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         flex_gd->count = flexbg_size;
 165
 166         flex_gd->groups = kmalloc(sizeof(struct ext4_new_group_data) *
 167                                   flexbg_size, GFP_NOFS);
 168         if (flex_gd->groups == NULL)
 169                 goto out2;
 170
 171         flex_gd->bg_flags = kmalloc(flexbg_size * sizeof(__u16), GFP_NOFS);
 172         if (flex_gd->bg_flags == NULL)
 173                 goto out1;
 174
 175         return flex_gd;
 176
 177 out1:
 178         kfree(flex_gd->groups);
 179 out2:
 180         kfree(flex_gd);
 181 out3:
 182         return NULL;
 183 }
 184
 185 static void free_flex_gd(struct ext4_new_flex_group_data *flex_gd)
 186 {
 187         kfree(flex_gd->bg_flags);
 188         kfree(flex_gd->groups);
 189         kfree(flex_gd);
 190 }
 191
 192 /*
 193  * ext4_alloc_group_tables() allocates block bitmaps, inode bitmaps
 194  * and inode tables for a flex group.
 195  *
 196  * This function is used by 64bit-resize.  Note that this function allocates
 197  * group tables from the 1st group of groups contained by @flexgd, which may
 198  * be a partial of a flex group.
 199  *
 200  * @sb: super block of fs to which the groups belongs
 201  */
 202 static void ext4_alloc_group_tables(struct super_block *sb,
 203                                 struct ext4_new_flex_group_data *flex_gd,
 204                                 int flexbg_size)
 205 {
 206         struct ext4_new_group_data *group_data = flex_gd->groups;
 207         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 208         ext4_fsblk_t start_blk;
 209         ext4_fsblk_t last_blk;
 210         ext4_group_t src_group;
 211         ext4_group_t bb_index = 0;
 212         ext4_group_t ib_index = 0;
 213         ext4_group_t it_index = 0;
 214         ext4_group_t group;
 215         ext4_group_t last_group;
 216         unsigned overhead;
 217
 218         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 219
 220         src_group = group_data[0].group;
 221         last_group  = src_group + flex_gd->count - 1;
 222
 223         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 224                (last_group & ~(flexbg_size - 1))));
 225 next_group:
 226         group = group_data[0].group;
 227         start_blk = ext4_group_first_block_no(sb, src_group);
 228         last_blk = start_blk + group_data[src_group - group].blocks_count;
 229
 230         overhead = ext4_bg_has_super(sb, src_group) ?
 231                    (1 + ext4_bg_num_gdb(sb, src_group) +
 232                     le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
 233
 234         start_blk += overhead;
 235
 236         BUG_ON(src_group >= group_data[0].group + flex_gd->count);
 237         /* We collect contiguous blocks as much as possible. */
 238         src_group++;
 239         for (; src_group <= last_group; src_group++)
 240                 if (!ext4_bg_has_super(sb, src_group))
 241                         last_blk += group_data[src_group - group].blocks_count;
 242                 else
 243                         break;
 244
 245         /* Allocate block bitmaps */
 246         for (; bb_index < flex_gd->count; bb_index++) {
 247                 if (start_blk >= last_blk)
 248                         goto next_group;
 249                 group_data[bb_index].block_bitmap = start_blk++;
 250                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 251                 group -= group_data[0].group;
 252                 group_data[group].free_blocks_count--;
 253                 if (flexbg_size > 1)
 254                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 255         }
 256
 257         /* Allocate inode bitmaps */
 258         for (; ib_index < flex_gd->count; ib_index++) {
 259                 if (start_blk >= last_blk)
 260                         goto next_group;
 261                 group_data[ib_index].inode_bitmap = start_blk++;
 262                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 263                 group -= group_data[0].group;
 264                 group_data[group].free_blocks_count--;
 265                 if (flexbg_size > 1)
 266                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 267         }
 268
 269         /* Allocate inode tables */
 270         for (; it_index < flex_gd->count; it_index++) {
 271                 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
 272                         goto next_group;
 273                 group_data[it_index].inode_table = start_blk;
 274                 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
 275                 group -= group_data[0].group;
 276                 group_data[group].free_blocks_count -=
 277                                         EXT4_SB(sb)->s_itb_per_group;
 278                 if (flexbg_size > 1)
 279                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 280
 281                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 282         }
 283
 284         if (test_opt(sb, DEBUG)) {
 285                 int i;
 286                 group = group_data[0].group;
 287
 288                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 289                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 290                        flexbg_size);
 291
 292                 for (i = 0; i < flex_gd->count; i++) {
 293                         printk(KERN_DEBUG "adding %s group %u: %u "
 294                                "blocks (%d free)\n",
 295                                ext4_bg_has_super(sb, group + i) ? "normal" :
 296                                "no-super", group + i,
 297                                group_data[i].blocks_count,
 298                                group_data[i].free_blocks_count);
 299                 }
 300         }
 301 }
 302
 303 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 304                                   ext4_fsblk_t blk)
 305 {
 306         struct buffer_head *bh;
 307         int err;
 308
 309         bh = sb_getblk(sb, blk);
 310         if (!bh)
 311                 return ERR_PTR(-EIO);
 312         if ((err = ext4_journal_get_write_access(handle, bh))) {
 313                 brelse(bh);
 314                 bh = ERR_PTR(err);
 315         } else {
 316                 memset(bh->b_data, 0, sb->s_blocksize);
 317                 set_buffer_uptodate(bh);
 318         }
 319
 320         return bh;
 321 }
 322
 323 /*
 324  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 325  * If that fails, restart the transaction & regain write access for the
 326  * buffer head which is used for block_bitmap modifications.
 327  */
 328 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 329 {
 330         int err;
 331
 332         if (ext4_handle_has_enough_credits(handle, thresh))
 333                 return 0;
 334
 335         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 336         if (err < 0)
 337                 return err;
 338         if (err) {
 339                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 340                 if (err)
 341                         return err;
 342         }
 343
 344         return 0;
 345 }
 346
 347 /*
 348  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 349  *
 350  * Helper function for ext4_setup_new_group_blocks() which set .
