fs/ext4/resize.c

   1 /*
   2  *  linux/fs/ext4/resize.c
   3  *
   4  * Support for resizing an ext4 filesystem while it is mounted.
   5  *
   6  * Copyright (C) 2001, 2002 Andreas Dilger <adilger@clusterfs.com>
   7  *
   8  * This could probably be made into a module, because it is not often in use.
   9  */
  10
  11
  12 #define EXT4FS_DEBUG
  13
  14 #include <linux/errno.h>
  15 #include <linux/slab.h>
  16
  17 #include "ext4_jbd2.h"
  18
  19 int ext4_resize_begin(struct super_block *sb)
  20 {
  21         int ret = 0;
  22
  23         if (!capable(CAP_SYS_RESOURCE))
  24                 return -EPERM;
  25
  26         /*
  27          * We are not allowed to do online-resizing on a filesystem mounted
  28          * with error, because it can destroy the filesystem easily.
  29          */
  30         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
  31                 ext4_warning(sb, "There are errors in the filesystem, "
  32                              "so online resizing is not allowed\n");
  33                 return -EPERM;
  34         }
  35
  36         if (test_and_set_bit_lock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags))
  37                 ret = -EBUSY;
  38
  39         return ret;
  40 }
  41
  42 void ext4_resize_end(struct super_block *sb)
  43 {
  44         clear_bit_unlock(EXT4_RESIZING, &EXT4_SB(sb)->s_resize_flags);
  45         smp_mb__after_clear_bit();
  46 }
  47
  48 #define outside(b, first, last) ((b) < (first) || (b) >= (last))
  49 #define inside(b, first, last)  ((b) >= (first) && (b) < (last))
  50
  51 static int verify_group_input(struct super_block *sb,
  52                               struct ext4_new_group_data *input)
  53 {
  54         struct ext4_sb_info *sbi = EXT4_SB(sb);
  55         struct ext4_super_block *es = sbi->s_es;
  56         ext4_fsblk_t start = ext4_blocks_count(es);
  57         ext4_fsblk_t end = start + input->blocks_count;
  58         ext4_group_t group = input->group;
  59         ext4_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
  60         unsigned overhead = ext4_bg_has_super(sb, group) ?
  61                 (1 + ext4_bg_num_gdb(sb, group) +
  62                  le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
  63         ext4_fsblk_t metaend = start + overhead;
  64         struct buffer_head *bh = NULL;
  65         ext4_grpblk_t free_blocks_count, offset;
  66         int err = -EINVAL;
  67
  68         input->free_blocks_count = free_blocks_count =
  69                 input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
  70
  71         if (test_opt(sb, DEBUG))
  72                 printk(KERN_DEBUG "EXT4-fs: adding %s group %u: %u blocks "
  73                        "(%d free, %u reserved)\n",
  74                        ext4_bg_has_super(sb, input->group) ? "normal" :
  75                        "no-super", input->group, input->blocks_count,
  76                        free_blocks_count, input->reserved_blocks);
  77
  78         ext4_get_group_no_and_offset(sb, start, NULL, &offset);
  79         if (group != sbi->s_groups_count)
  80                 ext4_warning(sb, "Cannot add at group %u (only %u groups)",
  81                              input->group, sbi->s_groups_count);
  82         else if (offset != 0)
  83                         ext4_warning(sb, "Last group not full");
  84         else if (input->reserved_blocks > input->blocks_count / 5)
  85                 ext4_warning(sb, "Reserved blocks too high (%u)",
  86                              input->reserved_blocks);
  87         else if (free_blocks_count < 0)
  88                 ext4_warning(sb, "Bad blocks count %u",
  89                              input->blocks_count);
  90         else if (!(bh = sb_bread(sb, end - 1)))
  91                 ext4_warning(sb, "Cannot read last block (%llu)",
  92                              end - 1);
  93         else if (outside(input->block_bitmap, start, end))
  94                 ext4_warning(sb, "Block bitmap not in group (block %llu)",
  95                              (unsigned long long)input->block_bitmap);
  96         else if (outside(input->inode_bitmap, start, end))
  97                 ext4_warning(sb, "Inode bitmap not in group (block %llu)",
  98                              (unsigned long long)input->inode_bitmap);
  99         else if (outside(input->inode_table, start, end) ||
 100                  outside(itend - 1, start, end))
 101                 ext4_warning(sb, "Inode table not in group (blocks %llu-%llu)",
 102                              (unsigned long long)input->inode_table, itend - 1);
 103         else if (input->inode_bitmap == input->block_bitmap)
 104                 ext4_warning(sb, "Block bitmap same as inode bitmap (%llu)",
 105                              (unsigned long long)input->block_bitmap);
 106         else if (inside(input->block_bitmap, input->inode_table, itend))
 107                 ext4_warning(sb, "Block bitmap (%llu) in inode table "
 108                              "(%llu-%llu)",
 109                              (unsigned long long)input->block_bitmap,
 110                              (unsigned long long)input->inode_table, itend - 1);
 111         else if (inside(input->inode_bitmap, input->inode_table, itend))
 112                 ext4_warning(sb, "Inode bitmap (%llu) in inode table "
 113                              "(%llu-%llu)",
 114                              (unsigned long long)input->inode_bitmap,
 115                              (unsigned long long)input->inode_table, itend - 1);
 116         else if (inside(input->block_bitmap, start, metaend))
 117                 ext4_warning(sb, "Block bitmap (%llu) in GDT table (%llu-%llu)",
 118                              (unsigned long long)input->block_bitmap,
 119                              start, metaend - 1);
 120         else if (inside(input->inode_bitmap, start, metaend))
 121                 ext4_warning(sb, "Inode bitmap (%llu) in GDT table (%llu-%llu)",
 122                              (unsigned long long)input->inode_bitmap,
 123                              start, metaend - 1);
 124         else if (inside(input->inode_table, start, metaend) ||
 125                  inside(itend - 1, start, metaend))
 126                 ext4_warning(sb, "Inode table (%llu-%llu) overlaps GDT table "
 127                              "(%llu-%llu)",
 128                              (unsigned long long)input->inode_table,
 129                              itend - 1, start, metaend - 1);
 130         else
 131                 err = 0;
 132         brelse(bh);
 133
 134         return err;
 135 }
 136
 137 /*
 138  * ext4_new_flex_group_data is used by 64bit-resize interface to add a flex
 139  * group each time.
 140  */
 141 struct ext4_new_flex_group_data {
 142         struct ext4_new_group_data *groups;     /* new_group_data for groups
 143                                                    in the flex group */
 144         __u16 *bg_flags;                        /* block group flags of groups
 145                                                    in @groups */
 146         ext4_group_t count;                     /* number of groups in @groups
 147                                                  */
 148 };
 149
 150 /*
 151  * alloc_flex_gd() allocates a ext4_new_flex_group_data with size of
 152  * @flexbg_size.
 153  *
 154  * Returns NULL on failure otherwise address of the allocated structure.
