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