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  * Returns 0 on a successful allocation of the metadata blocks in the
 205  * block group.
 206  */
 207 static int ext4_alloc_group_tables(struct super_block *sb,
 208                                 struct ext4_new_flex_group_data *flex_gd,
 209                                 int flexbg_size)
 210 {
 211         struct ext4_new_group_data *group_data = flex_gd->groups;
 212         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 213         ext4_fsblk_t start_blk;
 214         ext4_fsblk_t last_blk;
 215         ext4_group_t src_group;
 216         ext4_group_t bb_index = 0;
 217         ext4_group_t ib_index = 0;
 218         ext4_group_t it_index = 0;
 219         ext4_group_t group;
 220         ext4_group_t last_group;
 221         unsigned overhead;
 222
 223         BUG_ON(flex_gd->count == 0 || group_data == NULL);
 224
 225         src_group = group_data[0].group;
 226         last_group  = src_group + flex_gd->count - 1;
 227
 228         BUG_ON((flexbg_size > 1) && ((src_group & ~(flexbg_size - 1)) !=
 229                (last_group & ~(flexbg_size - 1))));
 230 next_group:
 231         group = group_data[0].group;
 232         if (src_group >= group_data[0].group + flex_gd->count)
 233                 return -ENOSPC;
 234         start_blk = ext4_group_first_block_no(sb, src_group);
 235         last_blk = start_blk + group_data[src_group - group].blocks_count;
 236
 237         overhead = ext4_bg_has_super(sb, src_group) ?
 238                    (1 + ext4_bg_num_gdb(sb, src_group) +
 239                     le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
 240
 241         start_blk += overhead;
 242
 243         /* We collect contiguous blocks as much as possible. */
 244         src_group++;
 245         for (; src_group <= last_group; src_group++)
 246                 if (!ext4_bg_has_super(sb, src_group))
 247                         last_blk += group_data[src_group - group].blocks_count;
 248                 else
 249                         break;
 250
 251         /* Allocate block bitmaps */
 252         for (; bb_index < flex_gd->count; bb_index++) {
 253                 if (start_blk >= last_blk)
 254                         goto next_group;
 255                 group_data[bb_index].block_bitmap = start_blk++;
 256                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 257                 group -= group_data[0].group;
 258                 group_data[group].free_blocks_count--;
 259                 if (flexbg_size > 1)
 260                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 261         }
 262
 263         /* Allocate inode bitmaps */
 264         for (; ib_index < flex_gd->count; ib_index++) {
 265                 if (start_blk >= last_blk)
 266                         goto next_group;
 267                 group_data[ib_index].inode_bitmap = start_blk++;
 268                 ext4_get_group_no_and_offset(sb, start_blk - 1, &group, NULL);
 269                 group -= group_data[0].group;
 270                 group_data[group].free_blocks_count--;
 271                 if (flexbg_size > 1)
 272                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 273         }
 274
 275         /* Allocate inode tables */
 276         for (; it_index < flex_gd->count; it_index++) {
 277                 if (start_blk + EXT4_SB(sb)->s_itb_per_group > last_blk)
 278                         goto next_group;
 279                 group_data[it_index].inode_table = start_blk;
 280                 ext4_get_group_no_and_offset(sb, start_blk, &group, NULL);
 281                 group -= group_data[0].group;
 282                 group_data[group].free_blocks_count -=
 283                                         EXT4_SB(sb)->s_itb_per_group;
 284                 if (flexbg_size > 1)
 285                         flex_gd->bg_flags[group] &= ~EXT4_BG_BLOCK_UNINIT;
 286
 287                 start_blk += EXT4_SB(sb)->s_itb_per_group;
 288         }
 289
 290         if (test_opt(sb, DEBUG)) {
 291                 int i;
 292                 group = group_data[0].group;
 293
 294                 printk(KERN_DEBUG "EXT4-fs: adding a flex group with "
 295                        "%d groups, flexbg size is %d:\n", flex_gd->count,
 296                        flexbg_size);
 297
 298                 for (i = 0; i < flex_gd->count; i++) {
 299                         printk(KERN_DEBUG "adding %s group %u: %u "
 300                                "blocks (%d free)\n",
 301                                ext4_bg_has_super(sb, group + i) ? "normal" :
 302                                "no-super", group + i,
 303                                group_data[i].blocks_count,
 304                                group_data[i].free_blocks_count);
 305                 }
 306         }
 307         return 0;
 308 }
 309
 310 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
 311                                   ext4_fsblk_t blk)
 312 {
 313         struct buffer_head *bh;
 314         int err;
 315
 316         bh = sb_getblk(sb, blk);
 317         if (!bh)
 318                 return ERR_PTR(-EIO);
 319         if ((err = ext4_journal_get_write_access(handle, bh))) {
 320                 brelse(bh);
 321                 bh = ERR_PTR(err);
 322         } else {
 323                 memset(bh->b_data, 0, sb->s_blocksize);
 324                 set_buffer_uptodate(bh);
 325         }
 326
 327         return bh;
 328 }
 329
 330 /*
 331  * If we have fewer than thresh credits, extend by EXT4_MAX_TRANS_DATA.
 332  * If that fails, restart the transaction & regain write access for the
 333  * buffer head which is used for block_bitmap modifications.
 334  */
 335 static int extend_or_restart_transaction(handle_t *handle, int thresh)
 336 {
 337         int err;
 338
 339         if (ext4_handle_has_enough_credits(handle, thresh))
 340                 return 0;
 341
 342         err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
 343         if (err < 0)
 344                 return err;
 345         if (err) {
 346                 err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA);
 347                 if (err)
 348                         return err;
 349         }
 350
 351         return 0;
 352 }
 353
 354 /*
 355  * set_flexbg_block_bitmap() mark @count blocks starting from @block used.
 356  *
 357  * Helper function for ext4_setup_new_group_blocks() which set .
 358  *
 359  * @sb: super block
 360  * @handle: journal handle
 361  * @flex_gd: flex group data
 362  */
 363 static int set_flexbg_block_bitmap(struct super_block *sb, handle_t *handle,
 364                         struct ext4_new_flex_group_data *flex_gd,
 365                         ext4_fsblk_t block, ext4_group_t count)
 366 {
 367         ext4_group_t count2;
 368
 369         ext4_debug("mark blocks [%llu/%u] used\n", block, count);
 370         for (count2 = count; count > 0; count -= count2, block += count2) {
 371                 ext4_fsblk_t start;
 372                 struct buffer_head *bh;
 373                 ext4_group_t group;
 374                 int err;
 375
 376                 ext4_get_group_no_and_offset(sb, block, &group, NULL);
 377                 start = ext4_group_first_block_no(sb, group);
 378                 group -= flex_gd->groups[0].group;
 379
 380                 count2 = sb->s_blocksize * 8 - (block - start);
 381                 if (count2 > count)
 382                         count2 = count;
 383
 384                 if (flex_gd->bg_flags[group] & EXT4_BG_BLOCK_UNINIT) {
 385                         BUG_ON(flex_gd->count > 1);
 386                         continue;
 387                 }
 388
 389                 err = extend_or_restart_transaction(handle, 1);
 390                 if (err)
 391                         return err;
 392
 393                 bh = sb_getblk(sb, flex_gd->groups[group].block_bitmap);
 394                 if (!bh)
 395                         return -EIO;
 396
 397                 err = ext4_journal_get_write_access(handle, bh);
 398                 if (err)
 399                         return err;
 400                 ext4_debug("mark block bitmap %#04llx (+%llu/%u)\n", block,
 401                            block - start, count2);
 402                 ext4_set_bits(bh->b_data, block - start, count2);
 403
 404                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 405                 if (unlikely(err))
 406                         return err;
 407                 brelse(bh);
 408         }
 409
 410         return 0;
 411 }
 412
 413 /*
 414  * Set up the block and inode bitmaps, and the inode table for the new groups.
