fs/ext4/resize.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
ext4_rcu_ptr_callback
ext4_kvfree_array_rcu
ext4_resize_begin
ext4_resize_end
ext4_meta_bg_first_group
ext4_meta_bg_first_block_no
ext4_group_overhead_blocks
verify_group_input
alloc_flex_gd
free_flex_gd
ext4_alloc_group_tables
bclean
extend_or_restart_transaction
set_flexbg_block_bitmap
setup_new_flex_group_blocks
ext4_list_backups
verify_reserved_gdb
add_new_gdb
add_new_gdb_meta_bg
reserve_backup_gdb
update_backups
ext4_add_new_descs
ext4_get_bitmap
ext4_set_bitmap_checksums
ext4_setup_new_descs
ext4_update_super
ext4_flex_group_add
ext4_setup_next_flex_gd
ext4_group_add
ext4_group_extend_no_check
ext4_group_extend
num_desc_blocks
ext4_convert_meta_bg
ext4_resize_fs
   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/resize.c
   4  *
   5  * Support for resizing an ext4 filesystem while it is mounted.
   6  *
   7  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   8  *
   9  * This could probably be made into a module, because it is not often in use.
  10  */
  11 
  12 
  13 #define EXT4FS_DEBUG
  14 
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 
  18 #include "ext4_jbd2.h"
  19 
  20 struct ext4_rcu_ptr {
  21         struct rcu_head rcu;
  22         void *ptr;
  23 };
  24 
  25 static void ext4_rcu_ptr_callback(struct rcu_head *head)
  26 {
  27         struct ext4_rcu_ptr *ptr;
  28 
  29         ptr = container_of(head, struct ext4_rcu_ptr, rcu);
  30         kvfree(ptr->ptr);
  31         kfree(ptr);
  32 }
  33 
  34 void ext4_kvfree_array_rcu(void *to_free)
  35 {
  36         struct ext4_rcu_ptr *ptr = kzalloc(sizeof(*ptr), GFP_KERNEL);
  37 
  38         if (ptr) {
  39                 ptr->ptr = to_free;
  40                 call_rcu(&ptr->rcu, ext4_rcu_ptr_callback);
  41                 return;
  42         }
  43         synchronize_rcu();
  44         kvfree(to_free);
  45 }
  46 
  47 int ext4_resize_begin(struct super_block *sb)
  48 {
  49         struct ext4_sb_info *sbi = EXT4_SB(sb);
  50         int ret = 0;
  51 
  52         if (!capable(CAP_SYS_RESOURCE))
  53                 return -EPERM;
  54 
  55         /*
  56          * If we are not using the primary superblock/GDT copy don't resize,
  57          * because the user tools have no way of handling this.  Probably a
  58          * bad time to do it anyways.
  59          */
  60         if (EXT4_B2C(sbi, sbi->s_sbh->b_blocknr) !=
  61             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
  62                 ext4_warning(sb, "won't resize using backup superblock at %llu",
  63                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
  64                 return -EPERM;
  65         }
  66 
  67         /*
  68          * We are not allowed to do online-resizing on a filesystem mounted
  69          * with error, because it can destroy the filesystem easily.
  70          */
  71         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  72                 ext4_warning(sb, "There are errors in the filesystem, "
  73                              "so online resizing is not allowed");
  74                 return -EPERM;
  75         }
  76 
  77         if (test_and_set_bit_lock(EXT4_FLAGS_RESIZING,
  78                                   &EXT4_SB(sb)->s_ext4_flags))
  79                 ret = -EBUSY;
  80 
  81         return ret;
  82 }
  83 
  84 void ext4_resize_end(struct super_block *sb)
  85 {
  86         clear_bit_unlock(EXT4_FLAGS_RESIZING, &EXT4_SB(sb)->s_ext4_flags);
  87         smp_mb__after_atomic();
  88 }
  89 
  90 static ext4_group_t ext4_meta_bg_first_group(struct super_block *sb,
  91                                              ext4_group_t group) {
  92         return (group >> EXT4_DESC_PER_BLOCK_BITS(sb)) <<
  93                EXT4_DESC_PER_BLOCK_BITS(sb);
  94 }
  95 
  96 static ext4_fsblk_t ext4_meta_bg_first_block_no(struct super_block *sb,
  97                                              ext4_group_t group) {
  98         group = ext4_meta_bg_first_group(sb, group);
  99         return ext4_group_first_block_no(sb, group);
 100 }
 101 
 102 static ext4_grpblk_t ext4_group_overhead_blocks(struct super_block *sb,
 103                                                 ext4_group_t group) {
 104         ext4_grpblk_t overhead;
 105         overhead = ext4_bg_num_gdb(sb, group);
 106         if (ext4_bg_has_super(sb, group))
 107                 overhead += 1 +
 108                           le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 109         return overhead;
 110 }
 111 
 112 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
 113 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
 114 
 115 static int verify_group_input(struct super_block *sb,
 116                               struct ext4_new_group_data *input)
 117 {
 118         struct ext4_sb_info *sbi = EXT4_SB(sb);
 119         struct ext4_super_block *es = sbi->s_es;
 120         ext4_fsblk_t start = ext4_blocks_count(es);
 121         ext4_fsblk_t end = start + input->blocks_count;
 122         ext4_group_t group = input->group;
 123         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
 124         unsigned overhead;
 125         ext4_fsblk_t metaend;
 126         struct buffer_head *bh = NULL;
 127         ext4_grpblk_t free_blocks_count, offset;
 128         int err = -EINVAL;
 129 
 130         if (group != sbi->s_groups_count) {
 131                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
 132                              input->group, sbi->s_groups_count);
 133                 return -EINVAL;
 134         }
 135 
 136         overhead = ext4_group_overhead_blocks(sb, group);
 137         metaend = start + overhead;
 138         input->free_clusters_count = free_blocks_count =
 139                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 140 
 141         if (test_opt(sb, DEBUG))
 142                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
 143                        "(%d free, %u reserved)\n",
 144                        ext4_bg_has_super(sb, input->group) ? "normal" :
 145                        "no-super", input->group, input->blocks_count,
 146                        free_blocks_count, input->reserved_blocks);
 147 
 148         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
 149         if (offset != 0)
 150                         ext4_warning(sb, "Last group not full");
 151         else if (input->reserved_blocks > input->blocks_count / 5)
 152                 ext4_warning(sb, "Reserved blocks too high (%u)",
 153                              input->reserved_blocks);
 154         else if (free_blocks_count < 0)
 155                 ext4_warning(sb, "Bad blocks count %u",
 156                              input->blocks_count);
 157         else if (IS_ERR(bh = ext4_sb_bread(sb, end - 1, 0))) {
 158                 err = PTR_ERR(bh);
 159                 bh = NULL;
 160                 ext4_warning(sb, "Cannot read last block (%llu)",
 161                              end - 1);
 162         } else if (outside(input->block_bitmap, start, end))
 163                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
 164                              (unsigned long long)input->block_bitmap);
 165         else if (outside(input->inode_bitmap, start, end))
 166                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
 167                              (unsigned long long)input->inode_bitmap);
 168         else if (outside(input->inode_table, start, end) ||
 169                  outside(itend - 1, start, end))
 170                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 171                              (unsigned long long)input->inode_table, itend - 1);
 172         else if (input->inode_bitmap == input->block_bitmap)
 173                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 174                              (unsigned long long)input->block_bitmap);
 175         else if (inside(input->block_bitmap, input->inode_table, itend))
 176                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 177                              "(%llu-%llu)",
 178                              (unsigned long long)input->block_bitmap,
 179                              (unsigned long long)input->inode_table, itend - 1);
 180         else if (inside(input->inode_bitmap, input->inode_table, itend))
 181                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 182                              "(%llu-%llu)",
 183                              (unsigned long long)input->inode_bitmap,
 184                              (unsigned long long)input->inode_table, itend - 1);
 185         else if (inside(input->block_bitmap, start, metaend))
 186                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 187                              (unsigned long long)input->block_bitmap,
 188                              start, metaend - 1);
 189         else if (inside(input->inode_bitmap, start, metaend))
 190                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 191                              (unsigned long long)input->inode_bitmap,
 192                              start, metaend - 1);
 193         else if (inside(input->inode_table, start, metaend) ||
 194                  inside(itend - 1, start, metaend))
 195                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 196                              "(%llu-%llu)",
 197                              (unsigned long long)input->inode_table,
 198                              itend - 1, start, metaend - 1);
 199         else
 200                 err = 0;
 201         brelse(bh);
 202 
 203         return err;
 204 }
 205 
 206 /*
 207  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 208  * group each time.
 209  */
 210 struct ext4_new_flex_group_data {
 211         struct ext4_new_group_data *groups;     /* new_group_data for groups
 212                                                    in the flex group */
 213         __u16 *bg_flags;                        /* block group flags of groups
 214                                                    in @groups */
 215         ext4_group_t count;                     /* number of groups in @groups
 216                                                  */
 217 };
 218 
 219 /*
 220  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 221  * @flexbg_size.
 222  *
 223  * Returns NULL on failure otherwise address of the allocated structure.
