fs/ext4/resize.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/resize.c
   4  *
   5  * Support for resizing an ext4 filesystem while it is mounted.
   6  *
   7  * Copyright (C) 2001, 2002 Andreas Dilger <[email protected]>
   8  *
   9  * This could probably be made into a module, because it is not often in use.
  10  */
  11
  12
  13 #define EXT4FS_DEBUG
  14
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17
  18 #include "ext4_jbd2.h"
  19
  20 int ext4_resize_begin(struct super_block *sb)
  21 {
  22         struct ext4_sb_info *sbi = EXT4_SB(sb);
  23         int ret = 0;
  24
  25         if (!capable(CAP_SYS_RESOURCE))
  26                 return -EPERM;
  27
  28         /*
  29          * If we are not using the primary superblock/GDT copy don't resize,
  30          * because the user tools have no way of handling this.  Probably a
  31          * bad time to do it anyways.
  32          */
  33         if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
  34             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  35                 ext4_warning(sb, "won't resize using backup superblock at %llu",
  36                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  37                 return -EPERM;
  38         }
  39
  40         /*
  41          * We are not allowed to do online-resizing on a filesystem mounted
  42          * with error, because it can destroy the filesystem easily.
  43          */
  44         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  45                 ext4_warning(sb, "There are errors in the filesystem, "
  46                              "so online resizing is not allowed");
  47                 return -EPERM;
  48         }
  49
  50         if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
  51                                   &EXT4_SB(sb)->s_ext4_flags))
  52                 ret = -EBUSY;
  53
  54         return ret;
  55 }
  56
  57 void ext4_resize_end(struct super_block *sb)
  58 {
  59         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
  60         smp_mb__after_atomic();
  61 }
  62
  63 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  64                                              ext4_group_t group) {
  65         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  66                EXT4_DESC_PER_BLOCK_BITS(sb);
  67 }
  68
  69 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
  70                                              ext4_group_t group) {
  71         group = ext4_meta_bg_first_group(sb, group);
  72         return ext4_group_first_block_no(sb, group);
  73 }
  74
  75 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
  76                                                 ext4_group_t group) {
  77         ext4_grpblk_t overhead;
  78         overhead = ext4_bg_num_gdb(sb, group);
  79         if (ext4_bg_has_super(sb, group))
  80                 overhead += 1 +
  81                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
  82         return overhead;
  83 }
  84
  85 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  86 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  87
  88 static int verify_group_input(struct super_block *sb,
  89                               struct ext4_new_group_data *input)
  90 {
  91         struct ext4_sb_info *sbi = EXT4_SB(sb);
  92         struct ext4_super_block *es = sbi->s_es;
  93         ext4_fsblk_t start = ext4_blocks_count(es);
  94         ext4_fsblk_t end = start + input->blocks_count;
  95         ext4_group_t group = input->group;
  96         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  97         unsigned overhead;
  98         ext4_fsblk_t metaend;
  99         struct buffer_head *bh = NULL;
 100         ext4_grpblk_t free_blocks_count, offset;
 101         int err = -EINVAL;
 102
 103         if (group != sbi->s_groups_count) {
 104                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 105                              input->group, sbi->s_groups_count);
 106                 return -EINVAL;
 107         }
 108
 109         overhead = ext4_group_overhead_blocks(sb, group);
 110         metaend = start + overhead;
 111         input->free_clusters_count = free_blocks_count =
 112                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 113
 114         if (test_opt(sb, DEBUG))
 115                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 116                        "(%d free, %u reserved)\n",
 117                        ext4_bg_has_super(sb, input->group) ? "normal" :
 118                        "no-super", input->group, input->blocks_count,
 119                        free_blocks_count, input->reserved_blocks);
 120
 121         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 122         if (offset != 0)
 123                         ext4_warning(sb, "Last group not full");
 124         else if (input->reserved_blocks > input->blocks_count / 5)
 125                 ext4_warning(sb, "Reserved blocks too high (%u)",
 126                              input->reserved_blocks);
 127         else if (free_blocks_count < 0)
 128                 ext4_warning(sb, "Bad blocks count %u",
 129                              input->blocks_count);
 130         else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
 131                 err = PTR_ERR(bh);
 132                 bh = NULL;
 133                 ext4_warning(sb, "Cannot read last block (%llu)",
 134                              end - 1);
 135         } else if (outside(input->block_bitmap, start, end))
 136                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 137                              (unsigned long long)input->block_bitmap);
 138         else if (outside(input->inode_bitmap, start, end))
 139                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 140                              (unsigned long long)input->inode_bitmap);
 141         else if (outside(input->inode_table, start, end) ||
 142                  outside(itend - 1, start, end))
 143                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 144                              (unsigned long long)input->inode_table, itend - 1);
 145         else if (input->inode_bitmap == input->block_bitmap)
 146                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 147                              (unsigned long long)input->block_bitmap);
 148         else if (inside(input->block_bitmap, input->inode_table, itend))
 149                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 150                              "(%llu-%llu)",
 151                              (unsigned long long)input->block_bitmap,
 152                              (unsigned long long)input->inode_table, itend - 1);
 153         else if (inside(input->inode_bitmap, input->inode_table, itend))
 154                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 155                              "(%llu-%llu)",
 156                              (unsigned long long)input->inode_bitmap,
 157                              (unsigned long long)input->inode_table, itend - 1);
 158         else if (inside(input->block_bitmap, start, metaend))
 159                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 160                              (unsigned long long)input->block_bitmap,
 161                              start, metaend - 1);
 162         else if (inside(input->inode_bitmap, start, metaend))
 163                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 164                              (unsigned long long)input->inode_bitmap,
 165                              start, metaend - 1);
 166         else if (inside(input->inode_table, start, metaend) ||
 167                  inside(itend - 1, start, metaend))
 168                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 169                              "(%llu-%llu)",
 170                              (unsigned long long)input->inode_table,
 171                              itend - 1, start, metaend - 1);
 172         else
 173                 err = 0;
 174         brelse(bh);
 175
 176         return err;
 177 }
 178
 179 /*
 180  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 181  * group each time.
 182  */
 183 struct ext4_new_flex_group_data {
 184         struct ext4_new_group_data *groups;     /* new_group_data for groups
 185                                                    in the flex group */
 186         __u16 *bg_flags;                        /* block group flags of groups
 187                                                    in @groups */
 188         ext4_group_t count;                     /* number of groups in @groups
 189                                                  */
 190 };
 191
 192 /*
 193  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 194  * @flexbg_size.
 195  *
 196  * Returns NULL on failure otherwise address of the allocated structure.
 197  */
 198 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 199 {
 200         struct ext4_new_flex_group_data *flex_gd;
 201
 202         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 203         if (flex_gd == NULL)
 204                 goto out3;
 205
 206         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
 207                 goto out2;
 208         flex_gd->count = flexbg_size;
 209
 210         flex_gd->groups = kmalloc_array(flexbg_size,
 211                                         sizeof(struct ext4_new_group_data),
 212                                         GFP_NOFS);
 213         if (flex_gd->groups == NULL)
 214                 goto out2;
 215
 216         flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
 217                                           GFP_NOFS);
 218         if (flex_gd->bg_flags == NULL)
 219                 goto out1;
 220
 221         return flex_gd;
 222
 223 out1:
 224         kfree(flex_gd->groups);
 225 out2:
 226         kfree(flex_gd);
 227 out3:
 228         return NULL;
 229 }
 230
 231 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 232 {
 233         kfree(flex_gd->bg_flags);
 234         kfree(flex_gd->groups);
 235         kfree(flex_gd);
 236 }
 237
 238 /*
 239  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 240  * and inode tables for a flex group.
 241  *
 242  * This function is used by 64bit-resize.  Note that this function allocates
 243  * group tables from the 1st group of groups contained by @flexgd, which may
 244  * be a partial of a flex group.
 245  *
 246  * @sb: super block of fs to which the groups belongs
 247  *
 248  * Returns 0 on a successful allocation of the metadata blocks in the
 249  * block group.