 351  *
 352  * @sb: super block
 353  * @handle: journal handle
 354  * @flex_gd: flex group data
 355  */
 356 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 357                         struct ext4_new_flex_group_data *flex_gd,
 358                         ext4_fsblk_t block, ext4_group_t count)
 359 {
 360         ext4_group_t count2;
 361
 362         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 363         for (count2 = count; count > 0; count -= count2, block += count2) {
 364                 ext4_fsblk_t start;
 365                 struct buffer_head *bh;
 366                 ext4_group_t group;
 367                 int err;
 368
 369                 ext4_get_group_no_and_offset(sb, block, &group, NULL);
 370                 start = ext4_group_first_block_no(sb, group);
 371                 group -= flex_gd->groups[0].group;
 372
 373                 count2 = sb->s_blocksize * 8 - (block - start);
 374                 if (count2 > count)
 375                         count2 = count;
 376
 377                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 378                         BUG_ON(flex_gd->count > 1);
 379                         continue;
 380                 }
 381
 382                 err = extend_or_restart_transaction(handle, 1);
 383                 if (err)
 384                         return err;
 385
 386                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 387                 if (!bh)
 388                         return -EIO;
 389
 390                 err = ext4_journal_get_write_access(handle, bh);
 391                 if (err)
 392                         return err;
 393                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 394                            block - start, count2);
 395                 ext4_set_bits(bh->b_data, block - start, count2);
 396
 397                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 398                 if (unlikely(err))
 399                         return err;
 400                 brelse(bh);
 401         }
 402
 403         return 0;
 404 }
 405
 406 /*
 407  * Set up the block and inode bitmaps, and the inode table for the new groups.
 408  * This doesn't need to be part of the main transaction, since we are only
 409  * changing blocks outside the actual filesystem.  We still do journaling to
 410  * ensure the recovery is correct in case of a failure just after resize.
 411  * If any part of this fails, we simply abort the resize.
 412  *
 413  * setup_new_flex_group_blocks handles a flex group as follow:
 414  *  1. copy super block and GDT, and initialize group tables if necessary.
 415  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 416  *     super block and GDT.
 417  *  2. allocate group tables in block bitmaps, that is, set bits in block
 418  *     bitmap for blocks taken by group tables.
 419  */
 420 static int setup_new_flex_group_blocks(struct super_block *sb,
 421                                 struct ext4_new_flex_group_data *flex_gd)
 422 {
 423         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 424         ext4_fsblk_t start;
 425         ext4_fsblk_t block;
 426         struct ext4_sb_info *sbi = EXT4_SB(sb);
 427         struct ext4_super_block *es = sbi->s_es;
 428         struct ext4_new_group_data *group_data = flex_gd->groups;
 429         __u16 *bg_flags = flex_gd->bg_flags;
 430         handle_t *handle;
 431         ext4_group_t group, count;
 432         struct buffer_head *bh = NULL;
 433         int reserved_gdb, i, j, err = 0, err2;
 434
 435         BUG_ON(!flex_gd->count || !group_data ||
 436                group_data[0].group != sbi->s_groups_count);
 437
 438         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 439
 440         /* This transaction may be extended/restarted along the way */
 441         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 442         if (IS_ERR(handle))
 443                 return PTR_ERR(handle);
 444
 445         group = group_data[0].group;
 446         for (i = 0; i < flex_gd->count; i++, group++) {
 447                 unsigned long gdblocks;
 448
 449                 gdblocks = ext4_bg_num_gdb(sb, group);
 450                 start = ext4_group_first_block_no(sb, group);
 451
 452                 /* Copy all of the GDT blocks into the backup in this group */
 453                 for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
 454                         struct buffer_head *gdb;
 455
 456                         ext4_debug("update backup group %#04llx\n", block);
 457                         err = extend_or_restart_transaction(handle, 1);
 458                         if (err)
 459                                 goto out;
 460
 461                         gdb = sb_getblk(sb, block);
 462                         if (!gdb) {
 463                                 err = -EIO;
 464                                 goto out;
 465                         }
 466
 467                         err = ext4_journal_get_write_access(handle, gdb);
 468                         if (err) {
 469                                 brelse(gdb);
 470                                 goto out;
 471                         }
 472                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 473                                gdb->b_size);
 474                         set_buffer_uptodate(gdb);
 475
 476                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 477                         if (unlikely(err)) {
 478                                 brelse(gdb);
 479                                 goto out;
 480                         }
 481                         brelse(gdb);
 482                 }
 483
 484                 /* Zero out all of the reserved backup group descriptor
 485                  * table blocks
 486                  */
 487                 if (ext4_bg_has_super(sb, group)) {
 488                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 489                                         reserved_gdb, GFP_NOFS);
 490                         if (err)
 491                                 goto out;
 492                 }
 493
 494                 /* Initialize group tables of the grop @group */
 495                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 496                         goto handle_bb;
 497
 498                 /* Zero out all of the inode table blocks */
 499                 block = group_data[i].inode_table;
 500                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 501                            block, sbi->s_itb_per_group);
 502                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 503                                        GFP_NOFS);
 504                 if (err)
 505                         goto out;
 506
 507 handle_bb:
 508                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 509                         goto handle_ib;
 510
 511                 /* Initialize block bitmap of the @group */
 512                 block = group_data[i].block_bitmap;
 513                 err = extend_or_restart_transaction(handle, 1);
 514                 if (err)
 515                         goto out;
 516
 517                 bh = bclean(handle, sb, block);
 518                 if (IS_ERR(bh)) {
 519                         err = PTR_ERR(bh);
 520                         goto out;
 521                 }
 522                 if (ext4_bg_has_super(sb, group)) {
 523                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 524                                    start);
 525                         ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
 526                                                      1);
 527                 }
 528                 ext4_mark_bitmap_end(group_data[i].blocks_count,
 529                                      sb->s_blocksize * 8, bh->b_data);
 530                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 531                 if (err)
 532                         goto out;
 533                 brelse(bh);
 534
 535 handle_ib:
 536                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 537                         continue;
 538
 539                 /* Initialize inode bitmap of the @group */
 540                 block = group_data[i].inode_bitmap;
 541                 err = extend_or_restart_transaction(handle, 1);
 542                 if (err)
 543                         goto out;
 544                 /* Mark unused entries in inode bitmap used */
 545                 bh = bclean(handle, sb, block);
 546                 if (IS_ERR(bh)) {
 547                         err = PTR_ERR(bh);
 548                         goto out;
 549                 }
 550
 551                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 552                                      sb->s_blocksize * 8, bh->b_data);
 553                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 554                 if (err)
 555                         goto out;
 556                 brelse(bh);
 557         }
 558         bh = NULL;
 559
 560         /* Mark group tables in block bitmap */
 561         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 562                 count = group_table_count[j];
 563                 start = (&group_data[0].block_bitmap)[j];
 564                 block = start;
 565                 for (i = 1; i < flex_gd->count; i++) {
 566                         block += group_table_count[j];
 567                         if (block == (&group_data[i].block_bitmap)[j]) {
 568                                 count += group_table_count[j];
 569                                 continue;
 570                         }
 571                         err = set_flexbg_block_bitmap(sb, handle,
 572                                                 flex_gd, start, count);
 573                         if (err)
 574                                 goto out;
 575                         count = group_table_count[j];
 576                         start = group_data[i].block_bitmap;
 577                         block = start;
 578                 }
 579
 580                 if (count) {
 581                         err = set_flexbg_block_bitmap(sb, handle,
 582                                                 flex_gd, start, count);
 583                         if (err)
 584                                 goto out;
 585                 }
 586         }
 587
 588 out:
 589         brelse(bh);
 590         err2 = ext4_journal_stop(handle);
 591         if (err2 && !err)
 592                 err = err2;
 593
 594         return err;
 595 }
 596
 597 /*
 598  * Set up the block and inode bitmaps, and the inode table for the new group.
