fs/ext4/resize.c

   1 /*
   2  *  linux/fs/ext4/resize.c
   3  *
   4  * Support for resizing an ext4 filesystem while it is mounted.
   5  *
   6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   7  *
   8  * This could probably be made into a module, because it is not often in use.
   9  */
  10
  11
  12 #define EXT4FS_DEBUG
  13
  14 #include <linux/errno.h>
  15 #include <linux/slab.h>
  16
  17 #include "ext4_jbd2.h"
  18
  19 int ext4_resize_begin(struct super_block *sb)
  20 {
  21         int ret = 0;
  22
  23         if (!capable(CAP_SYS_RESOURCE))
  24                 return -EPERM;
  25
  26         /*
  27          * We are not allowed to do online-resizing on a filesystem mounted
  28          * with error, because it can destroy the filesystem easily.
  29          */
  30         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  31                 ext4_warning(sb, "There are errors in the filesystem, "
  32                              "so online resizing is not allowed\n");
  33                 return -EPERM;
  34         }
  35
  36         if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
  37                 ret = -EBUSY;
  38
  39         return ret;
  40 }
  41
  42 void ext4_resize_end(struct super_block *sb)
  43 {
  44         clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
  45         smp_mb__after_atomic();
  46 }
  47
  48 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  49                                              ext4_group_t group) {
  50         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  51                EXT4_DESC_PER_BLOCK_BITS(sb);
  52 }
  53
  54 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
  55                                              ext4_group_t group) {
  56         group = ext4_meta_bg_first_group(sb, group);
  57         return ext4_group_first_block_no(sb, group);
  58 }
  59
  60 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
  61                                                 ext4_group_t group) {
  62         ext4_grpblk_t overhead;
  63         overhead = ext4_bg_num_gdb(sb, group);
  64         if (ext4_bg_has_super(sb, group))
  65                 overhead += 1 +
  66                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
  67         return overhead;
  68 }
  69
  70 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  71 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  72
  73 static int verify_group_input(struct super_block *sb,
  74                               struct ext4_new_group_data *input)
  75 {
  76         struct ext4_sb_info *sbi = EXT4_SB(sb);
  77         struct ext4_super_block *es = sbi->s_es;
  78         ext4_fsblk_t start = ext4_blocks_count(es);
  79         ext4_fsblk_t end = start + input->blocks_count;
  80         ext4_group_t group = input->group;
  81         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  82         unsigned overhead;
  83         ext4_fsblk_t metaend;
  84         struct buffer_head *bh = NULL;
  85         ext4_grpblk_t free_blocks_count, offset;
  86         int err = -EINVAL;
  87
  88         if (group != sbi->s_groups_count) {
  89                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
  90                              input->group, sbi->s_groups_count);
  91                 return -EINVAL;
  92         }
  93
  94         overhead = ext4_group_overhead_blocks(sb, group);
  95         metaend = start + overhead;
  96         input->free_blocks_count = free_blocks_count =
  97                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  98
  99         if (test_opt(sb, DEBUG))
 100                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 101                        "(%d free, %u reserved)\n",
 102                        ext4_bg_has_super(sb, input->group) ? "normal" :
 103                        "no-super", input->group, input->blocks_count,
 104                        free_blocks_count, input->reserved_blocks);
 105
 106         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 107         if (offset != 0)
 108                         ext4_warning(sb, "Last group not full");
 109         else if (input->reserved_blocks > input->blocks_count / 5)
 110                 ext4_warning(sb, "Reserved blocks too high (%u)",
 111                              input->reserved_blocks);
 112         else if (free_blocks_count < 0)
 113                 ext4_warning(sb, "Bad blocks count %u",
 114                              input->blocks_count);
 115         else if (!(bh = sb_bread(sb, end - 1)))
 116                 ext4_warning(sb, "Cannot read last block (%llu)",
 117                              end - 1);
 118         else if (outside(input->block_bitmap, start, end))
 119                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 120                              (unsigned long long)input->block_bitmap);
 121         else if (outside(input->inode_bitmap, start, end))
 122                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 123                              (unsigned long long)input->inode_bitmap);
 124         else if (outside(input->inode_table, start, end) ||
 125                  outside(itend - 1, start, end))
 126                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 127                              (unsigned long long)input->inode_table, itend - 1);
 128         else if (input->inode_bitmap == input->block_bitmap)
 129                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 130                              (unsigned long long)input->block_bitmap);
 131         else if (inside(input->block_bitmap, input->inode_table, itend))
 132                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 133                              "(%llu-%llu)",
 134                              (unsigned long long)input->block_bitmap,
 135                              (unsigned long long)input->inode_table, itend - 1);
 136         else if (inside(input->inode_bitmap, input->inode_table, itend))
 137                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 138                              "(%llu-%llu)",
 139                              (unsigned long long)input->inode_bitmap,
 140                              (unsigned long long)input->inode_table, itend - 1);
 141         else if (inside(input->block_bitmap, start, metaend))
 142                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 143                              (unsigned long long)input->block_bitmap,
 144                              start, metaend - 1);
 145         else if (inside(input->inode_bitmap, start, metaend))
 146                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 147                              (unsigned long long)input->inode_bitmap,
 148                              start, metaend - 1);
 149         else if (inside(input->inode_table, start, metaend) ||
 150                  inside(itend - 1, start, metaend))
 151                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 152                              "(%llu-%llu)",
 153                              (unsigned long long)input->inode_table,
 154                              itend - 1, start, metaend - 1);
 155         else
 156                 err = 0;
 157         brelse(bh);
 158
 159         return err;
 160 }
 161
 162 /*
 163  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 164  * group each time.
 165  */
 166 struct ext4_new_flex_group_data {
 167         struct ext4_new_group_data *groups;     /* new_group_data for groups
 168                                                    in the flex group */
 169         __u16 *bg_flags;                        /* block group flags of groups
 170                                                    in @groups */
 171         ext4_group_t count;                     /* number of groups in @groups
 172                                                  */
 173 };
 174
 175 /*
 176  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 177  * @flexbg_size.
 178  *
 179  * Returns NULL on failure otherwise address of the allocated structure.
 180  */
 181 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 182 {
 183         struct ext4_new_flex_group_data *flex_gd;
 184
 185         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 186         if (flex_gd == NULL)
 187                 goto out3;
 188
 189         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
 190                 goto out2;
 191         flex_gd->count = flexbg_size;
 192
 193         flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
 194                                   flexbg_size, GFP_NOFS);
 195         if (flex_gd->groups == NULL)
 196                 goto out2;
 197
 198         flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
 199         if (flex_gd->bg_flags == NULL)
 200                 goto out1;
 201
 202         return flex_gd;
 203
 204 out1:
 205         kfree(flex_gd->groups);
 206 out2:
 207         kfree(flex_gd);
 208 out3:
 209         return NULL;
 210 }
 211
 212 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 213 {
 214         kfree(flex_gd->bg_flags);
 215         kfree(flex_gd->groups);
 216         kfree(flex_gd);
 217 }
 218
 219 /*
 220  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 221  * and inode tables for a flex group.
 222  *
 223  * This function is used by 64bit-resize.  Note that this function allocates
 224  * group tables from the 1st group of groups contained by @flexgd, which may
 225  * be a partial of a flex group.
 226  *
 227  * @sb: super block of fs to which the groups belongs
 228  *
 229  * Returns 0 on a successful allocation of the metadata blocks in the
 230  * block group.
