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 #include <linux/jiffies.h>
  18
  19 #include "ext4_jbd2.h"
  20
  21 struct ext4_rcu_ptr {
  22         struct rcu_head rcu;
  23         void *ptr;
  24 };
  25
  26 static void ext4_rcu_ptr_callback(struct rcu_head *head)
  27 {
  28         struct ext4_rcu_ptr *ptr;
  29
  30         ptr = container_of(head, struct ext4_rcu_ptr, rcu);
  31         kvfree(ptr->ptr);
  32         kfree(ptr);
  33 }
  34
  35 void ext4_kvfree_array_rcu(void *to_free)
  36 {
  37         struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
  38
  39         if (ptr) {
  40                 ptr->ptr = to_free;
  41                 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
  42                 return;
  43         }
  44         synchronize_rcu();
  45         kvfree(to_free);
  46 }
  47
  48 int ext4_resize_begin(struct super_block *sb)
  49 {
  50         struct ext4_sb_info *sbi = EXT4_SB(sb);
  51         int ret = 0;
  52
  53         if (!capable(CAP_SYS_RESOURCE))
  54                 return -EPERM;
  55
  56         /*
  57          * If we are not using the primary superblock/GDT copy don't resize,
  58          * because the user tools have no way of handling this.  Probably a
  59          * bad time to do it anyways.
  60          */
  61         if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
  62             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  63                 ext4_warning(sb, "won't resize using backup superblock at %llu",
  64                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  65                 return -EPERM;
  66         }
  67
  68         /*
  69          * We are not allowed to do online-resizing on a filesystem mounted
  70          * with error, because it can destroy the filesystem easily.
  71          */
  72         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  73                 ext4_warning(sb, "There are errors in the filesystem, "
  74                              "so online resizing is not allowed");
  75                 return -EPERM;
  76         }
  77
  78         if (ext4_has_feature_sparse_super2(sb)) {
  79                 ext4_msg(sb, KERN_ERR, "Online resizing not supported with sparse_super2");
  80                 return -EOPNOTSUPP;
  81         }
  82
  83         if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
  84                                   &EXT4_SB(sb)->s_ext4_flags))
  85                 ret = -EBUSY;
  86
  87         return ret;
  88 }
  89
  90 void ext4_resize_end(struct super_block *sb)
  91 {
  92         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
  93         smp_mb__after_atomic();
  94 }
  95
  96 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  97                                              ext4_group_t group) {
  98         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  99                EXT4_DESC_PER_BLOCK_BITS(sb);
 100 }
 101
 102 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
 103                                              ext4_group_t group) {
 104         group = ext4_meta_bg_first_group(sb, group);
 105         return ext4_group_first_block_no(sb, group);
 106 }
 107
 108 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
 109                                                 ext4_group_t group) {
 110         ext4_grpblk_t overhead;
 111         overhead = ext4_bg_num_gdb(sb, group);
 112         if (ext4_bg_has_super(sb, group))
 113                 overhead += 1 +
 114                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 115         return overhead;
 116 }
 117
 118 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
 119 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
 120
 121 static int verify_group_input(struct super_block *sb,
 122                               struct ext4_new_group_data *input)
 123 {
 124         struct ext4_sb_info *sbi = EXT4_SB(sb);
 125         struct ext4_super_block *es = sbi->s_es;
 126         ext4_fsblk_t start = ext4_blocks_count(es);
 127         ext4_fsblk_t end = start + input->blocks_count;
 128         ext4_group_t group = input->group;
 129         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
 130         unsigned overhead;
 131         ext4_fsblk_t metaend;
 132         struct buffer_head *bh = NULL;
 133         ext4_grpblk_t free_blocks_count, offset;
 134         int err = -EINVAL;
 135
 136         if (group != sbi->s_groups_count) {
 137                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 138                              input->group, sbi->s_groups_count);
 139                 return -EINVAL;
 140         }
 141
 142         overhead = ext4_group_overhead_blocks(sb, group);
 143         metaend = start + overhead;
 144         input->free_clusters_count = free_blocks_count =
 145                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 146
 147         if (test_opt(sb, DEBUG))
 148                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 149                        "(%d free, %u reserved)\n",
 150                        ext4_bg_has_super(sb, input->group) ? "normal" :
 151                        "no-super", input->group, input->blocks_count,
 152                        free_blocks_count, input->reserved_blocks);
 153
 154         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 155         if (offset != 0)
 156                         ext4_warning(sb, "Last group not full");
 157         else if (input->reserved_blocks > input->blocks_count / 5)
 158                 ext4_warning(sb, "Reserved blocks too high (%u)",
 159                              input->reserved_blocks);
 160         else if (free_blocks_count < 0)
 161                 ext4_warning(sb, "Bad blocks count %u",
 162                              input->blocks_count);
 163         else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
 164                 err = PTR_ERR(bh);
 165                 bh = NULL;
 166                 ext4_warning(sb, "Cannot read last block (%llu)",
 167                              end - 1);
 168         } else if (outside(input->block_bitmap, start, end))
 169                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 170                              (unsigned long long)input->block_bitmap);
 171         else if (outside(input->inode_bitmap, start, end))
 172                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 173                              (unsigned long long)input->inode_bitmap);
 174         else if (outside(input->inode_table, start, end) ||
 175                  outside(itend - 1, start, end))
 176                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 177                              (unsigned long long)input->inode_table, itend - 1);
 178         else if (input->inode_bitmap == input->block_bitmap)
 179                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 180                              (unsigned long long)input->block_bitmap);
 181         else if (inside(input->block_bitmap, input->inode_table, itend))
 182                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 183                              "(%llu-%llu)",
 184                              (unsigned long long)input->block_bitmap,
 185                              (unsigned long long)input->inode_table, itend - 1);
 186         else if (inside(input->inode_bitmap, input->inode_table, itend))
 187                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 188                              "(%llu-%llu)",
 189                              (unsigned long long)input->inode_bitmap,
 190                              (unsigned long long)input->inode_table, itend - 1);
 191         else if (inside(input->block_bitmap, start, metaend))
 192                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 193                              (unsigned long long)input->block_bitmap,
 194                              start, metaend - 1);
 195         else if (inside(input->inode_bitmap, start, metaend))
 196                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 197                              (unsigned long long)input->inode_bitmap,
 198                              start, metaend - 1);
 199         else if (inside(input->inode_table, start, metaend) ||
 200                  inside(itend - 1, start, metaend))
 201                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 202                              "(%llu-%llu)",
 203                              (unsigned long long)input->inode_table,
 204                              itend - 1, start, metaend - 1);
 205         else
 206                 err = 0;
 207         brelse(bh);
 208
 209         return err;
 210 }
 211
 212 /*
 213  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 214  * group each time.
 215  */
 216 struct ext4_new_flex_group_data {
 217         struct ext4_new_group_data *groups;     /* new_group_data for groups
 218                                                    in the flex group */
 219         __u16 *bg_flags;                        /* block group flags of groups
 220                                                    in @groups */
 221         ext4_group_t count;                     /* number of groups in @groups
 222                                                  */
 223 };
 224
 225 /*
 226  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 227  * @flexbg_size.
 228  *
 229  * Returns NULL on failure otherwise address of the allocated structure.
 230  */
 231 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 232 {
 233         struct ext4_new_flex_group_data *flex_gd;
 234
 235         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 236         if (flex_gd == NULL)
 237                 goto out3;
 238
 239         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
 240                 goto out2;
 241         flex_gd->count = flexbg_size;
 242
 243         flex_gd->groups = kmalloc_array(flexbg_size,
 244                                         sizeof(struct ext4_new_group_data),
 245                                         GFP_NOFS);
 246         if (flex_gd->groups == NULL)
 247                 goto out2;
 248
 249         flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
 250                                           GFP_NOFS);
 251         if (flex_gd->bg_flags == NULL)
 252                 goto out1;
 253
 254         return flex_gd;
 255
 256 out1:
 257         kfree(flex_gd->groups);
 258 out2:
 259         kfree(flex_gd);
 260 out3:
 261         return NULL;
 262 }
 263
 264 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 265 {
 266         kfree(flex_gd->bg_flags);
 267         kfree(flex_gd->groups);
 268         kfree(flex_gd);
 269 }
 270
 271 /*
 272  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 273  * and inode tables for a flex group.
 274  *
 275  * This function is used by 64bit-resize.  Note that this function allocates
 276  * group tables from the 1st group of groups contained by @flexgd, which may
 277  * be a partial of a flex group.
 278  *
 279  * @sb: super block of fs to which the groups belongs
 280  *
 281  * Returns 0 on a successful allocation of the metadata blocks in the
 282  * block group.