 224  */
 225 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 226 {
 227         struct ext4_new_flex_group_data *flex_gd;
 228 
 229         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 230         if (flex_gd == NULL)
 231                 goto out3;
 232 
 233         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_group_data))
 234                 goto out2;
 235         flex_gd->count = flexbg_size;
 236 
 237         flex_gd->groups = kmalloc_array(flexbg_size,
 238                                         sizeof(struct ext4_new_group_data),
 239                                         GFP_NOFS);
 240         if (flex_gd->groups == NULL)
 241                 goto out2;
 242 
 243         flex_gd->bg_flags = kmalloc_array(flexbg_size, sizeof(__u16),
 244                                           GFP_NOFS);
 245         if (flex_gd->bg_flags == NULL)
 246                 goto out1;
 247 
 248         return flex_gd;
 249 
 250 out1:
 251         kfree(flex_gd->groups);
 252 out2:
 253         kfree(flex_gd);
 254 out3:
 255         return NULL;
 256 }
 257 
 258 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 259 {
 260         kfree(flex_gd->bg_flags);
 261         kfree(flex_gd->groups);
 262         kfree(flex_gd);
 263 }
 264 
 265 /*
 266  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 267  * and inode tables for a flex group.
 268  *
 269  * This function is used by 64bit-resize.  Note that this function allocates
 270  * group tables from the 1st group of groups contained by @flexgd, which may
 271  * be a partial of a flex group.
 272  *
 273  * @sb: super block of fs to which the groups belongs
 274  *
 275  * Returns 0 on a successful allocation of the metadata blocks in the
 276  * block group.
 277  */
 278 static int ext4_alloc_group_tables(struct super_block *sb,
 279                                 struct ext4_new_flex_group_data *flex_gd,
 280                                 int flexbg_size)
 281 {
 282         struct ext4_new_group_data *group_data = flex_gd->groups;
 283         ext4_fsblk_t start_blk;
 284         ext4_fsblk_t last_blk;
 285         ext4_group_t src_group;
 286         ext4_group_t bb_index = 0;
 287         ext4_group_t ib_index = 0;
 288         ext4_group_t it_index = 0;
 289         ext4_group_t group;
 290         ext4_group_t last_group;
 291         unsigned overhead;
 292         __u16 uninit_mask = (flexbg_size > 1) ? ~EXT4_BG_BLOCK_UNINIT : ~0;
 293         int i;
 294 
 295         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 296 
 297         src_group = group_data[0].group;
 298         last_group  = src_group + flex_gd->count - 1;
 299 
 300         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 301                (last_group & ~(flexbg_size - 1))));
 302 next_group:
 303         group = group_data[0].group;
 304         if (src_group >= group_data[0].group + flex_gd->count)
 305                 return -ENOSPC;
 306         start_blk = ext4_group_first_block_no(sb, src_group);
 307         last_blk = start_blk + group_data[src_group - group].blocks_count;
 308 
 309         overhead = ext4_group_overhead_blocks(sb, src_group);
 310 
 311         start_blk += overhead;
 312 
 313         /* We collect contiguous blocks as much as possible. */
 314         src_group++;
 315         for (; src_group <= last_group; src_group++) {
 316                 overhead = ext4_group_overhead_blocks(sb, src_group);
 317                 if (overhead == 0)
 318                         last_blk += group_data[src_group - group].blocks_count;
 319                 else
 320                         break;
 321         }
 322 
 323         /* Allocate block bitmaps */
 324         for (; bb_index < flex_gd->count; bb_index++) {
 325                 if (start_blk >= last_blk)
 326                         goto next_group;
 327                 group_data[bb_index].block_bitmap = start_blk++;
 328                 group = ext4_get_group_number(sb, start_blk - 1);
 329                 group -= group_data[0].group;
 330                 group_data[group].mdata_blocks++;
 331                 flex_gd->bg_flags[group] &= uninit_mask;
 332         }
 333 
 334         /* Allocate inode bitmaps */
 335         for (; ib_index < flex_gd->count; ib_index++) {
 336                 if (start_blk >= last_blk)
 337                         goto next_group;
 338                 group_data[ib_index].inode_bitmap = start_blk++;
 339                 group = ext4_get_group_number(sb, start_blk - 1);
 340                 group -= group_data[0].group;
 341                 group_data[group].mdata_blocks++;
 342                 flex_gd->bg_flags[group] &= uninit_mask;
 343         }
 344 
 345         /* Allocate inode tables */
 346         for (; it_index < flex_gd->count; it_index++) {
 347                 unsigned int itb = EXT4_SB(sb)->s_itb_per_group;
 348                 ext4_fsblk_t next_group_start;
 349 
 350                 if (start_blk + itb > last_blk)
 351                         goto next_group;
 352                 group_data[it_index].inode_table = start_blk;
 353                 group = ext4_get_group_number(sb, start_blk);
 354                 next_group_start = ext4_group_first_block_no(sb, group + 1);
 355                 group -= group_data[0].group;
 356 
 357                 if (start_blk + itb > next_group_start) {
 358                         flex_gd->bg_flags[group + 1] &= uninit_mask;
 359                         overhead = start_blk + itb - next_group_start;
 360                         group_data[group + 1].mdata_blocks += overhead;
 361                         itb -= overhead;
 362                 }
 363 
 364                 group_data[group].mdata_blocks += itb;
 365                 flex_gd->bg_flags[group] &= uninit_mask;
 366                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 367         }
 368 
 369         /* Update free clusters count to exclude metadata blocks */
 370         for (i = 0; i < flex_gd->count; i++) {
 371                 group_data[i].free_clusters_count -=
 372                                 EXT4_NUM_B2C(EXT4_SB(sb),
 373                                              group_data[i].mdata_blocks);
 374         }
 375 
 376         if (test_opt(sb, DEBUG)) {
 377                 int i;
 378                 group = group_data[0].group;
 379 
 380                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 381                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 382                        flexbg_size);
 383 
 384                 for (i = 0; i < flex_gd->count; i++) {
 385                         ext4_debug(
 386                                "adding %s group %u: %u blocks (%d free, %d mdata blocks)\n",
 387                                ext4_bg_has_super(sb, group + i) ? "normal" :
 388                                "no-super", group + i,
 389                                group_data[i].blocks_count,
 390                                group_data[i].free_clusters_count,
 391                                group_data[i].mdata_blocks);
 392                 }
 393         }
 394         return 0;
 395 }
 396 
 397 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 398                                   ext4_fsblk_t blk)
 399 {
 400         struct buffer_head *bh;
 401         int err;
 402 
 403         bh = sb_getblk(sb, blk);
 404         if (unlikely(!bh))
 405                 return ERR_PTR(-ENOMEM);
 406         BUFFER_TRACE(bh, "get_write_access");
 407         if ((err = ext4_journal_get_write_access(handle, bh))) {
 408                 brelse(bh);
 409                 bh = ERR_PTR(err);
 410         } else {
 411                 memset(bh->b_data, 0, sb->s_blocksize);
 412                 set_buffer_uptodate(bh);
 413         }
 414 
 415         return bh;
 416 }
 417 
 418 /*
 419  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 420  * If that fails, restart the transaction & regain write access for the
 421  * buffer head which is used for block_bitmap modifications.
 422  */
 423 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 424 {
 425         int err;
 426 
 427         if (ext4_handle_has_enough_credits(handle, thresh))
 428                 return 0;
 429 
 430         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 431         if (err < 0)
 432                 return err;
 433         if (err) {
 434                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 435                 if (err)
 436                         return err;
 437         }
 438 
 439         return 0;
 440 }
 441 
 442 /*
 443  * set_flexbg_block_bitmap() mark clusters [@first_cluster, @last_cluster] used.
 444  *
 445  * Helper function for ext4_setup_new_group_blocks() which set .
 446  *
 447  * @sb: super block
 448  * @handle: journal handle
 449  * @flex_gd: flex group data
 450  */
 451 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 452                         struct ext4_new_flex_group_data *flex_gd,
 453                         ext4_fsblk_t first_cluster, ext4_fsblk_t last_cluster)
 454 {
 455         struct ext4_sb_info *sbi = EXT4_SB(sb);
 456         ext4_group_t count = last_cluster - first_cluster + 1;
 457         ext4_group_t count2;
 458 
 459         ext4_debug("mark clusters [%llu-%llu] used\n", first_cluster,
 460                    last_cluster);
 461         for (count2 = count; count > 0;
 462              count -= count2, first_cluster += count2) {
 463                 ext4_fsblk_t start;
 464                 struct buffer_head *bh;
 465                 ext4_group_t group;
 466                 int err;
 467 
 468                 group = ext4_get_group_number(sb, EXT4_C2B(sbi, first_cluster));
 469                 start = EXT4_B2C(sbi, ext4_group_first_block_no(sb, group));
 470                 group -= flex_gd->groups[0].group;
 471 
 472                 count2 = EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start);
 473                 if (count2 > count)
 474                         count2 = count;
 475 
 476                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 477                         BUG_ON(flex_gd->count > 1);
 478                         continue;
 479                 }
 480 
 481                 err = extend_or_restart_transaction(handle, 1);
 482                 if (err)
 483                         return err;
 484 
 485                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 486                 if (unlikely(!bh))
 487                         return -ENOMEM;
 488 
 489                 BUFFER_TRACE(bh, "get_write_access");
 490                 err = ext4_journal_get_write_access(handle, bh);
 491                 if (err) {
 492                         brelse(bh);
 493                         return err;
 494                 }
 495                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n",
 496                            first_cluster, first_cluster - start, count2);
 497                 ext4_set_bits(bh->b_data, first_cluster - start, count2);
 498 
 499                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 500                 brelse(bh);
 501                 if (unlikely(err))
 502                         return err;
 503         }
 504 
 505         return 0;
 506 }
 507 
 508 /*
 509  * Set up the block and inode bitmaps, and the inode table for the new groups.