 155  */
 156 static struct ext4_new_flex_group_data *alloc_flex_gd(unsigned long flexbg_size)
 157 {
 158         struct ext4_new_flex_group_data *flex_gd;
 159
 160         flex_gd = kmalloc(sizeof(*flex_gd), GFP_NOFS);
 161         if (flex_gd == NULL)
 162                 goto out3;
 163
 164         if (flexbg_size >= UINT_MAX / sizeof(struct ext4_new_flex_group_data))
 165                 goto out2;
 166         flex_gd->count = flexbg_size;
 167
 168         flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
 169                                   flexbg_size, GFP_NOFS);
 170         if (flex_gd->groups == NULL)
 171                 goto out2;
 172
 173         flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
 174         if (flex_gd->bg_flags == NULL)
 175                 goto out1;
 176
 177         return flex_gd;
 178
 179 out1:
 180         kfree(flex_gd->groups);
 181 out2:
 182         kfree(flex_gd);
 183 out3:
 184         return NULL;
 185 }
 186
 187 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 188 {
 189         kfree(flex_gd->bg_flags);
 190         kfree(flex_gd->groups);
 191         kfree(flex_gd);
 192 }
 193
 194 /*
 195  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 196  * and inode tables for a flex group.
 197  *
 198  * This function is used by 64bit-resize.  Note that this function allocates
 199  * group tables from the 1st group of groups contained by @flexgd, which may
 200  * be a partial of a flex group.
 201  *
 202  * @sb: super block of fs to which the groups belongs
 203  */
 204 static void ext4_alloc_group_tables(struct super_block *sb,
 205                                 struct ext4_new_flex_group_data *flex_gd,
 206                                 int flexbg_size)
 207 {
 208         struct ext4_new_group_data *group_data = flex_gd->groups;
 209         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 210         ext4_fsblk_t start_blk;
 211         ext4_fsblk_t last_blk;
 212         ext4_group_t src_group;
 213         ext4_group_t bb_index = 0;
 214         ext4_group_t ib_index = 0;
 215         ext4_group_t it_index = 0;
 216         ext4_group_t group;
 217         ext4_group_t last_group;
 218         unsigned overhead;
 219
 220         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 221
 222         src_group = group_data[0].group;
 223         last_group  = src_group + flex_gd->count - 1;
 224
 225         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 226                (last_group & ~(flexbg_size - 1))));
 227 next_group:
 228         group = group_data[0].group;
 229         start_blk = ext4_group_first_block_no(sb, src_group);
 230         last_blk = start_blk + group_data[src_group - group].blocks_count;
 231
 232         overhead = ext4_bg_has_super(sb, src_group) ?
 233                    (1 + ext4_bg_num_gdb(sb, src_group) +
 234                     le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
 235
 236         start_blk += overhead;
 237
 238         BUG_ON(src_group >= group_data[0].group + flex_gd->count);
 239         /* We collect contiguous blocks as much as possible. */
 240         src_group++;
 241         for (; src_group <= last_group; src_group++)
 242                 if (!ext4_bg_has_super(sb, src_group))
 243                         last_blk += group_data[src_group - group].blocks_count;
 244                 else
 245                         break;
 246
 247         /* Allocate block bitmaps */
 248         for (; bb_index < flex_gd->count; bb_index++) {
 249                 if (start_blk >= last_blk)
 250                         goto next_group;
 251                 group_data[bb_index].block_bitmap = start_blk++;
 252                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 253                 group -= group_data[0].group;
 254                 group_data[group].free_blocks_count--;
 255                 if (flexbg_size > 1)
 256                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 257         }
 258
 259         /* Allocate inode bitmaps */
 260         for (; ib_index < flex_gd->count; ib_index++) {
 261                 if (start_blk >= last_blk)
 262                         goto next_group;
 263                 group_data[ib_index].inode_bitmap = start_blk++;
 264                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 265                 group -= group_data[0].group;
 266                 group_data[group].free_blocks_count--;
 267                 if (flexbg_size > 1)
 268                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 269         }
 270
 271         /* Allocate inode tables */
 272         for (; it_index < flex_gd->count; it_index++) {
 273                 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
 274                         goto next_group;
 275                 group_data[it_index].inode_table = start_blk;
 276                 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
 277                 group -= group_data[0].group;
 278                 group_data[group].free_blocks_count -=
 279                                         EXT4_SB(sb)->s_itb_per_group;
 280                 if (flexbg_size > 1)
 281                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 282
 283                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 284         }
 285
 286         if (test_opt(sb, DEBUG)) {
 287                 int i;
 288                 group = group_data[0].group;
 289
 290                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 291                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 292                        flexbg_size);
 293
 294                 for (i = 0; i < flex_gd->count; i++) {
 295                         printk(KERN_DEBUG "adding %s group %u: %u "
 296                                "blocks (%d free)\n",
 297                                ext4_bg_has_super(sb, group + i) ? "normal" :
 298                                "no-super", group + i,
 299                                group_data[i].blocks_count,
 300                                group_data[i].free_blocks_count);
 301                 }
 302         }
 303 }
 304
 305 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 306                                   ext4_fsblk_t blk)
 307 {
 308         struct buffer_head *bh;
 309         int err;
 310
 311         bh = sb_getblk(sb, blk);
 312         if (!bh)
 313                 return ERR_PTR(-EIO);
 314         if ((err = ext4_journal_get_write_access(handle, bh))) {
 315                 brelse(bh);
 316                 bh = ERR_PTR(err);
 317         } else {
 318                 memset(bh->b_data, 0, sb->s_blocksize);
 319                 set_buffer_uptodate(bh);
 320         }
 321
 322         return bh;
 323 }
 324
 325 /*
 326  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 327  * If that fails, restart the transaction & regain write access for the
 328  * buffer head which is used for block_bitmap modifications.
 329  */
 330 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 331 {
 332         int err;
 333
 334         if (ext4_handle_has_enough_credits(handle, thresh))
 335                 return 0;
 336
 337         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 338         if (err < 0)
 339                 return err;
 340         if (err) {
 341                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 342                 if (err)
 343                         return err;
 344         }
 345
 346         return 0;
 347 }
 348
 349 /*
 350  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 351  *
 352  * Helper function for ext4_setup_new_group_blocks() which set .
 353  *
 354  * @sb: super block
 355  * @handle: journal handle
 356  * @flex_gd: flex group data
 357  */
 358 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 359                         struct ext4_new_flex_group_data *flex_gd,
 360                         ext4_fsblk_t block, ext4_group_t count)
 361 {
 362         ext4_group_t count2;
 363
 364         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 365         for (count2 = count; count > 0; count -= count2, block += count2) {
 366                 ext4_fsblk_t start;
 367                 struct buffer_head *bh;
 368                 ext4_group_t group;
 369                 int err;
 370
 371                 ext4_get_group_no_and_offset(sb, block, &group, NULL);
 372                 start = ext4_group_first_block_no(sb, group);
 373                 group -= flex_gd->groups[0].group;
 374
 375                 count2 = sb->s_blocksize * 8 - (block - start);
 376                 if (count2 > count)
 377                         count2 = count;
 378
 379                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 380                         BUG_ON(flex_gd->count > 1);
 381                         continue;
 382                 }
 383
 384                 err = extend_or_restart_transaction(handle, 1);
 385                 if (err)
 386                         return err;
 387
 388                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 389                 if (!bh)
 390                         return -EIO;
 391
 392                 err = ext4_journal_get_write_access(handle, bh);
 393                 if (err)
 394                         return err;
 395                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 396                            block - start, count2);
 397                 ext4_set_bits(bh->b_data, block - start, count2);
 398
 399                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 400                 if (unlikely(err))
 401                         return err;
 402                 brelse(bh);
 403         }
 404
 405         return 0;
 406 }
 407
 408 /*
 409  * Set up the block and inode bitmaps, and the inode table for the new groups.