 250  */
 251 static int ext4_alloc_group_tables(struct super_block *sb,
 252                                 struct ext4_new_flex_group_data *flex_gd,
 253                                 int flexbg_size)
 254 {
 255         struct ext4_new_group_data *group_data = flex_gd->groups;
 256         ext4_fsblk_t start_blk;
 257         ext4_fsblk_t last_blk;
 258         ext4_group_t src_group;
 259         ext4_group_t bb_index = 0;
 260         ext4_group_t ib_index = 0;
 261         ext4_group_t it_index = 0;
 262         ext4_group_t group;
 263         ext4_group_t last_group;
 264         unsigned overhead;
 265         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 266         int i;
 267
 268         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 269
 270         src_group = group_data[0].group;
 271         last_group  = src_group + flex_gd->count - 1;
 272
 273         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 274                (last_group & ~(flexbg_size - 1))));
 275 next_group:
 276         group = group_data[0].group;
 277         if (src_group >= group_data[0].group + flex_gd->count)
 278                 return -ENOSPC;
 279         start_blk = ext4_group_first_block_no(sb, src_group);
 280         last_blk = start_blk + group_data[src_group - group].blocks_count;
 281
 282         overhead = ext4_group_overhead_blocks(sb, src_group);
 283
 284         start_blk += overhead;
 285
 286         /* We collect contiguous blocks as much as possible. */
 287         src_group++;
 288         for (; src_group <= last_group; src_group++) {
 289                 overhead = ext4_group_overhead_blocks(sb, src_group);
 290                 if (overhead == 0)
 291                         last_blk += group_data[src_group - group].blocks_count;
 292                 else
 293                         break;
 294         }
 295
 296         /* Allocate block bitmaps */
 297         for (; bb_index < flex_gd->count; bb_index++) {
 298                 if (start_blk >= last_blk)
 299                         goto next_group;
 300                 group_data[bb_index].block_bitmap = start_blk++;
 301                 group = ext4_get_group_number(sb, start_blk - 1);
 302                 group -= group_data[0].group;
 303                 group_data[group].mdata_blocks++;
 304                 flex_gd->bg_flags[group] &= uninit_mask;
 305         }
 306
 307         /* Allocate inode bitmaps */
 308         for (; ib_index < flex_gd->count; ib_index++) {
 309                 if (start_blk >= last_blk)
 310                         goto next_group;
 311                 group_data[ib_index].inode_bitmap = start_blk++;
 312                 group = ext4_get_group_number(sb, start_blk - 1);
 313                 group -= group_data[0].group;
 314                 group_data[group].mdata_blocks++;
 315                 flex_gd->bg_flags[group] &= uninit_mask;
 316         }
 317
 318         /* Allocate inode tables */
 319         for (; it_index < flex_gd->count; it_index++) {
 320                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 321                 ext4_fsblk_t next_group_start;
 322
 323                 if (start_blk + itb > last_blk)
 324                         goto next_group;
 325                 group_data[it_index].inode_table = start_blk;
 326                 group = ext4_get_group_number(sb, start_blk);
 327                 next_group_start = ext4_group_first_block_no(sb, group + 1);
 328                 group -= group_data[0].group;
 329
 330                 if (start_blk + itb > next_group_start) {
 331                         flex_gd->bg_flags[group + 1] &= uninit_mask;
 332                         overhead = start_blk + itb - next_group_start;
 333                         group_data[group + 1].mdata_blocks += overhead;
 334                         itb -= overhead;
 335                 }
 336
 337                 group_data[group].mdata_blocks += itb;
 338                 flex_gd->bg_flags[group] &= uninit_mask;
 339                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 340         }
 341
 342         /* Update free clusters count to exclude metadata blocks */
 343         for (i = 0; i < flex_gd->count; i++) {
 344                 group_data[i].free_clusters_count -=
 345                                 EXT4_NUM_B2C(EXT4_SB(sb),
 346                                              group_data[i].mdata_blocks);
 347         }
 348
 349         if (test_opt(sb, DEBUG)) {
 350                 int i;
 351                 group = group_data[0].group;
 352
 353                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 354                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 355                        flexbg_size);
 356
 357                 for (i = 0; i < flex_gd->count; i++) {
 358                         ext4_debug(
 359                                "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
 360                                ext4_bg_has_super(sb, group + i) ? "normal" :
 361                                "no-super", group + i,
 362                                group_data[i].blocks_count,
 363                                group_data[i].free_clusters_count,
 364                                group_data[i].mdata_blocks);
 365                 }
 366         }
 367         return 0;
 368 }
 369
 370 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 371                                   ext4_fsblk_t blk)
 372 {
 373         struct buffer_head *bh;
 374         int err;
 375
 376         bh = sb_getblk(sb, blk);
 377         if (unlikely(!bh))
 378                 return ERR_PTR(-ENOMEM);
 379         BUFFER_TRACE(bh, "get_write_access");
 380         if ((err = ext4_journal_get_write_access(handle, bh))) {
 381                 brelse(bh);
 382                 bh = ERR_PTR(err);
 383         } else {
 384                 memset(bh->b_data, 0, sb->s_blocksize);
 385                 set_buffer_uptodate(bh);
 386         }
 387
 388         return bh;
 389 }
 390
 391 /*
 392  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
 393  *
 394  * Helper function for ext4_setup_new_group_blocks() which set .
 395  *
 396  * @sb: super block
 397  * @handle: journal handle
 398  * @flex_gd: flex group data
 399  */
 400 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 401                         struct ext4_new_flex_group_data *flex_gd,
 402                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
 403 {
 404         struct ext4_sb_info *sbi = EXT4_SB(sb);
 405         ext4_group_t count = last_cluster - first_cluster + 1;
 406         ext4_group_t count2;
 407
 408         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
 409                    last_cluster);
 410         for (count2 = count; count > 0;
 411              count -= count2, first_cluster += count2) {
 412                 ext4_fsblk_t start;
 413                 struct buffer_head *bh;
 414                 ext4_group_t group;
 415                 int err;
 416
 417                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
 418                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
 419                 group -= flex_gd->groups[0].group;
 420
 421                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
 422                 if (count2 > count)
 423                         count2 = count;
 424
 425                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 426                         BUG_ON(flex_gd->count > 1);
 427                         continue;
 428                 }
 429
 430                 err = ext4_journal_ensure_credits_batch(handle, 1);
 431                 if (err < 0)
 432                         return err;
 433
 434                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 435                 if (unlikely(!bh))
 436                         return -ENOMEM;
 437
 438                 BUFFER_TRACE(bh, "get_write_access");
 439                 err = ext4_journal_get_write_access(handle, bh);
 440                 if (err) {
 441                         brelse(bh);
 442                         return err;
 443                 }
 444                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
 445                            first_cluster, first_cluster - start, count2);
 446                 ext4_set_bits(bh->b_data, first_cluster - start, count2);
 447
 448                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 449                 brelse(bh);
 450                 if (unlikely(err))
 451                         return err;
 452         }
 453
 454         return 0;
 455 }
 456
 457 /*
 458  * Set up the block and inode bitmaps, and the inode table for the new groups.
 459  * This doesn't need to be part of the main transaction, since we are only
 460  * changing blocks outside the actual filesystem.  We still do journaling to
 461  * ensure the recovery is correct in case of a failure just after resize.
 462  * If any part of this fails, we simply abort the resize.
 463  *
 464  * setup_new_flex_group_blocks handles a flex group as follow:
 465  *  1. copy super block and GDT, and initialize group tables if necessary.
 466  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 467  *     super block and GDT.
 468  *  2. allocate group tables in block bitmaps, that is, set bits in block
 469  *     bitmap for blocks taken by group tables.