 231  */
 232 static int ext4_alloc_group_tables(struct super_block *sb,
 233                                 struct ext4_new_flex_group_data *flex_gd,
 234                                 int flexbg_size)
 235 {
 236         struct ext4_new_group_data *group_data = flex_gd->groups;
 237         ext4_fsblk_t start_blk;
 238         ext4_fsblk_t last_blk;
 239         ext4_group_t src_group;
 240         ext4_group_t bb_index = 0;
 241         ext4_group_t ib_index = 0;
 242         ext4_group_t it_index = 0;
 243         ext4_group_t group;
 244         ext4_group_t last_group;
 245         unsigned overhead;
 246         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 247
 248         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 249
 250         src_group = group_data[0].group;
 251         last_group  = src_group + flex_gd->count - 1;
 252
 253         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 254                (last_group & ~(flexbg_size - 1))));
 255 next_group:
 256         group = group_data[0].group;
 257         if (src_group >= group_data[0].group + flex_gd->count)
 258                 return -ENOSPC;
 259         start_blk = ext4_group_first_block_no(sb, src_group);
 260         last_blk = start_blk + group_data[src_group - group].blocks_count;
 261
 262         overhead = ext4_group_overhead_blocks(sb, src_group);
 263
 264         start_blk += overhead;
 265
 266         /* We collect contiguous blocks as much as possible. */
 267         src_group++;
 268         for (; src_group <= last_group; src_group++) {
 269                 overhead = ext4_group_overhead_blocks(sb, src_group);
 270                 if (overhead == 0)
 271                         last_blk += group_data[src_group - group].blocks_count;
 272                 else
 273                         break;
 274         }
 275
 276         /* Allocate block bitmaps */
 277         for (; bb_index < flex_gd->count; bb_index++) {
 278                 if (start_blk >= last_blk)
 279                         goto next_group;
 280                 group_data[bb_index].block_bitmap = start_blk++;
 281                 group = ext4_get_group_number(sb, start_blk - 1);
 282                 group -= group_data[0].group;
 283                 group_data[group].free_blocks_count--;
 284                 flex_gd->bg_flags[group] &= uninit_mask;
 285         }
 286
 287         /* Allocate inode bitmaps */
 288         for (; ib_index < flex_gd->count; ib_index++) {
 289                 if (start_blk >= last_blk)
 290                         goto next_group;
 291                 group_data[ib_index].inode_bitmap = start_blk++;
 292                 group = ext4_get_group_number(sb, start_blk - 1);
 293                 group -= group_data[0].group;
 294                 group_data[group].free_blocks_count--;
 295                 flex_gd->bg_flags[group] &= uninit_mask;
 296         }
 297
 298         /* Allocate inode tables */
 299         for (; it_index < flex_gd->count; it_index++) {
 300                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 301                 ext4_fsblk_t next_group_start;
 302
 303                 if (start_blk + itb > last_blk)
 304                         goto next_group;
 305                 group_data[it_index].inode_table = start_blk;
 306                 group = ext4_get_group_number(sb, start_blk);
 307                 next_group_start = ext4_group_first_block_no(sb, group + 1);
 308                 group -= group_data[0].group;
 309
 310                 if (start_blk + itb > next_group_start) {
 311                         flex_gd->bg_flags[group + 1] &= uninit_mask;
 312                         overhead = start_blk + itb - next_group_start;
 313                         group_data[group + 1].free_blocks_count -= overhead;
 314                         itb -= overhead;
 315                 }
 316
 317                 group_data[group].free_blocks_count -= itb;
 318                 flex_gd->bg_flags[group] &= uninit_mask;
 319                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 320         }
 321
 322         if (test_opt(sb, DEBUG)) {
 323                 int i;
 324                 group = group_data[0].group;
 325
 326                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 327                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 328                        flexbg_size);
 329
 330                 for (i = 0; i < flex_gd->count; i++) {
 331                         printk(KERN_DEBUG "adding %s group %u: %u "
 332                                "blocks (%d free)\n",
 333                                ext4_bg_has_super(sb, group + i) ? "normal" :
 334                                "no-super", group + i,
 335                                group_data[i].blocks_count,
 336                                group_data[i].free_blocks_count);
 337                 }
 338         }
 339         return 0;
 340 }
 341
 342 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 343                                   ext4_fsblk_t blk)
 344 {
 345         struct buffer_head *bh;
 346         int err;
 347
 348         bh = sb_getblk(sb, blk);
 349         if (unlikely(!bh))
 350                 return ERR_PTR(-ENOMEM);
 351         BUFFER_TRACE(bh, "get_write_access");
 352         if ((err = ext4_journal_get_write_access(handle, bh))) {
 353                 brelse(bh);
 354                 bh = ERR_PTR(err);
 355         } else {
 356                 memset(bh->b_data, 0, sb->s_blocksize);
 357                 set_buffer_uptodate(bh);
 358         }
 359
 360         return bh;
 361 }
 362
 363 /*
 364  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 365  * If that fails, restart the transaction & regain write access for the
 366  * buffer head which is used for block_bitmap modifications.
 367  */
 368 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 369 {
 370         int err;
 371
 372         if (ext4_handle_has_enough_credits(handle, thresh))
 373                 return 0;
 374
 375         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 376         if (err < 0)
 377                 return err;
 378         if (err) {
 379                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 380                 if (err)
 381                         return err;
 382         }
 383
 384         return 0;
 385 }
 386
 387 /*
 388  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 389  *
 390  * Helper function for ext4_setup_new_group_blocks() which set .
 391  *
 392  * @sb: super block
 393  * @handle: journal handle
 394  * @flex_gd: flex group data
 395  */
 396 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 397                         struct ext4_new_flex_group_data *flex_gd,
 398                         ext4_fsblk_t block, ext4_group_t count)
 399 {
 400         ext4_group_t count2;
 401
 402         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 403         for (count2 = count; count > 0; count -= count2, block += count2) {
 404                 ext4_fsblk_t start;
 405                 struct buffer_head *bh;
 406                 ext4_group_t group;
 407                 int err;
 408
 409                 group = ext4_get_group_number(sb, block);
 410                 start = ext4_group_first_block_no(sb, group);
 411                 group -= flex_gd->groups[0].group;
 412
 413                 count2 = EXT4_BLOCKS_PER_GROUP(sb) - (block - start);
 414                 if (count2 > count)
 415                         count2 = count;
 416
 417                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 418                         BUG_ON(flex_gd->count > 1);
 419                         continue;
 420                 }
 421
 422                 err = extend_or_restart_transaction(handle, 1);
 423                 if (err)
 424                         return err;
 425
 426                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 427                 if (unlikely(!bh))
 428                         return -ENOMEM;
 429
 430                 BUFFER_TRACE(bh, "get_write_access");
 431                 err = ext4_journal_get_write_access(handle, bh);
 432                 if (err)
 433                         return err;
 434                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 435                            block - start, count2);
 436                 ext4_set_bits(bh->b_data, block - start, count2);
 437
 438                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 439                 if (unlikely(err))
 440                         return err;
 441                 brelse(bh);
 442         }
 443
 444         return 0;
 445 }
 446
 447 /*
 448  * Set up the block and inode bitmaps, and the inode table for the new groups.
 449  * This doesn't need to be part of the main transaction, since we are only
 450  * changing blocks outside the actual filesystem.  We still do journaling to
 451  * ensure the recovery is correct in case of a failure just after resize.
 452  * If any part of this fails, we simply abort the resize.
 453  *
 454  * setup_new_flex_group_blocks handles a flex group as follow:
 455  *  1. copy super block and GDT, and initialize group tables if necessary.
 456  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 457  *     super block and GDT.
 458  *  2. allocate group tables in block bitmaps, that is, set bits in block
 459  *     bitmap for blocks taken by group tables.