 410  * This doesn't need to be part of the main transaction, since we are only
 411  * changing blocks outside the actual filesystem.  We still do journaling to
 412  * ensure the recovery is correct in case of a failure just after resize.
 413  * If any part of this fails, we simply abort the resize.
 414  *
 415  * setup_new_flex_group_blocks handles a flex group as follow:
 416  *  1. copy super block and GDT, and initialize group tables if necessary.
 417  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 418  *     super block and GDT.
 419  *  2. allocate group tables in block bitmaps, that is, set bits in block
 420  *     bitmap for blocks taken by group tables.
 421  */
 422 static int setup_new_flex_group_blocks(struct super_block *sb,
 423                                 struct ext4_new_flex_group_data *flex_gd)
 424 {
 425         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 426         ext4_fsblk_t start;
 427         ext4_fsblk_t block;
 428         struct ext4_sb_info *sbi = EXT4_SB(sb);
 429         struct ext4_super_block *es = sbi->s_es;
 430         struct ext4_new_group_data *group_data = flex_gd->groups;
 431         __u16 *bg_flags = flex_gd->bg_flags;
 432         handle_t *handle;
 433         ext4_group_t group, count;
 434         struct buffer_head *bh = NULL;
 435         int reserved_gdb, i, j, err = 0, err2;
 436
 437         BUG_ON(!flex_gd->count || !group_data ||
 438                group_data[0].group != sbi->s_groups_count);
 439
 440         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 441
 442         /* This transaction may be extended/restarted along the way */
 443         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 444         if (IS_ERR(handle))
 445                 return PTR_ERR(handle);
 446
 447         group = group_data[0].group;
 448         for (i = 0; i < flex_gd->count; i++, group++) {
 449                 unsigned long gdblocks;
 450
 451                 gdblocks = ext4_bg_num_gdb(sb, group);
 452                 start = ext4_group_first_block_no(sb, group);
 453
 454                 /* Copy all of the GDT blocks into the backup in this group */
 455                 for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
 456                         struct buffer_head *gdb;
 457
 458                         ext4_debug("update backup group %#04llx\n", block);
 459                         err = extend_or_restart_transaction(handle, 1);
 460                         if (err)
 461                                 goto out;
 462
 463                         gdb = sb_getblk(sb, block);
 464                         if (!gdb) {
 465                                 err = -EIO;
 466                                 goto out;
 467                         }
 468
 469                         err = ext4_journal_get_write_access(handle, gdb);
 470                         if (err) {
 471                                 brelse(gdb);
 472                                 goto out;
 473                         }
 474                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 475                                gdb->b_size);
 476                         set_buffer_uptodate(gdb);
 477
 478                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 479                         if (unlikely(err)) {
 480                                 brelse(gdb);
 481                                 goto out;
 482                         }
 483                         brelse(gdb);
 484                 }
 485
 486                 /* Zero out all of the reserved backup group descriptor
 487                  * table blocks
 488                  */
 489                 if (ext4_bg_has_super(sb, group)) {
 490                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 491                                         reserved_gdb, GFP_NOFS);
 492                         if (err)
 493                                 goto out;
 494                 }
 495
 496                 /* Initialize group tables of the grop @group */
 497                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 498                         goto handle_bb;
 499
 500                 /* Zero out all of the inode table blocks */
 501                 block = group_data[i].inode_table;
 502                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 503                            block, sbi->s_itb_per_group);
 504                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 505                                        GFP_NOFS);
 506                 if (err)
 507                         goto out;
 508
 509 handle_bb:
 510                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 511                         goto handle_ib;
 512
 513                 /* Initialize block bitmap of the @group */
 514                 block = group_data[i].block_bitmap;
 515                 err = extend_or_restart_transaction(handle, 1);
 516                 if (err)
 517                         goto out;
 518
 519                 bh = bclean(handle, sb, block);
 520                 if (IS_ERR(bh)) {
 521                         err = PTR_ERR(bh);
 522                         goto out;
 523                 }
 524                 if (ext4_bg_has_super(sb, group)) {
 525                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 526                                    start);
 527                         ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
 528                                                      1);
 529                 }
 530                 ext4_mark_bitmap_end(group_data[i].blocks_count,
 531                                      sb->s_blocksize * 8, bh->b_data);
 532                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 533                 if (err)
 534                         goto out;
 535                 brelse(bh);
 536
 537 handle_ib:
 538                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 539                         continue;
 540
 541                 /* Initialize inode bitmap of the @group */
 542                 block = group_data[i].inode_bitmap;
 543                 err = extend_or_restart_transaction(handle, 1);
 544                 if (err)
 545                         goto out;
 546                 /* Mark unused entries in inode bitmap used */
 547                 bh = bclean(handle, sb, block);
 548                 if (IS_ERR(bh)) {
 549                         err = PTR_ERR(bh);
 550                         goto out;
 551                 }
 552
 553                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 554                                      sb->s_blocksize * 8, bh->b_data);
 555                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 556                 if (err)
 557                         goto out;
 558                 brelse(bh);
 559         }
 560         bh = NULL;
 561
 562         /* Mark group tables in block bitmap */
 563         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 564                 count = group_table_count[j];
 565                 start = (&group_data[0].block_bitmap)[j];
 566                 block = start;
 567                 for (i = 1; i < flex_gd->count; i++) {
 568                         block += group_table_count[j];
 569                         if (block == (&group_data[i].block_bitmap)[j]) {
 570                                 count += group_table_count[j];
 571                                 continue;
 572                         }
 573                         err = set_flexbg_block_bitmap(sb, handle,
 574                                                 flex_gd, start, count);
 575                         if (err)
 576                                 goto out;
 577                         count = group_table_count[j];
 578                         start = group_data[i].block_bitmap;
 579                         block = start;
 580                 }
 581
 582                 if (count) {
 583                         err = set_flexbg_block_bitmap(sb, handle,
 584                                                 flex_gd, start, count);
 585                         if (err)
 586                                 goto out;
 587                 }
 588         }
 589
 590 out:
 591         brelse(bh);
 592         err2 = ext4_journal_stop(handle);
 593         if (err2 && !err)
 594                 err = err2;
 595
 596         return err;
 597 }
 598
 599 /*
 600  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 601  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 602  * calling this for the first time.  In a sparse filesystem it will be the
 603  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 604  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 605  */
 606 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 607                                   unsigned *five, unsigned *seven)
 608 {
 609         unsigned *min = three;
 610         int mult = 3;
 611         unsigned ret;
 612
 613         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 614                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 615                 ret = *min;
 616                 *min += 1;
 617                 return ret;
 618         }
 619
 620         if (*five < *min) {
 621                 min = five;
 622                 mult = 5;
 623         }
 624         if (*seven < *min) {
 625                 min = seven;
 626                 mult = 7;
 627         }
 628
 629         ret = *min;
 630         *min *= mult;
 631
 632         return ret;
 633 }
 634
 635 /*
 636  * Check that all of the backup GDT blocks are held in the primary GDT block.