 470  */
 471 static int setup_new_flex_group_blocks(struct super_block *sb,
 472                                 struct ext4_new_flex_group_data *flex_gd)
 473 {
 474         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 475         ext4_fsblk_t start;
 476         ext4_fsblk_t block;
 477         struct ext4_sb_info *sbi = EXT4_SB(sb);
 478         struct ext4_super_block *es = sbi->s_es;
 479         struct ext4_new_group_data *group_data = flex_gd->groups;
 480         __u16 *bg_flags = flex_gd->bg_flags;
 481         handle_t *handle;
 482         ext4_group_t group, count;
 483         struct buffer_head *bh = NULL;
 484         int reserved_gdb, i, j, err = 0, err2;
 485         int meta_bg;
 486
 487         BUG_ON(!flex_gd->count || !group_data ||
 488                group_data[0].group != sbi->s_groups_count);
 489
 490         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 491         meta_bg = ext4_has_feature_meta_bg(sb);
 492
 493         /* This transaction may be extended/restarted along the way */
 494         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 495         if (IS_ERR(handle))
 496                 return PTR_ERR(handle);
 497
 498         group = group_data[0].group;
 499         for (i = 0; i < flex_gd->count; i++, group++) {
 500                 unsigned long gdblocks;
 501                 ext4_grpblk_t overhead;
 502
 503                 gdblocks = ext4_bg_num_gdb(sb, group);
 504                 start = ext4_group_first_block_no(sb, group);
 505
 506                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 507                         goto handle_itb;
 508
 509                 if (meta_bg == 1) {
 510                         ext4_group_t first_group;
 511                         first_group = ext4_meta_bg_first_group(sb, group);
 512                         if (first_group != group + 1 &&
 513                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 514                                 goto handle_itb;
 515                 }
 516
 517                 block = start + ext4_bg_has_super(sb, group);
 518                 /* Copy all of the GDT blocks into the backup in this group */
 519                 for (j = 0; j < gdblocks; j++, block++) {
 520                         struct buffer_head *gdb;
 521
 522                         ext4_debug("update backup group %#04llx\n", block);
 523                         err = ext4_journal_ensure_credits_batch(handle, 1);
 524                         if (err < 0)
 525                                 goto out;
 526
 527                         gdb = sb_getblk(sb, block);
 528                         if (unlikely(!gdb)) {
 529                                 err = -ENOMEM;
 530                                 goto out;
 531                         }
 532
 533                         BUFFER_TRACE(gdb, "get_write_access");
 534                         err = ext4_journal_get_write_access(handle, gdb);
 535                         if (err) {
 536                                 brelse(gdb);
 537                                 goto out;
 538                         }
 539                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 540                                gdb->b_size);
 541                         set_buffer_uptodate(gdb);
 542
 543                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 544                         if (unlikely(err)) {
 545                                 brelse(gdb);
 546                                 goto out;
 547                         }
 548                         brelse(gdb);
 549                 }
 550
 551                 /* Zero out all of the reserved backup group descriptor
 552                  * table blocks
 553                  */
 554                 if (ext4_bg_has_super(sb, group)) {
 555                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 556                                         reserved_gdb, GFP_NOFS);
 557                         if (err)
 558                                 goto out;
 559                 }
 560
 561 handle_itb:
 562                 /* Initialize group tables of the grop @group */
 563                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 564                         goto handle_bb;
 565
 566                 /* Zero out all of the inode table blocks */
 567                 block = group_data[i].inode_table;
 568                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 569                            block, sbi->s_itb_per_group);
 570                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 571                                        GFP_NOFS);
 572                 if (err)
 573                         goto out;
 574
 575 handle_bb:
 576                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 577                         goto handle_ib;
 578
 579                 /* Initialize block bitmap of the @group */
 580                 block = group_data[i].block_bitmap;
 581                 err = ext4_journal_ensure_credits_batch(handle, 1);
 582                 if (err < 0)
 583                         goto out;
 584
 585                 bh = bclean(handle, sb, block);
 586                 if (IS_ERR(bh)) {
 587                         err = PTR_ERR(bh);
 588                         goto out;
 589                 }
 590                 overhead = ext4_group_overhead_blocks(sb, group);
 591                 if (overhead != 0) {
 592                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 593                                    start);
 594                         ext4_set_bits(bh->b_data, 0,
 595                                       EXT4_NUM_B2C(sbi, overhead));
 596                 }
 597                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
 598                                      sb->s_blocksize * 8, bh->b_data);
 599                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 600                 brelse(bh);
 601                 if (err)
 602                         goto out;
 603
 604 handle_ib:
 605                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 606                         continue;
 607
 608                 /* Initialize inode bitmap of the @group */
 609                 block = group_data[i].inode_bitmap;
 610                 err = ext4_journal_ensure_credits_batch(handle, 1);
 611                 if (err < 0)
 612                         goto out;
 613                 /* Mark unused entries in inode bitmap used */
 614                 bh = bclean(handle, sb, block);
 615                 if (IS_ERR(bh)) {
 616                         err = PTR_ERR(bh);
 617                         goto out;
 618                 }
 619
 620                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 621                                      sb->s_blocksize * 8, bh->b_data);
 622                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 623                 brelse(bh);
 624                 if (err)
 625                         goto out;
 626         }
 627
 628         /* Mark group tables in block bitmap */
 629         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 630                 count = group_table_count[j];
 631                 start = (&group_data[0].block_bitmap)[j];
 632                 block = start;
 633                 for (i = 1; i < flex_gd->count; i++) {
 634                         block += group_table_count[j];
 635                         if (block == (&group_data[i].block_bitmap)[j]) {
 636                                 count += group_table_count[j];
 637                                 continue;
 638                         }
 639                         err = set_flexbg_block_bitmap(sb, handle,
 640                                                       flex_gd,
 641                                                       EXT4_B2C(sbi, start),
 642                                                       EXT4_B2C(sbi,
 643                                                                start + count
 644                                                                - 1));
 645                         if (err)
 646                                 goto out;
 647                         count = group_table_count[j];
 648                         start = (&group_data[i].block_bitmap)[j];
 649                         block = start;
 650                 }
 651
 652                 if (count) {
 653                         err = set_flexbg_block_bitmap(sb, handle,
 654                                                       flex_gd,
 655                                                       EXT4_B2C(sbi, start),
 656                                                       EXT4_B2C(sbi,
 657                                                                start + count
 658                                                                - 1));
 659                         if (err)
 660                                 goto out;
 661                 }
 662         }
 663
 664 out:
 665         err2 = ext4_journal_stop(handle);
 666         if (err2 && !err)
 667                 err = err2;
 668
 669         return err;
 670 }
 671
 672 /*
 673  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 674  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 675  * calling this for the first time.  In a sparse filesystem it will be the
 676  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 677  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 678  */
 679 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 680                                   unsigned *five, unsigned *seven)
 681 {
 682         unsigned *min = three;
 683         int mult = 3;
 684         unsigned ret;
 685
 686         if (!ext4_has_feature_sparse_super(sb)) {
 687                 ret = *min;
 688                 *min += 1;
 689                 return ret;
 690         }
 691
 692         if (*five < *min) {
 693                 min = five;
 694                 mult = 5;
 695         }
 696         if (*seven < *min) {
 697                 min = seven;
 698                 mult = 7;
 699         }
 700
 701         ret = *min;
 702         *min *= mult;
 703
 704         return ret;
 705 }
 706
 707 /*
 708  * Check that all of the backup GDT blocks are held in the primary GDT block.
 709  * It is assumed that they are stored in group order.  Returns the number of
 710  * groups in current filesystem that have BACKUPS, or -ve error code.
 711  */
 712 static int verify_reserved_gdb(struct super_block *sb,
 713                                ext4_group_t end,
 714                                struct buffer_head *primary)
 715 {
 716         const ext4_fsblk_t blk = primary->b_blocknr;
 717         unsigned three = 1;
 718         unsigned five = 5;
 719         unsigned seven = 7;
 720         unsigned grp;
 721         __le32 *p = (__le32 *)primary->b_data;
 722         int gdbackups = 0;
 723
 724         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 725                 if (le32_to_cpu(*p++) !=
 726                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 727                         ext4_warning(sb, "reserved GDT %llu"
 728                                      " missing grp %d (%llu)",
 729                                      blk, grp,
 730                                      grp *
 731                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 732                                      blk);
 733                         return -EINVAL;
 734                 }
 735                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 736                         return -EFBIG;
 737         }
 738
 739         return gdbackups;
 740 }
 741
 742 /*
 743  * Called when we need to bring a reserved group descriptor table block into
 744  * use from the resize inode.  The primary copy of the new GDT block currently
 745  * is an indirect block (under the double indirect block in the resize inode).
 746  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 747  * block, in group order.  Even though we know all the block numbers we need,
 748  * we check to ensure that the resize inode has actually reserved these blocks.
 749  *
 750  * Don't need to update the block bitmaps because the blocks are still in use.
 751  *
 752  * We get all of the error cases out of the way, so that we are sure to not
 753  * fail once we start modifying the data on disk, because JBD has no rollback.