 599  * This doesn't need to be part of the main transaction, since we are only
 600  * changing blocks outside the actual filesystem.  We still do journaling to
 601  * ensure the recovery is correct in case of a failure just after resize.
 602  * If any part of this fails, we simply abort the resize.
 603  */
 604 static int setup_new_group_blocks(struct super_block *sb,
 605                                   struct ext4_new_group_data *input)
 606 {
 607         struct ext4_sb_info *sbi = EXT4_SB(sb);
 608         ext4_fsblk_t start = ext4_group_first_block_no(sb, input->group);
 609         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
 610                 le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
 611         unsigned long gdblocks = ext4_bg_num_gdb(sb, input->group);
 612         struct buffer_head *bh;
 613         handle_t *handle;
 614         ext4_fsblk_t block;
 615         ext4_grpblk_t bit;
 616         int i;
 617         int err = 0, err2;
 618
 619         /* This transaction may be extended/restarted along the way */
 620         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 621
 622         if (IS_ERR(handle))
 623                 return PTR_ERR(handle);
 624
 625         BUG_ON(input->group != sbi->s_groups_count);
 626
 627         /* Copy all of the GDT blocks into the backup in this group */
 628         for (i = 0, bit = 1, block = start + 1;
 629              i < gdblocks; i++, block++, bit++) {
 630                 struct buffer_head *gdb;
 631
 632                 ext4_debug("update backup group %#04llx (+%d)\n", block, bit);
 633                 err = extend_or_restart_transaction(handle, 1);
 634                 if (err)
 635                         goto exit_journal;
 636
 637                 gdb = sb_getblk(sb, block);
 638                 if (!gdb) {
 639                         err = -EIO;
 640                         goto exit_journal;
 641                 }
 642                 if ((err = ext4_journal_get_write_access(handle, gdb))) {
 643                         brelse(gdb);
 644                         goto exit_journal;
 645                 }
 646                 memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
 647                 set_buffer_uptodate(gdb);
 648                 err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 649                 if (unlikely(err)) {
 650                         brelse(gdb);
 651                         goto exit_journal;
 652                 }
 653                 brelse(gdb);
 654         }
 655
 656         /* Zero out all of the reserved backup group descriptor table blocks */
 657         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 658                         block, sbi->s_itb_per_group);
 659         err = sb_issue_zeroout(sb, gdblocks + start + 1, reserved_gdb,
 660                                GFP_NOFS);
 661         if (err)
 662                 goto exit_journal;
 663
 664         err = extend_or_restart_transaction(handle, 2);
 665         if (err)
 666                 goto exit_journal;
 667
 668         bh = bclean(handle, sb, input->block_bitmap);
 669         if (IS_ERR(bh)) {
 670                 err = PTR_ERR(bh);
 671                 goto exit_journal;
 672         }
 673
 674         if (ext4_bg_has_super(sb, input->group)) {
 675                 ext4_debug("mark backup group tables %#04llx (+0)\n", start);
 676                 ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb + 1);
 677         }
 678
 679         ext4_debug("mark block bitmap %#04llx (+%llu)\n", input->block_bitmap,
 680                    input->block_bitmap - start);
 681         ext4_set_bit(input->block_bitmap - start, bh->b_data);
 682         ext4_debug("mark inode bitmap %#04llx (+%llu)\n", input->inode_bitmap,
 683                    input->inode_bitmap - start);
 684         ext4_set_bit(input->inode_bitmap - start, bh->b_data);
 685
 686         /* Zero out all of the inode table blocks */
 687         block = input->inode_table;
 688         ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 689                         block, sbi->s_itb_per_group);
 690         err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group, GFP_NOFS);
 691         if (err)
 692                 goto exit_bh;
 693         ext4_set_bits(bh->b_data, input->inode_table - start,
 694                       sbi->s_itb_per_group);
 695
 696
 697         ext4_mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8,
 698                              bh->b_data);
 699         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 700         if (unlikely(err)) {
 701                 ext4_std_error(sb, err);
 702                 goto exit_bh;
 703         }
 704         brelse(bh);
 705         /* Mark unused entries in inode bitmap used */
 706         ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
 707                    input->inode_bitmap, input->inode_bitmap - start);
 708         if (IS_ERR(bh = bclean(handle, sb, input->inode_bitmap))) {
 709                 err = PTR_ERR(bh);
 710                 goto exit_journal;
 711         }
 712
 713         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
 714                              bh->b_data);
 715         err = ext4_handle_dirty_metadata(handle, NULL, bh);
 716         if (unlikely(err))
 717                 ext4_std_error(sb, err);
 718 exit_bh:
 719         brelse(bh);
 720
 721 exit_journal:
 722         if ((err2 = ext4_journal_stop(handle)) && !err)
 723                 err = err2;
 724
 725         return err;
 726 }
 727
 728 /*
 729  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 730  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 731  * calling this for the first time.  In a sparse filesystem it will be the
 732  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 733  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 734  */
 735 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 736                                   unsigned *five, unsigned *seven)
 737 {
 738         unsigned *min = three;
 739         int mult = 3;
 740         unsigned ret;
 741
 742         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 743                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 744                 ret = *min;
 745                 *min += 1;
 746                 return ret;
 747         }
 748
 749         if (*five < *min) {
 750                 min = five;
 751                 mult = 5;
 752         }
 753         if (*seven < *min) {
 754                 min = seven;
 755                 mult = 7;
 756         }
 757
 758         ret = *min;
 759         *min *= mult;
 760
 761         return ret;
 762 }
 763
 764 /*
 765  * Check that all of the backup GDT blocks are held in the primary GDT block.
 766  * It is assumed that they are stored in group order.  Returns the number of
 767  * groups in current filesystem that have BACKUPS, or -ve error code.
