE(sb));
1290
1291                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1292                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1293                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1294                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1295                 if (err) {
1296                         ext4_std_error(sb, err);
1297                         break;
1298                 }
1299
1300                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1301                 ext4_free_group_clusters_set(sb, gdp,
1302                                              group_data->free_clusters_count);
1303                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1304                 if (ext4_has_group_desc_csum(sb))
1305                         ext4_itable_unused_set(sb, gdp,
1306                                                EXT4_INODES_PER_GROUP(sb));
1307                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1308                 ext4_group_desc_csum_set(sb, group, gdp);
1309
1310                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1311                 if (unlikely(err)) {
1312                         ext4_std_error(sb, err);
1313                         break;
1314                 }
1315
1316                 /*
1317                  * We can allocate memory for mb_alloc based on the new group
1318                  * descriptor
1319                  */
1320                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1321                 if (err)
1322                         break;
1323         }
1324         return err;
1325 }
1326
1327 /*
1328  * ext4_update_super() updates the super block so that the newly added
1329  * groups can be seen by the filesystem.
1330  *
1331  * @sb: super block
1332  * @flex_gd: new added groups
1333  */
1334 static void ext4_update_super(struct super_block *sb,
1335                              struct ext4_new_flex_group_data *flex_gd)
1336 {
1337         ext4_fsblk_t blocks_count = 0;
1338         ext4_fsblk_t free_blocks = 0;
1339         ext4_fsblk_t reserved_blocks = 0;
1340         struct ext4_new_group_data *group_data = flex_gd->groups;
1341         struct ext4_sb_info *sbi = EXT4_SB(sb);
1342         struct ext4_super_block *es = sbi->s_es;
1343         int i;
1344
1345         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1346         /*
1347          * Make the new blocks and inodes valid next.  We do this before
1348          * increasing the group count so that once the group is enabled,
1349          * all of its blocks and inodes are already valid.
1350          *
1351          * We always allocate group-by-group, then block-by-block or
1352          * inode-by-inode within a group, so enabling these
1353          * blocks/inodes before the group is live won't actually let us
1354          * allocate the new space yet.
1355          */
1356         for (i = 0; i < flex_gd->count; i++) {
1357                 blocks_count += group_data[i].blocks_count;
1358                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1359         }
1360
1361         reserved_blocks = ext4_r_blocks_count(es) * 100;
1362         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1363         reserved_blocks *= blocks_count;
1364         do_div(reserved_blocks, 100);
1365
1366         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1367         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1368         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1369                      flex_gd->count);
1370         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1371                      flex_gd->count);
1372
1373         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1374         /*
1375          * We need to protect s_groups_count against other CPUs seeing
1376          * inconsistent state in the superblock.
1377          *
1378          * The precise rules we use are:
1379          *
1380          * * Writers must perform a smp_wmb() after updating all
1381          *   dependent data and before modifying the groups count
1382          *
1383          * * Readers must perform an smp_rmb() after reading the groups
1384          *   count and before reading any dependent data.
1385          *
1386          * NB. These rules can be relaxed when checking the group count
1387          * while freeing data, as we can only allocate from a block
1388          * group after serialising against the group count, and we can
1389          * only then free after serialising in turn against that
1390          * allocation.
1391          */
1392         smp_wmb();
1393
1394         /* Update the global fs size fields */
1395         sbi->s_groups_count += flex_gd->count;
1396         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1397                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1398
1399         /* Update the reserved block counts only once the new group is
1400          * active. */
1401         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1402                                 reserved_blocks);
1403
1404         /* Update the free space counts */
1405         percpu_counter_add(&sbi->s_freeclusters_counter,
1406                            EXT4_NUM_B2C(sbi, free_blocks));
1407         percpu_counter_add(&sbi->s_freeinodes_counter,
1408                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1409
1410         ext4_debug("free blocks count %llu",
1411                    percpu_counter_read(&sbi->s_freeclusters_counter));
1412         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1413                 ext4_group_t flex_group;
1414                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1415                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1416                              &sbi->s_flex_groups[flex_group].free_clusters);
1417                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1418                            &sbi->s_flex_groups[flex_group].free_inodes);
1419         }
1420
1421         /*
1422          * Update the fs overhead information
1423          */
1424         ext4_calculate_overhead(sb);
1425
1426         if (test_opt(sb, DEBUG))
1427                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1428                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1429                        blocks_count, free_blocks, reserved_blocks);
1430 }
1431
1432 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1433  * _before_ we start modifying the filesystem, because we cannot abort the
1434  * transaction and not have it write the data to disk.