 754  */
 755 static int add_new_gdb(handle_t *handle, struct inode *inode,
 756                        ext4_group_t group)
 757 {
 758         struct super_block *sb = inode->i_sb;
 759         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 760         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 761         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 762         struct buffer_head **o_group_desc, **n_group_desc = NULL;
 763         struct buffer_head *dind = NULL;
 764         struct buffer_head *gdb_bh = NULL;
 765         int gdbackups;
 766         struct ext4_iloc iloc = { .bh = NULL };
 767         __le32 *data;
 768         int err;
 769
 770         if (test_opt(sb, DEBUG))
 771                 printk(KERN_DEBUG
 772                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 773                        gdb_num);
 774
 775         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 776         if (IS_ERR(gdb_bh))
 777                 return PTR_ERR(gdb_bh);
 778
 779         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 780         if (gdbackups < 0) {
 781                 err = gdbackups;
 782                 goto errout;
 783         }
 784
 785         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 786         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
 787         if (IS_ERR(dind)) {
 788                 err = PTR_ERR(dind);
 789                 dind = NULL;
 790                 goto errout;
 791         }
 792
 793         data = (__le32 *)dind->b_data;
 794         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 795                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 796                              group, gdblock);
 797                 err = -EINVAL;
 798                 goto errout;
 799         }
 800
 801         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 802         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 803         if (unlikely(err))
 804                 goto errout;
 805
 806         BUFFER_TRACE(gdb_bh, "get_write_access");
 807         err = ext4_journal_get_write_access(handle, gdb_bh);
 808         if (unlikely(err))
 809                 goto errout;
 810
 811         BUFFER_TRACE(dind, "get_write_access");
 812         err = ext4_journal_get_write_access(handle, dind);
 813         if (unlikely(err))
 814                 ext4_std_error(sb, err);
 815
 816         /* ext4_reserve_inode_write() gets a reference on the iloc */
 817         err = ext4_reserve_inode_write(handle, inode, &iloc);
 818         if (unlikely(err))
 819                 goto errout;
 820
 821         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 822                                      sizeof(struct buffer_head *),
 823                                      GFP_NOFS);
 824         if (!n_group_desc) {
 825                 err = -ENOMEM;
 826                 ext4_warning(sb, "not enough memory for %lu groups",
 827                              gdb_num + 1);
 828                 goto errout;
 829         }
 830
 831         /*
 832          * Finally, we have all of the possible failures behind us...
 833          *
 834          * Remove new GDT block from inode double-indirect block and clear out
 835          * the new GDT block for use (which also "frees" the backup GDT blocks
 836          * from the reserved inode).  We don't need to change the bitmaps for
 837          * these blocks, because they are marked as in-use from being in the
 838          * reserved inode, and will become GDT blocks (primary and backup).
 839          */
 840         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 841         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 842         if (unlikely(err)) {
 843                 ext4_std_error(sb, err);
 844                 goto errout;
 845         }
 846         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
 847                            (9 - EXT4_SB(sb)->s_cluster_bits);
 848         ext4_mark_iloc_dirty(handle, inode, &iloc);
 849         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 850         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 851         if (unlikely(err)) {
 852                 ext4_std_error(sb, err);
 853                 iloc.bh = NULL;
 854                 goto errout;
 855         }
 856         brelse(dind);
 857
 858         o_group_desc = EXT4_SB(sb)->s_group_desc;
 859         memcpy(n_group_desc, o_group_desc,
 860                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 861         n_group_desc[gdb_num] = gdb_bh;
 862         EXT4_SB(sb)->s_group_desc = n_group_desc;
 863         EXT4_SB(sb)->s_gdb_count++;
 864         kvfree(o_group_desc);
 865
 866         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 867         err = ext4_handle_dirty_super(handle, sb);
 868         if (err)
 869                 ext4_std_error(sb, err);
 870         return err;
 871 errout:
 872         kvfree(n_group_desc);
 873         brelse(iloc.bh);
 874         brelse(dind);
 875         brelse(gdb_bh);
 876
 877         ext4_debug("leaving with error %d\n", err);
 878         return err;
 879 }
 880
 881 /*
 882  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 883  */
 884 static int add_new_gdb_meta_bg(struct super_block *sb,
 885                                handle_t *handle, ext4_group_t group) {
 886         ext4_fsblk_t gdblock;
 887         struct buffer_head *gdb_bh;
 888         struct buffer_head **o_group_desc, **n_group_desc;
 889         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 890         int err;
 891
 892         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 893                    ext4_bg_has_super(sb, group);
 894         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 895         if (IS_ERR(gdb_bh))
 896                 return PTR_ERR(gdb_bh);
 897         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 898                                      sizeof(struct buffer_head *),
 899                                      GFP_NOFS);
 900         if (!n_group_desc) {
 901                 brelse(gdb_bh);
 902                 err = -ENOMEM;
 903                 ext4_warning(sb, "not enough memory for %lu groups",
 904                              gdb_num + 1);
 905                 return err;
 906         }
 907
 908         o_group_desc = EXT4_SB(sb)->s_group_desc;
 909         memcpy(n_group_desc, o_group_desc,
 910                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 911         n_group_desc[gdb_num] = gdb_bh;
 912
 913         BUFFER_TRACE(gdb_bh, "get_write_access");
 914         err = ext4_journal_get_write_access(handle, gdb_bh);
 915         if (err) {
 916                 kvfree(n_group_desc);
 917                 brelse(gdb_bh);
 918                 return err;
 919         }
 920
 921         EXT4_SB(sb)->s_group_desc = n_group_desc;
 922         EXT4_SB(sb)->s_gdb_count++;
 923         kvfree(o_group_desc);
 924         return err;
 925 }
 926
 927 /*
 928  * Called when we are adding a new group which has a backup copy of each of
 929  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 930  * We need to add these reserved backup GDT blocks to the resize inode, so
 931  * that they are kept for future resizing and not allocated to files.
 932  *
 933  * Each reserved backup GDT block will go into a different indirect block.
 934  * The indirect blocks are actually the primary reserved GDT blocks,
 935  * so we know in advance what their block numbers are.  We only get the
 936  * double-indirect block to verify it is pointing to the primary reserved
 937  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 938  * backup GDT blocks are stored in their reserved primary GDT block.
 939  */
 940 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 941                               ext4_group_t group)
 942 {
 943         struct super_block *sb = inode->i_sb;
 944         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 945         int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
 946         struct buffer_head **primary;
 947         struct buffer_head *dind;
 948         struct ext4_iloc iloc;
 949         ext4_fsblk_t blk;
 950         __le32 *data, *end;
 951         int gdbackups = 0;
 952         int res, i;
 953         int err;
 954
 955         primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
 956         if (!primary)
 957                 return -ENOMEM;
 958
 959         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 960         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
 961         if (IS_ERR(dind)) {
 962                 err = PTR_ERR(dind);
 963                 dind = NULL;
 964                 goto exit_free;
 965         }
 966
 967         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 968         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 969                                          EXT4_ADDR_PER_BLOCK(sb));
 970         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 971
 972         /* Get each reserved primary GDT block and verify it holds backups */
 973         for (res = 0; res < reserved_gdb; res++, blk++) {
 974                 if (le32_to_cpu(*data) != blk) {
 975                         ext4_warning(sb, "reserved block %llu"
 976                                      " not at offset %ld",
 977                                      blk,
 978                                      (long)(data - (__le32 *)dind->b_data));
 979                         err = -EINVAL;
 980                         goto exit_bh;
 981                 }
 982                 primary[res] = ext4_sb_bread(sb, blk, 0);
 983                 if (IS_ERR(primary[res])) {
 984                         err = PTR_ERR(primary[res]);
 985                         primary[res] = NULL;
 986                         goto exit_bh;
 987                 }
 988                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 989                 if (gdbackups < 0) {
 990                         brelse(primary[res]);
 991                         err = gdbackups;
 992                         goto exit_bh;
 993                 }
 994                 if (++data >= end)
 995                         data = (__le32 *)dind->b_data;
 996         }
 997
 998         for (i = 0; i < reserved_gdb; i++) {
 999                 BUFFER_TRACE(primary[i], "get_write_access");
1000                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1001                         goto exit_bh;
1002         }
1003
1004         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1005                 goto exit_bh;
1006
1007         /*
1008          * Finally we can add each of the reserved backup GDT blocks from
1009          * the new group to its reserved primary GDT block.