 637  * It is assumed that they are stored in group order.  Returns the number of
 638  * groups in current filesystem that have BACKUPS, or -ve error code.
 639  */
 640 static int verify_reserved_gdb(struct super_block *sb,
 641                                ext4_group_t end,
 642                                struct buffer_head *primary)
 643 {
 644         const ext4_fsblk_t blk = primary->b_blocknr;
 645         unsigned three = 1;
 646         unsigned five = 5;
 647         unsigned seven = 7;
 648         unsigned grp;
 649         __le32 *p = (__le32 *)primary->b_data;
 650         int gdbackups = 0;
 651
 652         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 653                 if (le32_to_cpu(*p++) !=
 654                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 655                         ext4_warning(sb, "reserved GDT %llu"
 656                                      " missing grp %d (%llu)",
 657                                      blk, grp,
 658                                      grp *
 659                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 660                                      blk);
 661                         return -EINVAL;
 662                 }
 663                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 664                         return -EFBIG;
 665         }
 666
 667         return gdbackups;
 668 }
 669
 670 /*
 671  * Called when we need to bring a reserved group descriptor table block into
 672  * use from the resize inode.  The primary copy of the new GDT block currently
 673  * is an indirect block (under the double indirect block in the resize inode).
 674  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 675  * block, in group order.  Even though we know all the block numbers we need,
 676  * we check to ensure that the resize inode has actually reserved these blocks.
 677  *
 678  * Don't need to update the block bitmaps because the blocks are still in use.
 679  *
 680  * We get all of the error cases out of the way, so that we are sure to not
 681  * fail once we start modifying the data on disk, because JBD has no rollback.
 682  */
 683 static int add_new_gdb(handle_t *handle, struct inode *inode,
 684                        ext4_group_t group)
 685 {
 686         struct super_block *sb = inode->i_sb;
 687         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 688         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 689         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 690         struct buffer_head **o_group_desc, **n_group_desc;
 691         struct buffer_head *dind;
 692         struct buffer_head *gdb_bh;
 693         int gdbackups;
 694         struct ext4_iloc iloc;
 695         __le32 *data;
 696         int err;
 697
 698         if (test_opt(sb, DEBUG))
 699                 printk(KERN_DEBUG
 700                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 701                        gdb_num);
 702
 703         /*
 704          * If we are not using the primary superblock/GDT copy don't resize,
 705          * because the user tools have no way of handling this.  Probably a
 706          * bad time to do it anyways.
 707          */
 708         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 709             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 710                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 711                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 712                 return -EPERM;
 713         }
 714
 715         gdb_bh = sb_bread(sb, gdblock);
 716         if (!gdb_bh)
 717                 return -EIO;
 718
 719         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 720         if (gdbackups < 0) {
 721                 err = gdbackups;
 722                 goto exit_bh;
 723         }
 724
 725         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 726         dind = sb_bread(sb, le32_to_cpu(*data));
 727         if (!dind) {
 728                 err = -EIO;
 729                 goto exit_bh;
 730         }
 731
 732         data = (__le32 *)dind->b_data;
 733         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 734                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 735                              group, gdblock);
 736                 err = -EINVAL;
 737                 goto exit_dind;
 738         }
 739
 740         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 741         if (unlikely(err))
 742                 goto exit_dind;
 743
 744         err = ext4_journal_get_write_access(handle, gdb_bh);
 745         if (unlikely(err))
 746                 goto exit_sbh;
 747
 748         err = ext4_journal_get_write_access(handle, dind);
 749         if (unlikely(err))
 750                 ext4_std_error(sb, err);
 751
 752         /* ext4_reserve_inode_write() gets a reference on the iloc */
 753         err = ext4_reserve_inode_write(handle, inode, &iloc);
 754         if (unlikely(err))
 755                 goto exit_dindj;
 756
 757         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 758                                      sizeof(struct buffer_head *),
 759                                      GFP_NOFS);
 760         if (!n_group_desc) {
 761                 err = -ENOMEM;
 762                 ext4_warning(sb, "not enough memory for %lu groups",
 763                              gdb_num + 1);
 764                 goto exit_inode;
 765         }
 766
 767         /*
 768          * Finally, we have all of the possible failures behind us...
 769          *
 770          * Remove new GDT block from inode double-indirect block and clear out
 771          * the new GDT block for use (which also "frees" the backup GDT blocks
 772          * from the reserved inode).  We don't need to change the bitmaps for
 773          * these blocks, because they are marked as in-use from being in the
 774          * reserved inode, and will become GDT blocks (primary and backup).
 775          */
 776         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 777         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 778         if (unlikely(err)) {
 779                 ext4_std_error(sb, err);
 780                 goto exit_inode;
 781         }
 782         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 783         ext4_mark_iloc_dirty(handle, inode, &iloc);
 784         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 785         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 786         if (unlikely(err)) {
 787                 ext4_std_error(sb, err);
 788                 goto exit_inode;
 789         }
 790         brelse(dind);
 791
 792         o_group_desc = EXT4_SB(sb)->s_group_desc;
 793         memcpy(n_group_desc, o_group_desc,
 794                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 795         n_group_desc[gdb_num] = gdb_bh;
 796         EXT4_SB(sb)->s_group_desc = n_group_desc;
 797         EXT4_SB(sb)->s_gdb_count++;
 798         ext4_kvfree(o_group_desc);
 799
 800         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 801         err = ext4_handle_dirty_super(handle, sb);
 802         if (err)
 803                 ext4_std_error(sb, err);
 804
 805         return err;
 806
 807 exit_inode:
 808         ext4_kvfree(n_group_desc);
 809         /* ext4_handle_release_buffer(handle, iloc.bh); */
 810         brelse(iloc.bh);
 811 exit_dindj:
 812         /* ext4_handle_release_buffer(handle, dind); */
 813 exit_sbh:
 814         /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
 815 exit_dind:
 816         brelse(dind);
 817 exit_bh:
 818         brelse(gdb_bh);
 819
 820         ext4_debug("leaving with error %d\n", err);
 821         return err;
 822 }
 823
 824 /*
 825  * Called when we are adding a new group which has a backup copy of each of
 826  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 827  * We need to add these reserved backup GDT blocks to the resize inode, so
 828  * that they are kept for future resizing and not allocated to files.
 829  *
 830  * Each reserved backup GDT block will go into a different indirect block.
 831  * The indirect blocks are actually the primary reserved GDT blocks,
 832  * so we know in advance what their block numbers are.  We only get the
 833  * double-indirect block to verify it is pointing to the primary reserved
 834  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 835  * backup GDT blocks are stored in their reserved primary GDT block.