 415  * This doesn't need to be part of the main transaction, since we are only
 416  * changing blocks outside the actual filesystem.  We still do journaling to
 417  * ensure the recovery is correct in case of a failure just after resize.
 418  * If any part of this fails, we simply abort the resize.
 419  *
 420  * setup_new_flex_group_blocks handles a flex group as follow:
 421  *  1. copy super block and GDT, and initialize group tables if necessary.
 422  *     In this step, we only set bits in blocks bitmaps for blocks taken by
 423  *     super block and GDT.
 424  *  2. allocate group tables in block bitmaps, that is, set bits in block
 425  *     bitmap for blocks taken by group tables.
 426  */
 427 static int setup_new_flex_group_blocks(struct super_block *sb,
 428                                 struct ext4_new_flex_group_data *flex_gd)
 429 {
 430         int group_table_count[] = {1, 1, EXT4_SB(sb)->s_itb_per_group};
 431         ext4_fsblk_t start;
 432         ext4_fsblk_t block;
 433         struct ext4_sb_info *sbi = EXT4_SB(sb);
 434         struct ext4_super_block *es = sbi->s_es;
 435         struct ext4_new_group_data *group_data = flex_gd->groups;
 436         __u16 *bg_flags = flex_gd->bg_flags;
 437         handle_t *handle;
 438         ext4_group_t group, count;
 439         struct buffer_head *bh = NULL;
 440         int reserved_gdb, i, j, err = 0, err2;
 441
 442         BUG_ON(!flex_gd->count || !group_data ||
 443                group_data[0].group != sbi->s_groups_count);
 444
 445         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
 446
 447         /* This transaction may be extended/restarted along the way */
 448         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 449         if (IS_ERR(handle))
 450                 return PTR_ERR(handle);
 451
 452         group = group_data[0].group;
 453         for (i = 0; i < flex_gd->count; i++, group++) {
 454                 unsigned long gdblocks;
 455
 456                 gdblocks = ext4_bg_num_gdb(sb, group);
 457                 start = ext4_group_first_block_no(sb, group);
 458
 459                 if (!ext4_bg_has_super(sb, group))
 460                         goto handle_itb;
 461
 462                 /* Copy all of the GDT blocks into the backup in this group */
 463                 for (j = 0, block = start + 1; j < gdblocks; j++, block++) {
 464                         struct buffer_head *gdb;
 465
 466                         ext4_debug("update backup group %#04llx\n", block);
 467                         err = extend_or_restart_transaction(handle, 1);
 468                         if (err)
 469                                 goto out;
 470
 471                         gdb = sb_getblk(sb, block);
 472                         if (!gdb) {
 473                                 err = -EIO;
 474                                 goto out;
 475                         }
 476
 477                         err = ext4_journal_get_write_access(handle, gdb);
 478                         if (err) {
 479                                 brelse(gdb);
 480                                 goto out;
 481                         }
 482                         memcpy(gdb->b_data, sbi->s_group_desc[j]->b_data,
 483                                gdb->b_size);
 484                         set_buffer_uptodate(gdb);
 485
 486                         err = ext4_handle_dirty_metadata(handle, NULL, gdb);
 487                         if (unlikely(err)) {
 488                                 brelse(gdb);
 489                                 goto out;
 490                         }
 491                         brelse(gdb);
 492                 }
 493
 494                 /* Zero out all of the reserved backup group descriptor
 495                  * table blocks
 496                  */
 497                 if (ext4_bg_has_super(sb, group)) {
 498                         err = sb_issue_zeroout(sb, gdblocks + start + 1,
 499                                         reserved_gdb, GFP_NOFS);
 500                         if (err)
 501                                 goto out;
 502                 }
 503
 504 handle_itb:
 505                 /* Initialize group tables of the grop @group */
 506                 if (!(bg_flags[i] & EXT4_BG_INODE_ZEROED))
 507                         goto handle_bb;
 508
 509                 /* Zero out all of the inode table blocks */
 510                 block = group_data[i].inode_table;
 511                 ext4_debug("clear inode table blocks %#04llx -> %#04lx\n",
 512                            block, sbi->s_itb_per_group);
 513                 err = sb_issue_zeroout(sb, block, sbi->s_itb_per_group,
 514                                        GFP_NOFS);
 515                 if (err)
 516                         goto out;
 517
 518 handle_bb:
 519                 if (bg_flags[i] & EXT4_BG_BLOCK_UNINIT)
 520                         goto handle_ib;
 521
 522                 /* Initialize block bitmap of the @group */
 523                 block = group_data[i].block_bitmap;
 524                 err = extend_or_restart_transaction(handle, 1);
 525                 if (err)
 526                         goto out;
 527
 528                 bh = bclean(handle, sb, block);
 529                 if (IS_ERR(bh)) {
 530                         err = PTR_ERR(bh);
 531                         goto out;
 532                 }
 533                 if (ext4_bg_has_super(sb, group)) {
 534                         ext4_debug("mark backup superblock %#04llx (+0)\n",
 535                                    start);
 536                         ext4_set_bits(bh->b_data, 0, gdblocks + reserved_gdb +
 537                                                      1);
 538                 }
 539                 ext4_mark_bitmap_end(group_data[i].blocks_count,
 540                                      sb->s_blocksize * 8, bh->b_data);
 541                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 542                 if (err)
 543                         goto out;
 544                 brelse(bh);
 545
 546 handle_ib:
 547                 if (bg_flags[i] & EXT4_BG_INODE_UNINIT)
 548                         continue;
 549
 550                 /* Initialize inode bitmap of the @group */
 551                 block = group_data[i].inode_bitmap;
 552                 err = extend_or_restart_transaction(handle, 1);
 553                 if (err)
 554                         goto out;
 555                 /* Mark unused entries in inode bitmap used */
 556                 bh = bclean(handle, sb, block);
 557                 if (IS_ERR(bh)) {
 558                         err = PTR_ERR(bh);
 559                         goto out;
 560                 }
 561
 562                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 563                                      sb->s_blocksize * 8, bh->b_data);
 564                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
 565                 if (err)
 566                         goto out;
 567                 brelse(bh);
 568         }
 569         bh = NULL;
 570
 571         /* Mark group tables in block bitmap */
 572         for (j = 0; j < GROUP_TABLE_COUNT; j++) {
 573                 count = group_table_count[j];
 574                 start = (&group_data[0].block_bitmap)[j];
 575                 block = start;
 576                 for (i = 1; i < flex_gd->count; i++) {
 577                         block += group_table_count[j];
 578                         if (block == (&group_data[i].block_bitmap)[j]) {
 579                                 count += group_table_count[j];
 580                                 continue;
 581                         }
 582                         err = set_flexbg_block_bitmap(sb, handle,
 583                                                 flex_gd, start, count);
 584                         if (err)
 585                                 goto out;
 586                         count = group_table_count[j];
 587                         start = group_data[i].block_bitmap;
 588                         block = start;
 589                 }
 590
 591                 if (count) {
 592                         err = set_flexbg_block_bitmap(sb, handle,
 593                                                 flex_gd, start, count);
 594                         if (err)
 595                                 goto out;
 596                 }
 597         }
 598
 599 out:
 600         brelse(bh);
 601         err2 = ext4_journal_stop(handle);
 602         if (err2 && !err)
 603                 err = err2;
 604
 605         return err;
 606 }
 607
 608 /*
 609  * Iterate through the groups which hold BACKUP superblock/GDT copies in an
 610  * ext4 filesystem.  The counters should be initialized to 1, 5, and 7 before
 611  * calling this for the first time.  In a sparse filesystem it will be the
 612  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
 613  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
 614  */
 615 static unsigned ext4_list_backups(struct super_block *sb, unsigned *three,
 616                                   unsigned *five, unsigned *seven)
 617 {
 618         unsigned *min = three;
 619         int mult = 3;
 620         unsigned ret;
 621
 622         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 623                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
 624                 ret = *min;
 625                 *min += 1;
 626                 return ret;
 627         }
 628
 629         if (*five < *min) {
 630                 min = five;
 631                 mult = 5;
 632         }
 633         if (*seven < *min) {
 634                 min = seven;
 635                 mult = 7;
 636         }
 637
 638         ret = *min;
 639         *min *= mult;
 640
 641         return ret;
 642 }
 643
 644 /*
 645  * Check that all of the backup GDT blocks are held in the primary GDT block.