 460  */
 461 static int setup_new_flex_group_blocks(struct super_block *sb,
 462                                 struct ext4_new_flex_group_data *flex_gd)
 463 {
 464         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 465         ext4_fsblk_t start;
 466         ext4_fsblk_t block;
 467         struct ext4_sb_info *sbi = EXT4_SB(sb);
 468         struct ext4_super_block *es = sbi->s_es;
 469         struct ext4_new_group_data *group_data = flex_gd->groups;
 470         __u16 *bg_flags = flex_gd->bg_flags;
 471         handle_t *handle;
 472         ext4_group_t group, count;
 473         struct buffer_head *bh = NULL;
 474         int reserved_gdb, i, j, err = 0, err2;
 475         int meta_bg;
 476
 477         BUG_ON(!flex_gd->count || !group_data ||
 478                group_data[0].group != sbi->s_groups_count);
 479
 480         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 481         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
 482
 483         /* This transaction may be extended/restarted along the way */
 484         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 485         if (IS_ERR(handle))
 486                 return PTR_ERR(handle);
 487
 488         group = group_data[0].group;
 489         for (i = 0; i < flex_gd->count; i++, group++) {
 490                 unsigned long gdblocks;
 491                 ext4_grpblk_t overhead;
 492
 493                 gdblocks = ext4_bg_num_gdb(sb, group);
 494                 start = ext4_group_first_block_no(sb, group);
 495
 496                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 497                         goto handle_itb;
 498
 499                 if (meta_bg == 1) {
 500                         ext4_group_t first_group;
 501                         first_group = ext4_meta_bg_first_group(sb, group);
 502                         if (first_group != group + 1 &&
 503                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 504                                 goto handle_itb;
 505                 }
 506
 507                 block = start + ext4_bg_has_super(sb, group);
 508                 /* Copy all of the GDT blocks into the backup in this group */
 509                 for (j = 0; j < gdblocks; j++, block++) {
 510                         struct buffer_head *gdb;
 511
 512                         ext4_debug("update backup group %#04llx\n", block);
 513                         err = extend_or_restart_transaction(handle, 1);
 514                         if (err)
 515                                 goto out;
 516
 517                         gdb = sb_getblk(sb, block);
 518                         if (unlikely(!gdb)) {
 519                                 err = -ENOMEM;
 520                                 goto out;
 521                         }
 522
 523                         BUFFER_TRACE(gdb, "get_write_access");
 524                         err = ext4_journal_get_write_access(handle, gdb);
 525                         if (err) {
 526                                 brelse(gdb);
 527                                 goto out;
 528                         }
 529                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 530                                gdb->b_size);
 531                         set_buffer_uptodate(gdb);
 532
 533                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 534                         if (unlikely(err)) {
 535                                 brelse(gdb);
 536                                 goto out;
 537                         }
 538                         brelse(gdb);
 539                 }
 540
 541                 /* Zero out all of the reserved backup group descriptor
 542                  * table blocks
 543                  */
 544                 if (ext4_bg_has_super(sb, group)) {
 545                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 546                                         reserved_gdb, GFP_NOFS);
 547                         if (err)
 548                                 goto out;
 549                 }
 550
 551 handle_itb:
 552                 /* Initialize group tables of the grop @group */
 553                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 554                         goto handle_bb;
 555
 556                 /* Zero out all of the inode table blocks */
 557                 block = group_data[i].inode_table;
 558                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 559                            block, sbi->s_itb_per_group);
 560                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 561                                        GFP_NOFS);
 562                 if (err)
 563                         goto out;
 564
 565 handle_bb:
 566                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 567                         goto handle_ib;
 568
 569                 /* Initialize block bitmap of the @group */
 570                 block = group_data[i].block_bitmap;
 571                 err = extend_or_restart_transaction(handle, 1);
 572                 if (err)
 573                         goto out;
 574
 575                 bh = bclean(handle, sb, block);
 576                 if (IS_ERR(bh)) {
 577                         err = PTR_ERR(bh);
 578                         goto out;
 579                 }
 580                 overhead = ext4_group_overhead_blocks(sb, group);
 581                 if (overhead != 0) {
 582                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 583                                    start);
 584                         ext4_set_bits(bh->b_data, 0, overhead);
 585                 }
 586                 ext4_mark_bitmap_end(group_data[i].blocks_count,
 587                                      sb->s_blocksize * 8, bh->b_data);
 588                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 589                 if (err)
 590                         goto out;
 591                 brelse(bh);
 592
 593 handle_ib:
 594                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 595                         continue;
 596
 597                 /* Initialize inode bitmap of the @group */
 598                 block = group_data[i].inode_bitmap;
 599                 err = extend_or_restart_transaction(handle, 1);
 600                 if (err)
 601                         goto out;
 602                 /* Mark unused entries in inode bitmap used */
 603                 bh = bclean(handle, sb, block);
 604                 if (IS_ERR(bh)) {
 605                         err = PTR_ERR(bh);
 606                         goto out;
 607                 }
 608
 609                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 610                                      sb->s_blocksize * 8, bh->b_data);
 611                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 612                 if (err)
 613                         goto out;
 614                 brelse(bh);
 615         }
 616         bh = NULL;
 617
 618         /* Mark group tables in block bitmap */
 619         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 620                 count = group_table_count[j];
 621                 start = (&group_data[0].block_bitmap)[j];
 622                 block = start;
 623                 for (i = 1; i < flex_gd->count; i++) {
 624                         block += group_table_count[j];
 625                         if (block == (&group_data[i].block_bitmap)[j]) {
 626                                 count += group_table_count[j];
 627                                 continue;
 628                         }
 629                         err = set_flexbg_block_bitmap(sb, handle,
 630                                                 flex_gd, start, count);
 631                         if (err)
 632                                 goto out;
 633                         count = group_table_count[j];
 634                         start = (&group_data[i].block_bitmap)[j];
 635                         block = start;
 636                 }
 637
 638                 if (count) {
 639                         err = set_flexbg_block_bitmap(sb, handle,
 640                                                 flex_gd, start, count);
 641                         if (err)
 642                                 goto out;
 643                 }
 644         }
 645
 646 out:
 647         brelse(bh);
 648         err2 = ext4_journal_stop(handle);
 649         if (err2 && !err)
 650                 err = err2;
 651
 652         return err;
 653 }
 654
 655 /*
 656  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 657  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 658  * calling this for the first time.  In a sparse filesystem it will be the
 659  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 660  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 661  */
 662 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 663                                   unsigned *five, unsigned *seven)
 664 {
 665         unsigned *min = three;
 666         int mult = 3;
 667         unsigned ret;
 668
 669         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 670                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 671                 ret = *min;
 672                 *min += 1;
 673                 return ret;
 674         }
 675
 676         if (*five < *min) {
 677                 min = five;
 678                 mult = 5;
 679         }
 680         if (*seven < *min) {
 681                 min = seven;
 682                 mult = 7;
 683         }
 684
 685         ret = *min;
 686         *min *= mult;
 687
 688         return ret;
 689 }
 690
 691 /*
 692  * Check that all of the backup GDT blocks are held in the primary GDT block.
 693  * It is assumed that they are stored in group order.  Returns the number of
 694  * groups in current filesystem that have BACKUPS, or -ve error code.
 695  */
 696 static int verify_reserved_gdb(struct super_block *sb,
 697                                ext4_group_t end,
 698                                struct buffer_head *primary)
 699 {
 700         const ext4_fsblk_t blk = primary->b_blocknr;
 701         unsigned three = 1;
 702         unsigned five = 5;
 703         unsigned seven = 7;
 704         unsigned grp;
 705         __le32 *p = (__le32 *)primary->b_data;
 706         int gdbackups = 0;
 707
 708         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 709                 if (le32_to_cpu(*p++) !=
 710                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 711                         ext4_warning(sb, "reserved GDT %llu"
 712                                      " missing grp %d (%llu)",
 713                                      blk, grp,
 714                                      grp *
 715                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 716                                      blk);
 717                         return -EINVAL;
 718                 }
 719                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 720                         return -EFBIG;
 721         }
 722
 723         return gdbackups;
 724 }
 725
 726 /*
 727  * Called when we need to bring a reserved group descriptor table block into
 728  * use from the resize inode.  The primary copy of the new GDT block currently
 729  * is an indirect block (under the double indirect block in the resize inode).
 730  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 731  * block, in group order.  Even though we know all the block numbers we need,
 732  * we check to ensure that the resize inode has actually reserved these blocks.
 733  *
 734  * Don't need to update the block bitmaps because the blocks are still in use.
 735  *
 736  * We get all of the error cases out of the way, so that we are sure to not
 737  * fail once we start modifying the data on disk, because JBD has no rollback.