 836  */
 837 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 838                               ext4_group_t group)
 839 {
 840         struct super_block *sb = inode->i_sb;
 841         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 842         struct buffer_head **primary;
 843         struct buffer_head *dind;
 844         struct ext4_iloc iloc;
 845         ext4_fsblk_t blk;
 846         __le32 *data, *end;
 847         int gdbackups = 0;
 848         int res, i;
 849         int err;
 850
 851         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 852         if (!primary)
 853                 return -ENOMEM;
 854
 855         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 856         dind = sb_bread(sb, le32_to_cpu(*data));
 857         if (!dind) {
 858                 err = -EIO;
 859                 goto exit_free;
 860         }
 861
 862         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 863         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 864                                          EXT4_ADDR_PER_BLOCK(sb));
 865         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 866
 867         /* Get each reserved primary GDT block and verify it holds backups */
 868         for (res = 0; res < reserved_gdb; res++, blk++) {
 869                 if (le32_to_cpu(*data) != blk) {
 870                         ext4_warning(sb, "reserved block %llu"
 871                                      " not at offset %ld",
 872                                      blk,
 873                                      (long)(data - (__le32 *)dind->b_data));
 874                         err = -EINVAL;
 875                         goto exit_bh;
 876                 }
 877                 primary[res] = sb_bread(sb, blk);
 878                 if (!primary[res]) {
 879                         err = -EIO;
 880                         goto exit_bh;
 881                 }
 882                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 883                 if (gdbackups < 0) {
 884                         brelse(primary[res]);
 885                         err = gdbackups;
 886                         goto exit_bh;
 887                 }
 888                 if (++data >= end)
 889                         data = (__le32 *)dind->b_data;
 890         }
 891
 892         for (i = 0; i < reserved_gdb; i++) {
 893                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
 894                         /*
 895                         int j;
 896                         for (j = 0; j < i; j++)
 897                                 ext4_handle_release_buffer(handle, primary[j]);
 898                          */
 899                         goto exit_bh;
 900                 }
 901         }
 902
 903         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 904                 goto exit_bh;
 905
 906         /*
 907          * Finally we can add each of the reserved backup GDT blocks from
 908          * the new group to its reserved primary GDT block.
 909          */
 910         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
 911         for (i = 0; i < reserved_gdb; i++) {
 912                 int err2;
 913                 data = (__le32 *)primary[i]->b_data;
 914                 /* printk("reserving backup %lu[%u] = %lu\n",
 915                        primary[i]->b_blocknr, gdbackups,
 916                        blk + primary[i]->b_blocknr); */
 917                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 918                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 919                 if (!err)
 920                         err = err2;
 921         }
 922         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 923         ext4_mark_iloc_dirty(handle, inode, &iloc);
 924
 925 exit_bh:
 926         while (--res >= 0)
 927                 brelse(primary[res]);
 928         brelse(dind);
 929
 930 exit_free:
 931         kfree(primary);
 932
 933         return err;
 934 }
 935
 936 /*
 937  * Update the backup copies of the ext4 metadata.  These don't need to be part
 938  * of the main resize transaction, because e2fsck will re-write them if there
 939  * is a problem (basically only OOM will cause a problem).  However, we
 940  * _should_ update the backups if possible, in case the primary gets trashed
 941  * for some reason and we need to run e2fsck from a backup superblock.  The
 942  * important part is that the new block and inode counts are in the backup
 943  * superblocks, and the location of the new group metadata in the GDT backups.
 944  *
 945  * We do not need take the s_resize_lock for this, because these
 946  * blocks are not otherwise touched by the filesystem code when it is
 947  * mounted.  We don't need to worry about last changing from
 948  * sbi->s_groups_count, because the worst that can happen is that we
 949  * do not copy the full number of backups at this time.  The resize
 950  * which changed s_groups_count will backup again.
 951  */
 952 static void update_backups(struct super_block *sb,
 953                            int blk_off, char *data, int size)
 954 {
 955         struct ext4_sb_info *sbi = EXT4_SB(sb);
 956         const ext4_group_t last = sbi->s_groups_count;
 957         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
 958         unsigned three = 1;
 959         unsigned five = 5;
 960         unsigned seven = 7;
 961         ext4_group_t group;
 962         int rest = sb->s_blocksize - size;
 963         handle_t *handle;
 964         int err = 0, err2;
 965
 966         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 967         if (IS_ERR(handle)) {
 968                 group = 1;
 969                 err = PTR_ERR(handle);
 970                 goto exit_err;
 971         }
 972
 973         ext4_superblock_csum_set(sb, (struct ext4_super_block *)data);
 974
 975         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
 976                 struct buffer_head *bh;
 977
 978                 /* Out of journal space, and can't get more - abort - so sad */
 979                 if (ext4_handle_valid(handle) &&
 980                     handle->h_buffer_credits == 0 &&
 981                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
 982                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 983                         break;
 984
 985                 bh = sb_getblk(sb, group * bpg + blk_off);
 986                 if (!bh) {
 987                         err = -EIO;
 988                         break;
 989                 }
 990                 ext4_debug("update metadata backup %#04lx\n",
 991                           (unsigned long)bh->b_blocknr);
 992                 if ((err = ext4_journal_get_write_access(handle, bh)))
 993                         break;
 994                 lock_buffer(bh);
 995                 memcpy(bh->b_data, data, size);
 996                 if (rest)
 997                         memset(bh->b_data + size, 0, rest);
 998                 set_buffer_uptodate(bh);
 999                 unlock_buffer(bh);
1000                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1001                 if (unlikely(err))
1002                         ext4_std_error(sb, err);
1003                 brelse(bh);
1004         }
1005         if ((err2 = ext4_journal_stop(handle)) && !err)
1006                 err = err2;
1007
1008         /*
1009          * Ugh! Need to have e2fsck write the backup copies.  It is too
1010          * late to revert the resize, we shouldn't fail just because of
1011          * the backup copies (they are only needed in case of corruption).
1012          *
1013          * However, if we got here we have a journal problem too, so we
1014          * can't really start a transaction to mark the superblock.
1015          * Chicken out and just set the flag on the hope it will be written
1016          * to disk, and if not - we will simply wait until next fsck.
1017          */
1018 exit_err:
1019         if (err) {
1020                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1021                              "forcing fsck on next reboot", group, err);
1022                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1023                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1024                 mark_buffer_dirty(sbi->s_sbh);
1025         }
1026 }
1027
1028 /*
1029  * ext4_add_new_descs() adds @count group descriptor of groups
1030  * starting at @group
1031  *
1032  * @handle: journal handle
1033  * @sb: super block
1034  * @group: the group no. of the first group desc to be added
1035  * @resize_inode: the resize inode
1036  * @count: number of group descriptors to be added
1037  */
1038 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1039                               ext4_group_t group, struct inode *resize_inode,
1040                               ext4_group_t count)
1041 {
1042         struct ext4_sb_info *sbi = EXT4_SB(sb);
1043         struct ext4_super_block *es = sbi->s_es;
1044         struct buffer_head *gdb_bh;
1045         int i, gdb_off, gdb_num, err = 0;
1046
1047         for (i = 0; i < count; i++, group++) {
1048                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1049                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1050
1051                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1052                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1053
1054                 /*
1055                  * We will only either add reserved group blocks to a backup group
1056                  * or remove reserved blocks for the first group in a new group block.
1057                  * Doing both would be mean more complex code, and sane people don't
1058                  * use non-sparse filesystems anymore.  This is already checked above.