1010          */
1011         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1012         for (i = 0; i < reserved_gdb; i++) {
1013                 int err2;
1014                 data = (__le32 *)primary[i]->b_data;
1015                 /* printk("reserving backup %lu[%u] = %lu\n",
1016                        primary[i]->b_blocknr, gdbackups,
1017                        blk + primary[i]->b_blocknr); */
1018                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1019                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1020                 if (!err)
1021                         err = err2;
1022         }
1023
1024         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1025         ext4_mark_iloc_dirty(handle, inode, &iloc);
1026
1027 exit_bh:
1028         while (--res >= 0)
1029                 brelse(primary[res]);
1030         brelse(dind);
1031
1032 exit_free:
1033         kfree(primary);
1034
1035         return err;
1036 }
1037
1038 /*
1039  * Update the backup copies of the ext4 metadata.  These don't need to be part
1040  * of the main resize transaction, because e2fsck will re-write them if there
1041  * is a problem (basically only OOM will cause a problem).  However, we
1042  * _should_ update the backups if possible, in case the primary gets trashed
1043  * for some reason and we need to run e2fsck from a backup superblock.  The
1044  * important part is that the new block and inode counts are in the backup
1045  * superblocks, and the location of the new group metadata in the GDT backups.
1046  *
1047  * We do not need take the s_resize_lock for this, because these
1048  * blocks are not otherwise touched by the filesystem code when it is
1049  * mounted.  We don't need to worry about last changing from
1050  * sbi->s_groups_count, because the worst that can happen is that we
1051  * do not copy the full number of backups at this time.  The resize
1052  * which changed s_groups_count will backup again.
1053  */
1054 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1055                            int size, int meta_bg)
1056 {
1057         struct ext4_sb_info *sbi = EXT4_SB(sb);
1058         ext4_group_t last;
1059         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1060         unsigned three = 1;
1061         unsigned five = 5;
1062         unsigned seven = 7;
1063         ext4_group_t group = 0;
1064         int rest = sb->s_blocksize - size;
1065         handle_t *handle;
1066         int err = 0, err2;
1067
1068         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1069         if (IS_ERR(handle)) {
1070                 group = 1;
1071                 err = PTR_ERR(handle);
1072                 goto exit_err;
1073         }
1074
1075         if (meta_bg == 0) {
1076                 group = ext4_list_backups(sb, &three, &five, &seven);
1077                 last = sbi->s_groups_count;
1078         } else {
1079                 group = ext4_get_group_number(sb, blk_off) + 1;
1080                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1081         }
1082
1083         while (group < sbi->s_groups_count) {
1084                 struct buffer_head *bh;
1085                 ext4_fsblk_t backup_block;
1086
1087                 /* Out of journal space, and can't get more - abort - so sad */
1088                 err = ext4_journal_ensure_credits_batch(handle, 1);
1089                 if (err < 0)
1090                         break;
1091
1092                 if (meta_bg == 0)
1093                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1094                 else
1095                         backup_block = (ext4_group_first_block_no(sb, group) +
1096                                         ext4_bg_has_super(sb, group));
1097
1098                 bh = sb_getblk(sb, backup_block);
1099                 if (unlikely(!bh)) {
1100                         err = -ENOMEM;
1101                         break;
1102                 }
1103                 ext4_debug("update metadata backup %llu(+%llu)\n",
1104                            backup_block, backup_block -
1105                            ext4_group_first_block_no(sb, group));
1106                 BUFFER_TRACE(bh, "get_write_access");
1107                 if ((err = ext4_journal_get_write_access(handle, bh))) {
1108                         brelse(bh);
1109                         break;
1110                 }
1111                 lock_buffer(bh);
1112                 memcpy(bh->b_data, data, size);
1113                 if (rest)
1114                         memset(bh->b_data + size, 0, rest);
1115                 set_buffer_uptodate(bh);
1116                 unlock_buffer(bh);
1117                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1118                 if (unlikely(err))
1119                         ext4_std_error(sb, err);
1120                 brelse(bh);
1121
1122                 if (meta_bg == 0)
1123                         group = ext4_list_backups(sb, &three, &five, &seven);
1124                 else if (group == last)
1125                         break;
1126                 else
1127                         group = last;
1128         }
1129         if ((err2 = ext4_journal_stop(handle)) && !err)
1130                 err = err2;
1131
1132         /*
1133          * Ugh! Need to have e2fsck write the backup copies.  It is too
1134          * late to revert the resize, we shouldn't fail just because of
1135          * the backup copies (they are only needed in case of corruption).
1136          *
1137          * However, if we got here we have a journal problem too, so we
1138          * can't really start a transaction to mark the superblock.
1139          * Chicken out and just set the flag on the hope it will be written
1140          * to disk, and if not - we will simply wait until next fsck.
1141          */
1142 exit_err:
1143         if (err) {
1144                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1145                              "forcing fsck on next reboot", group, err);
1146                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1147                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1148                 mark_buffer_dirty(sbi->s_sbh);
1149         }
1150 }
1151
1152 /*
1153  * ext4_add_new_descs() adds @count group descriptor of groups
1154  * starting at @group
1155  *
1156  * @handle: journal handle
1157  * @sb: super block
1158  * @group: the group no. of the first group desc to be added
1159  * @resize_inode: the resize inode
1160  * @count: number of group descriptors to be added
1161  */
1162 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1163                               ext4_group_t group, struct inode *resize_inode,
1164                               ext4_group_t count)
1165 {
1166         struct ext4_sb_info *sbi = EXT4_SB(sb);
1167         struct ext4_super_block *es = sbi->s_es;
1168         struct buffer_head *gdb_bh;
1169         int i, gdb_off, gdb_num, err = 0;
1170         int meta_bg;
1171
1172         meta_bg = ext4_has_feature_meta_bg(sb);
1173         for (i = 0; i < count; i++, group++) {
1174                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1175                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1176
1177                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1178                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1179
1180                 /*
1181                  * We will only either add reserved group blocks to a backup group
1182                  * or remove reserved blocks for the first group in a new group block.
1183                  * Doing both would be mean more complex code, and sane people don't
1184                  * use non-sparse filesystems anymore.  This is already checked above.
1185                  */
1186                 if (gdb_off) {
1187                         gdb_bh = sbi->s_group_desc[gdb_num];
1188                         BUFFER_TRACE(gdb_bh, "get_write_access");
1189                         err = ext4_journal_get_write_access(handle, gdb_bh);
1190
1191                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1192                                 err = reserve_backup_gdb(handle, resize_inode, group);
1193                 } else if (meta_bg != 0) {
1194                         err = add_new_gdb_meta_bg(sb, handle, group);
1195                 } else {
1196                         err = add_new_gdb(handle, resize_inode, group);
1197                 }
1198                 if (err)
1199                         break;
1200         }
1201         return err;
1202 }
1203
1204 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1205 {
1206         struct buffer_head *bh = sb_getblk(sb, block);
1207         if (unlikely(!bh))
1208                 return NULL;
1209         if (!bh_uptodate_or_lock(bh)) {
1210                 if (bh_submit_read(bh) < 0) {
1211                         brelse(bh);
1212                         return NULL;
1213                 }
1214         }
1215
1216         return bh;
1217 }
1218
1219 static int ext4_set_bitmap_checksums(struct super_block *sb,
1220                                      ext4_group_t group,
1221                                      struct ext4_group_desc *gdp,
1222                                      struct ext4_new_group_data *group_data)
1223 {
1224         struct buffer_head *bh;
1225
1226         if (!ext4_has_metadata_csum(sb))
1227                 return 0;
1228
1229         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1230         if (!bh)
1231                 return -EIO;
1232         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1233                                    EXT4_INODES_PER_GROUP(sb) / 8);
1234         brelse(bh);
1235
1236         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1237         if (!bh)
1238                 return -EIO;
1239         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1240         brelse(bh);
1241
1242         return 0;
1243 }
1244
1245 /*
1246  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1247  */
1248 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1249                                 struct ext4_new_flex_group_data *flex_gd)
1250 {
1251         struct ext4_new_group_data      *group_data = flex_gd->groups;
1252         struct ext4_group_desc          *gdp;
1253         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1254         struct buffer_head              *gdb_bh;
1255         ext4_group_t                    group;
1256         __u16                           *bg_flags = flex_gd->bg_flags;
1257         int                             i, gdb_off, gdb_num, err = 0;
1258
1259
1260         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1261                 group = group_data->group;
1262
1263                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1264                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1265
1266                 /*
1267                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1268                  */
1269                 gdb_bh = sbi->s_group_desc[gdb_num];
1270                 /* Update group descriptor block for new group */
1271                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1272                                                  gdb_off * EXT4_DESC_SIZE(sb));
1273
1274                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1275                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1276                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1277                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1278                 if (err) {
1279                         ext4_std_error(sb, err);
1280                         break;
1281                 }
1282
1283                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1284                 ext4_free_group_clusters_set(sb, gdp,
1285                                              group_data->free_clusters_count);
1286                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1287                 if (ext4_has_group_desc_csum(sb))
1288                         ext4_itable_unused_set(sb, gdp,
1289                                                EXT4_INODES_PER_GROUP(sb));
1290                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1291                 ext4_group_desc_csum_set(sb, group, gdp);
1292
1293                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1294                 if (unlikely(err)) {
1295                         ext4_std_error(sb, err);
1296                         break;
1297                 }
1298
1299                 /*
1300                  * We can allocate memory for mb_alloc based on the new group
1301                  * descriptor
1302                  */
1303                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1304                 if (err)
1305                         break;
1306         }
1307         return err;
1308 }
1309
1310 /*
1311  * ext4_update_super() updates the super block so that the newly added
1312  * groups can be seen by the filesystem.