1435  */
1436 static int ext4_flex_group_add(struct super_block *sb,
1437                                struct inode *resize_inode,
1438                                struct ext4_new_flex_group_data *flex_gd)
1439 {
1440         struct ext4_sb_info *sbi = EXT4_SB(sb);
1441         struct ext4_super_block *es = sbi->s_es;
1442         ext4_fsblk_t o_blocks_count;
1443         ext4_grpblk_t last;
1444         ext4_group_t group;
1445         handle_t *handle;
1446         unsigned reserved_gdb;
1447         int err = 0, err2 = 0, credit;
1448
1449         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1450
1451         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1452         o_blocks_count = ext4_blocks_count(es);
1453         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1454         BUG_ON(last);
1455
1456         err = setup_new_flex_group_blocks(sb, flex_gd);
1457         if (err)
1458                 goto exit;
1459         /*
1460          * We will always be modifying at least the superblock and  GDT
1461          * blocks.  If we are adding a group past the last current GDT block,
1462          * we will also modify the inode and the dindirect block.  If we
1463          * are adding a group with superblock/GDT backups  we will also
1464          * modify each of the reserved GDT dindirect blocks.
1465          */
1466         credit = 3;     /* sb, resize inode, resize inode dindirect */
1467         /* GDT blocks */
1468         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1469         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1470         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1471         if (IS_ERR(handle)) {
1472                 err = PTR_ERR(handle);
1473                 goto exit;
1474         }
1475
1476         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1477         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1478         if (err)
1479                 goto exit_journal;
1480
1481         group = flex_gd->groups[0].group;
1482         BUG_ON(group != sbi->s_groups_count);
1483         err = ext4_add_new_descs(handle, sb, group,
1484                                 resize_inode, flex_gd->count);
1485         if (err)
1486                 goto exit_journal;
1487
1488         err = ext4_setup_new_descs(handle, sb, flex_gd);
1489         if (err)
1490                 goto exit_journal;
1491
1492         ext4_update_super(sb, flex_gd);
1493
1494         err = ext4_handle_dirty_super(handle, sb);
1495
1496 exit_journal:
1497         err2 = ext4_journal_stop(handle);
1498         if (!err)
1499                 err = err2;
1500
1501         if (!err) {
1502                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1503                 int gdb_num_end = ((group + flex_gd->count - 1) /
1504                                    EXT4_DESC_PER_BLOCK(sb));
1505                 int meta_bg = ext4_has_feature_meta_bg(sb);
1506                 sector_t old_gdb = 0;
1507
1508                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1509                                sizeof(struct ext4_super_block), 0);
1510                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1511                         struct buffer_head *gdb_bh;
1512
1513                         gdb_bh = sbi->s_group_desc[gdb_num];
1514                         if (old_gdb == gdb_bh->b_blocknr)
1515                                 continue;
1516                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1517                                        gdb_bh->b_size, meta_bg);
1518                         old_gdb = gdb_bh->b_blocknr;
1519                 }
1520         }
1521 exit:
1522         return err;
1523 }
1524
1525 static int ext4_setup_next_flex_gd(struct super_block *sb,
1526                                     struct ext4_new_flex_group_data *flex_gd,
1527                                     ext4_fsblk_t n_blocks_count,
1528                                     unsigned long flexbg_size)
1529 {
1530         struct ext4_sb_info *sbi = EXT4_SB(sb);
1531         struct ext4_super_block *es = sbi->s_es;
1532         struct ext4_new_group_data *group_data = flex_gd->groups;
1533         ext4_fsblk_t o_blocks_count;
1534         ext4_group_t n_group;
1535         ext4_group_t group;
1536         ext4_group_t last_group;
1537         ext4_grpblk_t last;
1538         ext4_grpblk_t clusters_per_group;
1539         unsigned long i;
1540
1541         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1542
1543         o_blocks_count = ext4_blocks_count(es);
1544
1545         if (o_blocks_count == n_blocks_count)
1546                 return 0;
1547
1548         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1549         BUG_ON(last);
1550         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1551
1552         last_group = group | (flexbg_size - 1);