 738  */
 739 static int add_new_gdb(handle_t *handle, struct inode *inode,
 740                        ext4_group_t group)
 741 {
 742         struct super_block *sb = inode->i_sb;
 743         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 744         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 745         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 746         struct buffer_head **o_group_desc, **n_group_desc;
 747         struct buffer_head *dind;
 748         struct buffer_head *gdb_bh;
 749         int gdbackups;
 750         struct ext4_iloc iloc;
 751         __le32 *data;
 752         int err;
 753
 754         if (test_opt(sb, DEBUG))
 755                 printk(KERN_DEBUG
 756                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 757                        gdb_num);
 758
 759         /*
 760          * If we are not using the primary superblock/GDT copy don't resize,
 761          * because the user tools have no way of handling this.  Probably a
 762          * bad time to do it anyways.
 763          */
 764         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 765             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 766                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 767                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 768                 return -EPERM;
 769         }
 770
 771         gdb_bh = sb_bread(sb, gdblock);
 772         if (!gdb_bh)
 773                 return -EIO;
 774
 775         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 776         if (gdbackups < 0) {
 777                 err = gdbackups;
 778                 goto exit_bh;
 779         }
 780
 781         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 782         dind = sb_bread(sb, le32_to_cpu(*data));
 783         if (!dind) {
 784                 err = -EIO;
 785                 goto exit_bh;
 786         }
 787
 788         data = (__le32 *)dind->b_data;
 789         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 790                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 791                              group, gdblock);
 792                 err = -EINVAL;
 793                 goto exit_dind;
 794         }
 795
 796         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 797         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 798         if (unlikely(err))
 799                 goto exit_dind;
 800
 801         BUFFER_TRACE(gdb_bh, "get_write_access");
 802         err = ext4_journal_get_write_access(handle, gdb_bh);
 803         if (unlikely(err))
 804                 goto exit_dind;
 805
 806         BUFFER_TRACE(dind, "get_write_access");
 807         err = ext4_journal_get_write_access(handle, dind);
 808         if (unlikely(err))
 809                 ext4_std_error(sb, err);
 810
 811         /* ext4_reserve_inode_write() gets a reference on the iloc */
 812         err = ext4_reserve_inode_write(handle, inode, &iloc);
 813         if (unlikely(err))
 814                 goto exit_dind;
 815
 816         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 817                                      sizeof(struct buffer_head *),
 818                                      GFP_NOFS);
 819         if (!n_group_desc) {
 820                 err = -ENOMEM;
 821                 ext4_warning(sb, "not enough memory for %lu groups",
 822                              gdb_num + 1);
 823                 goto exit_inode;
 824         }
 825
 826         /*
 827          * Finally, we have all of the possible failures behind us...
 828          *
 829          * Remove new GDT block from inode double-indirect block and clear out
 830          * the new GDT block for use (which also "frees" the backup GDT blocks
 831          * from the reserved inode).  We don't need to change the bitmaps for
 832          * these blocks, because they are marked as in-use from being in the
 833          * reserved inode, and will become GDT blocks (primary and backup).
 834          */
 835         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 836         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 837         if (unlikely(err)) {
 838                 ext4_std_error(sb, err);
 839                 goto exit_inode;
 840         }
 841         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 842         ext4_mark_iloc_dirty(handle, inode, &iloc);
 843         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 844         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 845         if (unlikely(err)) {
 846                 ext4_std_error(sb, err);
 847                 goto exit_inode;
 848         }
 849         brelse(dind);
 850
 851         o_group_desc = EXT4_SB(sb)->s_group_desc;
 852         memcpy(n_group_desc, o_group_desc,
 853                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 854         n_group_desc[gdb_num] = gdb_bh;
 855         EXT4_SB(sb)->s_group_desc = n_group_desc;
 856         EXT4_SB(sb)->s_gdb_count++;
 857         ext4_kvfree(o_group_desc);
 858
 859         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 860         err = ext4_handle_dirty_super(handle, sb);
 861         if (err)
 862                 ext4_std_error(sb, err);
 863
 864         return err;
 865
 866 exit_inode:
 867         ext4_kvfree(n_group_desc);
 868         brelse(iloc.bh);
 869 exit_dind:
 870         brelse(dind);
 871 exit_bh:
 872         brelse(gdb_bh);
 873
 874         ext4_debug("leaving with error %d\n", err);
 875         return err;
 876 }
 877
 878 /*
 879  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 880  */
 881 static int add_new_gdb_meta_bg(struct super_block *sb,
 882                                handle_t *handle, ext4_group_t group) {
 883         ext4_fsblk_t gdblock;
 884         struct buffer_head *gdb_bh;
 885         struct buffer_head **o_group_desc, **n_group_desc;
 886         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 887         int err;
 888
 889         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 890                    ext4_bg_has_super(sb, group);
 891         gdb_bh = sb_bread(sb, gdblock);
 892         if (!gdb_bh)
 893                 return -EIO;
 894         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 895                                      sizeof(struct buffer_head *),
 896                                      GFP_NOFS);
 897         if (!n_group_desc) {
 898                 err = -ENOMEM;
 899                 ext4_warning(sb, "not enough memory for %lu groups",
 900                              gdb_num + 1);
 901                 return err;
 902         }
 903
 904         o_group_desc = EXT4_SB(sb)->s_group_desc;
 905         memcpy(n_group_desc, o_group_desc,
 906                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 907         n_group_desc[gdb_num] = gdb_bh;
 908         EXT4_SB(sb)->s_group_desc = n_group_desc;
 909         EXT4_SB(sb)->s_gdb_count++;
 910         ext4_kvfree(o_group_desc);
 911         BUFFER_TRACE(gdb_bh, "get_write_access");
 912         err = ext4_journal_get_write_access(handle, gdb_bh);
 913         if (unlikely(err))
 914                 brelse(gdb_bh);
 915         return err;
 916 }
 917
 918 /*
 919  * Called when we are adding a new group which has a backup copy of each of
 920  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 921  * We need to add these reserved backup GDT blocks to the resize inode, so
 922  * that they are kept for future resizing and not allocated to files.
 923  *
 924  * Each reserved backup GDT block will go into a different indirect block.
 925  * The indirect blocks are actually the primary reserved GDT blocks,
 926  * so we know in advance what their block numbers are.  We only get the
 927  * double-indirect block to verify it is pointing to the primary reserved
 928  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 929  * backup GDT blocks are stored in their reserved primary GDT block.
 930  */
 931 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 932                               ext4_group_t group)
 933 {
 934         struct super_block *sb = inode->i_sb;
 935         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 936         struct buffer_head **primary;
 937         struct buffer_head *dind;
 938         struct ext4_iloc iloc;
 939         ext4_fsblk_t blk;
 940         __le32 *data, *end;
 941         int gdbackups = 0;
 942         int res, i;
 943         int err;
 944
 945         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 946         if (!primary)
 947                 return -ENOMEM;
 948
 949         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 950         dind = sb_bread(sb, le32_to_cpu(*data));
 951         if (!dind) {
 952                 err = -EIO;
 953                 goto exit_free;
 954         }
 955
 956         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 957         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 958                                          EXT4_ADDR_PER_BLOCK(sb));
 959         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 960
 961         /* Get each reserved primary GDT block and verify it holds backups */
 962         for (res = 0; res < reserved_gdb; res++, blk++) {
 963                 if (le32_to_cpu(*data) != blk) {
 964                         ext4_warning(sb, "reserved block %llu"
 965                                      " not at offset %ld",
 966                                      blk,
 967                                      (long)(data - (__le32 *)dind->b_data));
 968                         err = -EINVAL;
 969                         goto exit_bh;
 970                 }
 971                 primary[res] = sb_bread(sb, blk);
 972                 if (!primary[res]) {
 973                         err = -EIO;
 974                         goto exit_bh;
 975                 }
 976                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 977                 if (gdbackups < 0) {
 978                         brelse(primary[res]);
 979                         err = gdbackups;
 980                         goto exit_bh;
 981                 }
 982                 if (++data >= end)
 983                         data = (__le32 *)dind->b_data;
 984         }
 985
 986         for (i = 0; i < reserved_gdb; i++) {
 987                 BUFFER_TRACE(primary[i], "get_write_access");
 988                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
 989                         goto exit_bh;
 990         }
 991
 992         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 993                 goto exit_bh;
 994
 995         /*
 996          * Finally we can add each of the reserved backup GDT blocks from
 997          * the new group to its reserved primary GDT block.