 646  * It is assumed that they are stored in group order.  Returns the number of
 647  * groups in current filesystem that have BACKUPS, or -ve error code.
 648  */
 649 static int verify_reserved_gdb(struct super_block *sb,
 650                                ext4_group_t end,
 651                                struct buffer_head *primary)
 652 {
 653         const ext4_fsblk_t blk = primary->b_blocknr;
 654         unsigned three = 1;
 655         unsigned five = 5;
 656         unsigned seven = 7;
 657         unsigned grp;
 658         __le32 *p = (__le32 *)primary->b_data;
 659         int gdbackups = 0;
 660
 661         while ((grp = ext4_list_backups(sb, &three, &five, &seven)) < end) {
 662                 if (le32_to_cpu(*p++) !=
 663                     grp * EXT4_BLOCKS_PER_GROUP(sb) + blk){
 664                         ext4_warning(sb, "reserved GDT %llu"
 665                                      " missing grp %d (%llu)",
 666                                      blk, grp,
 667                                      grp *
 668                                      (ext4_fsblk_t)EXT4_BLOCKS_PER_GROUP(sb) +
 669                                      blk);
 670                         return -EINVAL;
 671                 }
 672                 if (++gdbackups > EXT4_ADDR_PER_BLOCK(sb))
 673                         return -EFBIG;
 674         }
 675
 676         return gdbackups;
 677 }
 678
 679 /*
 680  * Called when we need to bring a reserved group descriptor table block into
 681  * use from the resize inode.  The primary copy of the new GDT block currently
 682  * is an indirect block (under the double indirect block in the resize inode).
 683  * The new backup GDT blocks will be stored as leaf blocks in this indirect
 684  * block, in group order.  Even though we know all the block numbers we need,
 685  * we check to ensure that the resize inode has actually reserved these blocks.
 686  *
 687  * Don't need to update the block bitmaps because the blocks are still in use.
 688  *
 689  * We get all of the error cases out of the way, so that we are sure to not
 690  * fail once we start modifying the data on disk, because JBD has no rollback.
 691  */
 692 static int add_new_gdb(handle_t *handle, struct inode *inode,
 693                        ext4_group_t group)
 694 {
 695         struct super_block *sb = inode->i_sb;
 696         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 697         unsigned long gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
 698         ext4_fsblk_t gdblock = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
 699         struct buffer_head **o_group_desc, **n_group_desc;
 700         struct buffer_head *dind;
 701         struct buffer_head *gdb_bh;
 702         int gdbackups;
 703         struct ext4_iloc iloc;
 704         __le32 *data;
 705         int err;
 706
 707         if (test_opt(sb, DEBUG))
 708                 printk(KERN_DEBUG
 709                        "EXT4-fs: ext4_add_new_gdb: adding group block %lu\n",
 710                        gdb_num);
 711
 712         /*
 713          * If we are not using the primary superblock/GDT copy don't resize,
 714          * because the user tools have no way of handling this.  Probably a
 715          * bad time to do it anyways.
 716          */
 717         if (EXT4_SB(sb)->s_sbh->b_blocknr !=
 718             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) {
 719                 ext4_warning(sb, "won't resize using backup superblock at %llu",
 720                         (unsigned long long)EXT4_SB(sb)->s_sbh->b_blocknr);
 721                 return -EPERM;
 722         }
 723
 724         gdb_bh = sb_bread(sb, gdblock);
 725         if (!gdb_bh)
 726                 return -EIO;
 727
 728         gdbackups = verify_reserved_gdb(sb, group, gdb_bh);
 729         if (gdbackups < 0) {
 730                 err = gdbackups;
 731                 goto exit_bh;
 732         }
 733
 734         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 735         dind = sb_bread(sb, le32_to_cpu(*data));
 736         if (!dind) {
 737                 err = -EIO;
 738                 goto exit_bh;
 739         }
 740
 741         data = (__le32 *)dind->b_data;
 742         if (le32_to_cpu(data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)]) != gdblock) {
 743                 ext4_warning(sb, "new group %u GDT block %llu not reserved",
 744                              group, gdblock);
 745                 err = -EINVAL;
 746                 goto exit_dind;
 747         }
 748
 749         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
 750         if (unlikely(err))
 751                 goto exit_dind;
 752
 753         err = ext4_journal_get_write_access(handle, gdb_bh);
 754         if (unlikely(err))
 755                 goto exit_sbh;
 756
 757         err = ext4_journal_get_write_access(handle, dind);
 758         if (unlikely(err))
 759                 ext4_std_error(sb, err);
 760
 761         /* ext4_reserve_inode_write() gets a reference on the iloc */
 762         err = ext4_reserve_inode_write(handle, inode, &iloc);
 763         if (unlikely(err))
 764                 goto exit_dindj;
 765
 766         n_group_desc = ext4_kvmalloc((gdb_num + 1) *
 767                                      sizeof(struct buffer_head *),
 768                                      GFP_NOFS);
 769         if (!n_group_desc) {
 770                 err = -ENOMEM;
 771                 ext4_warning(sb, "not enough memory for %lu groups",
 772                              gdb_num + 1);
 773                 goto exit_inode;
 774         }
 775
 776         /*
 777          * Finally, we have all of the possible failures behind us...
 778          *
 779          * Remove new GDT block from inode double-indirect block and clear out
 780          * the new GDT block for use (which also "frees" the backup GDT blocks
 781          * from the reserved inode).  We don't need to change the bitmaps for
 782          * these blocks, because they are marked as in-use from being in the
 783          * reserved inode, and will become GDT blocks (primary and backup).
 784          */
 785         data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
 786         err = ext4_handle_dirty_metadata(handle, NULL, dind);
 787         if (unlikely(err)) {
 788                 ext4_std_error(sb, err);
 789                 goto exit_inode;
 790         }
 791         inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
 792         ext4_mark_iloc_dirty(handle, inode, &iloc);
 793         memset(gdb_bh->b_data, 0, sb->s_blocksize);
 794         err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
 795         if (unlikely(err)) {
 796                 ext4_std_error(sb, err);
 797                 goto exit_inode;
 798         }
 799         brelse(dind);
 800
 801         o_group_desc = EXT4_SB(sb)->s_group_desc;
 802         memcpy(n_group_desc, o_group_desc,
 803                EXT4_SB(sb)->s_gdb_count * sizeof(struct buffer_head *));
 804         n_group_desc[gdb_num] = gdb_bh;
 805         EXT4_SB(sb)->s_group_desc = n_group_desc;
 806         EXT4_SB(sb)->s_gdb_count++;
 807         ext4_kvfree(o_group_desc);
 808
 809         le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
 810         err = ext4_handle_dirty_super(handle, sb);
 811         if (err)
 812                 ext4_std_error(sb, err);
 813
 814         return err;
 815
 816 exit_inode:
 817         ext4_kvfree(n_group_desc);
 818         /* ext4_handle_release_buffer(handle, iloc.bh); */
 819         brelse(iloc.bh);
 820 exit_dindj:
 821         /* ext4_handle_release_buffer(handle, dind); */
 822 exit_sbh:
 823         /* ext4_handle_release_buffer(handle, EXT4_SB(sb)->s_sbh); */
 824 exit_dind:
 825         brelse(dind);
 826 exit_bh:
 827         brelse(gdb_bh);
 828
 829         ext4_debug("leaving with error %d\n", err);
 830         return err;
 831 }
 832
 833 /*
 834  * Called when we are adding a new group which has a backup copy of each of
 835  * the GDT blocks (i.e. sparse group) and there are reserved GDT blocks.