1553         if (last_group > n_group)
1554                 last_group = n_group;
1555
1556         flex_gd->count = last_group - group + 1;
1557
1558         for (i = 0; i < flex_gd->count; i++) {
1559                 int overhead;
1560
1561                 group_data[i].group = group + i;
1562                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1563                 overhead = ext4_group_overhead_blocks(sb, group + i);
1564                 group_data[i].mdata_blocks = overhead;
1565                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1566                 if (ext4_has_group_desc_csum(sb)) {
1567                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1568                                                EXT4_BG_INODE_UNINIT;
1569                         if (!test_opt(sb, INIT_INODE_TABLE))
1570                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1571                 } else
1572                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1573         }
1574
1575         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1576                 /* We need to initialize block bitmap of last group. */
1577                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1578
1579         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1580                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1581                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1582                                                        last - 1;
1583         }
1584
1585         return 1;
1586 }
1587
1588 /* Add group descriptor data to an existing or new group descriptor block.
1589  * Ensure we handle all possible error conditions _before_ we start modifying
1590  * the filesystem, because we cannot abort the transaction and not have it
1591  * write the data to disk.
1592  *
1593  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1594  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1595  *
1596  * We only need to hold the superblock lock while we are actually adding
1597  * in the new group's counts to the superblock.  Prior to that we have
1598  * not really "added" the group at all.  We re-check that we are still
1599  * adding in the last group in case things have changed since verifying.
1600  */
1601 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1602 {
1603         struct ext4_new_flex_group_data flex_gd;
1604         struct ext4_sb_info *sbi = EXT4_SB(sb);
1605         struct ext4_super_block *es = sbi->s_es;
1606         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1607                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1608         struct inode *inode = NULL;
1609         int gdb_off;
1610         int err;
1611         __u16 bg_flags = 0;
1612
1613         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1614
1615         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1616                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1617                 return -EPERM;
1618         }
1619
1620         if (ext4_blocks_count(es) + input->blocks_count <
1621             ext4_blocks_count(es)) {
1622                 ext4_warning(sb, "blocks_count overflow");
1623                 return -EINVAL;
1624         }
1625
1626         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1627             le32_to_cpu(es->s_inodes_count)) {
1628                 ext4_warning(sb, "inodes_count overflow");
1629                 return -EINVAL;
1630         }
1631
1632         if (reserved_gdb || gdb_off == 0) {
1633                 if (ext4_has_feature_resize_inode(sb) ||
1634                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1635                         ext4_warning(sb,
1636                                      "No reserved GDT blocks, can't resize");
1637                         return -EPERM;
1638                 }
1639                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1640                 if (IS_ERR(inode)) {
1641                         ext4_warning(sb, "Error opening resize inode");
1642                         return PTR_ERR(inode);
1643                 }
1644         }
1645
1646
1647         err = verify_group_input(sb, input);
1648         if (err)
1649                 goto out;
1650
1651         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1652         if (err)
1653                 goto out;
1654
1655         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1656         if (err)
1657                 goto out;
1658
1659         flex_gd.count = 1;
1660         flex_gd.groups = input;
1661         flex_gd.bg_flags = &bg_flags;
1662         err = ext4_flex_group_add(sb, inode, &flex_gd);
1663 out:
1664         iput(inode);
1665         return err;
1666 } /* ext4_group_add */
1667
1668 /*
1669  * extend a group without checking assuming that checking has been done.