 283  */
 284 static int ext4_alloc_group_tables(struct super_block *sb,
 285                                 struct ext4_new_flex_group_data *flex_gd,
 286                                 int flexbg_size)
 287 {
 288         struct ext4_new_group_data *group_data = flex_gd->groups;
 289         ext4_fsblk_t start_blk;
 290         ext4_fsblk_t last_blk;
 291         ext4_group_t src_group;
 292         ext4_group_t bb_index = 0;
 293         ext4_group_t ib_index = 0;
 294         ext4_group_t it_index = 0;
 295         ext4_group_t group;
 296         ext4_group_t last_group;
 297         unsigned overhead;
 298         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 299         int i;
 300
 301         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 302
 303         src_group = group_data[0].group;
 304         last_group  = src_group + flex_gd->count - 1;
 305
 306         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 307                (last_group & ~(flexbg_size - 1))));
 308 next_group:
 309         group = group_data[0].group;
 310         if (src_group >= group_data[0].group + flex_gd->count)
 311                 return -ENOSPC;
 312         start_blk = ext4_group_first_block_no(sb, src_group);
 313         last_blk = start_blk + group_data[src_group - group].blocks_count;
 314
 315         overhead = ext4_group_overhead_blocks(sb, src_group);
 316
 317         start_blk += overhead;
 318
 319         /* We collect contiguous blocks as much as possible. */
 320         src_group++;
 321         for (; src_group <= last_group; src_group++) {
 322                 overhead = ext4_group_overhead_blocks(sb, src_group);
 323                 if (overhead == 0)
 324                         last_blk += group_data[src_group - group].blocks_count;
 325                 else
 326                         break;
 327         }
 328
 329         /* Allocate block bitmaps */
 330         for (; bb_index < flex_gd->count; bb_index++) {
 331                 if (start_blk >= last_blk)
 332                         goto next_group;
 333                 group_data[bb_index].block_bitmap = start_blk++;
 334                 group = ext4_get_group_number(sb, start_blk - 1);
 335                 group -= group_data[0].group;
 336                 group_data[group].mdata_blocks++;
 337                 flex_gd->bg_flags[group] &= uninit_mask;
 338         }
 339
 340         /* Allocate inode bitmaps */
 341         for (; ib_index < flex_gd->count; ib_index++) {
 342                 if (start_blk >= last_blk)
 343                         goto next_group;
 344                 group_data[ib_index].inode_bitmap = start_blk++;
 345                 group = ext4_get_group_number(sb, start_blk - 1);
 346                 group -= group_data[0].group;
 347                 group_data[group].mdata_blocks++;
 348                 flex_gd->bg_flags[group] &= uninit_mask;
 349         }
 350
 351         /* Allocate inode tables */
 352         for (; it_index < flex_gd->count; it_index++) {
 353                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 354                 ext4_fsblk_t next_group_start;
 355
 356                 if (start_blk + itb > last_blk)
 357                         goto next_group;
 358                 group_data[it_index].inode_table = start_blk;
 359                 group = ext4_get_group_number(sb, start_blk);
 360                 next_group_start = ext4_group_first_block_no(sb, group + 1);
 361                 group -= group_data[0].group;
 362
 363                 if (start_blk + itb > next_group_start) {
 364                         flex_gd->bg_flags[group + 1] &= uninit_mask;
 365                         overhead = start_blk + itb - next_group_start;
 366                         group_data[group + 1].mdata_blocks += overhead;
 367                         itb -= overhead;
 368                 }
 369
 370                 group_data[group].mdata_blocks += itb;
 371                 flex_gd->bg_flags[group] &= uninit_mask;
 372                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 373         }
 374
 375         /* Update free clusters count to exclude metadata blocks */
 376         for (i = 0; i < flex_gd->count; i++) {
 377                 group_data[i].free_clusters_count -=
 378                                 EXT4_NUM_B2C(EXT4_SB(sb),
 379                                              group_data[i].mdata_blocks);
 380         }
 381
 382         if (test_opt(sb, DEBUG)) {
 383                 int i;
 384                 group = group_data[0].group;
 385
 386                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 387                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 388                        flexbg_size);
 389
 390                 for (i = 0; i < flex_gd->count; i++) {
 391                         ext4_debug(
 392                                "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
 393                                ext4_bg_has_super(sb, group + i) ? "normal" :
 394                                "no-super", group + i,
 395                                group_data[i].blocks_count,
 396                                group_data[i].free_clusters_count,
 397                                group_data[i].mdata_blocks);
 398                 }
 399         }
 400         return 0;
 401 }
 402
 403 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 404                                   ext4_fsblk_t blk)
 405 {
 406         struct buffer_head *bh;
 407         int err;
 408
 409         bh = sb_getblk(sb, blk);
 410         if (unlikely(!bh))
 411                 return ERR_PTR(-ENOMEM);
 412         BUFFER_TRACE(bh, "get_write_access");
 413         err = ext4_journal_get_write_access(handle, sb, bh, EXT4_JTR_NONE);
 414         if (err) {
 415                 brelse(bh);
 416                 bh = ERR_PTR(err);
 417         } else {
 418                 memset(bh->b_data, 0, sb->s_blocksize);
 419                 set_buffer_uptodate(bh);
 420         }
 421
 422         return bh;
 423 }
 424
 425 static int ext4_resize_ensure_credits_batch(handle_t *handle, int credits)
 426 {
 427         return ext4_journal_ensure_credits_fn(handle, credits,
 428                 EXT4_MAX_TRANS_DATA, 0, 0);
 429 }
 430
 431 /*
 432  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
 433  *
 434  * Helper function for ext4_setup_new_group_blocks() which set .
 435  *
 436  * @sb: super block
 437  * @handle: journal handle
 438  * @flex_gd: flex group data
 439  */
 440 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 441                         struct ext4_new_flex_group_data *flex_gd,
 442                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
 443 {
 444         struct ext4_sb_info *sbi = EXT4_SB(sb);
 445         ext4_group_t count = last_cluster - first_cluster + 1;
 446         ext4_group_t count2;
 447
 448         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
 449                    last_cluster);
 450         for (count2 = count; count > 0;
 451              count -= count2, first_cluster += count2) {
 452                 ext4_fsblk_t start;
 453                 struct buffer_head *bh;
 454                 ext4_group_t group;
 455                 int err;
 456
 457                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
 458                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
 459                 group -= flex_gd->groups[0].group;
 460
 461                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
 462                 if (count2 > count)
 463                         count2 = count;
 464
 465                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 466                         BUG_ON(flex_gd->count > 1);
 467                         continue;
 468                 }
 469
 470                 err = ext4_resize_ensure_credits_batch(handle, 1);
 471                 if (err < 0)
 472                         return err;
 473
 474                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 475                 if (unlikely(!bh))
 476                         return -ENOMEM;
 477
 478                 BUFFER_TRACE(bh, "get_write_access");
 479                 err = ext4_journal_get_write_access(handle, sb, bh,
 480                                                     EXT4_JTR_NONE);
 481                 if (err) {
 482                         brelse(bh);
 483                         return err;
 484                 }
 485                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
 486                            first_cluster, first_cluster - start, count2);
 487                 mb_set_bits(bh->b_data, first_cluster - start, count2);
 488
 489                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 490                 brelse(bh);
 491                 if (unlikely(err))
 492                         return err;
 493         }
 494
 495         return 0;
 496 }
 497
 498 /*
 499  * Set up the block and inode bitmaps, and the inode table for the new groups.
 500  * This doesn't need to be part of the main transaction, since we are only
 501  * changing blocks outside the actual filesystem.  We still do journaling to
 502  * ensure the recovery is correct in case of a failure just after resize.
 503  * If any part of this fails, we simply abort the resize.
 504  *
 505  * setup_new_flex_group_blocks handles a flex group as follow:
 506  *  1. copy super block and GDT, and initialize group tables if necessary.
 507  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 508  *     super block and GDT.
 509  *  2. allocate group tables in block bitmaps, that is, set bits in block
 510  *     bitmap for blocks taken by group tables.