 998          */
 999         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1000         for (i = 0; i < reserved_gdb; i++) {
1001                 int err2;
1002                 data = (__le32 *)primary[i]->b_data;
1003                 /* printk("reserving backup %lu[%u] = %lu\n",
1004                        primary[i]->b_blocknr, gdbackups,
1005                        blk + primary[i]->b_blocknr); */
1006                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1007                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1008                 if (!err)
1009                         err = err2;
1010         }
1011         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1012         ext4_mark_iloc_dirty(handle, inode, &iloc);
1013
1014 exit_bh:
1015         while (--res >= 0)
1016                 brelse(primary[res]);
1017         brelse(dind);
1018
1019 exit_free:
1020         kfree(primary);
1021
1022         return err;
1023 }
1024
1025 /*
1026  * Update the backup copies of the ext4 metadata.  These don't need to be part
1027  * of the main resize transaction, because e2fsck will re-write them if there
1028  * is a problem (basically only OOM will cause a problem).  However, we
1029  * _should_ update the backups if possible, in case the primary gets trashed
1030  * for some reason and we need to run e2fsck from a backup superblock.  The
1031  * important part is that the new block and inode counts are in the backup
1032  * superblocks, and the location of the new group metadata in the GDT backups.
1033  *
1034  * We do not need take the s_resize_lock for this, because these
1035  * blocks are not otherwise touched by the filesystem code when it is
1036  * mounted.  We don't need to worry about last changing from
1037  * sbi->s_groups_count, because the worst that can happen is that we
1038  * do not copy the full number of backups at this time.  The resize
1039  * which changed s_groups_count will backup again.
1040  */
1041 static void update_backups(struct super_block *sb, int blk_off, char *data,
1042                            int size, int meta_bg)
1043 {
1044         struct ext4_sb_info *sbi = EXT4_SB(sb);
1045         ext4_group_t last;
1046         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1047         unsigned three = 1;
1048         unsigned five = 5;
1049         unsigned seven = 7;
1050         ext4_group_t group = 0;
1051         int rest = sb->s_blocksize - size;
1052         handle_t *handle;
1053         int err = 0, err2;
1054
1055         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1056         if (IS_ERR(handle)) {
1057                 group = 1;
1058                 err = PTR_ERR(handle);
1059                 goto exit_err;
1060         }
1061
1062         if (meta_bg == 0) {
1063                 group = ext4_list_backups(sb, &three, &five, &seven);
1064                 last = sbi->s_groups_count;
1065         } else {
1066                 group = ext4_meta_bg_first_group(sb, group) + 1;
1067                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1068         }
1069
1070         while (group < sbi->s_groups_count) {
1071                 struct buffer_head *bh;
1072                 ext4_fsblk_t backup_block;
1073
1074                 /* Out of journal space, and can't get more - abort - so sad */
1075                 if (ext4_handle_valid(handle) &&
1076                     handle->h_buffer_credits == 0 &&
1077                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1078                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1079                         break;
1080
1081                 if (meta_bg == 0)
1082                         backup_block = group * bpg + blk_off;
1083                 else
1084                         backup_block = (ext4_group_first_block_no(sb, group) +
1085                                         ext4_bg_has_super(sb, group));
1086
1087                 bh = sb_getblk(sb, backup_block);
1088                 if (unlikely(!bh)) {
1089                         err = -ENOMEM;
1090                         break;
1091                 }
1092                 ext4_debug("update metadata backup %llu(+%llu)\n",
1093                            backup_block, backup_block -
1094                            ext4_group_first_block_no(sb, group));
1095                 BUFFER_TRACE(bh, "get_write_access");
1096                 if ((err = ext4_journal_get_write_access(handle, bh)))
1097                         break;
1098                 lock_buffer(bh);
1099                 memcpy(bh->b_data, data, size);
1100                 if (rest)
1101                         memset(bh->b_data + size, 0, rest);
1102                 set_buffer_uptodate(bh);
1103                 unlock_buffer(bh);
1104                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1105                 if (unlikely(err))
1106                         ext4_std_error(sb, err);
1107                 brelse(bh);
1108
1109                 if (meta_bg == 0)
1110                         group = ext4_list_backups(sb, &three, &five, &seven);
1111                 else if (group == last)
1112                         break;
1113                 else
1114                         group = last;
1115         }
1116         if ((err2 = ext4_journal_stop(handle)) && !err)
1117                 err = err2;
1118
1119         /*
1120          * Ugh! Need to have e2fsck write the backup copies.  It is too
1121          * late to revert the resize, we shouldn't fail just because of
1122          * the backup copies (they are only needed in case of corruption).
1123          *
1124          * However, if we got here we have a journal problem too, so we
1125          * can't really start a transaction to mark the superblock.
1126          * Chicken out and just set the flag on the hope it will be written
1127          * to disk, and if not - we will simply wait until next fsck.
1128          */
1129 exit_err:
1130         if (err) {
1131                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1132                              "forcing fsck on next reboot", group, err);
1133                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1134                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1135                 mark_buffer_dirty(sbi->s_sbh);
1136         }
1137 }
1138
1139 /*
1140  * ext4_add_new_descs() adds @count group descriptor of groups
1141  * starting at @group
1142  *
1143  * @handle: journal handle
1144  * @sb: super block
1145  * @group: the group no. of the first group desc to be added
1146  * @resize_inode: the resize inode
1147  * @count: number of group descriptors to be added
1148  */
1149 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1150                               ext4_group_t group, struct inode *resize_inode,
1151                               ext4_group_t count)
1152 {
1153         struct ext4_sb_info *sbi = EXT4_SB(sb);
1154         struct ext4_super_block *es = sbi->s_es;
1155         struct buffer_head *gdb_bh;
1156         int i, gdb_off, gdb_num, err = 0;
1157         int meta_bg;
1158
1159         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1160         for (i = 0; i < count; i++, group++) {
1161                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1162                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1163
1164                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1165                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1166
1167                 /*
1168                  * We will only either add reserved group blocks to a backup group
1169                  * or remove reserved blocks for the first group in a new group block.
1170                  * Doing both would be mean more complex code, and sane people don't
1171                  * use non-sparse filesystems anymore.  This is already checked above.
1172                  */
1173                 if (gdb_off) {
1174                         gdb_bh = sbi->s_group_desc[gdb_num];
1175                         BUFFER_TRACE(gdb_bh, "get_write_access");
1176                         err = ext4_journal_get_write_access(handle, gdb_bh);
1177
1178                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1179                                 err = reserve_backup_gdb(handle, resize_inode, group);
1180                 } else if (meta_bg != 0) {
1181                         err = add_new_gdb_meta_bg(sb, handle, group);
1182                 } else {
1183                         err = add_new_gdb(handle, resize_inode, group);
1184                 }
1185                 if (err)
1186                         break;
1187         }
1188         return err;
1189 }
1190
1191 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1192 {
1193         struct buffer_head *bh = sb_getblk(sb, block);
1194         if (unlikely(!bh))
1195                 return NULL;
1196         if (!bh_uptodate_or_lock(bh)) {
1197                 if (bh_submit_read(bh) < 0) {
1198                         brelse(bh);
1199                         return NULL;
1200                 }
1201         }
1202
1203         return bh;
1204 }
1205
1206 static int ext4_set_bitmap_checksums(struct super_block *sb,
1207                                      ext4_group_t group,
1208                                      struct ext4_group_desc *gdp,
1209                                      struct ext4_new_group_data *group_data)
1210 {
1211         struct buffer_head *bh;
1212
1213         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1214                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1215                 return 0;
1216
1217         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1218         if (!bh)
1219                 return -EIO;
1220         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1221                                    EXT4_INODES_PER_GROUP(sb) / 8);
1222         brelse(bh);
1223
1224         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1225         if (!bh)
1226                 return -EIO;
1227         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1228         brelse(bh);
1229
1230         return 0;
1231 }
1232
1233 /*
1234  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1235  */
1236 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1237                                 struct ext4_new_flex_group_data *flex_gd)
1238 {
1239         struct ext4_new_group_data      *group_data = flex_gd->groups;
1240         struct ext4_group_desc          *gdp;
1241         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1242         struct buffer_head              *gdb_bh;
1243         ext4_group_t                    group;
1244         __u16                           *bg_flags = flex_gd->bg_flags;
1245         int                             i, gdb_off, gdb_num, err = 0;
1246
1247
1248         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1249                 group = group_data->group;
1250
1251                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1252                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1253
1254                 /*
1255                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1256                  */
1257                 gdb_bh = sbi->s_group_desc[gdb_num];
1258                 /* Update group descriptor block for new group */
1259                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1260                                                  gdb_off * EXT4_DESC_SIZE(sb));
1261
1262                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1263                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1264                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1265                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1266                 if (err) {
1267                         ext4_std_error(sb, err);
1268                         break;
1269                 }
1270
1271                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1272                 ext4_free_group_clusters_set(sb, gdp,
1273                         EXT4_NUM_B2C(sbi, group_data->free_blocks_count));
1274                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1275                 if (ext4_has_group_desc_csum(sb))
1276                         ext4_itable_unused_set(sb, gdp,
1277                                                EXT4_INODES_PER_GROUP(sb));
1278                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1279                 ext4_group_desc_csum_set(sb, group, gdp);
1280
1281                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1282                 if (unlikely(err)) {
1283                         ext4_std_error(sb, err);
1284                         break;
1285                 }
1286
1287                 /*
1288                  * We can allocate memory for mb_alloc based on the new group
1289                  * descriptor
1290                  */
1291                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1292                 if (err)
1293                         break;
1294         }
1295         return err;
1296 }
1297
1298 /*
1299  * ext4_update_super() updates the super block so that the newly added
1300  * groups can be seen by the filesystem.