1670  */
1671 static int ext4_group_extend_no_check(struct super_block *sb,
1672                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1673 {
1674         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1675         handle_t *handle;
1676         int err = 0, err2;
1677
1678         /* We will update the superblock, one block bitmap, and
1679          * one group descriptor via ext4_group_add_blocks().
1680          */
1681         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1682         if (IS_ERR(handle)) {
1683                 err = PTR_ERR(handle);
1684                 ext4_warning(sb, "error %d on journal start", err);
1685                 return err;
1686         }
1687
1688         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1689         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1690         if (err) {
1691                 ext4_warning(sb, "error %d on journal write access", err);
1692                 goto errout;
1693         }
1694
1695         ext4_blocks_count_set(es, o_blocks_count + add);
1696         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1697         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1698                    o_blocks_count + add);
1699         /* We add the blocks to the bitmap and set the group need init bit */
1700         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1701         if (err)
1702                 goto errout;
1703         ext4_handle_dirty_super(handle, sb);
1704         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1705                    o_blocks_count + add);
1706 errout:
1707         err2 = ext4_journal_stop(handle);
1708         if (err2 && !err)
1709                 err = err2;
1710
1711         if (!err) {
1712                 if (test_opt(sb, DEBUG))
1713                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1714                                "blocks\n", ext4_blocks_count(es));
1715                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1716                                (char *)es, sizeof(struct ext4_super_block), 0);
1717         }
1718         return err;
1719 }
1720
1721 /*
1722  * Extend the filesystem to the new number of blocks specified.  This entry
1723  * point is only used to extend the current filesystem to the end of the last
1724  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1725  * for emergencies (because it has no dependencies on reserved blocks).
1726  *
1727  * If we _really_ wanted, we could use default values to call ext4_group_add()
1728  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1729  * GDT blocks are reserved to grow to the desired size.
1730  */
1731 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1732                       ext4_fsblk_t n_blocks_count)
1733 {
1734         ext4_fsblk_t o_blocks_count;
1735         ext4_grpblk_t last;
1736         ext4_grpblk_t add;
1737         struct buffer_head *bh;
1738         int err;
1739         ext4_group_t group;
1740
1741         o_blocks_count = ext4_blocks_count(es);
1742
1743         if (test_opt(sb, DEBUG))
1744                 ext4_msg(sb, KERN_DEBUG,
1745                          "extending last group from %llu to %llu blocks",
1746                          o_blocks_count, n_blocks_count);
1747
1748         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1749                 return 0;
1750
1751         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1752                 ext4_msg(sb, KERN_ERR,
1753                          "filesystem too large to resize to %llu blocks safely",
1754                          n_blocks_count);
1755                 if (sizeof(sector_t) < 8)
1756                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1757                 return -EINVAL;
1758         }
1759
1760         if (n_blocks_count < o_blocks_count) {
1761                 ext4_warning(sb, "can't shrink FS - resize aborted");
1762                 return -EINVAL;
1763         }
1764
1765         /* Handle the remaining blocks in the last group only. */
1766         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1767
1768         if (last == 0) {
1769                 ext4_warning(sb, "need to use ext2online to resize further");
1770                 return -EPERM;
1771         }
1772
1773         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1774
1775         if (o_blocks_count + add < o_blocks_count) {
1776                 ext4_warning(sb, "blocks_count overflow");
1777                 return -EINVAL;
1778         }
1779
1780         if (o_blocks_count + add > n_blocks_count)
1781                 add = n_blocks_count - o_blocks_count;
1782
1783         if (o_blocks_count + add < n_blocks_count)
1784                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1785                              o_blocks_count + add, add);
1786
1787         /* See if the device is actually as big as what was requested */
1788         bh = sb_bread(sb, o_blocks_count + add - 1);
1789         if (!bh) {
1790                 ext4_warning(sb, "can't read last block, resize aborted");
1791                 return -ENOSPC;
1792         }
1793         brelse(bh);