 511  */
 512 static int setup_new_flex_group_blocks(struct super_block *sb,
 513                                 struct ext4_new_flex_group_data *flex_gd)
 514 {
 515         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 516         ext4_fsblk_t start;
 517         ext4_fsblk_t block;
 518         struct ext4_sb_info *sbi = EXT4_SB(sb);
 519         struct ext4_super_block *es = sbi->s_es;
 520         struct ext4_new_group_data *group_data = flex_gd->groups;
 521         __u16 *bg_flags = flex_gd->bg_flags;
 522         handle_t *handle;
 523         ext4_group_t group, count;
 524         struct buffer_head *bh = NULL;
 525         int reserved_gdb, i, j, err = 0, err2;
 526         int meta_bg;
 527
 528         BUG_ON(!flex_gd->count || !group_data ||
 529                group_data[0].group != sbi->s_groups_count);
 530
 531         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 532         meta_bg = ext4_has_feature_meta_bg(sb);
 533
 534         /* This transaction may be extended/restarted along the way */
 535         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 536         if (IS_ERR(handle))
 537                 return PTR_ERR(handle);
 538
 539         group = group_data[0].group;
 540         for (i = 0; i < flex_gd->count; i++, group++) {
 541                 unsigned long gdblocks;
 542                 ext4_grpblk_t overhead;
 543
 544                 gdblocks = ext4_bg_num_gdb(sb, group);
 545                 start = ext4_group_first_block_no(sb, group);
 546
 547                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 548                         goto handle_itb;
 549
 550                 if (meta_bg == 1) {
 551                         ext4_group_t first_group;
 552                         first_group = ext4_meta_bg_first_group(sb, group);
 553                         if (first_group != group + 1 &&
 554                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 555                                 goto handle_itb;
 556                 }
 557
 558                 block = start + ext4_bg_has_super(sb, group);
 559                 /* Copy all of the GDT blocks into the backup in this group */
 560                 for (j = 0; j < gdblocks; j++, block++) {
 561                         struct buffer_head *gdb;
 562
 563                         ext4_debug("update backup group %#04llx\n", block);
 564                         err = ext4_resize_ensure_credits_batch(handle, 1);
 565                         if (err < 0)
 566                                 goto out;
 567
 568                         gdb = sb_getblk(sb, block);
 569                         if (unlikely(!gdb)) {
 570                                 err = -ENOMEM;
 571                                 goto out;
 572                         }
 573
 574                         BUFFER_TRACE(gdb, "get_write_access");
 575                         err = ext4_journal_get_write_access(handle, sb, gdb,
 576                                                             EXT4_JTR_NONE);
 577                         if (err) {
 578                                 brelse(gdb);
 579                                 goto out;
 580                         }
 581                         memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
 582                                 s_group_desc, j)->b_data, gdb->b_size);
 583                         set_buffer_uptodate(gdb);
 584
 585                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 586                         if (unlikely(err)) {
 587                                 brelse(gdb);
 588                                 goto out;
 589                         }
 590                         brelse(gdb);
 591                 }
 592
 593                 /* Zero out all of the reserved backup group descriptor
 594                  * table blocks
 595                  */
 596                 if (ext4_bg_has_super(sb, group)) {
 597                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 598                                         reserved_gdb, GFP_NOFS);
 599                         if (err)
 600                                 goto out;
 601                 }
 602
 603 handle_itb:
 604                 /* Initialize group tables of the grop @group */
 605                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 606                         goto handle_bb;
 607
 608                 /* Zero out all of the inode table blocks */
 609                 block = group_data[i].inode_table;
 610                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 611                            block, sbi->s_itb_per_group);
 612                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 613                                        GFP_NOFS);
 614                 if (err)
 615                         goto out;
 616
 617 handle_bb:
 618                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 619                         goto handle_ib;
 620
 621                 /* Initialize block bitmap of the @group */
 622                 block = group_data[i].block_bitmap;
 623                 err = ext4_resize_ensure_credits_batch(handle, 1);
 624                 if (err < 0)
 625                         goto out;
 626
 627                 bh = bclean(handle, sb, block);
 628                 if (IS_ERR(bh)) {
 629                         err = PTR_ERR(bh);
 630                         goto out;
 631                 }
 632                 overhead = ext4_group_overhead_blocks(sb, group);
 633                 if (overhead != 0) {
 634                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 635                                    start);
 636                         mb_set_bits(bh->b_data, 0,
 637                                       EXT4_NUM_B2C(sbi, overhead));
 638                 }
 639                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
 640                                      sb->s_blocksize * 8, bh->b_data);
 641                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 642                 brelse(bh);
 643                 if (err)
 644                         goto out;
 645
 646 handle_ib:
 647                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 648                         continue;
 649
 650                 /* Initialize inode bitmap of the @group */
 651                 block = group_data[i].inode_bitmap;
 652                 err = ext4_resize_ensure_credits_batch(handle, 1);
 653                 if (err < 0)
 654                         goto out;
 655                 /* Mark unused entries in inode bitmap used */
 656                 bh = bclean(handle, sb, block);
 657                 if (IS_ERR(bh)) {
 658                         err = PTR_ERR(bh);
 659                         goto out;
 660                 }
 661
 662                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 663                                      sb->s_blocksize * 8, bh->b_data);
 664                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 665                 brelse(bh);
 666                 if (err)
 667                         goto out;
 668         }
 669
 670         /* Mark group tables in block bitmap */
 671         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 672                 count = group_table_count[j];
 673                 start = (&group_data[0].block_bitmap)[j];
 674                 block = start;
 675                 for (i = 1; i < flex_gd->count; i++) {
 676                         block += group_table_count[j];
 677                         if (block == (&group_data[i].block_bitmap)[j]) {
 678                                 count += group_table_count[j];
 679                                 continue;
 680                         }
 681                         err = set_flexbg_block_bitmap(sb, handle,
 682                                                       flex_gd,
 683                                                       EXT4_B2C(sbi, start),
 684                                                       EXT4_B2C(sbi,
 685                                                                start + count
 686                                                                - 1));
 687                         if (err)
 688                                 goto out;
 689                         count = group_table_count[j];
 690                         start = (&group_data[i].block_bitmap)[j];
 691                         block = start;
 692                 }
 693
 694                 if (count) {
 695                         err = set_flexbg_block_bitmap(sb, handle,
 696                                                       flex_gd,
 697                                                       EXT4_B2C(sbi, start),
 698                                                       EXT4_B2C(sbi,
 699                                                                start + count
 700                                                                - 1));
 701                         if (err)
 702                                 goto out;
 703                 }
 704         }
 705
 706 out:
 707         err2 = ext4_journal_stop(handle);
 708         if (err2 && !err)
 709                 err = err2;
 710
 711         return err;
 712 }
 713
 714 /*
 715  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 716  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 717  * calling this for the first time.  In a sparse filesystem it will be the
 718  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 719  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 720  */
 721 unsigned int ext4_list_backups(struct super_block *sb, unsigned int *three,
 722                                unsigned int *five, unsigned int *seven)
 723 {
 724         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 725         unsigned int *min = three;
 726         int mult = 3;
 727         unsigned int ret;
 728
 729         if (ext4_has_feature_sparse_super2(sb)) {
 730                 do {
 731                         if (*min > 2)
 732                                 return UINT_MAX;
 733                         ret = le32_to_cpu(es->s_backup_bgs[*min - 1]);
 734                         *min += 1;
 735                 } while (!ret);
 736                 return ret;
 737         }
 738
 739         if (!ext4_has_feature_sparse_super(sb)) {
 740                 ret = *min;
 741                 *min += 1;
 742                 return ret;
 743         }
 744
 745         if (*five < *min) {
 746                 min = five;
 747                 mult = 5;
 748         }
 749         if (*seven < *min) {
 750                 min = seven;
 751                 mult = 7;
 752         }
 753
 754         ret = *min;
 755         *min *= mult;
 756
 757         return ret;
 758 }
 759
 760 /*
 761  * Check that all of the backup GDT blocks are held in the primary GDT block.
 762  * It is assumed that they are stored in group order.  Returns the number of
 763  * groups in current filesystem that have BACKUPS, or -ve error code.
 764  */
 765 static int verify_reserved_gdb(struct super_block *sb,
 766                                ext4_group_t end,
 767                                struct buffer_head *primary)
 768 {
 769         const ext4_fsblk_t blk = primary->b_blocknr;
 770         unsigned three = 1;
 771         unsigned five = 5;
 772         unsigned seven = 7;
 773         unsigned grp;
 774         __le32 *p = (__le32 *)primary->b_data;
 775         int gdbackups = 0;
 776
 777         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 778                 if (le32_to_cpu(*p++) !=
 779                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 780                         ext4_warning(sb, "reserved GDT %llu"
 781                                      " missing grp %d (%llu)",
 782                                      blk, grp,
 783                                      grp *
 784                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 785                                      blk);
 786                         return -EINVAL;
 787                 }
 788                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 789                         return -EFBIG;
 790         }
 791
 792         return gdbackups;
 793 }
 794
 795 /*
 796  * Called when we need to bring a reserved group descriptor table block into
 797  * use from the resize inode.  The primary copy of the new GDT block currently
 798  * is an indirect block (under the double indirect block in the resize inode).