 768  */
 769 static int verify_reserved_gdb(struct super_block *sb,
 770                                ext4_group_t end,
 771                                struct buffer_head *primary)
 772 {
 773         const ext4_fsblk_t blk = primary->b_blocknr;
 774         unsigned three = 1;
 775         unsigned five = 5;
 776         unsigned seven = 7;
 777         unsigned grp;
 778         __le32 *p = (__le32 *)primary->b_data;
 779         int gdbackups = 0;
 780
 781         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 782                 if (le32_to_cpu(*p++) !=
 783                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 784                         ext4_warning(sb, "reserved GDT %llu"
 785                                      " missing grp %d (%llu)",
 786                                      blk, grp,
 787                                      grp *
 788                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 789                                      blk);
 790                         return -EINVAL;
 791                 }
 792                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 793                         return -EFBIG;
 794         }
 795
 796         return gdbackups;
 797 }
 798
 799 /*
 800  * Called when we need to bring a reserved group descriptor table block into
 801  * use from the resize inode.  The primary copy of the new GDT block currently
 802  * is an indirect block (under the double indirect block in the resize inode).
 803  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 804  * block, in group order.  Even though we know all the block numbers we need,
 805  * we check to ensure that the resize inode has actually reserved these blocks.
 806  *
 807  * Don't need to update the block bitmaps because the blocks are still in use.
 808  *
 809  * We get all of the error cases out of the way, so that we are sure to not
 810  * fail once we start modifying the data on disk, because JBD has no rollback.
 811  */
 812 static int add_new_gdb(handle_t *handle, struct inode *inode,
 813                        ext4_group_t group)
 814 {
 815         struct super_block *sb = inode->i_sb;
 816         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 817         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 818         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 819         struct buffer_head **o_group_desc, **n_group_desc;
 820         struct buffer_head *dind;
 821         struct buffer_head *gdb_bh;
 822         int gdbackups;
 823         struct ext4_iloc iloc;
 824         __le32 *data;
 825         int err;
 826
 827         if (test_opt(sb, DEBUG))
 828                 printk(KERN_DEBUG
 829                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 830                        gdb_num);
 831
 832         /*
 833          * If we are not using the primary superblock/GDT copy don't resize,
 834          * because the user tools have no way of handling this.  Probably a
 835          * bad time to do it anyways.
 836          */
 837         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 838             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 839                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 840                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 841                 return -EPERM;
 842         }
 843
 844         gdb_bh = sb_bread(sb, gdblock);
 845         if (!gdb_bh)
 846                 return -EIO;
 847
 848         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 849         if (gdbackups < 0) {
 850                 err = gdbackups;
 851                 goto exit_bh;
 852         }
 853
 854         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 855         dind = sb_bread(sb, le32_to_cpu(*data));
 856         if (!dind) {
 857                 err = -EIO;
 858                 goto exit_bh;
 859         }
 860
 861         data = (__le32 *)dind->b_data;
 862         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 863                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 864                              group, gdblock);
 865                 err = -EINVAL;
 866                 goto exit_dind;
 867         }
 868
 869         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 870         if (unlikely(err))
 871                 goto exit_dind;
 872
 873         err = ext4_journal_get_write_access(handle, gdb_bh);
 874         if (unlikely(err))
 875                 goto exit_sbh;
 876
 877         err = ext4_journal_get_write_access(handle, dind);
 878         if (unlikely(err))
 879                 ext4_std_error(sb, err);
 880
 881         /* ext4_reserve_inode_write() gets a reference on the iloc */
 882         err = ext4_reserve_inode_write(handle, inode, &iloc);
 883         if (unlikely(err))
 884                 goto exit_dindj;
 885
 886         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 887                                      sizeof(struct buffer_head *),
 888                                      GFP_NOFS);
 889         if (!n_group_desc) {
 890                 err = -ENOMEM;
 891                 ext4_warning(sb, "not enough memory for %lu groups",
 892                              gdb_num + 1);
 893                 goto exit_inode;
 894         }
 895
 896         /*
 897          * Finally, we have all of the possible failures behind us...
 898          *
 899          * Remove new GDT block from inode double-indirect block and clear out
 900          * the new GDT block for use (which also "frees" the backup GDT blocks
 901          * from the reserved inode).  We don't need to change the bitmaps for
 902          * these blocks, because they are marked as in-use from being in the
 903          * reserved inode, and will become GDT blocks (primary and backup).
 904          */
 905         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 906         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 907         if (unlikely(err)) {
 908                 ext4_std_error(sb, err);
 909                 goto exit_inode;
 910         }
 911         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 912         ext4_mark_iloc_dirty(handle, inode, &iloc);
 913         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 914         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 915         if (unlikely(err)) {
 916                 ext4_std_error(sb, err);
 917                 goto exit_inode;
 918         }
 919         brelse(dind);
 920
 921         o_group_desc = EXT4_SB(sb)->s_group_desc;
 922         memcpy(n_group_desc, o_group_desc,
 923                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 924         n_group_desc[gdb_num] = gdb_bh;
 925         EXT4_SB(sb)->s_group_desc = n_group_desc;
 926         EXT4_SB(sb)->s_gdb_count++;
 927         ext4_kvfree(o_group_desc);
 928
 929         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 930         err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 931         if (err)
 932                 ext4_std_error(sb, err);
 933
 934         return err;
 935
 936 exit_inode:
 937         ext4_kvfree(n_group_desc);
 938         /* ext4_handle_release_buffer(handle, iloc.bh); */
 939         brelse(iloc.bh);
 940 exit_dindj:
 941         /* ext4_handle_release_buffer(handle, dind); */
 942 exit_sbh:
 943         /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
 944 exit_dind:
 945         brelse(dind);
 946 exit_bh:
 947         brelse(gdb_bh);
 948
 949         ext4_debug("leaving with error %d\n", err);
 950         return err;
 951 }
 952
 953 /*
 954  * Called when we are adding a new group which has a backup copy of each of
 955  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 956  * We need to add these reserved backup GDT blocks to the resize inode, so
 957  * that they are kept for future resizing and not allocated to files.
 958  *
 959  * Each reserved backup GDT block will go into a different indirect block.
 960  * The indirect blocks are actually the primary reserved GDT blocks,
 961  * so we know in advance what their block numbers are.  We only get the
 962  * double-indirect block to verify it is pointing to the primary reserved
 963  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 964  * backup GDT blocks are stored in their reserved primary GDT block.