1313  *
1314  * @sb: super block
1315  * @flex_gd: new added groups
1316  */
1317 static void ext4_update_super(struct super_block *sb,
1318                              struct ext4_new_flex_group_data *flex_gd)
1319 {
1320         ext4_fsblk_t blocks_count = 0;
1321         ext4_fsblk_t free_blocks = 0;
1322         ext4_fsblk_t reserved_blocks = 0;
1323         struct ext4_new_group_data *group_data = flex_gd->groups;
1324         struct ext4_sb_info *sbi = EXT4_SB(sb);
1325         struct ext4_super_block *es = sbi->s_es;
1326         int i;
1327
1328         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1329         /*
1330          * Make the new blocks and inodes valid next.  We do this before
1331          * increasing the group count so that once the group is enabled,
1332          * all of its blocks and inodes are already valid.
1333          *
1334          * We always allocate group-by-group, then block-by-block or
1335          * inode-by-inode within a group, so enabling these
1336          * blocks/inodes before the group is live won't actually let us
1337          * allocate the new space yet.
1338          */
1339         for (i = 0; i < flex_gd->count; i++) {
1340                 blocks_count += group_data[i].blocks_count;
1341                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1342         }
1343
1344         reserved_blocks = ext4_r_blocks_count(es) * 100;
1345         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1346         reserved_blocks *= blocks_count;
1347         do_div(reserved_blocks, 100);
1348
1349         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1350         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1351         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1352                      flex_gd->count);
1353         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1354                      flex_gd->count);
1355
1356         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1357         /*
1358          * We need to protect s_groups_count against other CPUs seeing
1359          * inconsistent state in the superblock.
1360          *
1361          * The precise rules we use are:
1362          *
1363          * * Writers must perform a smp_wmb() after updating all
1364          *   dependent data and before modifying the groups count
1365          *
1366          * * Readers must perform an smp_rmb() after reading the groups
1367          *   count and before reading any dependent data.
1368          *
1369          * NB. These rules can be relaxed when checking the group count
1370          * while freeing data, as we can only allocate from a block
1371          * group after serialising against the group count, and we can
1372          * only then free after serialising in turn against that
1373          * allocation.
1374          */
1375         smp_wmb();
1376
1377         /* Update the global fs size fields */
1378         sbi->s_groups_count += flex_gd->count;
1379         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1380                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1381
1382         /* Update the reserved block counts only once the new group is
1383          * active. */
1384         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1385                                 reserved_blocks);
1386
1387         /* Update the free space counts */
1388         percpu_counter_add(&sbi->s_freeclusters_counter,
1389                            EXT4_NUM_B2C(sbi, free_blocks));
1390         percpu_counter_add(&sbi->s_freeinodes_counter,
1391                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1392
1393         ext4_debug("free blocks count %llu",
1394                    percpu_counter_read(&sbi->s_freeclusters_counter));
1395         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1396                 ext4_group_t flex_group;
1397                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1398                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1399                              &sbi->s_flex_groups[flex_group].free_clusters);
1400                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1401                            &sbi->s_flex_groups[flex_group].free_inodes);
1402         }
1403
1404         /*
1405          * Update the fs overhead information
1406          */
1407         ext4_calculate_overhead(sb);
1408
1409         if (test_opt(sb, DEBUG))
1410                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1411                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1412                        blocks_count, free_blocks, reserved_blocks);
1413 }
1414
1415 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1416  * _before_ we start modifying the filesystem, because we cannot abort the
1417  * transaction and not have it write the data to disk.
1418  */
1419 static int ext4_flex_group_add(struct super_block *sb,
1420                                struct inode *resize_inode,
1421                                struct ext4_new_flex_group_data *flex_gd)
1422 {
1423         struct ext4_sb_info *sbi = EXT4_SB(sb);
1424         struct ext4_super_block *es = sbi->s_es;
1425         ext4_fsblk_t o_blocks_count;
1426         ext4_grpblk_t last;
1427         ext4_group_t group;
1428         handle_t *handle;
1429         unsigned reserved_gdb;
1430         int err = 0, err2 = 0, credit;
1431
1432         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1433
1434         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1435         o_blocks_count = ext4_blocks_count(es);
1436         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1437         BUG_ON(last);
1438
1439         err = setup_new_flex_group_blocks(sb, flex_gd);
1440         if (err)
1441                 goto exit;
1442         /*
1443          * We will always be modifying at least the superblock and  GDT
1444          * blocks.  If we are adding a group past the last current GDT block,
1445          * we will also modify the inode and the dindirect block.  If we
1446          * are adding a group with superblock/GDT backups  we will also
1447          * modify each of the reserved GDT dindirect blocks.
1448          */
1449         credit = 3;     /* sb, resize inode, resize inode dindirect */
1450         /* GDT blocks */
1451         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1452         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1453         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1454         if (IS_ERR(handle)) {
1455                 err = PTR_ERR(handle);
1456                 goto exit;
1457         }
1458
1459         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1460         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1461         if (err)
1462                 goto exit_journal;
1463
1464         group = flex_gd->groups[0].group;
1465         BUG_ON(group != sbi->s_groups_count);
1466         err = ext4_add_new_descs(handle, sb, group,
1467                                 resize_inode, flex_gd->count);
1468         if (err)
1469                 goto exit_journal;
1470
1471         err = ext4_setup_new_descs(handle, sb, flex_gd);
1472         if (err)
1473                 goto exit_journal;
1474
1475         ext4_update_super(sb, flex_gd);
1476
1477         err = ext4_handle_dirty_super(handle, sb);
1478
1479 exit_journal:
1480         err2 = ext4_journal_stop(handle);
1481         if (!err)
1482                 err = err2;
1483
1484         if (!err) {
1485                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1486                 int gdb_num_end = ((group + flex_gd->count - 1) /
1487                                    EXT4_DESC_PER_BLOCK(sb));
1488                 int meta_bg = ext4_has_feature_meta_bg(sb);
1489                 sector_t old_gdb = 0;
1490
1491                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1492                                sizeof(struct ext4_super_block), 0);
1493                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1494                         struct buffer_head *gdb_bh;
1495
1496                         gdb_bh = sbi->s_group_desc[gdb_num];
1497                         if (old_gdb == gdb_bh->b_blocknr)
1498                                 continue;
1499                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1500                                        gdb_bh->b_size, meta_bg);
1501                         old_gdb = gdb_bh->b_blocknr;
1502                 }
1503         }
1504 exit:
1505         return err;
1506 }
1507
1508 static int ext4_setup_next_flex_gd(struct super_block *sb,
1509                                     struct ext4_new_flex_group_data *flex_gd,
1510                                     ext4_fsblk_t n_blocks_count,
1511                                     unsigned long flexbg_size)
1512 {
1513         struct ext4_sb_info *sbi = EXT4_SB(sb);
1514         struct ext4_super_block *es = sbi->s_es;
1515         struct ext4_new_group_data *group_data = flex_gd->groups;
1516         ext4_fsblk_t o_blocks_count;
1517         ext4_group_t n_group;
1518         ext4_group_t group;
1519         ext4_group_t last_group;
1520         ext4_grpblk_t last;
1521         ext4_grpblk_t clusters_per_group;
1522         unsigned long i;
1523
1524         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1525
1526         o_blocks_count = ext4_blocks_count(es);
1527
1528         if (o_blocks_count == n_blocks_count)
1529                 return 0;
1530
1531         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1532         BUG_ON(last);
1533         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1534
1535         last_group = group | (flexbg_size - 1);
1536         if (last_group > n_group)
1537                 last_group = n_group;
1538
1539         flex_gd->count = last_group - group + 1;
1540
1541         for (i = 0; i < flex_gd->count; i++) {
1542                 int overhead;
1543
1544                 group_data[i].group = group + i;
1545                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1546                 overhead = ext4_group_overhead_blocks(sb, group + i);
1547                 group_data[i].mdata_blocks = overhead;
1548                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1549                 if (ext4_has_group_desc_csum(sb)) {
1550                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1551                                                EXT4_BG_INODE_UNINIT;
1552                         if (!test_opt(sb, INIT_INODE_TABLE))
1553                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1554                 } else
1555                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1556         }
1557
1558         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1559                 /* We need to initialize block bitmap of last group. */
1560                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1561
1562         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1563                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1564                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1565                                                        last - 1;
1566         }
1567
1568         return 1;
1569 }
1570
1571 /* Add group descriptor data to an existing or new group descriptor block.