 799  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 800  * block, in group order.  Even though we know all the block numbers we need,
 801  * we check to ensure that the resize inode has actually reserved these blocks.
 802  *
 803  * Don't need to update the block bitmaps because the blocks are still in use.
 804  *
 805  * We get all of the error cases out of the way, so that we are sure to not
 806  * fail once we start modifying the data on disk, because JBD has no rollback.
 807  */
 808 static int add_new_gdb(handle_t *handle, struct inode *inode,
 809                        ext4_group_t group)
 810 {
 811         struct super_block *sb = inode->i_sb;
 812         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 813         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 814         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 815         struct buffer_head **o_group_desc, **n_group_desc = NULL;
 816         struct buffer_head *dind = NULL;
 817         struct buffer_head *gdb_bh = NULL;
 818         int gdbackups;
 819         struct ext4_iloc iloc = { .bh = NULL };
 820         __le32 *data;
 821         int err;
 822
 823         if (test_opt(sb, DEBUG))
 824                 printk(KERN_DEBUG
 825                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 826                        gdb_num);
 827
 828         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 829         if (IS_ERR(gdb_bh))
 830                 return PTR_ERR(gdb_bh);
 831
 832         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 833         if (gdbackups < 0) {
 834                 err = gdbackups;
 835                 goto errout;
 836         }
 837
 838         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 839         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
 840         if (IS_ERR(dind)) {
 841                 err = PTR_ERR(dind);
 842                 dind = NULL;
 843                 goto errout;
 844         }
 845
 846         data = (__le32 *)dind->b_data;
 847         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 848                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 849                              group, gdblock);
 850                 err = -EINVAL;
 851                 goto errout;
 852         }
 853
 854         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 855         err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
 856                                             EXT4_JTR_NONE);
 857         if (unlikely(err))
 858                 goto errout;
 859
 860         BUFFER_TRACE(gdb_bh, "get_write_access");
 861         err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
 862         if (unlikely(err))
 863                 goto errout;
 864
 865         BUFFER_TRACE(dind, "get_write_access");
 866         err = ext4_journal_get_write_access(handle, sb, dind, EXT4_JTR_NONE);
 867         if (unlikely(err)) {
 868                 ext4_std_error(sb, err);
 869                 goto errout;
 870         }
 871
 872         /* ext4_reserve_inode_write() gets a reference on the iloc */
 873         err = ext4_reserve_inode_write(handle, inode, &iloc);
 874         if (unlikely(err))
 875                 goto errout;
 876
 877         n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 878                                 GFP_KERNEL);
 879         if (!n_group_desc) {
 880                 err = -ENOMEM;
 881                 ext4_warning(sb, "not enough memory for %lu groups",
 882                              gdb_num + 1);
 883                 goto errout;
 884         }
 885
 886         /*
 887          * Finally, we have all of the possible failures behind us...
 888          *
 889          * Remove new GDT block from inode double-indirect block and clear out
 890          * the new GDT block for use (which also "frees" the backup GDT blocks
 891          * from the reserved inode).  We don't need to change the bitmaps for
 892          * these blocks, because they are marked as in-use from being in the
 893          * reserved inode, and will become GDT blocks (primary and backup).
 894          */
 895         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 896         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 897         if (unlikely(err)) {
 898                 ext4_std_error(sb, err);
 899                 goto errout;
 900         }
 901         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
 902                            (9 - EXT4_SB(sb)->s_cluster_bits);
 903         ext4_mark_iloc_dirty(handle, inode, &iloc);
 904         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 905         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 906         if (unlikely(err)) {
 907                 ext4_std_error(sb, err);
 908                 iloc.bh = NULL;
 909                 goto errout;
 910         }
 911         brelse(dind);
 912
 913         rcu_read_lock();
 914         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 915         memcpy(n_group_desc, o_group_desc,
 916                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 917         rcu_read_unlock();
 918         n_group_desc[gdb_num] = gdb_bh;
 919         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 920         EXT4_SB(sb)->s_gdb_count++;
 921         ext4_kvfree_array_rcu(o_group_desc);
 922
 923         lock_buffer(EXT4_SB(sb)->s_sbh);
 924         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 925         ext4_superblock_csum_set(sb);
 926         unlock_buffer(EXT4_SB(sb)->s_sbh);
 927         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 928         if (err)
 929                 ext4_std_error(sb, err);
 930         return err;
 931 errout:
 932         kvfree(n_group_desc);
 933         brelse(iloc.bh);
 934         brelse(dind);
 935         brelse(gdb_bh);
 936
 937         ext4_debug("leaving with error %d\n", err);
 938         return err;
 939 }
 940
 941 /*
 942  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 943  */
 944 static int add_new_gdb_meta_bg(struct super_block *sb,
 945                                handle_t *handle, ext4_group_t group) {
 946         ext4_fsblk_t gdblock;
 947         struct buffer_head *gdb_bh;
 948         struct buffer_head **o_group_desc, **n_group_desc;
 949         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 950         int err;
 951
 952         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 953                    ext4_bg_has_super(sb, group);
 954         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 955         if (IS_ERR(gdb_bh))
 956                 return PTR_ERR(gdb_bh);
 957         n_group_desc = kvmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
 958                                 GFP_KERNEL);
 959         if (!n_group_desc) {
 960                 brelse(gdb_bh);
 961                 err = -ENOMEM;
 962                 ext4_warning(sb, "not enough memory for %lu groups",
 963                              gdb_num + 1);
 964                 return err;
 965         }
 966
 967         rcu_read_lock();
 968         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 969         memcpy(n_group_desc, o_group_desc,
 970                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 971         rcu_read_unlock();
 972         n_group_desc[gdb_num] = gdb_bh;
 973
 974         BUFFER_TRACE(gdb_bh, "get_write_access");
 975         err = ext4_journal_get_write_access(handle, sb, gdb_bh, EXT4_JTR_NONE);
 976         if (err) {
 977                 kvfree(n_group_desc);
 978                 brelse(gdb_bh);
 979                 return err;
 980         }
 981
 982         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 983         EXT4_SB(sb)->s_gdb_count++;
 984         ext4_kvfree_array_rcu(o_group_desc);
 985         return err;
 986 }
 987
 988 /*
 989  * Called when we are adding a new group which has a backup copy of each of
 990  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 991  * We need to add these reserved backup GDT blocks to the resize inode, so
 992  * that they are kept for future resizing and not allocated to files.
 993  *
 994  * Each reserved backup GDT block will go into a different indirect block.
 995  * The indirect blocks are actually the primary reserved GDT blocks,
 996  * so we know in advance what their block numbers are.  We only get the
 997  * double-indirect block to verify it is pointing to the primary reserved
 998  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 999  * backup GDT blocks are stored in their reserved primary GDT block.
1000  */
1001 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
1002                               ext4_group_t group)
1003 {
1004         struct super_block *sb = inode->i_sb;
1005         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1006         int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1007         struct buffer_head **primary;
1008         struct buffer_head *dind;
1009         struct ext4_iloc iloc;
1010         ext4_fsblk_t blk;
1011         __le32 *data, *end;
1012         int gdbackups = 0;
1013         int res, i;
1014         int err;
1015
1016         primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1017         if (!primary)
1018                 return -ENOMEM;
1019
1020         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1021         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1022         if (IS_ERR(dind)) {
1023                 err = PTR_ERR(dind);
1024                 dind = NULL;
1025                 goto exit_free;
1026         }
1027
1028         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1029         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1030                                          EXT4_ADDR_PER_BLOCK(sb));
1031         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1032
1033         /* Get each reserved primary GDT block and verify it holds backups */
1034         for (res = 0; res < reserved_gdb; res++, blk++) {
1035                 if (le32_to_cpu(*data) != blk) {
1036                         ext4_warning(sb, "reserved block %llu"
1037                                      " not at offset %ld",
1038                                      blk,
1039                                      (long)(data - (__le32 *)dind->b_data));
1040                         err = -EINVAL;
1041                         goto exit_bh;
1042                 }
1043                 primary[res] = ext4_sb_bread(sb, blk, 0);
1044                 if (IS_ERR(primary[res])) {
1045                         err = PTR_ERR(primary[res]);
1046                         primary[res] = NULL;
1047                         goto exit_bh;
1048                 }
1049                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1050                 if (gdbackups < 0) {
1051                         brelse(primary[res]);
1052                         err = gdbackups;
1053                         goto exit_bh;
1054                 }
1055                 if (++data >= end)
1056                         data = (__le32 *)dind->b_data;
1057         }
1058
1059         for (i = 0; i < reserved_gdb; i++) {
1060                 BUFFER_TRACE(primary[i], "get_write_access");
1061                 if ((err = ext4_journal_get_write_access(handle, sb, primary[i],
1062                                                          EXT4_JTR_NONE)))
1063                         goto exit_bh;
1064         }
1065
1066         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1067                 goto exit_bh;
1068
1069         /*
1070          * Finally we can add each of the reserved backup GDT blocks from
1071          * the new group to its reserved primary GDT block.