1794
1795         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1796         return err;
1797 } /* ext4_group_extend */
1798
1799
1800 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1801 {
1802         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1803 }
1804
1805 /*
1806  * Release the resize inode and drop the resize_inode feature if there
1807  * are no more reserved gdt blocks, and then convert the file system
1808  * to enable meta_bg
1809  */
1810 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1811 {
1812         handle_t *handle;
1813         struct ext4_sb_info *sbi = EXT4_SB(sb);
1814         struct ext4_super_block *es = sbi->s_es;
1815         struct ext4_inode_info *ei = EXT4_I(inode);
1816         ext4_fsblk_t nr;
1817         int i, ret, err = 0;
1818         int credits = 1;
1819
1820         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1821         if (inode) {
1822                 if (es->s_reserved_gdt_blocks) {
1823                         ext4_error(sb, "Unexpected non-zero "
1824                                    "s_reserved_gdt_blocks");
1825                         return -EPERM;
1826                 }
1827
1828                 /* Do a quick sanity check of the resize inode */
1829                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1830                                              (9 - sbi->s_cluster_bits)))
1831                         goto invalid_resize_inode;
1832                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1833                         if (i == EXT4_DIND_BLOCK) {
1834                                 if (ei->i_data[i])
1835                                         continue;
1836                                 else
1837                                         goto invalid_resize_inode;
1838                         }
1839                         if (ei->i_data[i])
1840                                 goto invalid_resize_inode;
1841                 }
1842                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1843         }
1844
1845         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1846         if (IS_ERR(handle))
1847                 return PTR_ERR(handle);
1848
1849         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1850         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1851         if (err)
1852                 goto errout;
1853
1854         ext4_clear_feature_resize_inode(sb);
1855         ext4_set_feature_meta_bg(sb);
1856         sbi->s_es->s_first_meta_bg =
1857                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1858
1859         err = ext4_handle_dirty_super(handle, sb);
1860         if (err) {
1861                 ext4_std_error(sb, err);
1862                 goto errout;
1863         }
1864
1865         if (inode) {
1866                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1867                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1868                                  EXT4_FREE_BLOCKS_METADATA |
1869                                  EXT4_FREE_BLOCKS_FORGET);
1870                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1871                 inode->i_blocks = 0;
1872
1873                 err = ext4_mark_inode_dirty(handle, inode);
1874                 if (err)
1875                         ext4_std_error(sb, err);
1876         }
1877
1878 errout:
1879         ret = ext4_journal_stop(handle);
1880         if (!err)
1881                 err = ret;
1882         return ret;
1883
1884 invalid_resize_inode:
1885         ext4_error(sb, "corrupted/inconsistent resize inode");
1886         return -EINVAL;
1887 }
1888
1889 /*
1890  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1891  *
1892  * @sb: super block of the fs to be resized
1893  * @n_blocks_count: the number of blocks resides in the resized fs
1894  */
1895 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1896 {
1897         struct ext4_new_flex_group_data *flex_gd = NULL;
1898         struct ext4_sb_info *sbi = EXT4_SB(sb);
1899         struct ext4_super_block *es = sbi->s_es;
1900         struct buffer_head *bh;
1901         struct inode *resize_inode = NULL;
1902         ext4_grpblk_t add, offset;
1903         unsigned long n_desc_blocks;
1904         unsigned long o_desc_blocks;
1905         ext4_group_t o_group;
1906         ext4_group_t n_group;
1907         ext4_fsblk_t o_blocks_count;
1908         ext4_fsblk_t n_blocks_count_retry = 0;
1909         unsigned long last_update_time = 0;
1910         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1911         int meta_bg;
1912
1913         /* See if the device is actually as big as what was requested */
1914         bh = sb_bread(sb, n_blocks_count - 1);
1915         if (!bh) {
1916                 ext4_warning(sb, "can't read last block, resize aborted");
1917                 return -ENOSPC;
1918         }
1919         brelse(bh);
1920
1921 retry:
1922         o_blocks_count = ext4_blocks_count(es);
1923
1924         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1925                  "to %llu blocks", o_blocks_count, n_blocks_count);
1926