 836  * We need to add these reserved backup GDT blocks to the resize inode, so
 837  * that they are kept for future resizing and not allocated to files.
 838  *
 839  * Each reserved backup GDT block will go into a different indirect block.
 840  * The indirect blocks are actually the primary reserved GDT blocks,
 841  * so we know in advance what their block numbers are.  We only get the
 842  * double-indirect block to verify it is pointing to the primary reserved
 843  * GDT blocks so we don't overwrite a data block by accident.  The reserved
 844  * backup GDT blocks are stored in their reserved primary GDT block.
 845  */
 846 static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
 847                               ext4_group_t group)
 848 {
 849         struct super_block *sb = inode->i_sb;
 850         int reserved_gdb =le16_to_cpu(EXT4_SB(sb)->s_es->s_reserved_gdt_blocks);
 851         struct buffer_head **primary;
 852         struct buffer_head *dind;
 853         struct ext4_iloc iloc;
 854         ext4_fsblk_t blk;
 855         __le32 *data, *end;
 856         int gdbackups = 0;
 857         int res, i;
 858         int err;
 859
 860         primary = kmalloc(reserved_gdb * sizeof(*primary), GFP_NOFS);
 861         if (!primary)
 862                 return -ENOMEM;
 863
 864         data = EXT4_I(inode)->i_data + EXT4_DIND_BLOCK;
 865         dind = sb_bread(sb, le32_to_cpu(*data));
 866         if (!dind) {
 867                 err = -EIO;
 868                 goto exit_free;
 869         }
 870
 871         blk = EXT4_SB(sb)->s_sbh->b_blocknr + 1 + EXT4_SB(sb)->s_gdb_count;
 872         data = (__le32 *)dind->b_data + (EXT4_SB(sb)->s_gdb_count %
 873                                          EXT4_ADDR_PER_BLOCK(sb));
 874         end = (__le32 *)dind->b_data + EXT4_ADDR_PER_BLOCK(sb);
 875
 876         /* Get each reserved primary GDT block and verify it holds backups */
 877         for (res = 0; res < reserved_gdb; res++, blk++) {
 878                 if (le32_to_cpu(*data) != blk) {
 879                         ext4_warning(sb, "reserved block %llu"
 880                                      " not at offset %ld",
 881                                      blk,
 882                                      (long)(data - (__le32 *)dind->b_data));
 883                         err = -EINVAL;
 884                         goto exit_bh;
 885                 }
 886                 primary[res] = sb_bread(sb, blk);
 887                 if (!primary[res]) {
 888                         err = -EIO;
 889                         goto exit_bh;
 890                 }
 891                 gdbackups = verify_reserved_gdb(sb, group, primary[res]);
 892                 if (gdbackups < 0) {
 893                         brelse(primary[res]);
 894                         err = gdbackups;
 895                         goto exit_bh;
 896                 }
 897                 if (++data >= end)
 898                         data = (__le32 *)dind->b_data;
 899         }
 900
 901         for (i = 0; i < reserved_gdb; i++) {
 902                 if ((err = ext4_journal_get_write_access(handle, primary[i]))) {
 903                         /*
 904                         int j;
 905                         for (j = 0; j < i; j++)
 906                                 ext4_handle_release_buffer(handle, primary[j]);
 907                          */
 908                         goto exit_bh;
 909                 }
 910         }
 911
 912         if ((err = ext4_reserve_inode_write(handle, inode, &iloc)))
 913                 goto exit_bh;
 914
 915         /*
 916          * Finally we can add each of the reserved backup GDT blocks from
 917          * the new group to its reserved primary GDT block.
 918          */
 919         blk = group * EXT4_BLOCKS_PER_GROUP(sb);
 920         for (i = 0; i < reserved_gdb; i++) {
 921                 int err2;
 922                 data = (__le32 *)primary[i]->b_data;
 923                 /* printk("reserving backup %lu[%u] = %lu\n",
 924                        primary[i]->b_blocknr, gdbackups,
 925                        blk + primary[i]->b_blocknr); */
 926                 data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
 927                 err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
 928                 if (!err)
 929                         err = err2;
 930         }
 931         inode->i_blocks += reserved_gdb * sb->s_blocksize >> 9;
 932         ext4_mark_iloc_dirty(handle, inode, &iloc);
 933
 934 exit_bh:
 935         while (--res >= 0)
 936                 brelse(primary[res]);
 937         brelse(dind);
 938
 939 exit_free:
 940         kfree(primary);
 941
 942         return err;
 943 }
 944
 945 /*
 946  * Update the backup copies of the ext4 metadata.  These don't need to be part
 947  * of the main resize transaction, because e2fsck will re-write them if there
 948  * is a problem (basically only OOM will cause a problem).  However, we
 949  * _should_ update the backups if possible, in case the primary gets trashed
 950  * for some reason and we need to run e2fsck from a backup superblock.  The
 951  * important part is that the new block and inode counts are in the backup
 952  * superblocks, and the location of the new group metadata in the GDT backups.
 953  *
 954  * We do not need take the s_resize_lock for this, because these
 955  * blocks are not otherwise touched by the filesystem code when it is
 956  * mounted.  We don't need to worry about last changing from
 957  * sbi->s_groups_count, because the worst that can happen is that we
 958  * do not copy the full number of backups at this time.  The resize
 959  * which changed s_groups_count will backup again.
 960  */
 961 static void update_backups(struct super_block *sb,
 962                            int blk_off, char *data, int size)
 963 {
 964         struct ext4_sb_info *sbi = EXT4_SB(sb);
 965         const ext4_group_t last = sbi->s_groups_count;
 966         const int bpg = EXT4_BLOCKS_PER_GROUP(sb);
 967         unsigned three = 1;
 968         unsigned five = 5;
 969         unsigned seven = 7;
 970         ext4_group_t group;
 971         int rest = sb->s_blocksize - size;
 972         handle_t *handle;
 973         int err = 0, err2;
 974
 975         handle = ext4_journal_start_sb(sb, EXT4_MAX_TRANS_DATA);
 976         if (IS_ERR(handle)) {
 977                 group = 1;
 978                 err = PTR_ERR(handle);
 979                 goto exit_err;
 980         }
 981
 982         ext4_superblock_csum_set(sb, (struct ext4_super_block *)data);
 983
 984         while ((group = ext4_list_backups(sb, &three, &five, &seven)) < last) {
 985                 struct buffer_head *bh;
 986
 987                 /* Out of journal space, and can't get more - abort - so sad */
 988                 if (ext4_handle_valid(handle) &&
 989                     handle->h_buffer_credits == 0 &&
 990                     ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
 991                     (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
 992                         break;
 993
 994                 bh = sb_getblk(sb, group * bpg + blk_off);
 995                 if (!bh) {
 996                         err = -EIO;
 997                         break;
 998                 }
 999                 ext4_debug("update metadata backup %#04lx\n",
1000                           (unsigned long)bh->b_blocknr);
1001                 if ((err = ext4_journal_get_write_access(handle, bh)))
1002                         break;
1003                 lock_buffer(bh);
1004                 memcpy(bh->b_data, data, size);
1005                 if (rest)
1006                         memset(bh->b_data + size, 0, rest);
1007                 set_buffer_uptodate(bh);
1008                 unlock_buffer(bh);
1009                 err = ext4_handle_dirty_metadata(handle, NULL, bh);
1010                 if (unlikely(err))
1011                         ext4_std_error(sb, err);
1012                 brelse(bh);
1013         }
1014         if ((err2 = ext4_journal_stop(handle)) && !err)
1015                 err = err2;
1016
1017         /*
1018          * Ugh! Need to have e2fsck write the backup copies.  It is too
1019          * late to revert the resize, we shouldn't fail just because of
1020          * the backup copies (they are only needed in case of corruption).