1059                  */
1060                 if (gdb_off) {
1061                         gdb_bh = sbi->s_group_desc[gdb_num];
1062                         err = ext4_journal_get_write_access(handle, gdb_bh);
1063
1064                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1065                                 err = reserve_backup_gdb(handle, resize_inode, group);
1066                 } else
1067                         err = add_new_gdb(handle, resize_inode, group);
1068                 if (err)
1069                         break;
1070         }
1071         return err;
1072 }
1073
1074 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1075 {
1076         struct buffer_head *bh = sb_getblk(sb, block);
1077         if (!bh)
1078                 return NULL;
1079
1080         if (bitmap_uptodate(bh))
1081                 return bh;
1082
1083         lock_buffer(bh);
1084         if (bh_submit_read(bh) < 0) {
1085                 unlock_buffer(bh);
1086                 brelse(bh);
1087                 return NULL;
1088         }
1089         unlock_buffer(bh);
1090
1091         return bh;
1092 }
1093
1094 static int ext4_set_bitmap_checksums(struct super_block *sb,
1095                                      ext4_group_t group,
1096                                      struct ext4_group_desc *gdp,
1097                                      struct ext4_new_group_data *group_data)
1098 {
1099         struct buffer_head *bh;
1100
1101         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1102                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1103                 return 0;
1104
1105         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1106         if (!bh)
1107                 return -EIO;
1108         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1109                                    EXT4_INODES_PER_GROUP(sb) / 8);
1110         brelse(bh);
1111
1112         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1113         if (!bh)
1114                 return -EIO;
1115         ext4_block_bitmap_csum_set(sb, group, gdp, bh,
1116                                    EXT4_BLOCKS_PER_GROUP(sb) / 8);
1117         brelse(bh);
1118
1119         return 0;
1120 }
1121
1122 /*
1123  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1124  */
1125 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1126                                 struct ext4_new_flex_group_data *flex_gd)
1127 {
1128         struct ext4_new_group_data      *group_data = flex_gd->groups;
1129         struct ext4_group_desc          *gdp;
1130         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1131         struct buffer_head              *gdb_bh;
1132         ext4_group_t                    group;
1133         __u16                           *bg_flags = flex_gd->bg_flags;
1134         int                             i, gdb_off, gdb_num, err = 0;
1135
1136
1137         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1138                 group = group_data->group;
1139
1140                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1141                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1142
1143                 /*
1144                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1145                  */
1146                 gdb_bh = sbi->s_group_desc[gdb_num];
1147                 /* Update group descriptor block for new group */
1148                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1149                                                  gdb_off * EXT4_DESC_SIZE(sb));
1150
1151                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1152                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1153                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1154                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1155                 if (err) {
1156                         ext4_std_error(sb, err);
1157                         break;
1158                 }
1159
1160                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1161                 ext4_free_group_clusters_set(sb, gdp,
1162                                              EXT4_B2C(sbi, group_data->free_blocks_count));
1163                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1164                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1165                 ext4_group_desc_csum_set(sb, group, gdp);
1166
1167                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1168                 if (unlikely(err)) {
1169                         ext4_std_error(sb, err);
1170                         break;
1171                 }
1172
1173                 /*
1174                  * We can allocate memory for mb_alloc based on the new group
1175                  * descriptor
1176                  */
1177                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1178                 if (err)
1179                         break;
1180         }
1181         return err;
1182 }
1183
1184 /*
1185  * ext4_update_super() updates the super block so that the newly added
1186  * groups can be seen by the filesystem.
1187  *
1188  * @sb: super block
1189  * @flex_gd: new added groups
1190  */
1191 static void ext4_update_super(struct super_block *sb,
1192                              struct ext4_new_flex_group_data *flex_gd)
1193 {
1194         ext4_fsblk_t blocks_count = 0;
1195         ext4_fsblk_t free_blocks = 0;
1196         ext4_fsblk_t reserved_blocks = 0;
1197         struct ext4_new_group_data *group_data = flex_gd->groups;
1198         struct ext4_sb_info *sbi = EXT4_SB(sb);
1199         struct ext4_super_block *es = sbi->s_es;
1200         int i;
1201
1202         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1203         /*
1204          * Make the new blocks and inodes valid next.  We do this before
1205          * increasing the group count so that once the group is enabled,
1206          * all of its blocks and inodes are already valid.
1207          *
1208          * We always allocate group-by-group, then block-by-block or
1209          * inode-by-inode within a group, so enabling these
1210          * blocks/inodes before the group is live won't actually let us
1211          * allocate the new space yet.
1212          */
1213         for (i = 0; i < flex_gd->count; i++) {
1214                 blocks_count += group_data[i].blocks_count;
1215                 free_blocks += group_data[i].free_blocks_count;
1216         }
1217
1218         reserved_blocks = ext4_r_blocks_count(es) * 100;
1219         do_div(reserved_blocks, ext4_blocks_count(es));
1220         reserved_blocks *= blocks_count;
1221         do_div(reserved_blocks, 100);
1222
1223         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1224         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1225         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1226                      flex_gd->count);
1227         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1228                      flex_gd->count);
1229
1230         /*
1231          * We need to protect s_groups_count against other CPUs seeing
1232          * inconsistent state in the superblock.
1233          *
1234          * The precise rules we use are:
1235          *
1236          * * Writers must perform a smp_wmb() after updating all
1237          *   dependent data and before modifying the groups count
1238          *
1239          * * Readers must perform an smp_rmb() after reading the groups
1240          *   count and before reading any dependent data.
1241          *
1242          * NB. These rules can be relaxed when checking the group count
1243          * while freeing data, as we can only allocate from a block
1244          * group after serialising against the group count, and we can
1245          * only then free after serialising in turn against that
1246          * allocation.
1247          */
1248         smp_wmb();
1249
1250         /* Update the global fs size fields */
1251         sbi->s_groups_count += flex_gd->count;
1252
1253         /* Update the reserved block counts only once the new group is
1254          * active. */
1255         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1256                                 reserved_blocks);
1257
1258         /* Update the free space counts */
1259         percpu_counter_add(&sbi->s_freeclusters_counter,
1260                            EXT4_B2C(sbi, free_blocks));
1261         percpu_counter_add(&sbi->s_freeinodes_counter,
1262                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1263
1264         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1265                                       EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1266             sbi->s_log_groups_per_flex) {
1267                 ext4_group_t flex_group;
1268                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1269                 atomic_add(EXT4_B2C(sbi, free_blocks),
1270                            &sbi->s_flex_groups[flex_group].free_clusters);
1271                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1272                            &sbi->s_flex_groups[flex_group].free_inodes);
1273         }
1274
1275         /*
1276          * Update the fs overhead information
1277          */
1278         ext4_calculate_overhead(sb);
1279
1280         if (test_opt(sb, DEBUG))
1281                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1282                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1283                        blocks_count, free_blocks, reserved_blocks);
1284 }
1285
1286 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1287  * _before_ we start modifying the filesystem, because we cannot abort the
1288  * transaction and not have it write the data to disk.