1301  *
1302  * @sb: super block
1303  * @flex_gd: new added groups
1304  */
1305 static void ext4_update_super(struct super_block *sb,
1306                              struct ext4_new_flex_group_data *flex_gd)
1307 {
1308         ext4_fsblk_t blocks_count = 0;
1309         ext4_fsblk_t free_blocks = 0;
1310         ext4_fsblk_t reserved_blocks = 0;
1311         struct ext4_new_group_data *group_data = flex_gd->groups;
1312         struct ext4_sb_info *sbi = EXT4_SB(sb);
1313         struct ext4_super_block *es = sbi->s_es;
1314         int i;
1315
1316         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1317         /*
1318          * Make the new blocks and inodes valid next.  We do this before
1319          * increasing the group count so that once the group is enabled,
1320          * all of its blocks and inodes are already valid.
1321          *
1322          * We always allocate group-by-group, then block-by-block or
1323          * inode-by-inode within a group, so enabling these
1324          * blocks/inodes before the group is live won't actually let us
1325          * allocate the new space yet.
1326          */
1327         for (i = 0; i < flex_gd->count; i++) {
1328                 blocks_count += group_data[i].blocks_count;
1329                 free_blocks += group_data[i].free_blocks_count;
1330         }
1331
1332         reserved_blocks = ext4_r_blocks_count(es) * 100;
1333         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1334         reserved_blocks *= blocks_count;
1335         do_div(reserved_blocks, 100);
1336
1337         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1338         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1339         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1340                      flex_gd->count);
1341         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1342                      flex_gd->count);
1343
1344         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1345         /*
1346          * We need to protect s_groups_count against other CPUs seeing
1347          * inconsistent state in the superblock.
1348          *
1349          * The precise rules we use are:
1350          *
1351          * * Writers must perform a smp_wmb() after updating all
1352          *   dependent data and before modifying the groups count
1353          *
1354          * * Readers must perform an smp_rmb() after reading the groups
1355          *   count and before reading any dependent data.
1356          *
1357          * NB. These rules can be relaxed when checking the group count
1358          * while freeing data, as we can only allocate from a block
1359          * group after serialising against the group count, and we can
1360          * only then free after serialising in turn against that
1361          * allocation.
1362          */
1363         smp_wmb();
1364
1365         /* Update the global fs size fields */
1366         sbi->s_groups_count += flex_gd->count;
1367         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1368                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1369
1370         /* Update the reserved block counts only once the new group is
1371          * active. */
1372         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1373                                 reserved_blocks);
1374
1375         /* Update the free space counts */
1376         percpu_counter_add(&sbi->s_freeclusters_counter,
1377                            EXT4_NUM_B2C(sbi, free_blocks));
1378         percpu_counter_add(&sbi->s_freeinodes_counter,
1379                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1380
1381         ext4_debug("free blocks count %llu",
1382                    percpu_counter_read(&sbi->s_freeclusters_counter));
1383         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1384                                       EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1385             sbi->s_log_groups_per_flex) {
1386                 ext4_group_t flex_group;
1387                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1388                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1389                              &sbi->s_flex_groups[flex_group].free_clusters);
1390                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1391                            &sbi->s_flex_groups[flex_group].free_inodes);
1392         }
1393
1394         /*
1395          * Update the fs overhead information
1396          */
1397         ext4_calculate_overhead(sb);
1398
1399         if (test_opt(sb, DEBUG))
1400                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1401                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1402                        blocks_count, free_blocks, reserved_blocks);
1403 }
1404
1405 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1406  * _before_ we start modifying the filesystem, because we cannot abort the
1407  * transaction and not have it write the data to disk.
1408  */
1409 static int ext4_flex_group_add(struct super_block *sb,
1410                                struct inode *resize_inode,
1411                                struct ext4_new_flex_group_data *flex_gd)
1412 {
1413         struct ext4_sb_info *sbi = EXT4_SB(sb);
1414         struct ext4_super_block *es = sbi->s_es;
1415         ext4_fsblk_t o_blocks_count;
1416         ext4_grpblk_t last;
1417         ext4_group_t group;
1418         handle_t *handle;
1419         unsigned reserved_gdb;
1420         int err = 0, err2 = 0, credit;
1421
1422         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1423
1424         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1425         o_blocks_count = ext4_blocks_count(es);
1426         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1427         BUG_ON(last);
1428
1429         err = setup_new_flex_group_blocks(sb, flex_gd);
1430         if (err)
1431                 goto exit;
1432         /*
1433          * We will always be modifying at least the superblock and  GDT
1434          * block.  If we are adding a group past the last current GDT block,
1435          * we will also modify the inode and the dindirect block.  If we
1436          * are adding a group with superblock/GDT backups  we will also
1437          * modify each of the reserved GDT dindirect blocks.