 510  * This doesn't need to be part of the main transaction, since we are only
 511  * changing blocks outside the actual filesystem.  We still do journaling to
 512  * ensure the recovery is correct in case of a failure just after resize.
 513  * If any part of this fails, we simply abort the resize.
 514  *
 515  * setup_new_flex_group_blocks handles a flex group as follow:
 516  *  1. copy super block and GDT, and initialize group tables if necessary.
 517  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 518  *     super block and GDT.
 519  *  2. allocate group tables in block bitmaps, that is, set bits in block
 520  *     bitmap for blocks taken by group tables.
 521  */
 522 static int setup_new_flex_group_blocks(struct super_block *sb,
 523                                 struct ext4_new_flex_group_data *flex_gd)
 524 {
 525         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 526         ext4_fsblk_t start;
 527         ext4_fsblk_t block;
 528         struct ext4_sb_info *sbi = EXT4_SB(sb);
 529         struct ext4_super_block *es = sbi->s_es;
 530         struct ext4_new_group_data *group_data = flex_gd->groups;
 531         __u16 *bg_flags = flex_gd->bg_flags;
 532         handle_t *handle;
 533         ext4_group_t group, count;
 534         struct buffer_head *bh = NULL;
 535         int reserved_gdb, i, j, err = 0, err2;
 536         int meta_bg;
 537 
 538         BUG_ON(!flex_gd->count || !group_data ||
 539                group_data[0].group != sbi->s_groups_count);
 540 
 541         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 542         meta_bg = ext4_has_feature_meta_bg(sb);
 543 
 544         /* This transaction may be extended/restarted along the way */
 545         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
 546         if (IS_ERR(handle))
 547                 return PTR_ERR(handle);
 548 
 549         group = group_data[0].group;
 550         for (i = 0; i < flex_gd->count; i++, group++) {
 551                 unsigned long gdblocks;
 552                 ext4_grpblk_t overhead;
 553 
 554                 gdblocks = ext4_bg_num_gdb(sb, group);
 555                 start = ext4_group_first_block_no(sb, group);
 556 
 557                 if (meta_bg == 0 && !ext4_bg_has_super(sb, group))
 558                         goto handle_itb;
 559 
 560                 if (meta_bg == 1) {
 561                         ext4_group_t first_group;
 562                         first_group = ext4_meta_bg_first_group(sb, group);
 563                         if (first_group != group + 1 &&
 564                             first_group != group + EXT4_DESC_PER_BLOCK(sb) - 1)
 565                                 goto handle_itb;
 566                 }
 567 
 568                 block = start + ext4_bg_has_super(sb, group);
 569                 /* Copy all of the GDT blocks into the backup in this group */
 570                 for (j = 0; j < gdblocks; j++, block++) {
 571                         struct buffer_head *gdb;
 572 
 573                         ext4_debug("update backup group %#04llx\n", block);
 574                         err = extend_or_restart_transaction(handle, 1);
 575                         if (err)
 576                                 goto out;
 577 
 578                         gdb = sb_getblk(sb, block);
 579                         if (unlikely(!gdb)) {
 580                                 err = -ENOMEM;
 581                                 goto out;
 582                         }
 583 
 584                         BUFFER_TRACE(gdb, "get_write_access");
 585                         err = ext4_journal_get_write_access(handle, gdb);
 586                         if (err) {
 587                                 brelse(gdb);
 588                                 goto out;
 589                         }
 590                         memcpy(gdb->b_data, sbi_array_rcu_deref(sbi,
 591                                 s_group_desc, j)->b_data, gdb->b_size);
 592                         set_buffer_uptodate(gdb);
 593 
 594                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 595                         if (unlikely(err)) {
 596                                 brelse(gdb);
 597                                 goto out;
 598                         }
 599                         brelse(gdb);
 600                 }
 601 
 602                 /* Zero out all of the reserved backup group descriptor
 603                  * table blocks
 604                  */
 605                 if (ext4_bg_has_super(sb, group)) {
 606                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 607                                         reserved_gdb, GFP_NOFS);
 608                         if (err)
 609                                 goto out;
 610                 }
 611 
 612 handle_itb:
 613                 /* Initialize group tables of the grop @group */
 614                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 615                         goto handle_bb;
 616 
 617                 /* Zero out all of the inode table blocks */
 618                 block = group_data[i].inode_table;
 619                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 620                            block, sbi->s_itb_per_group);
 621                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 622                                        GFP_NOFS);
 623                 if (err)
 624                         goto out;
 625 
 626 handle_bb:
 627                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 628                         goto handle_ib;
 629 
 630                 /* Initialize block bitmap of the @group */
 631                 block = group_data[i].block_bitmap;
 632                 err = extend_or_restart_transaction(handle, 1);
 633                 if (err)
 634                         goto out;
 635 
 636                 bh = bclean(handle, sb, block);
 637                 if (IS_ERR(bh)) {
 638                         err = PTR_ERR(bh);
 639                         goto out;
 640                 }
 641                 overhead = ext4_group_overhead_blocks(sb, group);
 642                 if (overhead != 0) {
 643                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 644                                    start);
 645                         ext4_set_bits(bh->b_data, 0,
 646                                       EXT4_NUM_B2C(sbi, overhead));
 647                 }
 648                 ext4_mark_bitmap_end(EXT4_B2C(sbi, group_data[i].blocks_count),
 649                                      sb->s_blocksize * 8, bh->b_data);
 650                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 651                 brelse(bh);
 652                 if (err)
 653                         goto out;
 654 
 655 handle_ib:
 656                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 657                         continue;
 658 
 659                 /* Initialize inode bitmap of the @group */
 660                 block = group_data[i].inode_bitmap;
 661                 err = extend_or_restart_transaction(handle, 1);
 662                 if (err)
 663                         goto out;
 664                 /* Mark unused entries in inode bitmap used */
 665                 bh = bclean(handle, sb, block);
 666                 if (IS_ERR(bh)) {
 667                         err = PTR_ERR(bh);
 668                         goto out;
 669                 }
 670 
 671                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 672                                      sb->s_blocksize * 8, bh->b_data);
 673                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 674                 brelse(bh);
 675                 if (err)
 676                         goto out;
 677         }
 678 
 679         /* Mark group tables in block bitmap */
 680         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 681                 count = group_table_count[j];
 682                 start = (&group_data[0].block_bitmap)[j];
 683                 block = start;
 684                 for (i = 1; i < flex_gd->count; i++) {
 685                         block += group_table_count[j];
 686                         if (block == (&group_data[i].block_bitmap)[j]) {
 687                                 count += group_table_count[j];
 688                                 continue;
 689                         }
 690                         err = set_flexbg_block_bitmap(sb, handle,
 691                                                       flex_gd,
 692                                                       EXT4_B2C(sbi, start),
 693                                                       EXT4_B2C(sbi,
 694                                                                start + count
 695                                                                - 1));
 696                         if (err)
 697                                 goto out;
 698                         count = group_table_count[j];
 699                         start = (&group_data[i].block_bitmap)[j];
 700                         block = start;
 701                 }
 702 
 703                 if (count) {
 704                         err = set_flexbg_block_bitmap(sb, handle,
 705                                                       flex_gd,
 706                                                       EXT4_B2C(sbi, start),
 707                                                       EXT4_B2C(sbi,
 708                                                                start + count
 709                                                                - 1));
 710                         if (err)
 711                                 goto out;
 712                 }
 713         }
 714 
 715 out:
 716         err2 = ext4_journal_stop(handle);
 717         if (err2 && !err)
 718                 err = err2;
 719 
 720         return err;
 721 }
 722 
 723 /*
 724  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 725  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 726  * calling this for the first time.  In a sparse filesystem it will be the
 727  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 728  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 729  */
 730 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 731                                   unsigned *five, unsigned *seven)
 732 {
 733         unsigned *min = three;
 734         int mult = 3;
 735         unsigned ret;
 736 
 737         if (!ext4_has_feature_sparse_super(sb)) {
 738                 ret = *min;
 739                 *min += 1;
 740                 return ret;
 741         }
 742 
 743         if (*five < *min) {
 744                 min = five;
 745                 mult = 5;
 746         }
 747         if (*seven < *min) {
 748                 min = seven;
 749                 mult = 7;
 750         }
 751 
 752         ret = *min;
 753         *min *= mult;
 754 
 755         return ret;
 756 }
 757 
 758 /*
 759  * Check that all of the backup GDT blocks are held in the primary GDT block.
 760  * It is assumed that they are stored in group order.  Returns the number of
 761  * groups in current filesystem that have BACKUPS, or -ve error code.
 762  */
 763 static int verify_reserved_gdb(struct super_block *sb,
 764                                ext4_group_t end,
 765                                struct buffer_head *primary)
 766 {
 767         const ext4_fsblk_t blk = primary->b_blocknr;
 768         unsigned three = 1;
 769         unsigned five = 5;
 770         unsigned seven = 7;
 771         unsigned grp;
 772         __le32 *p = (__le32 *)primary->b_data;
 773         int gdbackups = 0;
 774 
 775         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 776                 if (le32_to_cpu(*p++) !=
 777                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 778                         ext4_warning(sb, "reserved GDT %llu"
 779                                      " missing grp %d (%llu)",
 780                                      blk, grp,
 781                                      grp *
 782                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 783                                      blk);
 784                         return -EINVAL;
 785                 }
 786                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 787                         return -EFBIG;
 788         }
 789 
 790         return gdbackups;
 791 }
 792 
 793 /*
 794  * Called when we need to bring a reserved group descriptor table block into
 795  * use from the resize inode.  The primary copy of the new GDT block currently
 796  * is an indirect block (under the double indirect block in the resize inode).