1289  */
1290 static int ext4_flex_group_add(struct super_block *sb,
1291                                struct inode *resize_inode,
1292                                struct ext4_new_flex_group_data *flex_gd)
1293 {
1294         struct ext4_sb_info *sbi = EXT4_SB(sb);
1295         struct ext4_super_block *es = sbi->s_es;
1296         ext4_fsblk_t o_blocks_count;
1297         ext4_grpblk_t last;
1298         ext4_group_t group;
1299         handle_t *handle;
1300         unsigned reserved_gdb;
1301         int err = 0, err2 = 0, credit;
1302
1303         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1304
1305         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1306         o_blocks_count = ext4_blocks_count(es);
1307         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1308         BUG_ON(last);
1309
1310         err = setup_new_flex_group_blocks(sb, flex_gd);
1311         if (err)
1312                 goto exit;
1313         /*
1314          * We will always be modifying at least the superblock and  GDT
1315          * block.  If we are adding a group past the last current GDT block,
1316          * we will also modify the inode and the dindirect block.  If we
1317          * are adding a group with superblock/GDT backups  we will also
1318          * modify each of the reserved GDT dindirect blocks.
1319          */
1320         credit = flex_gd->count * 4 + reserved_gdb;
1321         handle = ext4_journal_start_sb(sb, credit);
1322         if (IS_ERR(handle)) {
1323                 err = PTR_ERR(handle);
1324                 goto exit;
1325         }
1326
1327         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1328         if (err)
1329                 goto exit_journal;
1330
1331         group = flex_gd->groups[0].group;
1332         BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1333         err = ext4_add_new_descs(handle, sb, group,
1334                                 resize_inode, flex_gd->count);
1335         if (err)
1336                 goto exit_journal;
1337
1338         err = ext4_setup_new_descs(handle, sb, flex_gd);
1339         if (err)
1340                 goto exit_journal;
1341
1342         ext4_update_super(sb, flex_gd);
1343
1344         err = ext4_handle_dirty_super(handle, sb);
1345
1346 exit_journal:
1347         err2 = ext4_journal_stop(handle);
1348         if (!err)
1349                 err = err2;
1350
1351         if (!err) {
1352                 int i;
1353                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1354                                sizeof(struct ext4_super_block));
1355                 for (i = 0; i < flex_gd->count; i++, group++) {
1356                         struct buffer_head *gdb_bh;
1357                         int gdb_num;
1358                         gdb_num = group / EXT4_BLOCKS_PER_GROUP(sb);
1359                         gdb_bh = sbi->s_group_desc[gdb_num];
1360                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1361                                        gdb_bh->b_size);
1362                 }
1363         }
1364 exit:
1365         return err;
1366 }
1367
1368 static int ext4_setup_next_flex_gd(struct super_block *sb,
1369                                     struct ext4_new_flex_group_data *flex_gd,
1370                                     ext4_fsblk_t n_blocks_count,
1371                                     unsigned long flexbg_size)
1372 {
1373         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1374         struct ext4_new_group_data *group_data = flex_gd->groups;
1375         ext4_fsblk_t o_blocks_count;
1376         ext4_group_t n_group;
1377         ext4_group_t group;
1378         ext4_group_t last_group;
1379         ext4_grpblk_t last;
1380         ext4_grpblk_t blocks_per_group;
1381         unsigned long i;
1382
1383         blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1384
1385         o_blocks_count = ext4_blocks_count(es);
1386
1387         if (o_blocks_count == n_blocks_count)
1388                 return 0;
1389
1390         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1391         BUG_ON(last);
1392         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1393
1394         last_group = group | (flexbg_size - 1);
1395         if (last_group > n_group)
1396                 last_group = n_group;
1397
1398         flex_gd->count = last_group - group + 1;
1399
1400         for (i = 0; i < flex_gd->count; i++) {
1401                 int overhead;
1402
1403                 group_data[i].group = group + i;
1404                 group_data[i].blocks_count = blocks_per_group;
1405                 overhead = ext4_bg_has_super(sb, group + i) ?
1406                            (1 + ext4_bg_num_gdb(sb, group + i) +
1407                             le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
1408                 group_data[i].free_blocks_count = blocks_per_group - overhead;
1409                 if (ext4_has_group_desc_csum(sb))
1410                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1411                                                EXT4_BG_INODE_UNINIT;
1412                 else
1413                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1414         }
1415
1416         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1417                 /* We need to initialize block bitmap of last group. */
1418                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1419
1420         if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1421                 group_data[i - 1].blocks_count = last + 1;
1422                 group_data[i - 1].free_blocks_count -= blocks_per_group-
1423                                         last - 1;
1424         }
1425
1426         return 1;
1427 }
1428
1429 /* Add group descriptor data to an existing or new group descriptor block.
1430  * Ensure we handle all possible error conditions _before_ we start modifying
1431  * the filesystem, because we cannot abort the transaction and not have it
1432  * write the data to disk.
1433  *
1434  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1435  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1436  *
1437  * We only need to hold the superblock lock while we are actually adding
1438  * in the new group's counts to the superblock.  Prior to that we have
1439  * not really "added" the group at all.  We re-check that we are still
1440  * adding in the last group in case things have changed since verifying.
1441  */
1442 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1443 {
1444         struct ext4_new_flex_group_data flex_gd;
1445         struct ext4_sb_info *sbi = EXT4_SB(sb);
1446         struct ext4_super_block *es = sbi->s_es;
1447         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1448                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1449         struct inode *inode = NULL;
1450         int gdb_off, gdb_num;
1451         int err;
1452         __u16 bg_flags = 0;
1453
1454         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1455         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1456
1457         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1458                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1459                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1460                 return -EPERM;
1461         }
1462
1463         if (ext4_blocks_count(es) + input->blocks_count <
1464             ext4_blocks_count(es)) {
1465                 ext4_warning(sb, "blocks_count overflow");
1466                 return -EINVAL;
1467         }
1468
1469         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1470             le32_to_cpu(es->s_inodes_count)) {
1471                 ext4_warning(sb, "inodes_count overflow");
1472                 return -EINVAL;
1473         }
1474
1475         if (reserved_gdb || gdb_off == 0) {
1476                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1477                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
1478                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1479                         ext4_warning(sb,
1480                                      "No reserved GDT blocks, can't resize");
1481                         return -EPERM;
1482                 }
1483                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1484                 if (IS_ERR(inode)) {
1485                         ext4_warning(sb, "Error opening resize inode");
1486                         return PTR_ERR(inode);
1487                 }
1488         }
1489
1490
1491         err = verify_group_input(sb, input);
1492         if (err)
1493                 goto out;
1494
1495         flex_gd.count = 1;
1496         flex_gd.groups = input;
1497         flex_gd.bg_flags = &bg_flags;
1498         err = ext4_flex_group_add(sb, inode, &flex_gd);
1499 out:
1500         iput(inode);
1501         return err;
1502 } /* ext4_group_add */
1503
1504 /*
1505  * extend a group without checking assuming that checking has been done.
1506  */
1507 static int ext4_group_extend_no_check(struct super_block *sb,
1508                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1509 {
1510         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1511         handle_t *handle;
1512         int err = 0, err2;
1513
1514         /* We will update the superblock, one block bitmap, and
1515          * one group descriptor via ext4_group_add_blocks().