 965  */
 966 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 967                               ext4_group_t group)
 968 {
 969         struct super_block *sb = inode->i_sb;
 970         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 971         struct buffer_head **primary;
 972         struct buffer_head *dind;
 973         struct ext4_iloc iloc;
 974         ext4_fsblk_t blk;
 975         __le32 *data, *end;
 976         int gdbackups = 0;
 977         int res, i;
 978         int err;
 979
 980         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 981         if (!primary)
 982                 return -ENOMEM;
 983
 984         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 985         dind = sb_bread(sb, le32_to_cpu(*data));
 986         if (!dind) {
 987                 err = -EIO;
 988                 goto exit_free;
 989         }
 990
 991         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 992         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 993                                          EXT4_ADDR_PER_BLOCK(sb));
 994         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 995
 996         /* Get each reserved primary GDT block and verify it holds backups */
 997         for (res = 0; res < reserved_gdb; res++, blk++) {
 998                 if (le32_to_cpu(*data) != blk) {
 999                         ext4_warning(sb, "reserved block %llu"
1000                                      " not at offset %ld",
1001                                      blk,
1002                                      (long)(data - (__le32 *)dind->b_data));
1003                         err = -EINVAL;
1004                         goto exit_bh;
1005                 }
1006                 primary[res] = sb_bread(sb, blk);
1007                 if (!primary[res]) {
1008                         err = -EIO;
1009                         goto exit_bh;
1010                 }
1011                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
1012                 if (gdbackups < 0) {
1013                         brelse(primary[res]);
1014                         err = gdbackups;
1015                         goto exit_bh;
1016                 }
1017                 if (++data >= end)
1018                         data = (__le32 *)dind->b_data;
1019         }
1020
1021         for (i = 0; i < reserved_gdb; i++) {
1022                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
1023                         /*
1024                         int j;
1025                         for (j = 0; j < i; j++)
1026                                 ext4_handle_release_buffer(handle, primary[j]);
1027                          */
1028                         goto exit_bh;
1029                 }
1030         }
1031
1032         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
1033                 goto exit_bh;
1034
1035         /*
1036          * Finally we can add each of the reserved backup GDT blocks from
1037          * the new group to its reserved primary GDT block.
1038          */
1039         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
1040         for (i = 0; i < reserved_gdb; i++) {
1041                 int err2;
1042                 data = (__le32 *)primary[i]->b_data;
1043                 /* printk("reserving backup %lu[%u] = %lu\n",
1044                        primary[i]->b_blocknr, gdbackups,
1045                        blk + primary[i]->b_blocknr); */
1046                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
1047                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
1048                 if (!err)
1049                         err = err2;
1050         }
1051         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
1052         ext4_mark_iloc_dirty(handle, inode, &iloc);
1053
1054 exit_bh:
1055         while (--res >= 0)
1056                 brelse(primary[res]);
1057         brelse(dind);
1058
1059 exit_free:
1060         kfree(primary);
1061
1062         return err;
1063 }
1064
1065 /*
1066  * Update the backup copies of the ext4 metadata.  These don't need to be part
1067  * of the main resize transaction, because e2fsck will re-write them if there
1068  * is a problem (basically only OOM will cause a problem).  However, we
1069  * _should_ update the backups if possible, in case the primary gets trashed
1070  * for some reason and we need to run e2fsck from a backup superblock.  The
1071  * important part is that the new block and inode counts are in the backup
1072  * superblocks, and the location of the new group metadata in the GDT backups.
1073  *
1074  * We do not need take the s_resize_lock for this, because these
1075  * blocks are not otherwise touched by the filesystem code when it is
1076  * mounted.  We don't need to worry about last changing from
1077  * sbi->s_groups_count, because the worst that can happen is that we
1078  * do not copy the full number of backups at this time.  The resize
1079  * which changed s_groups_count will backup again.
1080  */
1081 static void update_backups(struct super_block *sb,
1082                            int blk_off, char *data, int size)
1083 {
1084         struct ext4_sb_info *sbi = EXT4_SB(sb);
1085         const ext4_group_t last = sbi->s_groups_count;
1086         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
1087         unsigned three = 1;
1088         unsigned five = 5;
1089         unsigned seven = 7;
1090         ext4_group_t group;
1091         int rest = sb->s_blocksize - size;
1092         handle_t *handle;
1093         int err = 0, err2;
1094
1095         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
1096         if (IS_ERR(handle)) {
1097                 group = 1;
1098                 err = PTR_ERR(handle);
1099                 goto exit_err;
1100         }
1101
1102         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
1103                 struct buffer_head *bh;
1104
1105                 /* Out of journal space, and can't get more - abort - so sad */
1106                 if (ext4_handle_valid(handle) &&
1107                     handle->h_buffer_credits == 0 &&
1108                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
1109                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
1110                         break;
1111
1112                 bh = sb_getblk(sb, group * bpg + blk_off);
1113                 if (!bh) {
1114                         err = -EIO;
1115                         break;
1116                 }
1117                 ext4_debug("update metadata backup %#04lx\n",
1118                           (unsigned long)bh->b_blocknr);
1119                 if ((err = ext4_journal_get_write_access(handle, bh)))
1120                         break;
1121                 lock_buffer(bh);
1122                 memcpy(bh->b_data, data, size);
1123                 if (rest)
1124                         memset(bh->b_data + size, 0, rest);
1125                 set_buffer_uptodate(bh);
1126                 unlock_buffer(bh);
1127                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1128                 if (unlikely(err))
1129                         ext4_std_error(sb, err);
1130                 brelse(bh);
1131         }
1132         if ((err2 = ext4_journal_stop(handle)) && !err)
1133                 err = err2;
1134
1135         /*
1136          * Ugh! Need to have e2fsck write the backup copies.  It is too
1137          * late to revert the resize, we shouldn't fail just because of
1138          * the backup copies (they are only needed in case of corruption).
1139          *
1140          * However, if we got here we have a journal problem too, so we
1141          * can't really start a transaction to mark the superblock.
1142          * Chicken out and just set the flag on the hope it will be written
1143          * to disk, and if not - we will simply wait until next fsck.
1144          */
1145 exit_err:
1146         if (err) {
1147                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1148                              "forcing fsck on next reboot", group, err);
1149                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1150                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1151                 mark_buffer_dirty(sbi->s_sbh);
1152         }
1153 }
1154
1155 /*
1156  * ext4_add_new_descs() adds @count group descriptor of groups
1157  * starting at @group
1158  *
1159  * @handle: journal handle
1160  * @sb: super block
1161  * @group: the group no. of the first group desc to be added
1162  * @resize_inode: the resize inode
1163  * @count: number of group descriptors to be added
1164  */
1165 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1166                               ext4_group_t group, struct inode *resize_inode,
1167                               ext4_group_t count)
1168 {
1169         struct ext4_sb_info *sbi = EXT4_SB(sb);
1170         struct ext4_super_block *es = sbi->s_es;
1171         struct buffer_head *gdb_bh;
1172         int i, gdb_off, gdb_num, err = 0;
1173
1174         for (i = 0; i < count; i++, group++) {
1175                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1176                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1177
1178                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1179                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1180
1181                 /*
1182                  * We will only either add reserved group blocks to a backup group
1183                  * or remove reserved blocks for the first group in a new group block.