1021          *
1022          * However, if we got here we have a journal problem too, so we
1023          * can't really start a transaction to mark the superblock.
1024          * Chicken out and just set the flag on the hope it will be written
1025          * to disk, and if not - we will simply wait until next fsck.
1026          */
1027 exit_err:
1028         if (err) {
1029                 ext4_warning(sb, "can't update backup for group %u (err %d), "
1030                              "forcing fsck on next reboot", group, err);
1031                 sbi->s_mount_state &= ~EXT4_VALID_FS;
1032                 sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1033                 mark_buffer_dirty(sbi->s_sbh);
1034         }
1035 }
1036
1037 /*
1038  * ext4_add_new_descs() adds @count group descriptor of groups
1039  * starting at @group
1040  *
1041  * @handle: journal handle
1042  * @sb: super block
1043  * @group: the group no. of the first group desc to be added
1044  * @resize_inode: the resize inode
1045  * @count: number of group descriptors to be added
1046  */
1047 static int ext4_add_new_descs(handle_t *handle, struct super_block *sb,
1048                               ext4_group_t group, struct inode *resize_inode,
1049                               ext4_group_t count)
1050 {
1051         struct ext4_sb_info *sbi = EXT4_SB(sb);
1052         struct ext4_super_block *es = sbi->s_es;
1053         struct buffer_head *gdb_bh;
1054         int i, gdb_off, gdb_num, err = 0;
1055
1056         for (i = 0; i < count; i++, group++) {
1057                 int reserved_gdb = ext4_bg_has_super(sb, group) ?
1058                         le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1059
1060                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1061                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1062
1063                 /*
1064                  * We will only either add reserved group blocks to a backup group
1065                  * or remove reserved blocks for the first group in a new group block.
1066                  * Doing both would be mean more complex code, and sane people don't
1067                  * use non-sparse filesystems anymore.  This is already checked above.
1068                  */
1069                 if (gdb_off) {
1070                         gdb_bh = sbi->s_group_desc[gdb_num];
1071                         err = ext4_journal_get_write_access(handle, gdb_bh);
1072
1073                         if (!err && reserved_gdb && ext4_bg_num_gdb(sb, group))
1074                                 err = reserve_backup_gdb(handle, resize_inode, group);
1075                 } else
1076                         err = add_new_gdb(handle, resize_inode, group);
1077                 if (err)
1078                         break;
1079         }
1080         return err;
1081 }
1082
1083 static struct buffer_head *ext4_get_bitmap(struct super_block *sb, __u64 block)
1084 {
1085         struct buffer_head *bh = sb_getblk(sb, block);
1086         if (!bh)
1087                 return NULL;
1088
1089         if (bitmap_uptodate(bh))
1090                 return bh;
1091
1092         lock_buffer(bh);
1093         if (bh_submit_read(bh) < 0) {
1094                 unlock_buffer(bh);
1095                 brelse(bh);
1096                 return NULL;
1097         }
1098         unlock_buffer(bh);
1099
1100         return bh;
1101 }
1102
1103 static int ext4_set_bitmap_checksums(struct super_block *sb,
1104                                      ext4_group_t group,
1105                                      struct ext4_group_desc *gdp,
1106                                      struct ext4_new_group_data *group_data)
1107 {
1108         struct buffer_head *bh;
1109
1110         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
1111                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1112                 return 0;
1113
1114         bh = ext4_get_bitmap(sb, group_data->inode_bitmap);
1115         if (!bh)
1116                 return -EIO;
1117         ext4_inode_bitmap_csum_set(sb, group, gdp, bh,
1118                                    EXT4_INODES_PER_GROUP(sb) / 8);
1119         brelse(bh);
1120
1121         bh = ext4_get_bitmap(sb, group_data->block_bitmap);
1122         if (!bh)
1123                 return -EIO;
1124         ext4_block_bitmap_csum_set(sb, group, gdp, bh,
1125                                    EXT4_BLOCKS_PER_GROUP(sb) / 8);
1126         brelse(bh);
1127
1128         return 0;
1129 }
1130
1131 /*
1132  * ext4_setup_new_descs() will set up the group descriptor descriptors of a flex bg
1133  */
1134 static int ext4_setup_new_descs(handle_t *handle, struct super_block *sb,
1135                                 struct ext4_new_flex_group_data *flex_gd)
1136 {
1137         struct ext4_new_group_data      *group_data = flex_gd->groups;
1138         struct ext4_group_desc          *gdp;
1139         struct ext4_sb_info             *sbi = EXT4_SB(sb);
1140         struct buffer_head              *gdb_bh;
1141         ext4_group_t                    group;
1142         __u16                           *bg_flags = flex_gd->bg_flags;
1143         int                             i, gdb_off, gdb_num, err = 0;
1144
1145
1146         for (i = 0; i < flex_gd->count; i++, group_data++, bg_flags++) {
1147                 group = group_data->group;
1148
1149                 gdb_off = group % EXT4_DESC_PER_BLOCK(sb);
1150                 gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1151
1152                 /*
1153                  * get_write_access() has been called on gdb_bh by ext4_add_new_desc().
1154                  */
1155                 gdb_bh = sbi->s_group_desc[gdb_num];
1156                 /* Update group descriptor block for new group */
1157                 gdp = (struct ext4_group_desc *)(gdb_bh->b_data +
1158                                                  gdb_off * EXT4_DESC_SIZE(sb));
1159
1160                 memset(gdp, 0, EXT4_DESC_SIZE(sb));
1161                 ext4_block_bitmap_set(sb, gdp, group_data->block_bitmap);
1162                 ext4_inode_bitmap_set(sb, gdp, group_data->inode_bitmap);
1163                 err = ext4_set_bitmap_checksums(sb, group, gdp, group_data);
1164                 if (err) {
1165                         ext4_std_error(sb, err);
1166                         break;
1167                 }
1168
1169                 ext4_inode_table_set(sb, gdp, group_data->inode_table);
1170                 ext4_free_group_clusters_set(sb, gdp,
1171                                              EXT4_B2C(sbi, group_data->free_blocks_count));
1172                 ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
1173                 gdp->bg_flags = cpu_to_le16(*bg_flags);
1174                 ext4_group_desc_csum_set(sb, group, gdp);
1175
1176                 err = ext4_handle_dirty_metadata(handle, NULL, gdb_bh);
1177                 if (unlikely(err)) {
1178                         ext4_std_error(sb, err);
1179                         break;
1180                 }
1181
1182                 /*
1183                  * We can allocate memory for mb_alloc based on the new group
1184                  * descriptor
1185                  */
1186                 err = ext4_mb_add_groupinfo(sb, group, gdp);
1187                 if (err)
1188                         break;
1189         }
1190         return err;
1191 }
1192
1193 /*
1194  * ext4_update_super() updates the super block so that the newly added
1195  * groups can be seen by the filesystem.