1072          */
1073         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1074         for (i = 0; i < reserved_gdb; i++) {
1075                 int err2;
1076                 data = (__le32 *)primary[i]->b_data;
1077                 /* printk("reserving backup %lu[%u] = %lu\n",
1078                        primary[i]->b_blocknr, gdbackups,
1079                        blk + primary[i]->b_blocknr); */
1080                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1081                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1082                 if (!err)
1083                         err = err2;
1084         }
1085
1086         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1087         ext4_mark_iloc_dirty(handle, inode, &iloc);
1088
1089 exit_bh:
1090         while (--res >= 0)
1091                 brelse(primary[res]);
1092         brelse(dind);
1093
1094 exit_free:
1095         kfree(primary);
1096
1097         return err;
1098 }
1099
1100 /*
1101  * Update the backup copies of the ext4 metadata.  These don't need to be part
1102  * of the main resize transaction, because e2fsck will re-write them if there
1103  * is a problem (basically only OOM will cause a problem).  However, we
1104  * _should_ update the backups if possible, in case the primary gets trashed
1105  * for some reason and we need to run e2fsck from a backup superblock.  The
1106  * important part is that the new block and inode counts are in the backup
1107  * superblocks, and the location of the new group metadata in the GDT backups.
1108  *
1109  * We do not need take the s_resize_lock for this, because these
1110  * blocks are not otherwise touched by the filesystem code when it is
1111  * mounted.  We don't need to worry about last changing from
1112  * sbi->s_groups_count, because the worst that can happen is that we
1113  * do not copy the full number of backups at this time.  The resize
1114  * which changed s_groups_count will backup again.
1115  */
1116 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1117                            int size, int meta_bg)
1118 {
1119         struct ext4_sb_info *sbi = EXT4_SB(sb);
1120         ext4_group_t last;
1121         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1122         unsigned three = 1;
1123         unsigned five = 5;
1124         unsigned seven = 7;
1125         ext4_group_t group = 0;
1126         int rest = sb->s_blocksize - size;
1127         handle_t *handle;
1128         int err = 0, err2;
1129
1130         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1131         if (IS_ERR(handle)) {
1132                 group = 1;
1133                 err = PTR_ERR(handle);
1134                 goto exit_err;
1135         }
1136
1137         if (meta_bg == 0) {
1138                 group = ext4_list_backups(sb, &three, &five, &seven);
1139                 last = sbi->s_groups_count;
1140         } else {
1141                 group = ext4_get_group_number(sb, blk_off) + 1;
1142                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1143         }
1144
1145         while (group < sbi->s_groups_count) {
1146                 struct buffer_head *bh;
1147                 ext4_fsblk_t backup_block;
1148
1149                 /* Out of journal space, and can't get more - abort - so sad */
1150                 err = ext4_resize_ensure_credits_batch(handle, 1);
1151                 if (err < 0)
1152                         break;
1153
1154                 if (meta_bg == 0)
1155                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1156                 else
1157                         backup_block = (ext4_group_first_block_no(sb, group) +
1158                                         ext4_bg_has_super(sb, group));
1159
1160                 bh = sb_getblk(sb, backup_block);
1161                 if (unlikely(!bh)) {
1162                         err = -ENOMEM;
1163                         break;
1164                 }
1165                 ext4_debug("update metadata backup %llu(+%llu)\n",
1166                            backup_block, backup_block -
1167                            ext4_group_first_block_no(sb, group));
1168                 BUFFER_TRACE(bh, "get_write_access");
1169                 if ((err = ext4_journal_get_write_access(handle, sb, bh,
1170                                                          EXT4_JTR_NONE)))
1171                         break;
1172                 lock_buffer(bh);
1173                 memcpy(bh->b_data, data, size);
1174                 if (rest)
1175                         memset(bh->b_data + size, 0, rest);
1176                 set_buffer_uptodate(bh);
1177                 unlock_buffer(bh);
1178                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1179                 if (unlikely(err))
1180                         ext4_std_error(sb, err);
1181                 brelse(bh);
1182
1183                 if (meta_bg == 0)
1184                         group = ext4_list_backups(sb, &three, &five, &seven);
1185                 else if (group == last)
1186                         break;
1187                 else
1188                         group = last;
1189         }
1190         if ((err2 = ext4_journal_stop(handle)) && !err)
1191                 err = err2;
1192
1193         /*
1194          * Ugh! Need to have e2fsck write the backup copies.  It is too
1195          * late to revert the resize, we shouldn't fail just because of
1196          * the backup copies (they are only needed in case of corruption).
1197          *
1198          * However, if we got here we have a journal problem too, so we
1199          * can't really start a transaction to mark the superblock.
1200          * Chicken out and just set the flag on the hope it will be written
1201          * to disk, and if not - we will simply wait until next fsck.
1202          */
1203 exit_err:
1204         if (err) {
1205                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1206                              "forcing fsck on next reboot", group, err);
1207                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1208                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1209                 mark_buffer_dirty(sbi->s_sbh);
1210         }
1211 }
1212
1213 /*
1214  * ext4_add_new_descs() adds @count group descriptor of groups
1215  * starting at @group
1216  *
1217  * @handle: journal handle
1218  * @sb: super block
1219  * @group: the group no. of the first group desc to be added
1220  * @resize_inode: the resize inode
1221  * @count: number of group descriptors to be added
1222  */
1223 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1224                               ext4_group_t group, struct inode *resize_inode,
1225                               ext4_group_t count)
1226 {
1227         struct ext4_sb_info *sbi = EXT4_SB(sb);
1228         struct ext4_super_block *es = sbi->s_es;
1229         struct buffer_head *gdb_bh;
1230         int i, gdb_off, gdb_num, err = 0;
1231         int meta_bg;
1232
1233         meta_bg = ext4_has_feature_meta_bg(sb);
1234         for (i = 0; i < count; i++, group++) {
1235                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1236                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1237
1238                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1239                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1240
1241                 /*
1242                  * We will only either add reserved group blocks to a backup group
1243                  * or remove reserved blocks for the first group in a new group block.
1244                  * Doing both would be mean more complex code, and sane people don't
1245                  * use non-sparse filesystems anymore.  This is already checked above.
1246                  */
1247                 if (gdb_off) {
1248                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1249                                                      gdb_num);
1250                         BUFFER_TRACE(gdb_bh, "get_write_access");
1251                         err = ext4_journal_get_write_access(handle, sb, gdb_bh,
1252                                                             EXT4_JTR_NONE);
1253
1254                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1255                                 err = reserve_backup_gdb(handle, resize_inode, group);
1256                 } else if (meta_bg != 0) {
1257                         err = add_new_gdb_meta_bg(sb, handle, group);
1258                 } else {
1259                         err = add_new_gdb(handle, resize_inode, group);
1260                 }
1261                 if (err)
1262                         break;
1263         }
1264         return err;
1265 }
1266
1267 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1268 {
1269         struct buffer_head *bh = sb_getblk(sb, block);
1270         if (unlikely(!bh))
1271                 return NULL;
1272         if (!bh_uptodate_or_lock(bh)) {
1273                 if (ext4_read_bh(bh, 0, NULL) < 0) {
1274                         brelse(bh);
1275                         return NULL;
1276                 }
1277         }
1278
1279         return bh;
1280 }
1281
1282 static int ext4_set_bitmap_checksums(struct super_block *sb,
1283                                      ext4_group_t group,
1284                                      struct ext4_group_desc *gdp,
1285                                      struct ext4_new_group_data *group_data)
1286 {
1287         struct buffer_head *bh;
1288
1289         if (!ext4_has_metadata_csum(sb))
1290                 return 0;
1291
1292         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1293         if (!bh)
1294                 return -EIO;
1295         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1296                                    EXT4_INODES_PER_GROUP(sb) / 8);
1297         brelse(bh);
1298
1299         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1300         if (!bh)
1301                 return -EIO;
1302         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1303         brelse(bh);
1304
1305         return 0;
1306 }
1307
1308 /*
1309  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1310  */
1311 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1312                                 struct ext4_new_flex_group_data *flex_gd)
1313 {
1314         struct ext4_new_group_data      *group_data = flex_gd->groups;
1315         struct ext4_group_desc          *gdp;
1316         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1317         struct buffer_head              *gdb_bh;
1318         ext4_group_t                    group;
1319         __u16                           *bg_flags = flex_gd->bg_flags;
1320         int                             i, gdb_off, gdb_num, err = 0;
1321
1322
1323         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1324                 group = group_data->group;
1325
1326                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1327                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1328
1329                 /*
1330                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1331                  */
1332                 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1333                 /* Update group descriptor block for new group */
1334                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1335                                                  gdb_off * EXT4_DESC_SIZE(sb));
1336
1337                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1338                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1339                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1340                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1341                 if (err) {
1342                         ext4_std_error(sb, err);
1343                         break;
1344                 }
1345
1346                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1347                 ext4_free_group_clusters_set(sb, gdp,
1348                                              group_data->free_clusters_count);
1349                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1350                 if (ext4_has_group_desc_csum(sb))
1351                         ext4_itable_unused_set(sb, gdp,
1352                                                EXT4_INODES_PER_GROUP(sb));
1353                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1354                 ext4_group_desc_csum_set(sb, group, gdp);
1355
1356                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1357                 if (unlikely(err)) {
1358                         ext4_std_error(sb, err);
1359                         break;
1360                 }
1361
1362                 /*
1363                  * We can allocate memory for mb_alloc based on the new group
1364                  * descriptor
1365                  */
1366                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1367                 if (err)
1368                         break;
1369         }
1370         return err;
1371 }
1372
1373 /*
1374  * ext4_update_super() updates the super block so that the newly added
1375  * groups can be seen by the filesystem.