1184                  * Doing both would be mean more complex code, and sane people don't
1185                  * use non-sparse filesystems anymore.  This is already checked above.
1186                  */
1187                 if (gdb_off) {
1188                         gdb_bh = sbi->s_group_desc[gdb_num];
1189                         err = ext4_journal_get_write_access(handle, gdb_bh);
1190
1191                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1192                                 err = reserve_backup_gdb(handle, resize_inode, group);
1193                 } else
1194                         err = add_new_gdb(handle, resize_inode, group);
1195                 if (err)
1196                         break;
1197         }
1198         return err;
1199 }
1200
1201 /*
1202  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1203  */
1204 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1205                                 struct ext4_new_flex_group_data *flex_gd)
1206 {
1207         struct ext4_new_group_data      *group_data = flex_gd->groups;
1208         struct ext4_group_desc          *gdp;
1209         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1210         struct buffer_head              *gdb_bh;
1211         ext4_group_t                    group;
1212         __u16                           *bg_flags = flex_gd->bg_flags;
1213         int                             i, gdb_off, gdb_num, err = 0;
1214
1215
1216         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1217                 group = group_data->group;
1218
1219                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1220                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1221
1222                 /*
1223                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1224                  */
1225                 gdb_bh = sbi->s_group_desc[gdb_num];
1226                 /* Update group descriptor block for new group */
1227                 gdp = (struct ext4_group_desc *)((char *)gdb_bh->b_data +
1228                                                  gdb_off * EXT4_DESC_SIZE(sb));
1229
1230                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1231                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1232                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1233                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1234                 ext4_free_group_clusters_set(sb, gdp,
1235                                              EXT4_B2C(sbi, group_data->free_blocks_count));
1236                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1237                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1238                 gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1239
1240                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1241                 if (unlikely(err)) {
1242                         ext4_std_error(sb, err);
1243                         break;
1244                 }
1245
1246                 /*
1247                  * We can allocate memory for mb_alloc based on the new group
1248                  * descriptor
1249                  */
1250                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1251                 if (err)
1252                         break;
1253         }
1254         return err;
1255 }
1256
1257 /*
1258  * ext4_update_super() updates the super block so that the newly added
1259  * groups can be seen by the filesystem.
1260  *
1261  * @sb: super block
1262  * @flex_gd: new added groups
1263  */
1264 static void ext4_update_super(struct super_block *sb,
1265                              struct ext4_new_flex_group_data *flex_gd)
1266 {
1267         ext4_fsblk_t blocks_count = 0;
1268         ext4_fsblk_t free_blocks = 0;
1269         ext4_fsblk_t reserved_blocks = 0;
1270         struct ext4_new_group_data *group_data = flex_gd->groups;
1271         struct ext4_sb_info *sbi = EXT4_SB(sb);
1272         struct ext4_super_block *es = sbi->s_es;
1273         int i;
1274
1275         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1276         /*
1277          * Make the new blocks and inodes valid next.  We do this before
1278          * increasing the group count so that once the group is enabled,
1279          * all of its blocks and inodes are already valid.
1280          *
1281          * We always allocate group-by-group, then block-by-block or
1282          * inode-by-inode within a group, so enabling these
1283          * blocks/inodes before the group is live won't actually let us
1284          * allocate the new space yet.
1285          */
1286         for (i = 0; i < flex_gd->count; i++) {
1287                 blocks_count += group_data[i].blocks_count;
1288                 free_blocks += group_data[i].free_blocks_count;
1289         }
1290
1291         reserved_blocks = ext4_r_blocks_count(es) * 100;
1292         do_div(reserved_blocks, ext4_blocks_count(es));
1293         reserved_blocks *= blocks_count;
1294         do_div(reserved_blocks, 100);
1295
1296         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1297         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1298                      flex_gd->count);
1299
1300         /*
1301          * We need to protect s_groups_count against other CPUs seeing
1302          * inconsistent state in the superblock.
1303          *
1304          * The precise rules we use are:
1305          *
1306          * * Writers must perform a smp_wmb() after updating all
1307          *   dependent data and before modifying the groups count
1308          *
1309          * * Readers must perform an smp_rmb() after reading the groups
1310          *   count and before reading any dependent data.
1311          *
1312          * NB. These rules can be relaxed when checking the group count
1313          * while freeing data, as we can only allocate from a block
1314          * group after serialising against the group count, and we can
1315          * only then free after serialising in turn against that
1316          * allocation.
1317          */
1318         smp_wmb();
1319
1320         /* Update the global fs size fields */
1321         sbi->s_groups_count += flex_gd->count;
1322
1323         /* Update the reserved block counts only once the new group is
1324          * active. */
1325         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1326                                 reserved_blocks);
1327
1328         /* Update the free space counts */
1329         percpu_counter_add(&sbi->s_freeclusters_counter,
1330                            EXT4_B2C(sbi, free_blocks));
1331         percpu_counter_add(&sbi->s_freeinodes_counter,
1332                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1333
1334         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1335                                       EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1336             sbi->s_log_groups_per_flex) {
1337                 ext4_group_t flex_group;
1338                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1339                 atomic_add(EXT4_B2C(sbi, free_blocks),
1340                            &sbi->s_flex_groups[flex_group].free_clusters);
1341                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1342                            &sbi->s_flex_groups[flex_group].free_inodes);
1343         }
1344
1345         if (test_opt(sb, DEBUG))
1346                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1347                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1348                        blocks_count, free_blocks, reserved_blocks);
1349 }
1350
1351 /* Add group descriptor data to an existing or new group descriptor block.
1352  * Ensure we handle all possible error conditions _before_ we start modifying
1353  * the filesystem, because we cannot abort the transaction and not have it
1354  * write the data to disk.
1355  *
1356  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1357  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1358  *
1359  * We only need to hold the superblock lock while we are actually adding
1360  * in the new group's counts to the superblock.  Prior to that we have
1361  * not really "added" the group at all.  We re-check that we are still
1362  * adding in the last group in case things have changed since verifying.