1196  *
1197  * @sb: super block
1198  * @flex_gd: new added groups
1199  */
1200 static void ext4_update_super(struct super_block *sb,
1201                              struct ext4_new_flex_group_data *flex_gd)
1202 {
1203         ext4_fsblk_t blocks_count = 0;
1204         ext4_fsblk_t free_blocks = 0;
1205         ext4_fsblk_t reserved_blocks = 0;
1206         struct ext4_new_group_data *group_data = flex_gd->groups;
1207         struct ext4_sb_info *sbi = EXT4_SB(sb);
1208         struct ext4_super_block *es = sbi->s_es;
1209         int i;
1210
1211         BUG_ON(flex_gd->count == 0 || group_data == NULL);
1212         /*
1213          * Make the new blocks and inodes valid next.  We do this before
1214          * increasing the group count so that once the group is enabled,
1215          * all of its blocks and inodes are already valid.
1216          *
1217          * We always allocate group-by-group, then block-by-block or
1218          * inode-by-inode within a group, so enabling these
1219          * blocks/inodes before the group is live won't actually let us
1220          * allocate the new space yet.
1221          */
1222         for (i = 0; i < flex_gd->count; i++) {
1223                 blocks_count += group_data[i].blocks_count;
1224                 free_blocks += group_data[i].free_blocks_count;
1225         }
1226
1227         reserved_blocks = ext4_r_blocks_count(es) * 100;
1228         do_div(reserved_blocks, ext4_blocks_count(es));
1229         reserved_blocks *= blocks_count;
1230         do_div(reserved_blocks, 100);
1231
1232         ext4_blocks_count_set(es, ext4_blocks_count(es) + blocks_count);
1233         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + free_blocks);
1234         le32_add_cpu(&es->s_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1235                      flex_gd->count);
1236         le32_add_cpu(&es->s_free_inodes_count, EXT4_INODES_PER_GROUP(sb) *
1237                      flex_gd->count);
1238
1239         /*
1240          * We need to protect s_groups_count against other CPUs seeing
1241          * inconsistent state in the superblock.
1242          *
1243          * The precise rules we use are:
1244          *
1245          * * Writers must perform a smp_wmb() after updating all
1246          *   dependent data and before modifying the groups count
1247          *
1248          * * Readers must perform an smp_rmb() after reading the groups
1249          *   count and before reading any dependent data.
1250          *
1251          * NB. These rules can be relaxed when checking the group count
1252          * while freeing data, as we can only allocate from a block
1253          * group after serialising against the group count, and we can
1254          * only then free after serialising in turn against that
1255          * allocation.
1256          */
1257         smp_wmb();
1258
1259         /* Update the global fs size fields */
1260         sbi->s_groups_count += flex_gd->count;
1261
1262         /* Update the reserved block counts only once the new group is
1263          * active. */
1264         ext4_r_blocks_count_set(es, ext4_r_blocks_count(es) +
1265                                 reserved_blocks);
1266
1267         /* Update the free space counts */
1268         percpu_counter_add(&sbi->s_freeclusters_counter,
1269                            EXT4_B2C(sbi, free_blocks));
1270         percpu_counter_add(&sbi->s_freeinodes_counter,
1271                            EXT4_INODES_PER_GROUP(sb) * flex_gd->count);
1272
1273         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
1274                                       EXT4_FEATURE_INCOMPAT_FLEX_BG) &&
1275             sbi->s_log_groups_per_flex) {
1276                 ext4_group_t flex_group;
1277                 flex_group = ext4_flex_group(sbi, group_data[0].group);
1278                 atomic_add(EXT4_B2C(sbi, free_blocks),
1279                            &sbi->s_flex_groups[flex_group].free_clusters);
1280                 atomic_add(EXT4_INODES_PER_GROUP(sb) * flex_gd->count,
1281                            &sbi->s_flex_groups[flex_group].free_inodes);
1282         }
1283
1284         /*
1285          * Update the fs overhead information
1286          */
1287         ext4_calculate_overhead(sb);
1288
1289         if (test_opt(sb, DEBUG))
1290                 printk(KERN_DEBUG "EXT4-fs: added group %u:"
1291                        "%llu blocks(%llu free %llu reserved)\n", flex_gd->count,
1292                        blocks_count, free_blocks, reserved_blocks);
1293 }
1294
1295 /* Add a flex group to an fs. Ensure we handle all possible error conditions
1296  * _before_ we start modifying the filesystem, because we cannot abort the
1297  * transaction and not have it write the data to disk.
1298  */
1299 static int ext4_flex_group_add(struct super_block *sb,
1300                                struct inode *resize_inode,
1301                                struct ext4_new_flex_group_data *flex_gd)
1302 {
1303         struct ext4_sb_info *sbi = EXT4_SB(sb);
1304         struct ext4_super_block *es = sbi->s_es;
1305         ext4_fsblk_t o_blocks_count;
1306         ext4_grpblk_t last;
1307         ext4_group_t group;
1308         handle_t *handle;
1309         unsigned reserved_gdb;
1310         int err = 0, err2 = 0, credit;
1311
1312         BUG_ON(!flex_gd->count || !flex_gd->groups || !flex_gd->bg_flags);
1313
1314         reserved_gdb = le16_to_cpu(es->s_reserved_gdt_blocks);
1315         o_blocks_count = ext4_blocks_count(es);
1316         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1317         BUG_ON(last);
1318
1319         err = setup_new_flex_group_blocks(sb, flex_gd);
1320         if (err)
1321                 goto exit;
1322         /*
1323          * We will always be modifying at least the superblock and  GDT
1324          * block.  If we are adding a group past the last current GDT block,
1325          * we will also modify the inode and the dindirect block.  If we
1326          * are adding a group with superblock/GDT backups  we will also
1327          * modify each of the reserved GDT dindirect blocks.
1328          */
1329         credit = flex_gd->count * 4 + reserved_gdb;
1330         handle = ext4_journal_start_sb(sb, credit);
1331         if (IS_ERR(handle)) {
1332                 err = PTR_ERR(handle);
1333                 goto exit;
1334         }
1335
1336         err = ext4_journal_get_write_access(handle, sbi->s_sbh);
1337         if (err)
1338                 goto exit_journal;
1339
1340         group = flex_gd->groups[0].group;
1341         BUG_ON(group != EXT4_SB(sb)->s_groups_count);
1342         err = ext4_add_new_descs(handle, sb, group,
1343                                 resize_inode, flex_gd->count);
1344         if (err)
1345                 goto exit_journal;
1346
1347         err = ext4_setup_new_descs(handle, sb, flex_gd);
1348         if (err)
1349                 goto exit_journal;
1350
1351         ext4_update_super(sb, flex_gd);
1352
1353         err = ext4_handle_dirty_super(handle, sb);
1354
1355 exit_journal:
1356         err2 = ext4_journal_stop(handle);
1357         if (!err)
1358                 err = err2;
1359
1360         if (!err) {
1361                 int gdb_num = group / EXT4_DESC_PER_BLOCK(sb);
1362                 int gdb_num_end = ((group + flex_gd->count - 1) /
1363                                    EXT4_DESC_PER_BLOCK(sb));
1364
1365                 update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
1366                                sizeof(struct ext4_super_block));
1367                 for (; gdb_num <= gdb_num_end; gdb_num++) {
1368                         struct buffer_head *gdb_bh;
1369
1370                         gdb_bh = sbi->s_group_desc[gdb_num];
1371                         update_backups(sb, gdb_bh->b_blocknr, gdb_bh->b_data,
1372                                        gdb_bh->b_size);
1373                 }
1374         }
1375 exit:
1376         return err;
1377 }
1378
1379 static int ext4_setup_next_flex_gd(struct super_block *sb,
1380                                     struct ext4_new_flex_group_data *flex_gd,
1381                                     ext4_fsblk_t n_blocks_count,
1382                                     unsigned long flexbg_size)
1383 {
1384         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1385         struct ext4_new_group_data *group_data = flex_gd->groups;
1386         ext4_fsblk_t o_blocks_count;
1387         ext4_group_t n_group;
1388         ext4_group_t group;
1389         ext4_group_t last_group;
1390         ext4_grpblk_t last;
1391         ext4_grpblk_t blocks_per_group;
1392         unsigned long i;
1393
1394         blocks_per_group = EXT4_BLOCKS_PER_GROUP(sb);
1395
1396         o_blocks_count = ext4_blocks_count(es);
1397
1398         if (o_blocks_count == n_blocks_count)
1399                 return 0;
1400
1401         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1402         BUG_ON(last);
1403         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &last);
1404
1405         last_group = group | (flexbg_size - 1);
1406         if (last_group > n_group)
1407                 last_group = n_group;
1408
1409         flex_gd->count = last_group - group + 1;
1410
1411         for (i = 0; i < flex_gd->count; i++) {
1412                 int overhead;
1413
1414                 group_data[i].group = group + i;
1415                 group_data[i].blocks_count = blocks_per_group;
1416                 overhead = ext4_bg_has_super(sb, group + i) ?