1376  *
1377  * @sb: super block
1378  * @flex_gd: new added groups
1379  */
1380 static void ext4_update_super(struct super_block *sb,
1381                              struct ext4_new_flex_group_data *flex_gd)
1382 {
1383         ext4_fsblk_t blocks_count = 0;
1384         ext4_fsblk_t free_blocks = 0;
1385         ext4_fsblk_t reserved_blocks = 0;
1386         struct ext4_new_group_data *group_data = flex_gd->groups;
1387         struct ext4_sb_info *sbi = EXT4_SB(sb);
1388         struct ext4_super_block *es = sbi->s_es;
1389         int i;
1390
1391         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1392         /*
1393          * Make the new blocks and inodes valid next.  We do this before
1394          * increasing the group count so that once the group is enabled,
1395          * all of its blocks and inodes are already valid.
1396          *
1397          * We always allocate group-by-group, then block-by-block or
1398          * inode-by-inode within a group, so enabling these
1399          * blocks/inodes before the group is live won't actually let us
1400          * allocate the new space yet.
1401          */
1402         for (i = 0; i < flex_gd->count; i++) {
1403                 blocks_count += group_data[i].blocks_count;
1404                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1405         }
1406
1407         reserved_blocks = ext4_r_blocks_count(es) * 100;
1408         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1409         reserved_blocks *= blocks_count;
1410         do_div(reserved_blocks, 100);
1411
1412         lock_buffer(sbi->s_sbh);
1413         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1414         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1415         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1416                      flex_gd->count);
1417         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1418                      flex_gd->count);
1419
1420         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1421         /*
1422          * We need to protect s_groups_count against other CPUs seeing
1423          * inconsistent state in the superblock.
1424          *
1425          * The precise rules we use are:
1426          *
1427          * * Writers must perform a smp_wmb() after updating all
1428          *   dependent data and before modifying the groups count
1429          *
1430          * * Readers must perform an smp_rmb() after reading the groups
1431          *   count and before reading any dependent data.
1432          *
1433          * NB. These rules can be relaxed when checking the group count
1434          * while freeing data, as we can only allocate from a block
1435          * group after serialising against the group count, and we can
1436          * only then free after serialising in turn against that
1437          * allocation.
1438          */
1439         smp_wmb();
1440
1441         /* Update the global fs size fields */
1442         sbi->s_groups_count += flex_gd->count;
1443         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1444                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1445
1446         /* Update the reserved block counts only once the new group is
1447          * active. */
1448         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1449                                 reserved_blocks);
1450         ext4_superblock_csum_set(sb);
1451         unlock_buffer(sbi->s_sbh);
1452
1453         /* Update the free space counts */
1454         percpu_counter_add(&sbi->s_freeclusters_counter,
1455                            EXT4_NUM_B2C(sbi, free_blocks));
1456         percpu_counter_add(&sbi->s_freeinodes_counter,
1457                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1458
1459         ext4_debug("free blocks count %llu",
1460                    percpu_counter_read(&sbi->s_freeclusters_counter));
1461         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1462                 ext4_group_t flex_group;
1463                 struct flex_groups *fg;
1464
1465                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1466                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1467                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1468                              &fg->free_clusters);
1469                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1470                            &fg->free_inodes);
1471         }
1472
1473         /*
1474          * Update the fs overhead information
1475          */
1476         ext4_calculate_overhead(sb);
1477
1478         if (test_opt(sb, DEBUG))
1479                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1480                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1481                        blocks_count, free_blocks, reserved_blocks);
1482 }
1483
1484 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1485  * _before_ we start modifying the filesystem, because we cannot abort the
1486  * transaction and not have it write the data to disk.
1487  */
1488 static int ext4_flex_group_add(struct super_block *sb,
1489                                struct inode *resize_inode,
1490                                struct ext4_new_flex_group_data *flex_gd)
1491 {
1492         struct ext4_sb_info *sbi = EXT4_SB(sb);
1493         struct ext4_super_block *es = sbi->s_es;
1494         ext4_fsblk_t o_blocks_count;
1495         ext4_grpblk_t last;
1496         ext4_group_t group;
1497         handle_t *handle;
1498         unsigned reserved_gdb;
1499         int err = 0, err2 = 0, credit;
1500
1501         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1502
1503         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1504         o_blocks_count = ext4_blocks_count(es);
1505         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1506         BUG_ON(last);
1507
1508         err = setup_new_flex_group_blocks(sb, flex_gd);
1509         if (err)
1510                 goto exit;
1511         /*
1512          * We will always be modifying at least the superblock and  GDT
1513          * blocks.  If we are adding a group past the last current GDT block,
1514          * we will also modify the inode and the dindirect block.  If we
1515          * are adding a group with superblock/GDT backups  we will also
1516          * modify each of the reserved GDT dindirect blocks.
1517          */
1518         credit = 3;     /* sb, resize inode, resize inode dindirect */
1519         /* GDT blocks */
1520         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1521         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1522         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1523         if (IS_ERR(handle)) {
1524                 err = PTR_ERR(handle);
1525                 goto exit;
1526         }
1527
1528         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1529         err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1530                                             EXT4_JTR_NONE);
1531         if (err)
1532                 goto exit_journal;
1533
1534         group = flex_gd->groups[0].group;
1535         BUG_ON(group != sbi->s_groups_count);
1536         err = ext4_add_new_descs(handle, sb, group,
1537                                 resize_inode, flex_gd->count);
1538         if (err)
1539                 goto exit_journal;
1540
1541         err = ext4_setup_new_descs(handle, sb, flex_gd);
1542         if (err)
1543                 goto exit_journal;
1544
1545         ext4_update_super(sb, flex_gd);
1546
1547         err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1548
1549 exit_journal:
1550         err2 = ext4_journal_stop(handle);
1551         if (!err)
1552                 err = err2;
1553
1554         if (!err) {
1555                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1556                 int gdb_num_end = ((group + flex_gd->count - 1) /
1557                                    EXT4_DESC_PER_BLOCK(sb));
1558                 int meta_bg = ext4_has_feature_meta_bg(sb);
1559                 sector_t old_gdb = 0;
1560
1561                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1562                                sizeof(struct ext4_super_block), 0);
1563                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1564                         struct buffer_head *gdb_bh;
1565
1566                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1567                                                      gdb_num);
1568                         if (old_gdb == gdb_bh->b_blocknr)
1569                                 continue;
1570                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1571                                        gdb_bh->b_size, meta_bg);
1572                         old_gdb = gdb_bh->b_blocknr;
1573                 }
1574         }
1575 exit:
1576         return err;
1577 }
1578
1579 static int ext4_setup_next_flex_gd(struct super_block *sb,
1580                                     struct ext4_new_flex_group_data *flex_gd,
1581                                     ext4_fsblk_t n_blocks_count,
1582                                     unsigned long flexbg_size)
1583 {
1584         struct ext4_sb_info *sbi = EXT4_SB(sb);
1585         struct ext4_super_block *es = sbi->s_es;
1586         struct ext4_new_group_data *group_data = flex_gd->groups;
1587         ext4_fsblk_t o_blocks_count;
1588         ext4_group_t n_group;
1589         ext4_group_t group;
1590         ext4_group_t last_group;
1591         ext4_grpblk_t last;
1592         ext4_grpblk_t clusters_per_group;
1593         unsigned long i;
1594
1595         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1596
1597         o_blocks_count = ext4_blocks_count(es);
1598
1599         if (o_blocks_count == n_blocks_count)
1600                 return 0;
1601
1602         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1603         BUG_ON(last);
1604         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1605
1606         last_group = group | (flexbg_size - 1);
1607         if (last_group > n_group)
1608                 last_group = n_group;
1609
1610         flex_gd->count = last_group - group + 1;
1611
1612         for (i = 0; i < flex_gd->count; i++) {
1613                 int overhead;
1614
1615                 group_data[i].group = group + i;
1616                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1617                 overhead = ext4_group_overhead_blocks(sb, group + i);
1618                 group_data[i].mdata_blocks = overhead;
1619                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1620                 if (ext4_has_group_desc_csum(sb)) {
1621                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1622                                                EXT4_BG_INODE_UNINIT;
1623                         if (!test_opt(sb, INIT_INODE_TABLE))
1624                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1625                 } else
1626                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1627         }
1628
1629         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1630                 /* We need to initialize block bitmap of last group. */
1631                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1632
1633         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1634                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1635                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1636                                                        last - 1;
1637         }
1638
1639         return 1;
1640 }
1641
1642 /* Add group descriptor data to an existing or new group descriptor block.