1363  */
1364 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1365 {
1366         struct ext4_sb_info *sbi = EXT4_SB(sb);
1367         struct ext4_super_block *es = sbi->s_es;
1368         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1369                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1370         struct buffer_head *primary = NULL;
1371         struct ext4_group_desc *gdp;
1372         struct inode *inode = NULL;
1373         handle_t *handle;
1374         int gdb_off, gdb_num;
1375         int err, err2;
1376
1377         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1378         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1379
1380         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1381                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1382                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1383                 return -EPERM;
1384         }
1385
1386         if (ext4_blocks_count(es) + input->blocks_count <
1387             ext4_blocks_count(es)) {
1388                 ext4_warning(sb, "blocks_count overflow");
1389                 return -EINVAL;
1390         }
1391
1392         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1393             le32_to_cpu(es->s_inodes_count)) {
1394                 ext4_warning(sb, "inodes_count overflow");
1395                 return -EINVAL;
1396         }
1397
1398         if (reserved_gdb || gdb_off == 0) {
1399                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1400                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
1401                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1402                         ext4_warning(sb,
1403                                      "No reserved GDT blocks, can't resize");
1404                         return -EPERM;
1405                 }
1406                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1407                 if (IS_ERR(inode)) {
1408                         ext4_warning(sb, "Error opening resize inode");
1409                         return PTR_ERR(inode);
1410                 }
1411         }
1412
1413
1414         if ((err = verify_group_input(sb, input)))
1415                 goto exit_put;
1416
1417         if ((err = setup_new_group_blocks(sb, input)))
1418                 goto exit_put;
1419
1420         /*
1421          * We will always be modifying at least the superblock and a GDT
1422          * block.  If we are adding a group past the last current GDT block,
1423          * we will also modify the inode and the dindirect block.  If we
1424          * are adding a group with superblock/GDT backups  we will also
1425          * modify each of the reserved GDT dindirect blocks.
1426          */
1427         handle = ext4_journal_start_sb(sb,
1428                                        ext4_bg_has_super(sb, input->group) ?
1429                                        3 + reserved_gdb : 4);
1430         if (IS_ERR(handle)) {
1431                 err = PTR_ERR(handle);
1432                 goto exit_put;
1433         }
1434
1435         if ((err = ext4_journal_get_write_access(handle, sbi->s_sbh)))
1436                 goto exit_journal;
1437
1438         /*
1439          * We will only either add reserved group blocks to a backup group
1440          * or remove reserved blocks for the first group in a new group block.
1441          * Doing both would be mean more complex code, and sane people don't
1442          * use non-sparse filesystems anymore.  This is already checked above.
1443          */
1444         if (gdb_off) {
1445                 primary = sbi->s_group_desc[gdb_num];
1446                 if ((err = ext4_journal_get_write_access(handle, primary)))
1447                         goto exit_journal;
1448
1449                 if (reserved_gdb && ext4_bg_num_gdb(sb, input->group)) {
1450                         err = reserve_backup_gdb(handle, inode, input->group);
1451                         if (err)
1452                                 goto exit_journal;
1453                 }
1454         } else {
1455                 /*
1456                  * Note that we can access new group descriptor block safely
1457                  * only if add_new_gdb() succeeds.
1458                  */
1459                 err = add_new_gdb(handle, inode, input->group);
1460                 if (err)
1461                         goto exit_journal;
1462                 primary = sbi->s_group_desc[gdb_num];
1463         }
1464
1465         /*
1466          * OK, now we've set up the new group.  Time to make it active.
1467          *
1468          * so we have to be safe wrt. concurrent accesses the group
1469          * data.  So we need to be careful to set all of the relevant
1470          * group descriptor data etc. *before* we enable the group.
1471          *
1472          * The key field here is sbi->s_groups_count: as long as
1473          * that retains its old value, nobody is going to access the new
1474          * group.
1475          *
1476          * So first we update all the descriptor metadata for the new
1477          * group; then we update the total disk blocks count; then we
1478          * update the groups count to enable the group; then finally we
1479          * update the free space counts so that the system can start
1480          * using the new disk blocks.
1481          */
1482
1483         /* Update group descriptor block for new group */
1484         gdp = (struct ext4_group_desc *)((char *)primary->b_data +
1485                                          gdb_off * EXT4_DESC_SIZE(sb));
1486
1487         memset(gdp, 0, EXT4_DESC_SIZE(sb));
1488         ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
1489         ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
1490         ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
1491         ext4_free_group_clusters_set(sb, gdp, input->free_blocks_count);
1492         ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1493         gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
1494         gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
1495
1496         /*
1497          * We can allocate memory for mb_alloc based on the new group
1498          * descriptor
1499          */
1500         err = ext4_mb_add_groupinfo(sb, input->group, gdp);
1501         if (err)
1502                 goto exit_journal;
1503
1504         /*
1505          * Make the new blocks and inodes valid next.  We do this before
1506          * increasing the group count so that once the group is enabled,
1507          * all of its blocks and inodes are already valid.
1508          *
1509          * We always allocate group-by-group, then block-by-block or
1510          * inode-by-inode within a group, so enabling these
1511          * blocks/inodes before the group is live won't actually let us
1512          * allocate the new space yet.
1513          */
1514         ext4_blocks_count_set(es, ext4_blocks_count(es) +
1515                 input->blocks_count);
1516         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb));
1517
1518         /*
1519          * We need to protect s_groups_count against other CPUs seeing
1520          * inconsistent state in the superblock.
1521          *
1522          * The precise rules we use are:
1523          *
1524          * * Writers must perform a smp_wmb() after updating all dependent
1525          *   data and before modifying the groups count
1526          *
1527          * * Readers must perform an smp_rmb() after reading the groups count
1528          *   and before reading any dependent data.
1529          *
1530          * NB. These rules can be relaxed when checking the group count
1531          * while freeing data, as we can only allocate from a block
1532          * group after serialising against the group count, and we can
1533          * only then free after serialising in turn against that
1534          * allocation.