 797  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 798  * block, in group order.  Even though we know all the block numbers we need,
 799  * we check to ensure that the resize inode has actually reserved these blocks.
 800  *
 801  * Don't need to update the block bitmaps because the blocks are still in use.
 802  *
 803  * We get all of the error cases out of the way, so that we are sure to not
 804  * fail once we start modifying the data on disk, because JBD has no rollback.
 805  */
 806 static int add_new_gdb(handle_t *handle, struct inode *inode,
 807                        ext4_group_t group)
 808 {
 809         struct super_block *sb = inode->i_sb;
 810         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 811         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 812         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 813         struct buffer_head **o_group_desc, **n_group_desc = NULL;
 814         struct buffer_head *dind = NULL;
 815         struct buffer_head *gdb_bh = NULL;
 816         int gdbackups;
 817         struct ext4_iloc iloc = { .bh = NULL };
 818         __le32 *data;
 819         int err;
 820 
 821         if (test_opt(sb, DEBUG))
 822                 printk(KERN_DEBUG
 823                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 824                        gdb_num);
 825 
 826         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 827         if (IS_ERR(gdb_bh))
 828                 return PTR_ERR(gdb_bh);
 829 
 830         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 831         if (gdbackups < 0) {
 832                 err = gdbackups;
 833                 goto errout;
 834         }
 835 
 836         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 837         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
 838         if (IS_ERR(dind)) {
 839                 err = PTR_ERR(dind);
 840                 dind = NULL;
 841                 goto errout;
 842         }
 843 
 844         data = (__le32 *)dind->b_data;
 845         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 846                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 847                              group, gdblock);
 848                 err = -EINVAL;
 849                 goto errout;
 850         }
 851 
 852         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 853         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 854         if (unlikely(err))
 855                 goto errout;
 856 
 857         BUFFER_TRACE(gdb_bh, "get_write_access");
 858         err = ext4_journal_get_write_access(handle, gdb_bh);
 859         if (unlikely(err))
 860                 goto errout;
 861 
 862         BUFFER_TRACE(dind, "get_write_access");
 863         err = ext4_journal_get_write_access(handle, dind);
 864         if (unlikely(err))
 865                 ext4_std_error(sb, err);
 866 
 867         /* ext4_reserve_inode_write() gets a reference on the iloc */
 868         err = ext4_reserve_inode_write(handle, inode, &iloc);
 869         if (unlikely(err))
 870                 goto errout;
 871 
 872         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 873                                      sizeof(struct buffer_head *),
 874                                      GFP_NOFS);
 875         if (!n_group_desc) {
 876                 err = -ENOMEM;
 877                 ext4_warning(sb, "not enough memory for %lu groups",
 878                              gdb_num + 1);
 879                 goto errout;
 880         }
 881 
 882         /*
 883          * Finally, we have all of the possible failures behind us...
 884          *
 885          * Remove new GDT block from inode double-indirect block and clear out
 886          * the new GDT block for use (which also "frees" the backup GDT blocks
 887          * from the reserved inode).  We don't need to change the bitmaps for
 888          * these blocks, because they are marked as in-use from being in the
 889          * reserved inode, and will become GDT blocks (primary and backup).
 890          */
 891         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 892         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 893         if (unlikely(err)) {
 894                 ext4_std_error(sb, err);
 895                 goto errout;
 896         }
 897         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >>
 898                            (9 - EXT4_SB(sb)->s_cluster_bits);
 899         ext4_mark_iloc_dirty(handle, inode, &iloc);
 900         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 901         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 902         if (unlikely(err)) {
 903                 ext4_std_error(sb, err);
 904                 iloc.bh = NULL;
 905                 goto errout;
 906         }
 907         brelse(dind);
 908 
 909         rcu_read_lock();
 910         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 911         memcpy(n_group_desc, o_group_desc,
 912                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 913         rcu_read_unlock();
 914         n_group_desc[gdb_num] = gdb_bh;
 915         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 916         EXT4_SB(sb)->s_gdb_count++;
 917         ext4_kvfree_array_rcu(o_group_desc);
 918 
 919         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 920         err = ext4_handle_dirty_super(handle, sb);
 921         if (err)
 922                 ext4_std_error(sb, err);
 923         return err;
 924 errout:
 925         kvfree(n_group_desc);
 926         brelse(iloc.bh);
 927         brelse(dind);
 928         brelse(gdb_bh);
 929 
 930         ext4_debug("leaving with error %d\n", err);
 931         return err;
 932 }
 933 
 934 /*
 935  * add_new_gdb_meta_bg is the sister of add_new_gdb.
 936  */
 937 static int add_new_gdb_meta_bg(struct super_block *sb,
 938                                handle_t *handle, ext4_group_t group) {
 939         ext4_fsblk_t gdblock;
 940         struct buffer_head *gdb_bh;
 941         struct buffer_head **o_group_desc, **n_group_desc;
 942         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 943         int err;
 944 
 945         gdblock = ext4_meta_bg_first_block_no(sb, group) +
 946                    ext4_bg_has_super(sb, group);
 947         gdb_bh = ext4_sb_bread(sb, gdblock, 0);
 948         if (IS_ERR(gdb_bh))
 949                 return PTR_ERR(gdb_bh);
 950         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 951                                      sizeof(struct buffer_head *),
 952                                      GFP_NOFS);
 953         if (!n_group_desc) {
 954                 brelse(gdb_bh);
 955                 err = -ENOMEM;
 956                 ext4_warning(sb, "not enough memory for %lu groups",
 957                              gdb_num + 1);
 958                 return err;
 959         }
 960 
 961         rcu_read_lock();
 962         o_group_desc = rcu_dereference(EXT4_SB(sb)->s_group_desc);
 963         memcpy(n_group_desc, o_group_desc,
 964                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 965         rcu_read_unlock();
 966         n_group_desc[gdb_num] = gdb_bh;
 967 
 968         BUFFER_TRACE(gdb_bh, "get_write_access");
 969         err = ext4_journal_get_write_access(handle, gdb_bh);
 970         if (err) {
 971                 kvfree(n_group_desc);
 972                 brelse(gdb_bh);
 973                 return err;
 974         }
 975 
 976         rcu_assign_pointer(EXT4_SB(sb)->s_group_desc, n_group_desc);
 977         EXT4_SB(sb)->s_gdb_count++;
 978         ext4_kvfree_array_rcu(o_group_desc);
 979         return err;
 980 }
 981 
 982 /*
 983  * Called when we are adding a new group which has a backup copy of each of
 984  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 985  * We need to add these reserved backup GDT blocks to the resize inode, so
 986  * that they are kept for future resizing and not allocated to files.
 987  *
 988  * Each reserved backup GDT block will go into a different indirect block.
 989  * The indirect blocks are actually the primary reserved GDT blocks,
 990  * so we know in advance what their block numbers are.  We only get the
 991  * double-indirect block to verify it is pointing to the primary reserved
 992  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 993  * backup GDT blocks are stored in their reserved primary GDT block.
 994  */
 995 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 996                               ext4_group_t group)
 997 {
 998         struct super_block *sb = inode->i_sb;
 999         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
1000         int cluster_bits = EXT4_SB(sb)->s_cluster_bits;
1001         struct buffer_head **primary;
1002         struct buffer_head *dind;
1003         struct ext4_iloc iloc;
1004         ext4_fsblk_t blk;
1005         __le32 *data, *end;
1006         int gdbackups = 0;
1007         int res, i;
1008         int err;
1009 
1010         primary = kmalloc_array(reserved_gdb, sizeof(*primary), GFP_NOFS);
1011         if (!primary)
1012                 return -ENOMEM;
1013 
1014         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
1015         dind = ext4_sb_bread(sb, le32_to_cpu(*data), 0);
1016         if (IS_ERR(dind)) {
1017                 err = PTR_ERR(dind);
1018                 dind = NULL;
1019                 goto exit_free;
1020         }
1021 
1022         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
1023         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
1024                                          EXT4_ADDR_PER_BLOCK(sb));
1025         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
1026 
1027         /* Get each reserved primary GDT block and verify it holds backups */
1028         for (res = 0; res < reserved_gdb; res++, blk++) {
1029                 if (le32_to_cpu(*data) != blk) {
1030                         ext4_warning(sb, "reserved block %llu"
1031                                      " not at offset %ld",
1032                                      blk,
1033                                      (long)(data - (__le32 *)dind->b_data));
1034                         err = -EINVAL;
1035                         goto exit_bh;
1036                 }
1037                 primary[res] = ext4_sb_bread(sb, blk, 0);
1038                 if (IS_ERR(primary[res])) {
1039                         err = PTR_ERR(primary[res]);
1040                         primary[res] = NULL;
1041                         goto exit_bh;
1042                 }
1043                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1044                 if (gdbackups < 0) {
1045                         brelse(primary[res]);
1046                         err = gdbackups;
1047                         goto exit_bh;
1048                 }
1049                 if (++data >= end)
1050                         data = (__le32 *)dind->b_data;
1051         }
1052 
1053         for (i = 0; i < reserved_gdb; i++) {
1054                 BUFFER_TRACE(primary[i], "get_write_access");
1055                 if ((err = ext4_journal_get_write_access(handle, primary[i])))
1056                         goto exit_bh;
1057         }
1058 
1059         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1060                 goto exit_bh;
1061 
1062         /*
1063          * Finally we can add each of the reserved backup GDT blocks from
1064          * the new group to its reserved primary GDT block.