1643  * Ensure we handle all possible error conditions _before_ we start modifying
1644  * the filesystem, because we cannot abort the transaction and not have it
1645  * write the data to disk.
1646  *
1647  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1648  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1649  *
1650  * We only need to hold the superblock lock while we are actually adding
1651  * in the new group's counts to the superblock.  Prior to that we have
1652  * not really "added" the group at all.  We re-check that we are still
1653  * adding in the last group in case things have changed since verifying.
1654  */
1655 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1656 {
1657         struct ext4_new_flex_group_data flex_gd;
1658         struct ext4_sb_info *sbi = EXT4_SB(sb);
1659         struct ext4_super_block *es = sbi->s_es;
1660         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1661                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1662         struct inode *inode = NULL;
1663         int gdb_off;
1664         int err;
1665         __u16 bg_flags = 0;
1666
1667         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1668
1669         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1670                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1671                 return -EPERM;
1672         }
1673
1674         if (ext4_blocks_count(es) + input->blocks_count <
1675             ext4_blocks_count(es)) {
1676                 ext4_warning(sb, "blocks_count overflow");
1677                 return -EINVAL;
1678         }
1679
1680         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1681             le32_to_cpu(es->s_inodes_count)) {
1682                 ext4_warning(sb, "inodes_count overflow");
1683                 return -EINVAL;
1684         }
1685
1686         if (reserved_gdb || gdb_off == 0) {
1687                 if (!ext4_has_feature_resize_inode(sb) ||
1688                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1689                         ext4_warning(sb,
1690                                      "No reserved GDT blocks, can't resize");
1691                         return -EPERM;
1692                 }
1693                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1694                 if (IS_ERR(inode)) {
1695                         ext4_warning(sb, "Error opening resize inode");
1696                         return PTR_ERR(inode);
1697                 }
1698         }
1699
1700
1701         err = verify_group_input(sb, input);
1702         if (err)
1703                 goto out;
1704
1705         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1706         if (err)
1707                 goto out;
1708
1709         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1710         if (err)
1711                 goto out;
1712
1713         flex_gd.count = 1;
1714         flex_gd.groups = input;
1715         flex_gd.bg_flags = &bg_flags;
1716         err = ext4_flex_group_add(sb, inode, &flex_gd);
1717 out:
1718         iput(inode);
1719         return err;
1720 } /* ext4_group_add */
1721
1722 /*
1723  * extend a group without checking assuming that checking has been done.
1724  */
1725 static int ext4_group_extend_no_check(struct super_block *sb,
1726                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1727 {
1728         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1729         handle_t *handle;
1730         int err = 0, err2;
1731
1732         /* We will update the superblock, one block bitmap, and
1733          * one group descriptor via ext4_group_add_blocks().
1734          */
1735         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1736         if (IS_ERR(handle)) {
1737                 err = PTR_ERR(handle);
1738                 ext4_warning(sb, "error %d on journal start", err);
1739                 return err;
1740         }
1741
1742         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1743         err = ext4_journal_get_write_access(handle, sb, EXT4_SB(sb)->s_sbh,
1744                                             EXT4_JTR_NONE);
1745         if (err) {
1746                 ext4_warning(sb, "error %d on journal write access", err);
1747                 goto errout;
1748         }
1749
1750         lock_buffer(EXT4_SB(sb)->s_sbh);
1751         ext4_blocks_count_set(es, o_blocks_count + add);
1752         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1753         ext4_superblock_csum_set(sb);
1754         unlock_buffer(EXT4_SB(sb)->s_sbh);
1755         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1756                    o_blocks_count + add);
1757         /* We add the blocks to the bitmap and set the group need init bit */
1758         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1759         if (err)
1760                 goto errout;
1761         ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
1762         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1763                    o_blocks_count + add);
1764 errout:
1765         err2 = ext4_journal_stop(handle);
1766         if (err2 && !err)
1767                 err = err2;
1768
1769         if (!err) {
1770                 if (test_opt(sb, DEBUG))
1771                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1772                                "blocks\n", ext4_blocks_count(es));
1773                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1774                                (char *)es, sizeof(struct ext4_super_block), 0);
1775         }
1776         return err;
1777 }
1778
1779 /*
1780  * Extend the filesystem to the new number of blocks specified.  This entry
1781  * point is only used to extend the current filesystem to the end of the last
1782  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1783  * for emergencies (because it has no dependencies on reserved blocks).
1784  *
1785  * If we _really_ wanted, we could use default values to call ext4_group_add()
1786  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1787  * GDT blocks are reserved to grow to the desired size.
1788  */
1789 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1790                       ext4_fsblk_t n_blocks_count)
1791 {
1792         ext4_fsblk_t o_blocks_count;
1793         ext4_grpblk_t last;
1794         ext4_grpblk_t add;
1795         struct buffer_head *bh;
1796         int err;
1797         ext4_group_t group;
1798
1799         o_blocks_count = ext4_blocks_count(es);
1800
1801         if (test_opt(sb, DEBUG))
1802                 ext4_msg(sb, KERN_DEBUG,
1803                          "extending last group from %llu to %llu blocks",
1804                          o_blocks_count, n_blocks_count);
1805
1806         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1807                 return 0;
1808
1809         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1810                 ext4_msg(sb, KERN_ERR,
1811                          "filesystem too large to resize to %llu blocks safely",
1812                          n_blocks_count);
1813                 return -EINVAL;
1814         }
1815
1816         if (n_blocks_count < o_blocks_count) {
1817                 ext4_warning(sb, "can't shrink FS - resize aborted");
1818                 return -EINVAL;
1819         }
1820
1821         /* Handle the remaining blocks in the last group only. */
1822         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1823
1824         if (last == 0) {
1825                 ext4_warning(sb, "need to use ext2online to resize further");
1826                 return -EPERM;
1827         }
1828
1829         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1830
1831         if (o_blocks_count + add < o_blocks_count) {
1832                 ext4_warning(sb, "blocks_count overflow");
1833                 return -EINVAL;
1834         }
1835
1836         if (o_blocks_count + add > n_blocks_count)
1837                 add = n_blocks_count - o_blocks_count;
1838
1839         if (o_blocks_count + add < n_blocks_count)
1840                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1841                              o_blocks_count + add, add);
1842
1843         /* See if the device is actually as big as what was requested */
1844         bh = ext4_sb_bread(sb, o_blocks_count + add - 1, 0);
1845         if (IS_ERR(bh)) {
1846                 ext4_warning(sb, "can't read last block, resize aborted");
1847                 return -ENOSPC;
1848         }
1849         brelse(bh);
1850
1851         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1852         return err;
1853 } /* ext4_group_extend */
1854
1855
1856 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1857 {
1858         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1859 }
1860
1861 /*
1862  * Release the resize inode and drop the resize_inode feature if there
1863  * are no more reserved gdt blocks, and then convert the file system
1864  * to enable meta_bg
1865  */
1866 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1867 {
1868         handle_t *handle;
1869         struct ext4_sb_info *sbi = EXT4_SB(sb);
1870         struct ext4_super_block *es = sbi->s_es;
1871         struct ext4_inode_info *ei = EXT4_I(inode);
1872         ext4_fsblk_t nr;
1873         int i, ret, err = 0;
1874         int credits = 1;
1875
1876         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1877         if (inode) {
1878                 if (es->s_reserved_gdt_blocks) {
1879                         ext4_error(sb, "Unexpected non-zero "
1880                                    "s_reserved_gdt_blocks");
1881                         return -EPERM;
1882                 }
1883
1884                 /* Do a quick sanity check of the resize inode */
1885                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1886                                              (9 - sbi->s_cluster_bits)))
1887                         goto invalid_resize_inode;
1888                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1889                         if (i == EXT4_DIND_BLOCK) {
1890                                 if (ei->i_data[i])
1891                                         continue;
1892                                 else