1417                            (1 + ext4_bg_num_gdb(sb, group + i) +
1418                             le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
1419                 group_data[i].free_blocks_count = blocks_per_group - overhead;
1420                 if (ext4_has_group_desc_csum(sb))
1421                         flex_gd->bg_flags[i] = EXT4_BG_BLOCK_UNINIT |
1422                                                EXT4_BG_INODE_UNINIT;
1423                 else
1424                         flex_gd->bg_flags[i] = EXT4_BG_INODE_ZEROED;
1425         }
1426
1427         if (last_group == n_group && ext4_has_group_desc_csum(sb))
1428                 /* We need to initialize block bitmap of last group. */
1429                 flex_gd->bg_flags[i - 1] &= ~EXT4_BG_BLOCK_UNINIT;
1430
1431         if ((last_group == n_group) && (last != blocks_per_group - 1)) {
1432                 group_data[i - 1].blocks_count = last + 1;
1433                 group_data[i - 1].free_blocks_count -= blocks_per_group-
1434                                         last - 1;
1435         }
1436
1437         return 1;
1438 }
1439
1440 /* Add group descriptor data to an existing or new group descriptor block.
1441  * Ensure we handle all possible error conditions _before_ we start modifying
1442  * the filesystem, because we cannot abort the transaction and not have it
1443  * write the data to disk.
1444  *
1445  * If we are on a GDT block boundary, we need to get the reserved GDT block.
1446  * Otherwise, we may need to add backup GDT blocks for a sparse group.
1447  *
1448  * We only need to hold the superblock lock while we are actually adding
1449  * in the new group's counts to the superblock.  Prior to that we have
1450  * not really "added" the group at all.  We re-check that we are still
1451  * adding in the last group in case things have changed since verifying.
1452  */
1453 int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
1454 {
1455         struct ext4_new_flex_group_data flex_gd;
1456         struct ext4_sb_info *sbi = EXT4_SB(sb);
1457         struct ext4_super_block *es = sbi->s_es;
1458         int reserved_gdb = ext4_bg_has_super(sb, input->group) ?
1459                 le16_to_cpu(es->s_reserved_gdt_blocks) : 0;
1460         struct inode *inode = NULL;
1461         int gdb_off, gdb_num;
1462         int err;
1463         __u16 bg_flags = 0;
1464
1465         gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
1466         gdb_off = input->group % EXT4_DESC_PER_BLOCK(sb);
1467
1468         if (gdb_off == 0 && !EXT4_HAS_RO_COMPAT_FEATURE(sb,
1469                                         EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
1470                 ext4_warning(sb, "Can't resize non-sparse filesystem further");
1471                 return -EPERM;
1472         }
1473
1474         if (ext4_blocks_count(es) + input->blocks_count <
1475             ext4_blocks_count(es)) {
1476                 ext4_warning(sb, "blocks_count overflow");
1477                 return -EINVAL;
1478         }
1479
1480         if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
1481             le32_to_cpu(es->s_inodes_count)) {
1482                 ext4_warning(sb, "inodes_count overflow");
1483                 return -EINVAL;
1484         }
1485
1486         if (reserved_gdb || gdb_off == 0) {
1487                 if (!EXT4_HAS_COMPAT_FEATURE(sb,
1488                                              EXT4_FEATURE_COMPAT_RESIZE_INODE)
1489                     || !le16_to_cpu(es->s_reserved_gdt_blocks)) {
1490                         ext4_warning(sb,
1491                                      "No reserved GDT blocks, can't resize");
1492                         return -EPERM;
1493                 }
1494                 inode = ext4_iget(sb, EXT4_RESIZE_INO);
1495                 if (IS_ERR(inode)) {
1496                         ext4_warning(sb, "Error opening resize inode");
1497                         return PTR_ERR(inode);
1498                 }
1499         }
1500
1501
1502         err = verify_group_input(sb, input);
1503         if (err)
1504                 goto out;
1505
1506         err = ext4_alloc_flex_bg_array(sb, input->group + 1);
1507         if (err)
1508                 return err;
1509
1510         err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
1511         if (err)
1512                 goto out;
1513
1514         flex_gd.count = 1;
1515         flex_gd.groups = input;
1516         flex_gd.bg_flags = &bg_flags;
1517         err = ext4_flex_group_add(sb, inode, &flex_gd);
1518 out:
1519         iput(inode);
1520         return err;
1521 } /* ext4_group_add */
1522
1523 /*
1524  * extend a group without checking assuming that checking has been done.
1525  */
1526 static int ext4_group_extend_no_check(struct super_block *sb,
1527                                       ext4_fsblk_t o_blocks_count, ext4_grpblk_t add)
1528 {
1529         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
1530         handle_t *handle;
1531         int err = 0, err2;
1532
1533         /* We will update the superblock, one block bitmap, and
1534          * one group descriptor via ext4_group_add_blocks().
1535          */
1536         handle = ext4_journal_start_sb(sb, 3);
1537         if (IS_ERR(handle)) {
1538                 err = PTR_ERR(handle);
1539                 ext4_warning(sb, "error %d on journal start", err);
1540                 return err;
1541         }
1542
1543         err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
1544         if (err) {
1545                 ext4_warning(sb, "error %d on journal write access", err);
1546                 goto errout;
1547         }
1548
1549         ext4_blocks_count_set(es, o_blocks_count + add);
1550         ext4_free_blocks_count_set(es, ext4_free_blocks_count(es) + add);
1551         ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
1552                    o_blocks_count + add);
1553         /* We add the blocks to the bitmap and set the group need init bit */
1554         err = ext4_group_add_blocks(handle, sb, o_blocks_count, add);
1555         if (err)
1556                 goto errout;
1557         ext4_handle_dirty_super(handle, sb);
1558         ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
1559                    o_blocks_count + add);
1560 errout:
1561         err2 = ext4_journal_stop(handle);
1562         if (err2 && !err)
1563                 err = err2;
1564
1565         if (!err) {
1566                 if (test_opt(sb, DEBUG))
1567                         printk(KERN_DEBUG "EXT4-fs: extended group to %llu "
1568                                "blocks\n", ext4_blocks_count(es));
1569                 update_backups(sb, EXT4_SB(sb)->s_sbh->b_blocknr, (char *)es,
1570                                sizeof(struct ext4_super_block));
1571         }
1572         return err;
1573 }
1574
1575 /*
1576  * Extend the filesystem to the new number of blocks specified.  This entry
1577  * point is only used to extend the current filesystem to the end of the last
1578  * existing group.  It can be accessed via ioctl, or by "remount,resize=<size>"
1579  * for emergencies (because it has no dependencies on reserved blocks).