1516          */
1517         handle = ext4_journal_start_sb(sb, 3);
1518         if (IS_ERR(handle)) {
1519                 err = PTR_ERR(handle);
1520                 ext4_warning(sb, "error %d on journal start", err);
1521                 return err;
1522         }
1523
1524         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1525         if (err) {
1526                 ext4_warning(sb, "error %d on journal write access", err);
1527                 goto errout;
1528         }
1529
1530         ext4_blocks_count_set(es, o_blocks_count + add);
1531         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1532         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1533                    o_blocks_count + add);
1534         /* We add the blocks to the bitmap and set the group need init bit */
1535         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1536         if (err)
1537                 goto errout;
1538         ext4_handle_dirty_super(handle, sb);
1539         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1540                    o_blocks_count + add);
1541 errout:
1542         err2 = ext4_journal_stop(handle);
1543         if (err2 && !err)
1544                 err = err2;
1545
1546         if (!err) {
1547                 if (test_opt(sb, DEBUG))
1548                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1549                                "blocks\n", ext4_blocks_count(es));
1550                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1551                                sizeof(struct ext4_super_block));
1552         }
1553         return err;
1554 }
1555
1556 /*
1557  * Extend the filesystem to the new number of blocks specified.  This entry
1558  * point is only used to extend the current filesystem to the end of the last
1559  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1560  * for emergencies (because it has no dependencies on reserved blocks).
1561  *
1562  * If we _really_ wanted, we could use default values to call ext4_group_add()
1563  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1564  * GDT blocks are reserved to grow to the desired size.
1565  */
1566 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1567                       ext4_fsblk_t n_blocks_count)
1568 {
1569         ext4_fsblk_t o_blocks_count;
1570         ext4_grpblk_t last;
1571         ext4_grpblk_t add;
1572         struct buffer_head *bh;
1573         int err;
1574         ext4_group_t group;
1575
1576         o_blocks_count = ext4_blocks_count(es);
1577
1578         if (test_opt(sb, DEBUG))
1579                 ext4_msg(sb, KERN_DEBUG,
1580                          "extending last group from %llu to %llu blocks",
1581                          o_blocks_count, n_blocks_count);
1582
1583         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1584                 return 0;
1585
1586         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1587                 ext4_msg(sb, KERN_ERR,
1588                          "filesystem too large to resize to %llu blocks safely",
1589                          n_blocks_count);
1590                 if (sizeof(sector_t) < 8)
1591                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1592                 return -EINVAL;
1593         }
1594
1595         if (n_blocks_count < o_blocks_count) {
1596                 ext4_warning(sb, "can't shrink FS - resize aborted");
1597                 return -EINVAL;
1598         }
1599
1600         /* Handle the remaining blocks in the last group only. */
1601         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1602
1603         if (last == 0) {
1604                 ext4_warning(sb, "need to use ext2online to resize further");
1605                 return -EPERM;
1606         }
1607
1608         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1609
1610         if (o_blocks_count + add < o_blocks_count) {
1611                 ext4_warning(sb, "blocks_count overflow");
1612                 return -EINVAL;
1613         }
1614
1615         if (o_blocks_count + add > n_blocks_count)
1616                 add = n_blocks_count - o_blocks_count;
1617
1618         if (o_blocks_count + add < n_blocks_count)
1619                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1620                              o_blocks_count + add, add);
1621
1622         /* See if the device is actually as big as what was requested */
1623         bh = sb_bread(sb, o_blocks_count + add - 1);
1624         if (!bh) {
1625                 ext4_warning(sb, "can't read last block, resize aborted");
1626                 return -ENOSPC;
1627         }
1628         brelse(bh);
1629
1630         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1631         return err;
1632 } /* ext4_group_extend */
1633
1634 /*
1635  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1636  *
1637  * @sb: super block of the fs to be resized
1638  * @n_blocks_count: the number of blocks resides in the resized fs
1639  */
1640 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1641 {
1642         struct ext4_new_flex_group_data *flex_gd = NULL;
1643         struct ext4_sb_info *sbi = EXT4_SB(sb);
1644         struct ext4_super_block *es = sbi->s_es;
1645         struct buffer_head *bh;
1646         struct inode *resize_inode;
1647         ext4_fsblk_t o_blocks_count;
1648         ext4_group_t o_group;
1649         ext4_group_t n_group;
1650         ext4_grpblk_t offset, add;
1651         unsigned long n_desc_blocks;
1652         unsigned long o_desc_blocks;
1653         unsigned long desc_blocks;
1654         int err = 0, flexbg_size = 1;
1655
1656         o_blocks_count = ext4_blocks_count(es);
1657
1658         if (test_opt(sb, DEBUG))
1659                 ext4_msg(sb, KERN_DEBUG, "resizing filesystem from %llu "
1660                        "to %llu blocks", o_blocks_count, n_blocks_count);
1661
1662         if (n_blocks_count < o_blocks_count) {
1663                 /* On-line shrinking not supported */
1664                 ext4_warning(sb, "can't shrink FS - resize aborted");
1665                 return -EINVAL;
1666         }
1667
1668         if (n_blocks_count == o_blocks_count)
1669                 /* Nothing need to do */
1670                 return 0;
1671
1672         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &offset);
1673         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1674
1675         n_desc_blocks = (n_group + EXT4_DESC_PER_BLOCK(sb)) /
1676                         EXT4_DESC_PER_BLOCK(sb);
1677         o_desc_blocks = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1678                         EXT4_DESC_PER_BLOCK(sb);
1679         desc_blocks = n_desc_blocks - o_desc_blocks;
1680
1681         if (desc_blocks &&
1682             (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE) ||
1683              le16_to_cpu(es->s_reserved_gdt_blocks) < desc_blocks)) {
1684                 ext4_warning(sb, "No reserved GDT blocks, can't resize");
1685                 return -EPERM;
1686         }
1687
1688         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1689         if (IS_ERR(resize_inode)) {
1690                 ext4_warning(sb, "Error opening resize inode");
1691                 return PTR_ERR(resize_inode);
1692         }
1693
1694         /* See if the device is actually as big as what was requested */
1695         bh = sb_bread(sb, n_blocks_count - 1);
1696         if (!bh) {
1697                 ext4_warning(sb, "can't read last block, resize aborted");
1698                 return -ENOSPC;
1699         }
1700         brelse(bh);
1701
1702         /* extend the last group */
1703         if (n_group == o_group)
1704                 add = n_blocks_count - o_blocks_count;
1705         else
1706                 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1707         if (add > 0) {
1708                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1709                 if (err)
1710                         goto out;
1711         }
1712
1713         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1714             es->s_log_groups_per_flex)
1715                 flexbg_size = 1 << es->s_log_groups_per_flex;
1716
1717         o_blocks_count = ext4_blocks_count(es);
1718         if (o_blocks_count == n_blocks_count)
1719                 goto out;
1720
1721         flex_gd = alloc_flex_gd(flexbg_size);
1722         if (flex_gd == NULL) {
1723                 err = -ENOMEM;
1724                 goto out;
1725         }
1726
1727         /* Add flex groups. Note that a regular group is a
1728          * flex group with 1 group.
1729          */
1730         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1731                                               flexbg_size)) {
1732                 ext4_alloc_group_tables(sb, flex_gd, flexbg_size);
1733                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1734                 if (unlikely(err))
1735                         break;
1736         }
1737
1738 out:
1739         if (flex_gd)
1740                 free_flex_gd(flex_gd);
1741
1742         iput(resize_inode);
1743         if (test_opt(sb, DEBUG))
1744                 ext4_msg(sb, KERN_DEBUG, "resized filesystem from %llu "
1745                        "upto %llu blocks", o_blocks_count, n_blocks_count);
1746         return err;
1747 }