1065          */
1066         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1067         for (i = 0; i < reserved_gdb; i++) {
1068                 int err2;
1069                 data = (__le32 *)primary[i]->b_data;
1070                 /* printk("reserving backup %lu[%u] = %lu\n",
1071                        primary[i]->b_blocknr, gdbackups,
1072                        blk + primary[i]->b_blocknr); */
1073                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1074                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1075                 if (!err)
1076                         err = err2;
1077         }
1078 
1079         inode->i_blocks += reserved_gdb * sb->s_blocksize >> (9 - cluster_bits);
1080         ext4_mark_iloc_dirty(handle, inode, &iloc);
1081 
1082 exit_bh:
1083         while (--res >= 0)
1084                 brelse(primary[res]);
1085         brelse(dind);
1086 
1087 exit_free:
1088         kfree(primary);
1089 
1090         return err;
1091 }
1092 
1093 /*
1094  * Update the backup copies of the ext4 metadata.  These don't need to be part
1095  * of the main resize transaction, because e2fsck will re-write them if there
1096  * is a problem (basically only OOM will cause a problem).  However, we
1097  * _should_ update the backups if possible, in case the primary gets trashed
1098  * for some reason and we need to run e2fsck from a backup superblock.  The
1099  * important part is that the new block and inode counts are in the backup
1100  * superblocks, and the location of the new group metadata in the GDT backups.
1101  *
1102  * We do not need take the s_resize_lock for this, because these
1103  * blocks are not otherwise touched by the filesystem code when it is
1104  * mounted.  We don't need to worry about last changing from
1105  * sbi->s_groups_count, because the worst that can happen is that we
1106  * do not copy the full number of backups at this time.  The resize
1107  * which changed s_groups_count will backup again.
1108  */
1109 static void update_backups(struct super_block *sb, sector_t blk_off, char *data,
1110                            int size, int meta_bg)
1111 {
1112         struct ext4_sb_info *sbi = EXT4_SB(sb);
1113         ext4_group_t last;
1114         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1115         unsigned three = 1;
1116         unsigned five = 5;
1117         unsigned seven = 7;
1118         ext4_group_t group = 0;
1119         int rest = sb->s_blocksize - size;
1120         handle_t *handle;
1121         int err = 0, err2;
1122 
1123         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, EXT4_MAX_TRANS_DATA);
1124         if (IS_ERR(handle)) {
1125                 group = 1;
1126                 err = PTR_ERR(handle);
1127                 goto exit_err;
1128         }
1129 
1130         if (meta_bg == 0) {
1131                 group = ext4_list_backups(sb, &three, &five, &seven);
1132                 last = sbi->s_groups_count;
1133         } else {
1134                 group = ext4_get_group_number(sb, blk_off) + 1;
1135                 last = (ext4_group_t)(group + EXT4_DESC_PER_BLOCK(sb) - 2);
1136         }
1137 
1138         while (group < sbi->s_groups_count) {
1139                 struct buffer_head *bh;
1140                 ext4_fsblk_t backup_block;
1141 
1142                 /* Out of journal space, and can't get more - abort - so sad */
1143                 if (ext4_handle_valid(handle) &&
1144                     handle->h_buffer_credits == 0 &&
1145                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1146                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1147                         break;
1148 
1149                 if (meta_bg == 0)
1150                         backup_block = ((ext4_fsblk_t)group) * bpg + blk_off;
1151                 else
1152                         backup_block = (ext4_group_first_block_no(sb, group) +
1153                                         ext4_bg_has_super(sb, group));
1154 
1155                 bh = sb_getblk(sb, backup_block);
1156                 if (unlikely(!bh)) {
1157                         err = -ENOMEM;
1158                         break;
1159                 }
1160                 ext4_debug("update metadata backup %llu(+%llu)\n",
1161                            backup_block, backup_block -
1162                            ext4_group_first_block_no(sb, group));
1163                 BUFFER_TRACE(bh, "get_write_access");
1164                 if ((err = ext4_journal_get_write_access(handle, bh))) {
1165                         brelse(bh);
1166                         break;
1167                 }
1168                 lock_buffer(bh);
1169                 memcpy(bh->b_data, data, size);
1170                 if (rest)
1171                         memset(bh->b_data + size, 0, rest);
1172                 set_buffer_uptodate(bh);
1173                 unlock_buffer(bh);
1174                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1175                 if (unlikely(err))
1176                         ext4_std_error(sb, err);
1177                 brelse(bh);
1178 
1179                 if (meta_bg == 0)
1180                         group = ext4_list_backups(sb, &three, &five, &seven);
1181                 else if (group == last)
1182                         break;
1183                 else
1184                         group = last;
1185         }
1186         if ((err2 = ext4_journal_stop(handle)) && !err)
1187                 err = err2;
1188 
1189         /*
1190          * Ugh! Need to have e2fsck write the backup copies.  It is too
1191          * late to revert the resize, we shouldn't fail just because of
1192          * the backup copies (they are only needed in case of corruption).
1193          *
1194          * However, if we got here we have a journal problem too, so we
1195          * can't really start a transaction to mark the superblock.
1196          * Chicken out and just set the flag on the hope it will be written
1197          * to disk, and if not - we will simply wait until next fsck.
1198          */
1199 exit_err:
1200         if (err) {
1201                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1202                              "forcing fsck on next reboot", group, err);
1203                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1204                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1205                 mark_buffer_dirty(sbi->s_sbh);
1206         }
1207 }
1208 
1209 /*
1210  * ext4_add_new_descs() adds @count group descriptor of groups
1211  * starting at @group
1212  *
1213  * @handle: journal handle
1214  * @sb: super block
1215  * @group: the group no. of the first group desc to be added
1216  * @resize_inode: the resize inode
1217  * @count: number of group descriptors to be added
1218  */
1219 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1220                               ext4_group_t group, struct inode *resize_inode,
1221                               ext4_group_t count)
1222 {
1223         struct ext4_sb_info *sbi = EXT4_SB(sb);
1224         struct ext4_super_block *es = sbi->s_es;
1225         struct buffer_head *gdb_bh;
1226         int i, gdb_off, gdb_num, err = 0;
1227         int meta_bg;
1228 
1229         meta_bg = ext4_has_feature_meta_bg(sb);
1230         for (i = 0; i < count; i++, group++) {
1231                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1232                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1233 
1234                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1235                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1236 
1237                 /*
1238                  * We will only either add reserved group blocks to a backup group
1239                  * or remove reserved blocks for the first group in a new group block.
1240                  * Doing both would be mean more complex code, and sane people don't
1241                  * use non-sparse filesystems anymore.  This is already checked above.
1242                  */
1243                 if (gdb_off) {
1244                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1245                                                      gdb_num);
1246                         BUFFER_TRACE(gdb_bh, "get_write_access");
1247                         err = ext4_journal_get_write_access(handle, gdb_bh);
1248 
1249                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1250                                 err = reserve_backup_gdb(handle, resize_inode, group);
1251                 } else if (meta_bg != 0) {
1252                         err = add_new_gdb_meta_bg(sb, handle, group);
1253                 } else {
1254                         err = add_new_gdb(handle, resize_inode, group);
1255                 }
1256                 if (err)
1257                         break;
1258         }
1259         return err;
1260 }
1261 
1262 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1263 {
1264         struct buffer_head *bh = sb_getblk(sb, block);
1265         if (unlikely(!bh))
1266                 return NULL;
1267         if (!bh_uptodate_or_lock(bh)) {
1268                 if (bh_submit_read(bh) < 0) {
1269                         brelse(bh);
1270                         return NULL;
1271                 }
1272         }
1273 
1274         return bh;
1275 }
1276 
1277 static int ext4_set_bitmap_checksums(struct super_block *sb,
1278                                      ext4_group_t group,
1279                                      struct ext4_group_desc *gdp,
1280                                      struct ext4_new_group_data *group_data)
1281 {
1282         struct buffer_head *bh;
1283 
1284         if (!ext4_has_metadata_csum(sb))
1285                 return 0;
1286 
1287         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1288         if (!bh)
1289                 return -EIO;
1290         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1291                                    EXT4_INODES_PER_GROUP(sb) / 8);
1292         brelse(bh);
1293 
1294         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1295         if (!bh)
1296                 return -EIO;
1297         ext4_block_bitmap_csum_set(sb, group, gdp, bh);
1298         brelse(bh);
1299 
1300         return 0;
1301 }
1302 
1303 /*
1304  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1305  */
1306 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1307                                 struct ext4_new_flex_group_data *flex_gd)
1308 {
1309         struct ext4_new_group_data      *group_data = flex_gd->groups;
1310         struct ext4_group_desc          *gdp;
1311         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1312         struct buffer_head              *gdb_bh;
1313         ext4_group_t                    group;
1314         __u16                           *bg_flags = flex_gd->bg_flags;
1315         int                             i, gdb_off, gdb_num, err = 0;
1316 
1317 
1318         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1319                 group = group_data->group;
1320 
1321                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1322                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1323 
1324                 /*
1325                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1326                  */
1327                 gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc, gdb_num);
1328                 /* Update group descriptor block for new group */
1329                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1330                                                  gdb_off * EXT4_DESC_SIZE(sb));
1331 
1332                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1333                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1334                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1335                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1336                 if (err) {
1337                         ext4_std_error(sb, err);
1338                         break;
1339                 }
1340 
1341                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1342                 ext4_free_group_clusters_set(sb, gdp,
1343                                              group_data->free_clusters_count);
1344                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1345                 if (ext4_has_group_desc_csum(sb))
1346                         ext4_itable_unused_set(sb, gdp,
1347                                                EXT4_INODES_PER_GROUP(sb));
1348                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1349                 ext4_group_desc_csum_set(sb, group, gdp);
1350 
1351                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1352                 if (unlikely(err)) {
1353                         ext4_std_error(sb, err);
1354                         break;
1355                 }
1356 
1357                 /*
1358                  * We can allocate memory for mb_alloc based on the new group
1359                  * descriptor
1360                  */
1361                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1362                 if (err)
1363                         break;
1364         }
1365         return err;
1366 }
1367 
1368 /*
1369  * ext4_update_super() updates the super block so that the newly added
1370  * groups can be seen by the filesystem.