1572  * Ensure we handle all possible error conditions _before_ we start modifying
1573  * the filesystem, because we cannot abort the transaction and not have it
1574  * write the data to disk.
1575  *
1576  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1577  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1578  *
1579  * We only need to hold the superblock lock while we are actually adding
1580  * in the new group's counts to the superblock.  Prior to that we have
1581  * not really "added" the group at all.  We re-check that we are still
1582  * adding in the last group in case things have changed since verifying.
1583  */
1584 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1585 {
1586         struct ext4_new_flex_group_data flex_gd;
1587         struct ext4_sb_info *sbi = EXT4_SB(sb);
1588         struct ext4_super_block *es = sbi->s_es;
1589         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1590                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1591         struct inode *inode = NULL;
1592         int gdb_off;
1593         int err;
1594         __u16 bg_flags = 0;
1595
1596         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1597
1598         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1599                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1600                 return -EPERM;
1601         }
1602
1603         if (ext4_blocks_count(es) + input->blocks_count <
1604             ext4_blocks_count(es)) {
1605                 ext4_warning(sb, "blocks_count overflow");
1606                 return -EINVAL;
1607         }
1608
1609         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1610             le32_to_cpu(es->s_inodes_count)) {
1611                 ext4_warning(sb, "inodes_count overflow");
1612                 return -EINVAL;
1613         }
1614
1615         if (reserved_gdb || gdb_off == 0) {
1616                 if (!ext4_has_feature_resize_inode(sb) ||
1617                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1618                         ext4_warning(sb,
1619                                      "No reserved GDT blocks, can't resize");
1620                         return -EPERM;
1621                 }
1622                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1623                 if (IS_ERR(inode)) {
1624                         ext4_warning(sb, "Error opening resize inode");
1625                         return PTR_ERR(inode);
1626                 }
1627         }
1628
1629
1630         err = verify_group_input(sb, input);
1631         if (err)
1632                 goto out;
1633
1634         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1635         if (err)
1636                 goto out;
1637
1638         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1639         if (err)
1640                 goto out;
1641
1642         flex_gd.count = 1;
1643         flex_gd.groups = input;
1644         flex_gd.bg_flags = &bg_flags;
1645         err = ext4_flex_group_add(sb, inode, &flex_gd);
1646 out:
1647         iput(inode);
1648         return err;
1649 } /* ext4_group_add */
1650
1651 /*
1652  * extend a group without checking assuming that checking has been done.
1653  */
1654 static int ext4_group_extend_no_check(struct super_block *sb,
1655                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1656 {
1657         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1658         handle_t *handle;
1659         int err = 0, err2;
1660
1661         /* We will update the superblock, one block bitmap, and
1662          * one group descriptor via ext4_group_add_blocks().
1663          */
1664         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1665         if (IS_ERR(handle)) {
1666                 err = PTR_ERR(handle);
1667                 ext4_warning(sb, "error %d on journal start", err);
1668                 return err;
1669         }
1670
1671         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1672         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1673         if (err) {
1674                 ext4_warning(sb, "error %d on journal write access", err);
1675                 goto errout;
1676         }
1677
1678         ext4_blocks_count_set(es, o_blocks_count + add);
1679         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1680         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1681                    o_blocks_count + add);
1682         /* We add the blocks to the bitmap and set the group need init bit */
1683         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1684         if (err)
1685                 goto errout;
1686         ext4_handle_dirty_super(handle, sb);
1687         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1688                    o_blocks_count + add);
1689 errout:
1690         err2 = ext4_journal_stop(handle);
1691         if (err2 && !err)
1692                 err = err2;
1693
1694         if (!err) {
1695                 if (test_opt(sb, DEBUG))
1696                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1697                                "blocks\n", ext4_blocks_count(es));
1698                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1699                                (char *)es, sizeof(struct ext4_super_block), 0);
1700         }
1701         return err;
1702 }
1703
1704 /*
1705  * Extend the filesystem to the new number of blocks specified.  This entry
1706  * point is only used to extend the current filesystem to the end of the last
1707  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1708  * for emergencies (because it has no dependencies on reserved blocks).
1709  *
1710  * If we _really_ wanted, we could use default values to call ext4_group_add()
1711  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1712  * GDT blocks are reserved to grow to the desired size.
1713  */
1714 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1715                       ext4_fsblk_t n_blocks_count)
1716 {
1717         ext4_fsblk_t o_blocks_count;
1718         ext4_grpblk_t last;
1719         ext4_grpblk_t add;
1720         struct buffer_head *bh;
1721         int err;
1722         ext4_group_t group;
1723
1724         o_blocks_count = ext4_blocks_count(es);
1725
1726         if (test_opt(sb, DEBUG))
1727                 ext4_msg(sb, KERN_DEBUG,
1728                          "extending last group from %llu to %llu blocks",
1729                          o_blocks_count, n_blocks_count);
1730
1731         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1732                 return 0;
1733
1734         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1735                 ext4_msg(sb, KERN_ERR,
1736                          "filesystem too large to resize to %llu blocks safely",
1737                          n_blocks_count);
1738                 return -EINVAL;
1739         }
1740
1741         if (n_blocks_count < o_blocks_count) {
1742                 ext4_warning(sb, "can't shrink FS - resize aborted");
1743                 return -EINVAL;
1744         }
1745
1746         /* Handle the remaining blocks in the last group only. */
1747         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1748
1749         if (last == 0) {
1750                 ext4_warning(sb, "need to use ext2online to resize further");
1751                 return -EPERM;
1752         }
1753
1754         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1755
1756         if (o_blocks_count + add < o_blocks_count) {
1757                 ext4_warning(sb, "blocks_count overflow");
1758                 return -EINVAL;
1759         }
1760
1761         if (o_blocks_count + add > n_blocks_count)
1762                 add = n_blocks_count - o_blocks_count;
1763
1764         if (o_blocks_count + add < n_blocks_count)
1765                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1766                              o_blocks_count + add, add);
1767
1768         /* See if the device is actually as big as what was requested */
1769         bh = sb_bread(sb, o_blocks_count + add - 1);
1770         if (!bh) {
1771                 ext4_warning(sb, "can't read last block, resize aborted");
1772                 return -ENOSPC;
1773         }
1774         brelse(bh);
1775
1776         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1777         return err;
1778 } /* ext4_group_extend */
1779
1780
1781 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1782 {
1783         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1784 }
1785
1786 /*
1787  * Release the resize inode and drop the resize_inode feature if there
1788  * are no more reserved gdt blocks, and then convert the file system
1789  * to enable meta_bg
1790  */
1791 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1792 {
1793         handle_t *handle;
1794         struct ext4_sb_info *sbi = EXT4_SB(sb);
1795         struct ext4_super_block *es = sbi->s_es;
1796         struct ext4_inode_info *ei = EXT4_I(inode);
1797         ext4_fsblk_t nr;
1798         int i, ret, err = 0;
1799         int credits = 1;
1800
1801         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1802         if (inode) {
1803                 if (es->s_reserved_gdt_blocks) {
1804                         ext4_error(sb, "Unexpected non-zero "
1805                                    "s_reserved_gdt_blocks");
1806                         return -EPERM;
1807                 }
1808
1809                 /* Do a quick sanity check of the resize inode */
1810                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1811                                              (9 - sbi->s_cluster_bits)))
1812                         goto invalid_resize_inode;
1813                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1814                         if (i == EXT4_DIND_BLOCK) {
1815                                 if (ei->i_data[i])
1816                                         continue;
1817                                 else