1927         if (n_blocks_count < o_blocks_count) {
1928                 /* On-line shrinking not supported */
1929                 ext4_warning(sb, "can't shrink FS - resize aborted");
1930                 return -EINVAL;
1931         }
1932
1933         if (n_blocks_count == o_blocks_count)
1934                 /* Nothing need to do */
1935                 return 0;
1936
1937         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1938         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1939                 ext4_warning(sb, "resize would cause inodes_count overflow");
1940                 return -EINVAL;
1941         }
1942         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1943
1944         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1945         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1946
1947         meta_bg = ext4_has_feature_meta_bg(sb);
1948
1949         if (ext4_has_feature_resize_inode(sb)) {
1950                 if (meta_bg) {
1951                         ext4_error(sb, "resize_inode and meta_bg enabled "
1952                                    "simultaneously");
1953                         return -EINVAL;
1954                 }
1955                 if (n_desc_blocks > o_desc_blocks +
1956                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
1957                         n_blocks_count_retry = n_blocks_count;
1958                         n_desc_blocks = o_desc_blocks +
1959                                 le16_to_cpu(es->s_reserved_gdt_blocks);
1960                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1961                         n_blocks_count = (ext4_fsblk_t)n_group *
1962                                 EXT4_BLOCKS_PER_GROUP(sb);
1963                         n_group--; /* set to last group number */
1964                 }
1965
1966                 if (!resize_inode)
1967                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1968                 if (IS_ERR(resize_inode)) {
1969                         ext4_warning(sb, "Error opening resize inode");
1970                         return PTR_ERR(resize_inode);
1971                 }
1972         }
1973
1974         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1975                 err = ext4_convert_meta_bg(sb, resize_inode);
1976                 if (err)
1977                         goto out;
1978                 if (resize_inode) {
1979                         iput(resize_inode);
1980                         resize_inode = NULL;
1981                 }
1982                 if (n_blocks_count_retry) {
1983                         n_blocks_count = n_blocks_count_retry;
1984                         n_blocks_count_retry = 0;
1985                         goto retry;
1986                 }
1987         }
1988
1989         /* extend the last group */
1990         if (n_group == o_group)
1991                 add = n_blocks_count - o_blocks_count;
1992         else
1993                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
1994         if (add > 0) {
1995                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1996                 if (err)
1997                         goto out;
1998         }
1999
2000         if (ext4_blocks_count(es) == n_blocks_count)
2001                 goto out;
2002
2003         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2004         if (err)
2005                 return err;
2006
2007         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2008         if (err)
2009                 goto out;
2010
2011         flex_gd = alloc_flex_gd(flexbg_size);
2012         if (flex_gd == NULL) {
2013                 err = -ENOMEM;
2014                 goto out;
2015         }
2016
2017         /* Add flex groups. Note that a regular group is a
2018          * flex group with 1 group.
2019          */
2020         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2021                                               flexbg_size)) {
2022                 if (jiffies - last_update_time > HZ * 10) {
2023                         if (last_update_time)
2024                                 ext4_msg(sb, KERN_INFO,
2025                                          "resized to %llu blocks",
2026                                          ext4_blocks_count(es));
2027                         last_update_time = jiffies;
2028                 }
2029                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2030                         break;
2031                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2032                 if (unlikely(err))
2033                         break;
2034         }
2035
2036         if (!err && n_blocks_count_retry) {
2037                 n_blocks_count = n_blocks_count_retry;
2038                 n_blocks_count_retry = 0;
2039                 free_flex_gd(flex_gd);
2040                 flex_gd = NULL;
2041                 goto retry;
2042         }
2043
2044 out:
2045         if (flex_gd)
2046                 free_flex_gd(flex_gd);
2047         if (resize_inode != NULL)
2048                 iput(resize_inode);
2049         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2050         return err;
2051 }