1371  *
1372  * @sb: super block
1373  * @flex_gd: new added groups
1374  */
1375 static void ext4_update_super(struct super_block *sb,
1376                              struct ext4_new_flex_group_data *flex_gd)
1377 {
1378         ext4_fsblk_t blocks_count = 0;
1379         ext4_fsblk_t free_blocks = 0;
1380         ext4_fsblk_t reserved_blocks = 0;
1381         struct ext4_new_group_data *group_data = flex_gd->groups;
1382         struct ext4_sb_info *sbi = EXT4_SB(sb);
1383         struct ext4_super_block *es = sbi->s_es;
1384         int i;
1385 
1386         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1387         /*
1388          * Make the new blocks and inodes valid next.  We do this before
1389          * increasing the group count so that once the group is enabled,
1390          * all of its blocks and inodes are already valid.
1391          *
1392          * We always allocate group-by-group, then block-by-block or
1393          * inode-by-inode within a group, so enabling these
1394          * blocks/inodes before the group is live won't actually let us
1395          * allocate the new space yet.
1396          */
1397         for (i = 0; i < flex_gd->count; i++) {
1398                 blocks_count += group_data[i].blocks_count;
1399                 free_blocks += EXT4_C2B(sbi, group_data[i].free_clusters_count);
1400         }
1401 
1402         reserved_blocks = ext4_r_blocks_count(es) * 100;
1403         reserved_blocks = div64_u64(reserved_blocks, ext4_blocks_count(es));
1404         reserved_blocks *= blocks_count;
1405         do_div(reserved_blocks, 100);
1406 
1407         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1408         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1409         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1410                      flex_gd->count);
1411         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1412                      flex_gd->count);
1413 
1414         ext4_debug("free blocks count %llu", ext4_free_blocks_count(es));
1415         /*
1416          * We need to protect s_groups_count against other CPUs seeing
1417          * inconsistent state in the superblock.
1418          *
1419          * The precise rules we use are:
1420          *
1421          * * Writers must perform a smp_wmb() after updating all
1422          *   dependent data and before modifying the groups count
1423          *
1424          * * Readers must perform an smp_rmb() after reading the groups
1425          *   count and before reading any dependent data.
1426          *
1427          * NB. These rules can be relaxed when checking the group count
1428          * while freeing data, as we can only allocate from a block
1429          * group after serialising against the group count, and we can
1430          * only then free after serialising in turn against that
1431          * allocation.
1432          */
1433         smp_wmb();
1434 
1435         /* Update the global fs size fields */
1436         sbi->s_groups_count += flex_gd->count;
1437         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
1438                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
1439 
1440         /* Update the reserved block counts only once the new group is
1441          * active. */
1442         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1443                                 reserved_blocks);
1444 
1445         /* Update the free space counts */
1446         percpu_counter_add(&sbi->s_freeclusters_counter,
1447                            EXT4_NUM_B2C(sbi, free_blocks));
1448         percpu_counter_add(&sbi->s_freeinodes_counter,
1449                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1450 
1451         ext4_debug("free blocks count %llu",
1452                    percpu_counter_read(&sbi->s_freeclusters_counter));
1453         if (ext4_has_feature_flex_bg(sb) && sbi->s_log_groups_per_flex) {
1454                 ext4_group_t flex_group;
1455                 struct flex_groups *fg;
1456 
1457                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1458                 fg = sbi_array_rcu_deref(sbi, s_flex_groups, flex_group);
1459                 atomic64_add(EXT4_NUM_B2C(sbi, free_blocks),
1460                              &fg->free_clusters);
1461                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1462                            &fg->free_inodes);
1463         }
1464 
1465         /*
1466          * Update the fs overhead information
1467          */
1468         ext4_calculate_overhead(sb);
1469 
1470         if (test_opt(sb, DEBUG))
1471                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1472                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1473                        blocks_count, free_blocks, reserved_blocks);
1474 }
1475 
1476 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1477  * _before_ we start modifying the filesystem, because we cannot abort the
1478  * transaction and not have it write the data to disk.
1479  */
1480 static int ext4_flex_group_add(struct super_block *sb,
1481                                struct inode *resize_inode,
1482                                struct ext4_new_flex_group_data *flex_gd)
1483 {
1484         struct ext4_sb_info *sbi = EXT4_SB(sb);
1485         struct ext4_super_block *es = sbi->s_es;
1486         ext4_fsblk_t o_blocks_count;
1487         ext4_grpblk_t last;
1488         ext4_group_t group;
1489         handle_t *handle;
1490         unsigned reserved_gdb;
1491         int err = 0, err2 = 0, credit;
1492 
1493         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1494 
1495         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1496         o_blocks_count = ext4_blocks_count(es);
1497         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1498         BUG_ON(last);
1499 
1500         err = setup_new_flex_group_blocks(sb, flex_gd);
1501         if (err)
1502                 goto exit;
1503         /*
1504          * We will always be modifying at least the superblock and  GDT
1505          * blocks.  If we are adding a group past the last current GDT block,
1506          * we will also modify the inode and the dindirect block.  If we
1507          * are adding a group with superblock/GDT backups  we will also
1508          * modify each of the reserved GDT dindirect blocks.
1509          */
1510         credit = 3;     /* sb, resize inode, resize inode dindirect */
1511         /* GDT blocks */
1512         credit += 1 + DIV_ROUND_UP(flex_gd->count, EXT4_DESC_PER_BLOCK(sb));
1513         credit += reserved_gdb; /* Reserved GDT dindirect blocks */
1514         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credit);
1515         if (IS_ERR(handle)) {
1516                 err = PTR_ERR(handle);
1517                 goto exit;
1518         }
1519 
1520         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1521         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1522         if (err)
1523                 goto exit_journal;
1524 
1525         group = flex_gd->groups[0].group;
1526         BUG_ON(group != sbi->s_groups_count);
1527         err = ext4_add_new_descs(handle, sb, group,
1528                                 resize_inode, flex_gd->count);
1529         if (err)
1530                 goto exit_journal;
1531 
1532         err = ext4_setup_new_descs(handle, sb, flex_gd);
1533         if (err)
1534                 goto exit_journal;
1535 
1536         ext4_update_super(sb, flex_gd);
1537 
1538         err = ext4_handle_dirty_super(handle, sb);
1539 
1540 exit_journal:
1541         err2 = ext4_journal_stop(handle);
1542         if (!err)
1543                 err = err2;
1544 
1545         if (!err) {
1546                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1547                 int gdb_num_end = ((group + flex_gd->count - 1) /
1548                                    EXT4_DESC_PER_BLOCK(sb));
1549                 int meta_bg = ext4_has_feature_meta_bg(sb);
1550                 sector_t old_gdb = 0;
1551 
1552                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1553                                sizeof(struct ext4_super_block), 0);
1554                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1555                         struct buffer_head *gdb_bh;
1556 
1557                         gdb_bh = sbi_array_rcu_deref(sbi, s_group_desc,
1558                                                      gdb_num);
1559                         if (old_gdb == gdb_bh->b_blocknr)
1560                                 continue;
1561                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1562                                        gdb_bh->b_size, meta_bg);
1563                         old_gdb = gdb_bh->b_blocknr;
1564                 }
1565         }
1566 exit:
1567         return err;
1568 }
1569 
1570 static int ext4_setup_next_flex_gd(struct super_block *sb,
1571                                     struct ext4_new_flex_group_data *flex_gd,
1572                                     ext4_fsblk_t n_blocks_count,
1573                                     unsigned long flexbg_size)
1574 {
1575         struct ext4_sb_info *sbi = EXT4_SB(sb);
1576         struct ext4_super_block *es = sbi->s_es;
1577         struct ext4_new_group_data *group_data = flex_gd->groups;
1578         ext4_fsblk_t o_blocks_count;
1579         ext4_group_t n_group;
1580         ext4_group_t group;
1581         ext4_group_t last_group;
1582         ext4_grpblk_t last;
1583         ext4_grpblk_t clusters_per_group;
1584         unsigned long i;
1585 
1586         clusters_per_group = EXT4_CLUSTERS_PER_GROUP(sb);
1587 
1588         o_blocks_count = ext4_blocks_count(es);
1589 
1590         if (o_blocks_count == n_blocks_count)
1591                 return 0;
1592 
1593         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1594         BUG_ON(last);
1595         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1596 
1597         last_group = group | (flexbg_size - 1);
1598         if (last_group > n_group)
1599                 last_group = n_group;
1600 
1601         flex_gd->count = last_group - group + 1;
1602 
1603         for (i = 0; i < flex_gd->count; i++) {
1604                 int overhead;
1605 
1606                 group_data[i].group = group + i;
1607                 group_data[i].blocks_count = EXT4_BLOCKS_PER_GROUP(sb);
1608                 overhead = ext4_group_overhead_blocks(sb, group + i);
1609                 group_data[i].mdata_blocks = overhead;
1610                 group_data[i].free_clusters_count = EXT4_CLUSTERS_PER_GROUP(sb);
1611                 if (ext4_has_group_desc_csum(sb)) {
1612                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1613                                                EXT4_BG_INODE_UNINIT;
1614                         if (!test_opt(sb, INIT_INODE_TABLE))
1615                                 flex_gd->bg_flags[i] |= EXT4_BG_INODE_ZEROED;
1616                 } else
1617                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1618         }
1619 
1620         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1621                 /* We need to initialize block bitmap of last group. */
1622                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1623 
1624         if ((last_group == n_group) && (last != clusters_per_group - 1)) {
1625                 group_data[i - 1].blocks_count = EXT4_C2B(sbi, last + 1);
1626                 group_data[i - 1].free_clusters_count -= clusters_per_group -
1627                                                        last - 1;
1628         }
1629 
1630         return 1;
1631 }
1632 
1633 /* Add group descriptor data to an existing or new group descriptor block.
