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