1893                                         goto invalid_resize_inode;
1894                         }
1895                         if (ei->i_data[i])
1896                                 goto invalid_resize_inode;
1897                 }
1898                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1899         }
1900
1901         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1902         if (IS_ERR(handle))
1903                 return PTR_ERR(handle);
1904
1905         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1906         err = ext4_journal_get_write_access(handle, sb, sbi->s_sbh,
1907                                             EXT4_JTR_NONE);
1908         if (err)
1909                 goto errout;
1910
1911         lock_buffer(sbi->s_sbh);
1912         ext4_clear_feature_resize_inode(sb);
1913         ext4_set_feature_meta_bg(sb);
1914         sbi->s_es->s_first_meta_bg =
1915                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1916         ext4_superblock_csum_set(sb);
1917         unlock_buffer(sbi->s_sbh);
1918
1919         err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
1920         if (err) {
1921                 ext4_std_error(sb, err);
1922                 goto errout;
1923         }
1924
1925         if (inode) {
1926                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1927                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1928                                  EXT4_FREE_BLOCKS_METADATA |
1929                                  EXT4_FREE_BLOCKS_FORGET);
1930                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1931                 inode->i_blocks = 0;
1932
1933                 err = ext4_mark_inode_dirty(handle, inode);
1934                 if (err)
1935                         ext4_std_error(sb, err);
1936         }
1937
1938 errout:
1939         ret = ext4_journal_stop(handle);
1940         if (!err)
1941                 err = ret;
1942         return ret;
1943
1944 invalid_resize_inode:
1945         ext4_error(sb, "corrupted/inconsistent resize inode");
1946         return -EINVAL;
1947 }
1948
1949 /*
1950  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1951  *
1952  * @sb: super block of the fs to be resized
1953  * @n_blocks_count: the number of blocks resides in the resized fs
1954  */
1955 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1956 {
1957         struct ext4_new_flex_group_data *flex_gd = NULL;
1958         struct ext4_sb_info *sbi = EXT4_SB(sb);
1959         struct ext4_super_block *es = sbi->s_es;
1960         struct buffer_head *bh;
1961         struct inode *resize_inode = NULL;
1962         ext4_grpblk_t add, offset;
1963         unsigned long n_desc_blocks;
1964         unsigned long o_desc_blocks;
1965         ext4_group_t o_group;
1966         ext4_group_t n_group;
1967         ext4_fsblk_t o_blocks_count;
1968         ext4_fsblk_t n_blocks_count_retry = 0;
1969         unsigned long last_update_time = 0;
1970         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1971         int meta_bg;
1972
1973         /* See if the device is actually as big as what was requested */
1974         bh = ext4_sb_bread(sb, n_blocks_count - 1, 0);
1975         if (IS_ERR(bh)) {
1976                 ext4_warning(sb, "can't read last block, resize aborted");
1977                 return -ENOSPC;
1978         }
1979         brelse(bh);
1980
1981 retry:
1982         o_blocks_count = ext4_blocks_count(es);
1983
1984         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1985                  "to %llu blocks", o_blocks_count, n_blocks_count);
1986
1987         if (n_blocks_count < o_blocks_count) {
1988                 /* On-line shrinking not supported */
1989                 ext4_warning(sb, "can't shrink FS - resize aborted");
1990                 return -EINVAL;
1991         }
1992
1993         if (n_blocks_count == o_blocks_count)
1994                 /* Nothing need to do */
1995                 return 0;
1996
1997         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1998         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1999                 ext4_warning(sb, "resize would cause inodes_count overflow");
2000                 return -EINVAL;
2001         }
2002         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
2003
2004         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
2005         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
2006
2007         meta_bg = ext4_has_feature_meta_bg(sb);
2008
2009         if (ext4_has_feature_resize_inode(sb)) {
2010                 if (meta_bg) {
2011                         ext4_error(sb, "resize_inode and meta_bg enabled "
2012                                    "simultaneously");
2013                         return -EINVAL;
2014                 }
2015                 if (n_desc_blocks > o_desc_blocks +
2016                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
2017                         n_blocks_count_retry = n_blocks_count;
2018                         n_desc_blocks = o_desc_blocks +
2019                                 le16_to_cpu(es->s_reserved_gdt_blocks);
2020                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2021                         n_blocks_count = (ext4_fsblk_t)n_group *
2022                                 EXT4_BLOCKS_PER_GROUP(sb) +
2023                                 le32_to_cpu(es->s_first_data_block);
2024                         n_group--; /* set to last group number */
2025                 }
2026
2027                 if (!resize_inode)
2028                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2029                                                  EXT4_IGET_SPECIAL);
2030                 if (IS_ERR(resize_inode)) {
2031                         ext4_warning(sb, "Error opening resize inode");
2032                         return PTR_ERR(resize_inode);
2033                 }
2034         }
2035
2036         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2037                 err = ext4_convert_meta_bg(sb, resize_inode);
2038                 if (err)
2039                         goto out;
2040                 if (resize_inode) {
2041                         iput(resize_inode);
2042                         resize_inode = NULL;
2043                 }
2044                 if (n_blocks_count_retry) {
2045                         n_blocks_count = n_blocks_count_retry;
2046                         n_blocks_count_retry = 0;
2047                         goto retry;
2048                 }
2049         }
2050
2051         /*
2052          * Make sure the last group has enough space so that it's
2053          * guaranteed to have enough space for all metadata blocks
2054          * that it might need to hold.  (We might not need to store
2055          * the inode table blocks in the last block group, but there
2056          * will be cases where this might be needed.)
2057          */
2058         if ((ext4_group_first_block_no(sb, n_group) +
2059              ext4_group_overhead_blocks(sb, n_group) + 2 +
2060              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2061                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2062                 n_group--;
2063                 n_blocks_count_retry = 0;
2064                 if (resize_inode) {
2065                         iput(resize_inode);
2066                         resize_inode = NULL;
2067                 }
2068                 goto retry;
2069         }
2070
2071         /* extend the last group */
2072         if (n_group == o_group)
2073                 add = n_blocks_count - o_blocks_count;
2074         else
2075                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2076         if (add > 0) {
2077                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2078                 if (err)
2079                         goto out;
2080         }
2081
2082         if (ext4_blocks_count(es) == n_blocks_count)
2083                 goto out;
2084
2085         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2086         if (err)
2087                 goto out;
2088
2089         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2090         if (err)
2091                 goto out;
2092
2093         flex_gd = alloc_flex_gd(flexbg_size);
2094         if (flex_gd == NULL) {
2095                 err = -ENOMEM;
2096                 goto out;
2097         }
2098
2099         /* Add flex groups. Note that a regular group is a
2100          * flex group with 1 group.
2101          */
2102         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2103                                               flexbg_size)) {
2104                 if (time_is_before_jiffies(last_update_time + HZ * 10)) {
2105                         if (last_update_time)
2106                                 ext4_msg(sb, KERN_INFO,
2107                                          "resized to %llu blocks",
2108                                          ext4_blocks_count(es));
2109                         last_update_time = jiffies;
2110                 }
2111                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2112                         break;
2113                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2114                 if (unlikely(err))
2115                         break;
2116         }
2117
2118         if (!err && n_blocks_count_retry) {
2119                 n_blocks_count = n_blocks_count_retry;
2120                 n_blocks_count_retry = 0;
2121                 free_flex_gd(flex_gd);
2122                 flex_gd = NULL;
2123                 if (resize_inode) {
2124                         iput(resize_inode);
2125                         resize_inode = NULL;
2126                 }
2127                 goto retry;
2128         }
2129
2130 out:
2131         if (flex_gd)
2132                 free_flex_gd(flex_gd);
2133         if (resize_inode != NULL)
2134                 iput(resize_inode);
2135         if (err)
2136                 ext4_warning(sb, "error (%d) occurred during "
2137                              "file system resize", err);
2138         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2139                  ext4_blocks_count(es));
2140         return err;
2141 }