1634  * Ensure we handle all possible error conditions _before_ we start modifying
1635  * the filesystem, because we cannot abort the transaction and not have it
1636  * write the data to disk.
1637  *
1638  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1639  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1640  *
1641  * We only need to hold the superblock lock while we are actually adding
1642  * in the new group's counts to the superblock.  Prior to that we have
1643  * not really "added" the group at all.  We re-check that we are still
1644  * adding in the last group in case things have changed since verifying.
1645  */
1646 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1647 {
1648         struct ext4_new_flex_group_data flex_gd;
1649         struct ext4_sb_info *sbi = EXT4_SB(sb);
1650         struct ext4_super_block *es = sbi->s_es;
1651         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1652                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1653         struct inode *inode = NULL;
1654         int gdb_off;
1655         int err;
1656         __u16 bg_flags = 0;
1657 
1658         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1659 
1660         if (gdb_off == 0 && !ext4_has_feature_sparse_super(sb)) {
1661                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1662                 return -EPERM;
1663         }
1664 
1665         if (ext4_blocks_count(es) + input->blocks_count <
1666             ext4_blocks_count(es)) {
1667                 ext4_warning(sb, "blocks_count overflow");
1668                 return -EINVAL;
1669         }
1670 
1671         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1672             le32_to_cpu(es->s_inodes_count)) {
1673                 ext4_warning(sb, "inodes_count overflow");
1674                 return -EINVAL;
1675         }
1676 
1677         if (reserved_gdb || gdb_off == 0) {
1678                 if (!ext4_has_feature_resize_inode(sb) ||
1679                     !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1680                         ext4_warning(sb,
1681                                      "No reserved GDT blocks, can't resize");
1682                         return -EPERM;
1683                 }
1684                 inode = ext4_iget(sb, EXT4_RESIZE_INO, EXT4_IGET_SPECIAL);
1685                 if (IS_ERR(inode)) {
1686                         ext4_warning(sb, "Error opening resize inode");
1687                         return PTR_ERR(inode);
1688                 }
1689         }
1690 
1691 
1692         err = verify_group_input(sb, input);
1693         if (err)
1694                 goto out;
1695 
1696         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1697         if (err)
1698                 goto out;
1699 
1700         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1701         if (err)
1702                 goto out;
1703 
1704         flex_gd.count = 1;
1705         flex_gd.groups = input;
1706         flex_gd.bg_flags = &bg_flags;
1707         err = ext4_flex_group_add(sb, inode, &flex_gd);
1708 out:
1709         iput(inode);
1710         return err;
1711 } /* ext4_group_add */
1712 
1713 /*
1714  * extend a group without checking assuming that checking has been done.
1715  */
1716 static int ext4_group_extend_no_check(struct super_block *sb,
1717                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1718 {
1719         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1720         handle_t *handle;
1721         int err = 0, err2;
1722 
1723         /* We will update the superblock, one block bitmap, and
1724          * one group descriptor via ext4_group_add_blocks().
1725          */
1726         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, 3);
1727         if (IS_ERR(handle)) {
1728                 err = PTR_ERR(handle);
1729                 ext4_warning(sb, "error %d on journal start", err);
1730                 return err;
1731         }
1732 
1733         BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1734         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1735         if (err) {
1736                 ext4_warning(sb, "error %d on journal write access", err);
1737                 goto errout;
1738         }
1739 
1740         ext4_blocks_count_set(es, o_blocks_count + add);
1741         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1742         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1743                    o_blocks_count + add);
1744         /* We add the blocks to the bitmap and set the group need init bit */
1745         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1746         if (err)
1747                 goto errout;
1748         ext4_handle_dirty_super(handle, sb);
1749         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1750                    o_blocks_count + add);
1751 errout:
1752         err2 = ext4_journal_stop(handle);
1753         if (err2 && !err)
1754                 err = err2;
1755 
1756         if (!err) {
1757                 if (test_opt(sb, DEBUG))
1758                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1759                                "blocks\n", ext4_blocks_count(es));
1760                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr,
1761                                (char *)es, sizeof(struct ext4_super_block), 0);
1762         }
1763         return err;
1764 }
1765 
1766 /*
1767  * Extend the filesystem to the new number of blocks specified.  This entry
1768  * point is only used to extend the current filesystem to the end of the last
1769  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1770  * for emergencies (because it has no dependencies on reserved blocks).
1771  *
1772  * If we _really_ wanted, we could use default values to call ext4_group_add()
1773  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1774  * GDT blocks are reserved to grow to the desired size.
1775  */
1776 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1777                       ext4_fsblk_t n_blocks_count)
1778 {
1779         ext4_fsblk_t o_blocks_count;
1780         ext4_grpblk_t last;
1781         ext4_grpblk_t add;
1782         struct buffer_head *bh;
1783         int err;
1784         ext4_group_t group;
1785 
1786         o_blocks_count = ext4_blocks_count(es);
1787 
1788         if (test_opt(sb, DEBUG))
1789                 ext4_msg(sb, KERN_DEBUG,
1790                          "extending last group from %llu to %llu blocks",
1791                          o_blocks_count, n_blocks_count);
1792 
1793         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1794                 return 0;
1795 
1796         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1797                 ext4_msg(sb, KERN_ERR,
1798                          "filesystem too large to resize to %llu blocks safely",
1799                          n_blocks_count);
1800                 return -EINVAL;
1801         }
1802 
1803         if (n_blocks_count < o_blocks_count) {
1804                 ext4_warning(sb, "can't shrink FS - resize aborted");
1805                 return -EINVAL;
1806         }
1807 
1808         /* Handle the remaining blocks in the last group only. */
1809         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1810 
1811         if (last == 0) {
1812                 ext4_warning(sb, "need to use ext2online to resize further");
1813                 return -EPERM;
1814         }
1815 
1816         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1817 
1818         if (o_blocks_count + add < o_blocks_count) {
1819                 ext4_warning(sb, "blocks_count overflow");
1820                 return -EINVAL;
1821         }
1822 
1823         if (o_blocks_count + add > n_blocks_count)
1824                 add = n_blocks_count - o_blocks_count;
1825 
1826         if (o_blocks_count + add < n_blocks_count)
1827                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1828                              o_blocks_count + add, add);
1829 
1830         /* See if the device is actually as big as what was requested */
1831         bh = sb_bread(sb, o_blocks_count + add - 1);
1832         if (!bh) {
1833                 ext4_warning(sb, "can't read last block, resize aborted");
1834                 return -ENOSPC;
1835         }
1836         brelse(bh);
1837 
1838         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1839         return err;
1840 } /* ext4_group_extend */
1841 
1842 
1843 static int num_desc_blocks(struct super_block *sb, ext4_group_t groups)
1844 {
1845         return (groups + EXT4_DESC_PER_BLOCK(sb) - 1) / EXT4_DESC_PER_BLOCK(sb);
1846 }
1847 
1848 /*
1849  * Release the resize inode and drop the resize_inode feature if there
1850  * are no more reserved gdt blocks, and then convert the file system
1851  * to enable meta_bg
1852  */
1853 static int ext4_convert_meta_bg(struct super_block *sb, struct inode *inode)
1854 {
1855         handle_t *handle;
1856         struct ext4_sb_info *sbi = EXT4_SB(sb);
1857         struct ext4_super_block *es = sbi->s_es;
1858         struct ext4_inode_info *ei = EXT4_I(inode);
1859         ext4_fsblk_t nr;
1860         int i, ret, err = 0;
1861         int credits = 1;
1862 
1863         ext4_msg(sb, KERN_INFO, "Converting file system to meta_bg");
1864         if (inode) {
1865                 if (es->s_reserved_gdt_blocks) {
1866                         ext4_error(sb, "Unexpected non-zero "
1867                                    "s_reserved_gdt_blocks");
1868                         return -EPERM;
1869                 }
1870 
1871                 /* Do a quick sanity check of the resize inode */
1872                 if (inode->i_blocks != 1 << (inode->i_blkbits -
1873                                              (9 - sbi->s_cluster_bits)))
1874                         goto invalid_resize_inode;
1875                 for (i = 0; i < EXT4_N_BLOCKS; i++) {
1876                         if (i == EXT4_DIND_BLOCK) {
1877                                 if (ei->i_data[i])
1878                                         continue;
1879                                 else
1880                                         goto invalid_resize_inode;