1438          */
1439         credit = flex_gd->count * 4 + reserved_gdb;
1440         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1441         if (IS_ERR(handle)) {
1442                 err = PTR_ERR(handle);
1443                 goto exit;
1444         }
1445
1446         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1447         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1448         if (err)
1449                 goto exit_journal;
1450
1451         group = flex_gd->groups[0].group;
1452         BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1453         err = ext4_add_new_descs(handle, sb, group,
1454                                 resize_inode, flex_gd->count);
1455         if (err)
1456                 goto exit_journal;
1457
1458         err = ext4_setup_new_descs(handle, sb, flex_gd);
1459         if (err)
1460                 goto exit_journal;
1461
1462         ext4_update_super(sb, flex_gd);
1463
1464         err = ext4_handle_dirty_super(handle, sb);
1465
1466 exit_journal:
1467         err2 = ext4_journal_stop(handle);
1468         if (!err)
1469                 err = err2;
1470
1471         if (!err) {
1472                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1473                 int gdb_num_end = ((group + flex_gd->count - 1) /
1474                                    EXT4_DESC_PER_BLOCK(sb));
1475                 int meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb,
1476                                 EXT4_FEATURE_INCOMPAT_META_BG);
1477                 sector_t old_gdb = 0;
1478
1479                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1480                                sizeof(struct ext4_super_block), 0);
1481                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1482                         struct buffer_head *gdb_bh;
1483
1484                         gdb_bh = sbi->s_group_desc[gdb_num];
1485                         if (old_gdb == gdb_bh->b_blocknr)
1486                                 continue;
1487                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1488                                        gdb_bh->b_size, meta_bg);
1489                         old_gdb = gdb_bh->b_blocknr;
1490                 }
1491         }
1492 exit:
1493         return err;
1494 }
1495
1496 static int ext4_setup_next_flex_gd(struct super_block *sb,
1497                                     struct ext4_new_flex_group_data *flex_gd,
1498                                     ext4_fsblk_t n_blocks_count,
1499                                     unsigned long flexbg_size)
1500 {
1501         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1502         struct ext4_new_group_data *group_data = flex_gd->groups;
1503         ext4_fsblk_t o_blocks_count;
1504         ext4_group_t n_group;
1505         ext4_group_t group;
1506         ext4_group_t last_group;
1507         ext4_grpblk_t last;
1508         ext4_grpblk_t blocks_per_group;
1509         unsigned long i;
1510
1511         blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1512
1513         o_blocks_count = ext4_blocks_count(es);
1514
1515         if (o_blocks_count == n_blocks_count)
1516                 return 0;
1517
1518         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1519         BUG_ON(last);
1520         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1521
1522         last_group = group | (flexbg_size - 1);
1523         if (last_group > n_group)
1524                 last_group = n_group;
1525
1526         flex_gd->count = last_group - group + 1;
1527
1528         for (i = 0; i < flex_gd->count; i++) {
1529                 int overhead;
1530
1531                 group_data[i].group = group + i;
1532                 group_data[i].blocks_count = blocks_per_group;
1533                 overhead = ext4_group_overhead_blocks(sb, group + i);
1534                 group_data[i].free_blocks_count = blocks_per_group - overhead;
1535                 if (ext4_has_group_desc_csum(sb)) {
1536                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1537                                                EXT4_BG_INODE_UNINIT;
1538                         if (!test_opt(sb, INIT_INODE_TABLE))
1539                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1540                 } else
1541                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1542         }
1543
1544         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1545                 /* We need to initialize block bitmap of last group. */
1546                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1547
1548         if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1549                 group_data[i - 1].blocks_count = last + 1;
1550                 group_data[i - 1].free_blocks_count -= blocks_per_group-
1551                                         last - 1;
1552         }
1553
1554         return 1;
1555 }
1556
1557 /* Add group descriptor data to an existing or new group descriptor block.
1558  * Ensure we handle all possible error conditions _before_ we start modifying
1559  * the filesystem, because we cannot abort the transaction and not have it
1560  * write the data to disk.
1561  *
1562  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1563  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1564  *
1565  * We only need to hold the superblock lock while we are actually adding
1566  * in the new group's counts to the superblock.  Prior to that we have
1567  * not really "added" the group at all.  We re-check that we are still
1568  * adding in the last group in case things have changed since verifying.
1569  */
1570 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1571 {
1572         struct ext4_new_flex_group_data flex_gd;
1573         struct ext4_sb_info *sbi = EXT4_SB(sb);
1574         struct ext4_super_block *es = sbi->s_es;
1575         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1576                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1577         struct inode *inode = NULL;
1578         int gdb_off;
1579         int err;
1580         __u16 bg_flags = 0;
1581
1582         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1583
1584         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1585                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1586                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1587                 return -EPERM;
1588         }
1589
1590         if (ext4_blocks_count(es) + input->blocks_count <
1591             ext4_blocks_count(es)) {
1592                 ext4_warning(sb, "blocks_count overflow");
1593                 return -EINVAL;
1594         }
1595
1596         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1597             le32_to_cpu(es->s_inodes_count)) {
1598                 ext4_warning(sb, "inodes_count overflow");
1599                 return -EINVAL;
1600         }
1601
1602         if (reserved_gdb || gdb_off == 0) {
1603                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1604                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
1605                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1606                         ext4_warning(sb,
1607                                      "No reserved GDT blocks, can't resize");
1608                         return -EPERM;
1609                 }
1610                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1611                 if (IS_ERR(inode)) {
1612                         ext4_warning(sb, "Error opening resize inode");
1613                         return PTR_ERR(inode);
1614                 }
1615         }
1616
1617
1618         err = verify_group_input(sb, input);
1619         if (err)
1620                 goto out;
1621
1622         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1623         if (err)
1624                 goto out;
1625
1626         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1627         if (err)
1628                 goto out;
1629
1630         flex_gd.count = 1;
1631         flex_gd.groups = input;
1632         flex_gd.bg_flags = &bg_flags;
1633         err = ext4_flex_group_add(sb, inode, &flex_gd);
1634 out:
1635         iput(inode);
1636         return err;
1637 } /* ext4_group_add */
1638
1639 /*
1640  * extend a group without checking assuming that checking has been done.
1641  */
1642 static int ext4_group_extend_no_check(struct super_block *sb,
1643                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1644 {
1645         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1646         handle_t *handle;
1647         int err = 0, err2;
1648
1649         /* We will update the superblock, one block bitmap, and
1650          * one group descriptor via ext4_group_add_blocks().
1651          */
1652         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1653         if (IS_ERR(handle)) {
1654                 err = PTR_ERR(handle);
1655                 ext4_warning(sb, "error %d on journal start", err);
1656                 return err;
1657         }
1658
1659         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1660         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1661         if (err) {
1662                 ext4_warning(sb, "error %d on journal write access", err);
1663                 goto errout;
1664         }
1665
1666         ext4_blocks_count_set(es, o_blocks_count + add);
1667         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1668         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1669                    o_blocks_count + add);
1670         /* We add the blocks to the bitmap and set the group need init bit */
1671         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1672         if (err)
1673                 goto errout;
1674         ext4_handle_dirty_super(handle, sb);
1675         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1676                    o_blocks_count + add);
1677 errout:
1678         err2 = ext4_journal_stop(handle);
1679         if (err2 && !err)
1680                 err = err2;
1681
1682         if (!err) {
1683                 if (test_opt(sb, DEBUG))
1684                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1685                                "blocks\n", ext4_blocks_count(es));
1686                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1687                                (char *)es, sizeof(struct ext4_super_block), 0);
1688         }
1689         return err;
1690 }
1691
1692 /*
1693  * Extend the filesystem to the new number of blocks specified.  This entry
1694  * point is only used to extend the current filesystem to the end of the last
1695  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1696  * for emergencies (because it has no dependencies on reserved blocks).
1697  *
1698  * If we _really_ wanted, we could use default values to call ext4_group_add()
1699  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1700  * GDT blocks are reserved to grow to the desired size.