1818                                         goto invalid_resize_inode;
1819                         }
1820                         if (ei->i_data[i])
1821                                 goto invalid_resize_inode;
1822                 }
1823                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1824         }
1825
1826         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1827         if (IS_ERR(handle))
1828                 return PTR_ERR(handle);
1829
1830         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1831         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1832         if (err)
1833                 goto errout;
1834
1835         ext4_clear_feature_resize_inode(sb);
1836         ext4_set_feature_meta_bg(sb);
1837         sbi->s_es->s_first_meta_bg =
1838                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1839
1840         err = ext4_handle_dirty_super(handle, sb);
1841         if (err) {
1842                 ext4_std_error(sb, err);
1843                 goto errout;
1844         }
1845
1846         if (inode) {
1847                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1848                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1849                                  EXT4_FREE_BLOCKS_METADATA |
1850                                  EXT4_FREE_BLOCKS_FORGET);
1851                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1852                 inode->i_blocks = 0;
1853
1854                 err = ext4_mark_inode_dirty(handle, inode);
1855                 if (err)
1856                         ext4_std_error(sb, err);
1857         }
1858
1859 errout:
1860         ret = ext4_journal_stop(handle);
1861         if (!err)
1862                 err = ret;
1863         return ret;
1864
1865 invalid_resize_inode:
1866         ext4_error(sb, "corrupted/inconsistent resize inode");
1867         return -EINVAL;
1868 }
1869
1870 /*
1871  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1872  *
1873  * @sb: super block of the fs to be resized
1874  * @n_blocks_count: the number of blocks resides in the resized fs
1875  */
1876 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1877 {
1878         struct ext4_new_flex_group_data *flex_gd = NULL;
1879         struct ext4_sb_info *sbi = EXT4_SB(sb);
1880         struct ext4_super_block *es = sbi->s_es;
1881         struct buffer_head *bh;
1882         struct inode *resize_inode = NULL;
1883         ext4_grpblk_t add, offset;
1884         unsigned long n_desc_blocks;
1885         unsigned long o_desc_blocks;
1886         ext4_group_t o_group;
1887         ext4_group_t n_group;
1888         ext4_fsblk_t o_blocks_count;
1889         ext4_fsblk_t n_blocks_count_retry = 0;
1890         unsigned long last_update_time = 0;
1891         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1892         int meta_bg;
1893
1894         /* See if the device is actually as big as what was requested */
1895         bh = sb_bread(sb, n_blocks_count - 1);
1896         if (!bh) {
1897                 ext4_warning(sb, "can't read last block, resize aborted");
1898                 return -ENOSPC;
1899         }
1900         brelse(bh);
1901
1902 retry:
1903         o_blocks_count = ext4_blocks_count(es);
1904
1905         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1906                  "to %llu blocks", o_blocks_count, n_blocks_count);
1907
1908         if (n_blocks_count < o_blocks_count) {
1909                 /* On-line shrinking not supported */
1910                 ext4_warning(sb, "can't shrink FS - resize aborted");
1911                 return -EINVAL;
1912         }
1913
1914         if (n_blocks_count == o_blocks_count)
1915                 /* Nothing need to do */
1916                 return 0;
1917
1918         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1919         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1920                 ext4_warning(sb, "resize would cause inodes_count overflow");
1921                 return -EINVAL;
1922         }
1923         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1924
1925         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1926         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1927
1928         meta_bg = ext4_has_feature_meta_bg(sb);
1929
1930         if (ext4_has_feature_resize_inode(sb)) {
1931                 if (meta_bg) {
1932                         ext4_error(sb, "resize_inode and meta_bg enabled "
1933                                    "simultaneously");
1934                         return -EINVAL;
1935                 }
1936                 if (n_desc_blocks > o_desc_blocks +
1937                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
1938                         n_blocks_count_retry = n_blocks_count;
1939                         n_desc_blocks = o_desc_blocks +
1940                                 le16_to_cpu(es->s_reserved_gdt_blocks);
1941                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
1942                         n_blocks_count = (ext4_fsblk_t)n_group *
1943                                 EXT4_BLOCKS_PER_GROUP(sb) +
1944                                 le32_to_cpu(es->s_first_data_block);
1945                         n_group--; /* set to last group number */
1946                 }
1947
1948                 if (!resize_inode)
1949                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
1950                                                  EXT4_IGET_SPECIAL);
1951                 if (IS_ERR(resize_inode)) {
1952                         ext4_warning(sb, "Error opening resize inode");
1953                         return PTR_ERR(resize_inode);
1954                 }
1955         }
1956
1957         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
1958                 err = ext4_convert_meta_bg(sb, resize_inode);
1959                 if (err)
1960                         goto out;
1961                 if (resize_inode) {
1962                         iput(resize_inode);
1963                         resize_inode = NULL;
1964                 }
1965                 if (n_blocks_count_retry) {
1966                         n_blocks_count = n_blocks_count_retry;
1967                         n_blocks_count_retry = 0;
1968                         goto retry;
1969                 }
1970         }
1971
1972         /*
1973          * Make sure the last group has enough space so that it's
1974          * guaranteed to have enough space for all metadata blocks
1975          * that it might need to hold.  (We might not need to store
1976          * the inode table blocks in the last block group, but there
1977          * will be cases where this might be needed.)
1978          */
1979         if ((ext4_group_first_block_no(sb, n_group) +
1980              ext4_group_overhead_blocks(sb, n_group) + 2 +
1981              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
1982                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
1983                 n_group--;
1984                 n_blocks_count_retry = 0;
1985                 if (resize_inode) {
1986                         iput(resize_inode);
1987                         resize_inode = NULL;
1988                 }
1989                 goto retry;
1990         }
1991
1992         /* extend the last group */
1993         if (n_group == o_group)
1994                 add = n_blocks_count - o_blocks_count;
1995         else
1996                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
1997         if (add > 0) {
1998                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1999                 if (err)
2000                         goto out;
2001         }
2002
2003         if (ext4_blocks_count(es) == n_blocks_count)
2004                 goto out;
2005
2006         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2007         if (err)
2008                 goto out;
2009
2010         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2011         if (err)
2012                 goto out;
2013
2014         flex_gd = alloc_flex_gd(flexbg_size);
2015         if (flex_gd == NULL) {
2016                 err = -ENOMEM;
2017                 goto out;
2018         }
2019
2020         /* Add flex groups. Note that a regular group is a
2021          * flex group with 1 group.
2022          */
2023         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2024                                               flexbg_size)) {
2025                 if (jiffies - last_update_time > HZ * 10) {
2026                         if (last_update_time)
2027                                 ext4_msg(sb, KERN_INFO,
2028                                          "resized to %llu blocks",
2029                                          ext4_blocks_count(es));
2030                         last_update_time = jiffies;
2031                 }
2032                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2033                         break;
2034                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2035                 if (unlikely(err))
2036                         break;
2037         }
2038
2039         if (!err && n_blocks_count_retry) {
2040                 n_blocks_count = n_blocks_count_retry;
2041                 n_blocks_count_retry = 0;
2042                 free_flex_gd(flex_gd);
2043                 flex_gd = NULL;
2044                 if (resize_inode) {
2045                         iput(resize_inode);
2046                         resize_inode = NULL;
2047                 }
2048                 goto retry;
2049         }
2050
2051 out:
2052         if (flex_gd)
2053                 free_flex_gd(flex_gd);
2054         if (resize_inode != NULL)
2055                 iput(resize_inode);
2056         if (err)
2057                 ext4_warning(sb, "error (%d) occurred during "
2058                              "file system resize", err);
2059         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2060                  ext4_blocks_count(es));
2061         return err;
2062 }