1881                         }
1882                         if (ei->i_data[i])
1883                                 goto invalid_resize_inode;
1884                 }
1885                 credits += 3;   /* block bitmap, bg descriptor, resize inode */
1886         }
1887 
1888         handle = ext4_journal_start_sb(sb, EXT4_HT_RESIZE, credits);
1889         if (IS_ERR(handle))
1890                 return PTR_ERR(handle);
1891 
1892         BUFFER_TRACE(sbi->s_sbh, "get_write_access");
1893         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1894         if (err)
1895                 goto errout;
1896 
1897         ext4_clear_feature_resize_inode(sb);
1898         ext4_set_feature_meta_bg(sb);
1899         sbi->s_es->s_first_meta_bg =
1900                 cpu_to_le32(num_desc_blocks(sb, sbi->s_groups_count));
1901 
1902         err = ext4_handle_dirty_super(handle, sb);
1903         if (err) {
1904                 ext4_std_error(sb, err);
1905                 goto errout;
1906         }
1907 
1908         if (inode) {
1909                 nr = le32_to_cpu(ei->i_data[EXT4_DIND_BLOCK]);
1910                 ext4_free_blocks(handle, inode, NULL, nr, 1,
1911                                  EXT4_FREE_BLOCKS_METADATA |
1912                                  EXT4_FREE_BLOCKS_FORGET);
1913                 ei->i_data[EXT4_DIND_BLOCK] = 0;
1914                 inode->i_blocks = 0;
1915 
1916                 err = ext4_mark_inode_dirty(handle, inode);
1917                 if (err)
1918                         ext4_std_error(sb, err);
1919         }
1920 
1921 errout:
1922         ret = ext4_journal_stop(handle);
1923         if (!err)
1924                 err = ret;
1925         return ret;
1926 
1927 invalid_resize_inode:
1928         ext4_error(sb, "corrupted/inconsistent resize inode");
1929         return -EINVAL;
1930 }
1931 
1932 /*
1933  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1934  *
1935  * @sb: super block of the fs to be resized
1936  * @n_blocks_count: the number of blocks resides in the resized fs
1937  */
1938 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1939 {
1940         struct ext4_new_flex_group_data *flex_gd = NULL;
1941         struct ext4_sb_info *sbi = EXT4_SB(sb);
1942         struct ext4_super_block *es = sbi->s_es;
1943         struct buffer_head *bh;
1944         struct inode *resize_inode = NULL;
1945         ext4_grpblk_t add, offset;
1946         unsigned long n_desc_blocks;
1947         unsigned long o_desc_blocks;
1948         ext4_group_t o_group;
1949         ext4_group_t n_group;
1950         ext4_fsblk_t o_blocks_count;
1951         ext4_fsblk_t n_blocks_count_retry = 0;
1952         unsigned long last_update_time = 0;
1953         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1954         int meta_bg;
1955 
1956         /* See if the device is actually as big as what was requested */
1957         bh = sb_bread(sb, n_blocks_count - 1);
1958         if (!bh) {
1959                 ext4_warning(sb, "can't read last block, resize aborted");
1960                 return -ENOSPC;
1961         }
1962         brelse(bh);
1963 
1964 retry:
1965         o_blocks_count = ext4_blocks_count(es);
1966 
1967         ext4_msg(sb, KERN_INFO, "resizing filesystem from %llu "
1968                  "to %llu blocks", o_blocks_count, n_blocks_count);
1969 
1970         if (n_blocks_count < o_blocks_count) {
1971                 /* On-line shrinking not supported */
1972                 ext4_warning(sb, "can't shrink FS - resize aborted");
1973                 return -EINVAL;
1974         }
1975 
1976         if (n_blocks_count == o_blocks_count)
1977                 /* Nothing need to do */
1978                 return 0;
1979 
1980         n_group = ext4_get_group_number(sb, n_blocks_count - 1);
1981         if (n_group >= (0xFFFFFFFFUL / EXT4_INODES_PER_GROUP(sb))) {
1982                 ext4_warning(sb, "resize would cause inodes_count overflow");
1983                 return -EINVAL;
1984         }
1985         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1986 
1987         n_desc_blocks = num_desc_blocks(sb, n_group + 1);
1988         o_desc_blocks = num_desc_blocks(sb, sbi->s_groups_count);
1989 
1990         meta_bg = ext4_has_feature_meta_bg(sb);
1991 
1992         if (ext4_has_feature_resize_inode(sb)) {
1993                 if (meta_bg) {
1994                         ext4_error(sb, "resize_inode and meta_bg enabled "
1995                                    "simultaneously");
1996                         return -EINVAL;
1997                 }
1998                 if (n_desc_blocks > o_desc_blocks +
1999                     le16_to_cpu(es->s_reserved_gdt_blocks)) {
2000                         n_blocks_count_retry = n_blocks_count;
2001                         n_desc_blocks = o_desc_blocks +
2002                                 le16_to_cpu(es->s_reserved_gdt_blocks);
2003                         n_group = n_desc_blocks * EXT4_DESC_PER_BLOCK(sb);
2004                         n_blocks_count = (ext4_fsblk_t)n_group *
2005                                 EXT4_BLOCKS_PER_GROUP(sb) +
2006                                 le32_to_cpu(es->s_first_data_block);
2007                         n_group--; /* set to last group number */
2008                 }
2009 
2010                 if (!resize_inode)
2011                         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO,
2012                                                  EXT4_IGET_SPECIAL);
2013                 if (IS_ERR(resize_inode)) {
2014                         ext4_warning(sb, "Error opening resize inode");
2015                         return PTR_ERR(resize_inode);
2016                 }
2017         }
2018 
2019         if ((!resize_inode && !meta_bg) || n_blocks_count == o_blocks_count) {
2020                 err = ext4_convert_meta_bg(sb, resize_inode);
2021                 if (err)
2022                         goto out;
2023                 if (resize_inode) {
2024                         iput(resize_inode);
2025                         resize_inode = NULL;
2026                 }
2027                 if (n_blocks_count_retry) {
2028                         n_blocks_count = n_blocks_count_retry;
2029                         n_blocks_count_retry = 0;
2030                         goto retry;
2031                 }
2032         }
2033 
2034         /*
2035          * Make sure the last group has enough space so that it's
2036          * guaranteed to have enough space for all metadata blocks
2037          * that it might need to hold.  (We might not need to store
2038          * the inode table blocks in the last block group, but there
2039          * will be cases where this might be needed.)
2040          */
2041         if ((ext4_group_first_block_no(sb, n_group) +
2042              ext4_group_overhead_blocks(sb, n_group) + 2 +
2043              sbi->s_itb_per_group + sbi->s_cluster_ratio) >= n_blocks_count) {
2044                 n_blocks_count = ext4_group_first_block_no(sb, n_group);
2045                 n_group--;
2046                 n_blocks_count_retry = 0;
2047                 if (resize_inode) {
2048                         iput(resize_inode);
2049                         resize_inode = NULL;
2050                 }
2051                 goto retry;
2052         }
2053 
2054         /* extend the last group */
2055         if (n_group == o_group)
2056                 add = n_blocks_count - o_blocks_count;
2057         else
2058                 add = EXT4_C2B(sbi, EXT4_CLUSTERS_PER_GROUP(sb) - (offset + 1));
2059         if (add > 0) {
2060                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
2061                 if (err)
2062                         goto out;
2063         }
2064 
2065         if (ext4_blocks_count(es) == n_blocks_count)
2066                 goto out;
2067 
2068         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
2069         if (err)
2070                 goto out;
2071 
2072         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
2073         if (err)
2074                 goto out;
2075 
2076         flex_gd = alloc_flex_gd(flexbg_size);
2077         if (flex_gd == NULL) {
2078                 err = -ENOMEM;
2079                 goto out;
2080         }
2081 
2082         /* Add flex groups. Note that a regular group is a
2083          * flex group with 1 group.
2084          */
2085         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
2086                                               flexbg_size)) {
2087                 if (jiffies - last_update_time > HZ * 10) {
2088                         if (last_update_time)
2089                                 ext4_msg(sb, KERN_INFO,
2090                                          "resized to %llu blocks",
2091                                          ext4_blocks_count(es));
2092                         last_update_time = jiffies;
2093                 }
2094                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
2095                         break;
2096                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
2097                 if (unlikely(err))
2098                         break;
2099         }
2100 
2101         if (!err && n_blocks_count_retry) {
2102                 n_blocks_count = n_blocks_count_retry;
2103                 n_blocks_count_retry = 0;
2104                 free_flex_gd(flex_gd);
2105                 flex_gd = NULL;
2106                 if (resize_inode) {
2107                         iput(resize_inode);
2108                         resize_inode = NULL;
2109                 }
2110                 goto retry;
2111         }
2112 
2113 out:
2114         if (flex_gd)
2115                 free_flex_gd(flex_gd);
2116         if (resize_inode != NULL)
2117                 iput(resize_inode);
2118         if (err)
2119                 ext4_warning(sb, "error (%d) occurred during "
2120                              "file system resize", err);
2121         ext4_msg(sb, KERN_INFO, "resized filesystem to %llu",
2122                  ext4_blocks_count(es));
2123         return err;
2124 }
/* [<][>][^][v][top][bottom][index][help] */
root/fs/ext4/resize.c

DEFINITIONS