1535          */
1536         smp_wmb();
1537
1538         /* Update the global fs size fields */
1539         sbi->s_groups_count++;
1540
1541         err = ext4_handle_dirty_metadata(handle, NULL, primary);
1542         if (unlikely(err)) {
1543                 ext4_std_error(sb, err);
1544                 goto exit_journal;
1545         }
1546
1547         /* Update the reserved block counts only once the new group is
1548          * active. */
1549         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1550                 input->reserved_blocks);
1551
1552         /* Update the free space counts */
1553         percpu_counter_add(&sbi->s_freeclusters_counter,
1554                            EXT4_B2C(sbi, input->free_blocks_count));
1555         percpu_counter_add(&sbi->s_freeinodes_counter,
1556                            EXT4_INODES_PER_GROUP(sb));
1557
1558         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1559             sbi->s_log_groups_per_flex) {
1560                 ext4_group_t flex_group;
1561                 flex_group = ext4_flex_group(sbi, input->group);
1562                 atomic_add(EXT4_B2C(sbi, input->free_blocks_count),
1563                            &sbi->s_flex_groups[flex_group].free_clusters);
1564                 atomic_add(EXT4_INODES_PER_GROUP(sb),
1565                            &sbi->s_flex_groups[flex_group].free_inodes);
1566         }
1567
1568         ext4_handle_dirty_super(handle, sb);
1569
1570 exit_journal:
1571         if ((err2 = ext4_journal_stop(handle)) && !err)
1572                 err = err2;
1573         if (!err && primary) {
1574                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1575                                sizeof(struct ext4_super_block));
1576                 update_backups(sb, primary->b_blocknr, primary->b_data,
1577                                primary->b_size);
1578         }
1579 exit_put:
1580         iput(inode);
1581         return err;
1582 } /* ext4_group_add */
1583
1584 /*
1585  * extend a group without checking assuming that checking has been done.
1586  */
1587 static int ext4_group_extend_no_check(struct super_block *sb,
1588                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1589 {
1590         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1591         handle_t *handle;
1592         int err = 0, err2;
1593
1594         /* We will update the superblock, one block bitmap, and
1595          * one group descriptor via ext4_group_add_blocks().
1596          */
1597         handle = ext4_journal_start_sb(sb, 3);
1598         if (IS_ERR(handle)) {
1599                 err = PTR_ERR(handle);
1600                 ext4_warning(sb, "error %d on journal start", err);
1601                 return err;
1602         }
1603
1604         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1605         if (err) {
1606                 ext4_warning(sb, "error %d on journal write access", err);
1607                 goto errout;
1608         }
1609
1610         ext4_blocks_count_set(es, o_blocks_count + add);
1611         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1612                    o_blocks_count + add);
1613         /* We add the blocks to the bitmap and set the group need init bit */
1614         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1615         if (err)
1616                 goto errout;
1617         ext4_handle_dirty_super(handle, sb);
1618         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1619                    o_blocks_count + add);
1620 errout:
1621         err2 = ext4_journal_stop(handle);
1622         if (err2 && !err)
1623                 err = err2;
1624
1625         if (!err) {
1626                 if (test_opt(sb, DEBUG))
1627                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1628                                "blocks\n", ext4_blocks_count(es));
1629                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1630                                sizeof(struct ext4_super_block));
1631         }
1632         return err;
1633 }
1634
1635 /*
1636  * Extend the filesystem to the new number of blocks specified.  This entry
1637  * point is only used to extend the current filesystem to the end of the last
1638  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1639  * for emergencies (because it has no dependencies on reserved blocks).
1640  *
1641  * If we _really_ wanted, we could use default values to call ext4_group_add()
1642  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1643  * GDT blocks are reserved to grow to the desired size.
1644  */
1645 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1646                       ext4_fsblk_t n_blocks_count)
1647 {
1648         ext4_fsblk_t o_blocks_count;
1649         ext4_grpblk_t last;
1650         ext4_grpblk_t add;
1651         struct buffer_head *bh;
1652         handle_t *handle;
1653         int err, err2;
1654         ext4_group_t group;
1655
1656         o_blocks_count = ext4_blocks_count(es);
1657
1658         if (test_opt(sb, DEBUG))
1659                 printk(KERN_DEBUG "EXT4-fs: extending last group from %llu to %llu blocks\n",
1660                        o_blocks_count, n_blocks_count);
1661
1662         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1663                 return 0;
1664
1665         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1666                 printk(KERN_ERR "EXT4-fs: filesystem on %s:"
1667                         " too large to resize to %llu blocks safely\n",
1668                         sb->s_id, n_blocks_count);
1669                 if (sizeof(sector_t) < 8)
1670                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1671                 return -EINVAL;
1672         }
1673
1674         if (n_blocks_count < o_blocks_count) {
1675                 ext4_warning(sb, "can't shrink FS - resize aborted");
1676                 return -EINVAL;
1677         }
1678
1679         /* Handle the remaining blocks in the last group only. */
1680         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1681
1682         if (last == 0) {
1683                 ext4_warning(sb, "need to use ext2online to resize further");
1684                 return -EPERM;
1685         }
1686
1687         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1688
1689         if (o_blocks_count + add < o_blocks_count) {
1690                 ext4_warning(sb, "blocks_count overflow");
1691                 return -EINVAL;
1692         }
1693
1694         if (o_blocks_count + add > n_blocks_count)
1695                 add = n_blocks_count - o_blocks_count;
1696
1697         if (o_blocks_count + add < n_blocks_count)
1698                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1699                              o_blocks_count + add, add);
1700
1701         /* See if the device is actually as big as what was requested */
1702         bh = sb_bread(sb, o_blocks_count + add - 1);
1703         if (!bh) {
1704                 ext4_warning(sb, "can't read last block, resize aborted");
1705                 return -ENOSPC;
1706         }
1707         brelse(bh);
1708
1709         /* We will update the superblock, one block bitmap, and
1710          * one group descriptor via ext4_free_blocks().
1711          */
1712         handle = ext4_journal_start_sb(sb, 3);
1713         if (IS_ERR(handle)) {
1714                 err = PTR_ERR(handle);
1715                 ext4_warning(sb, "error %d on journal start", err);
1716                 goto exit_put;
1717         }
1718
1719         if ((err = ext4_journal_get_write_access(handle,
1720                                                  EXT4_SB(sb)->s_sbh))) {
1721                 ext4_warning(sb, "error %d on journal write access", err);
1722                 ext4_journal_stop(handle);
1723                 goto exit_put;
1724         }
1725         ext4_blocks_count_set(es, o_blocks_count + add);
1726         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1727                    o_blocks_count + add);
1728         /* We add the blocks to the bitmap and set the group need init bit */
1729         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1730         ext4_handle_dirty_super(handle, sb);
1731         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1732                    o_blocks_count + add);
1733         err2 = ext4_journal_stop(handle);
1734         if (!err && err2)
1735                 err = err2;
1736
1737         if (err)
1738                 goto exit_put;
1739
1740         if (test_opt(sb, DEBUG))
1741                 printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
1742                        ext4_blocks_count(es));
1743         update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1744                        sizeof(struct ext4_super_block));
1745 exit_put:
1746         return err;
1747 } /* ext4_group_extend */