1701  */
1702 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1703                       ext4_fsblk_t n_blocks_count)
1704 {
1705         ext4_fsblk_t o_blocks_count;
1706         ext4_grpblk_t last;
1707         ext4_grpblk_t add;
1708         struct buffer_head *bh;
1709         int err;
1710         ext4_group_t group;
1711
1712         o_blocks_count = ext4_blocks_count(es);
1713
1714         if (test_opt(sb, DEBUG))
1715                 ext4_msg(sb, KERN_DEBUG,
1716                          "extending last group from %llu to %llu blocks",
1717                          o_blocks_count, n_blocks_count);
1718
1719         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1720                 return 0;
1721
1722         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1723                 ext4_msg(sb, KERN_ERR,
1724                          "filesystem too large to resize to %llu blocks safely",
1725                          n_blocks_count);
1726                 if (sizeof(sector_t) < 8)
1727                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1728                 return -EINVAL;
1729         }
1730
1731         if (n_blocks_count < o_blocks_count) {
1732                 ext4_warning(sb, "can't shrink FS - resize aborted");
1733                 return -EINVAL;
1734         }
1735
1736         /* Handle the remaining blocks in the last group only. */
1737         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1738
1739         if (last == 0) {
1740                 ext4_warning(sb, "need to use ext2online to resize further");
1741                 return -EPERM;
1742         }
1743
1744         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1745
1746         if (o_blocks_count + add < o_blocks_count) {
1747                 ext4_warning(sb, "blocks_count overflow");
1748                 return -EINVAL;
1749         }
1750
1751         if (o_blocks_count + add > n_blocks_count)
1752                 add = n_blocks_count - o_blocks_count;
1753
1754         if (o_blocks_count + add < n_blocks_count)
1755                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1756                              o_blocks_count + add, add);
1757
1758         /* See if the device is actually as big as what was requested */
1759         bh = sb_bread(sb, o_blocks_count + add - 1);
1760         if (!bh) {
1761                 ext4_warning(sb, "can't read last block, resize aborted");
1762                 return -ENOSPC;
1763         }
1764         brelse(bh);
1765
1766         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1767         return err;
1768 } /* ext4_group_extend */
1769
1770
1771 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1772 {
1773         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1774 }
1775
1776 /*
1777  * Release the resize inode and drop the resize_inode feature if there
1778  * are no more reserved gdt blocks, and then convert the file system
1779  * to enable meta_bg
1780  */
1781 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1782 {
1783         handle_t *handle;
1784         struct ext4_sb_info *sbi = EXT4_SB(sb);
1785         struct ext4_super_block *es = sbi->s_es;
1786         struct ext4_inode_info *ei = EXT4_I(inode);
1787         ext4_fsblk_t nr;
1788         int i, ret, err = 0;
1789         int credits = 1;
1790
1791         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1792         if (inode) {
1793                 if (es->s_reserved_gdt_blocks) {
1794                         ext4_error(sb, "Unexpected non-zero "
1795                                    "s_reserved_gdt_blocks");
1796                         return -EPERM;
1797                 }
1798
1799                 /* Do a quick sanity check of the resize inode */
1800                 if (inode->i_blocks != 1 << (inode->i_blkbits - 9))
1801                         goto invalid_resize_inode;
1802                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1803                         if (i == EXT4_DIND_BLOCK) {
1804                                 if (ei->i_data[i])
1805                                         continue;
1806                                 else
1807                                         goto invalid_resize_inode;
1808                         }
1809                         if (ei->i_data[i])
1810                                 goto invalid_resize_inode;
1811                 }
1812                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1813         }
1814
1815         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1816         if (IS_ERR(handle))
1817                 return PTR_ERR(handle);
1818
1819         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1820         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1821         if (err)
1822                 goto errout;
1823
1824         EXT4_CLEAR_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE);
1825         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1826         sbi->s_es->s_first_meta_bg =
1827                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1828
1829         err = ext4_handle_dirty_super(handle, sb);
1830         if (err) {
1831                 ext4_std_error(sb, err);
1832                 goto errout;
1833         }
1834
1835         if (inode) {
1836                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1837                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1838                                  EXT4_FREE_BLOCKS_METADATA |
1839                                  EXT4_FREE_BLOCKS_FORGET);
1840                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1841                 inode->i_blocks = 0;
1842
1843                 err = ext4_mark_inode_dirty(handle, inode);
1844                 if (err)
1845                         ext4_std_error(sb, err);
1846         }
1847
1848 errout:
1849         ret = ext4_journal_stop(handle);
1850         if (!err)
1851                 err = ret;
1852         return ret;
1853
1854 invalid_resize_inode:
1855         ext4_error(sb, "corrupted/inconsistent resize inode");
1856         return -EINVAL;
1857 }
1858
1859 /*
1860  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1861  *
1862  * @sb: super block of the fs to be resized
1863  * @n_blocks_count: the number of blocks resides in the resized fs
1864  */
1865 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1866 {
1867         struct ext4_new_flex_group_data *flex_gd = NULL;
1868         struct ext4_sb_info *sbi = EXT4_SB(sb);
1869         struct ext4_super_block *es = sbi->s_es;
1870         struct buffer_head *bh;
1871         struct inode *resize_inode = NULL;
1872         ext4_grpblk_t add, offset;
1873         unsigned long n_desc_blocks;
1874         unsigned long o_desc_blocks;
1875         ext4_group_t o_group;
1876         ext4_group_t n_group;
1877         ext4_fsblk_t o_blocks_count;
1878         ext4_fsblk_t n_blocks_count_retry = 0;
1879         unsigned long last_update_time = 0;
1880         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1881         int meta_bg;
1882
1883         /* See if the device is actually as big as what was requested */
1884         bh = sb_bread(sb, n_blocks_count - 1);
1885         if (!bh) {
1886                 ext4_warning(sb, "can't read last block, resize aborted");
1887                 return -ENOSPC;
1888         }
1889         brelse(bh);
1890
1891 retry:
1892         o_blocks_count = ext4_blocks_count(es);
1893
1894         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1895                  "to %llu blocks", o_blocks_count, n_blocks_count);
1896
1897         if (n_blocks_count < o_blocks_count) {
1898                 /* On-line shrinking not supported */
1899                 ext4_warning(sb, "can't shrink FS - resize aborted");
1900                 return -EINVAL;
1901         }
1902
1903         if (n_blocks_count == o_blocks_count)
1904                 /* Nothing need to do */
1905                 return 0;
1906
1907         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1908         if (n_group > (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1909                 ext4_warning(sb, "resize would cause inodes_count overflow");
1910                 return -EINVAL;
1911         }
1912         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1913
1914         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1915         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1916
1917         meta_bg = EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG);
1918
1919         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE)) {
1920                 if (meta_bg) {
1921                         ext4_error(sb, "resize_inode and meta_bg enabled "
1922                                    "simultaneously");
1923                         return -EINVAL;
1924                 }
1925                 if (n_desc_blocks > o_desc_blocks +
1926                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
1927                         n_blocks_count_retry = n_blocks_count;
1928                         n_desc_blocks = o_desc_blocks +
1929                                 le16_to_cpu(es->s_reserved_gdt_blocks);
1930                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1931                         n_blocks_count = n_group * EXT4_BLOCKS_PER_GROUP(sb);
1932                         n_group--; /* set to last group number */
1933                 }
1934
1935                 if (!resize_inode)
1936                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1937                 if (IS_ERR(resize_inode)) {
1938                         ext4_warning(sb, "Error opening resize inode");
1939                         return PTR_ERR(resize_inode);
1940                 }
1941         }
1942
1943         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1944                 err = ext4_convert_meta_bg(sb, resize_inode);
1945                 if (err)
1946                         goto out;
1947                 if (resize_inode) {
1948                         iput(resize_inode);
1949                         resize_inode = NULL;
1950                 }
1951                 if (n_blocks_count_retry) {
1952                         n_blocks_count = n_blocks_count_retry;
1953                         n_blocks_count_retry = 0;
1954                         goto retry;
1955                 }
1956         }
1957
1958         /* extend the last group */
1959         if (n_group == o_group)
1960                 add = n_blocks_count - o_blocks_count;
1961         else
1962                 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1963         if (add > 0) {
1964                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1965                 if (err)
1966                         goto out;
1967         }
1968
1969         if (ext4_blocks_count(es) == n_blocks_count)
1970                 goto out;
1971
1972         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1973         if (err)
1974                 return err;
1975
1976         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1977         if (err)
1978                 goto out;
1979
1980         flex_gd = alloc_flex_gd(flexbg_size);
1981         if (flex_gd == NULL) {
1982                 err = -ENOMEM;
1983                 goto out;
1984         }
1985
1986         /* Add flex groups. Note that a regular group is a
1987          * flex group with 1 group.
1988          */
1989         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1990                                               flexbg_size)) {
1991                 if (jiffies - last_update_time > HZ * 10) {
1992                         if (last_update_time)
1993                                 ext4_msg(sb, KERN_INFO,
1994                                          "resized to %llu blocks",
1995                                          ext4_blocks_count(es));
1996                         last_update_time = jiffies;
1997                 }
1998                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1999                         break;
2000                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2001                 if (unlikely(err))
2002                         break;
2003         }
2004
2005         if (!err && n_blocks_count_retry) {
2006                 n_blocks_count = n_blocks_count_retry;
2007                 n_blocks_count_retry = 0;
2008                 free_flex_gd(flex_gd);
2009                 flex_gd = NULL;
2010                 goto retry;
2011         }
2012
2013 out:
2014         if (flex_gd)
2015                 free_flex_gd(flex_gd);
2016         if (resize_inode != NULL)
2017                 iput(resize_inode);
2018         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu", n_blocks_count);
2019         return err;
2020 }