1580  *
1581  * If we _really_ wanted, we could use default values to call ext4_group_add()
1582  * allow the "remount" trick to work for arbitrary resizing, assuming enough
1583  * GDT blocks are reserved to grow to the desired size.
1584  */
1585 int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
1586                       ext4_fsblk_t n_blocks_count)
1587 {
1588         ext4_fsblk_t o_blocks_count;
1589         ext4_grpblk_t last;
1590         ext4_grpblk_t add;
1591         struct buffer_head *bh;
1592         int err;
1593         ext4_group_t group;
1594
1595         o_blocks_count = ext4_blocks_count(es);
1596
1597         if (test_opt(sb, DEBUG))
1598                 ext4_msg(sb, KERN_DEBUG,
1599                          "extending last group from %llu to %llu blocks",
1600                          o_blocks_count, n_blocks_count);
1601
1602         if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
1603                 return 0;
1604
1605         if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
1606                 ext4_msg(sb, KERN_ERR,
1607                          "filesystem too large to resize to %llu blocks safely",
1608                          n_blocks_count);
1609                 if (sizeof(sector_t) < 8)
1610                         ext4_warning(sb, "CONFIG_LBDAF not enabled");
1611                 return -EINVAL;
1612         }
1613
1614         if (n_blocks_count < o_blocks_count) {
1615                 ext4_warning(sb, "can't shrink FS - resize aborted");
1616                 return -EINVAL;
1617         }
1618
1619         /* Handle the remaining blocks in the last group only. */
1620         ext4_get_group_no_and_offset(sb, o_blocks_count, &group, &last);
1621
1622         if (last == 0) {
1623                 ext4_warning(sb, "need to use ext2online to resize further");
1624                 return -EPERM;
1625         }
1626
1627         add = EXT4_BLOCKS_PER_GROUP(sb) - last;
1628
1629         if (o_blocks_count + add < o_blocks_count) {
1630                 ext4_warning(sb, "blocks_count overflow");
1631                 return -EINVAL;
1632         }
1633
1634         if (o_blocks_count + add > n_blocks_count)
1635                 add = n_blocks_count - o_blocks_count;
1636
1637         if (o_blocks_count + add < n_blocks_count)
1638                 ext4_warning(sb, "will only finish group (%llu blocks, %u new)",
1639                              o_blocks_count + add, add);
1640
1641         /* See if the device is actually as big as what was requested */
1642         bh = sb_bread(sb, o_blocks_count + add - 1);
1643         if (!bh) {
1644                 ext4_warning(sb, "can't read last block, resize aborted");
1645                 return -ENOSPC;
1646         }
1647         brelse(bh);
1648
1649         err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1650         return err;
1651 } /* ext4_group_extend */
1652
1653 /*
1654  * ext4_resize_fs() resizes a fs to new size specified by @n_blocks_count
1655  *
1656  * @sb: super block of the fs to be resized
1657  * @n_blocks_count: the number of blocks resides in the resized fs
1658  */
1659 int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
1660 {
1661         struct ext4_new_flex_group_data *flex_gd = NULL;
1662         struct ext4_sb_info *sbi = EXT4_SB(sb);
1663         struct ext4_super_block *es = sbi->s_es;
1664         struct buffer_head *bh;
1665         struct inode *resize_inode;
1666         ext4_fsblk_t o_blocks_count;
1667         ext4_group_t o_group;
1668         ext4_group_t n_group;
1669         ext4_grpblk_t offset, add;
1670         unsigned long n_desc_blocks;
1671         unsigned long o_desc_blocks;
1672         unsigned long desc_blocks;
1673         int err = 0, flexbg_size = 1 << sbi->s_log_groups_per_flex;
1674
1675         o_blocks_count = ext4_blocks_count(es);
1676
1677         if (test_opt(sb, DEBUG))
1678                 ext4_msg(sb, KERN_DEBUG, "resizing filesystem from %llu "
1679                        "to %llu blocks", o_blocks_count, n_blocks_count);
1680
1681         if (n_blocks_count < o_blocks_count) {
1682                 /* On-line shrinking not supported */
1683                 ext4_warning(sb, "can't shrink FS - resize aborted");
1684                 return -EINVAL;
1685         }
1686
1687         if (n_blocks_count == o_blocks_count)
1688                 /* Nothing need to do */
1689                 return 0;
1690
1691         ext4_get_group_no_and_offset(sb, n_blocks_count - 1, &n_group, &offset);
1692         ext4_get_group_no_and_offset(sb, o_blocks_count - 1, &o_group, &offset);
1693
1694         n_desc_blocks = (n_group + EXT4_DESC_PER_BLOCK(sb)) /
1695                         EXT4_DESC_PER_BLOCK(sb);
1696         o_desc_blocks = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
1697                         EXT4_DESC_PER_BLOCK(sb);
1698         desc_blocks = n_desc_blocks - o_desc_blocks;
1699
1700         if (desc_blocks &&
1701             (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_RESIZE_INODE) ||
1702              le16_to_cpu(es->s_reserved_gdt_blocks) < desc_blocks)) {
1703                 ext4_warning(sb, "No reserved GDT blocks, can't resize");
1704                 return -EPERM;
1705         }
1706
1707         resize_inode = ext4_iget(sb, EXT4_RESIZE_INO);
1708         if (IS_ERR(resize_inode)) {
1709                 ext4_warning(sb, "Error opening resize inode");
1710                 return PTR_ERR(resize_inode);
1711         }
1712
1713         /* See if the device is actually as big as what was requested */
1714         bh = sb_bread(sb, n_blocks_count - 1);
1715         if (!bh) {
1716                 ext4_warning(sb, "can't read last block, resize aborted");
1717                 return -ENOSPC;
1718         }
1719         brelse(bh);
1720
1721         /* extend the last group */
1722         if (n_group == o_group)
1723                 add = n_blocks_count - o_blocks_count;
1724         else
1725                 add = EXT4_BLOCKS_PER_GROUP(sb) - (offset + 1);
1726         if (add > 0) {
1727                 err = ext4_group_extend_no_check(sb, o_blocks_count, add);
1728                 if (err)
1729                         goto out;
1730         }
1731
1732         if (ext4_blocks_count(es) == n_blocks_count)
1733                 goto out;
1734
1735         err = ext4_alloc_flex_bg_array(sb, n_group + 1);
1736         if (err)
1737                 return err;
1738
1739         err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
1740         if (err)
1741                 goto out;
1742
1743         flex_gd = alloc_flex_gd(flexbg_size);
1744         if (flex_gd == NULL) {
1745                 err = -ENOMEM;
1746                 goto out;
1747         }
1748
1749         /* Add flex groups. Note that a regular group is a
1750          * flex group with 1 group.
1751          */
1752         while (ext4_setup_next_flex_gd(sb, flex_gd, n_blocks_count,
1753                                               flexbg_size)) {
1754                 if (ext4_alloc_group_tables(sb, flex_gd, flexbg_size) != 0)
1755                         break;
1756                 err = ext4_flex_group_add(sb, resize_inode, flex_gd);
1757                 if (unlikely(err))
1758                         break;
1759         }
1760
1761 out:
1762         if (flex_gd)
1763                 free_flex_gd(flex_gd);
1764
1765         iput(resize_inode);
1766         if (test_opt(sb, DEBUG))
1767                 ext4_msg(sb, KERN_DEBUG, "resized filesystem from %llu "
1768                        "upto %llu blocks", o_blocks_count, n_blocks_count);
1769         return err;
1770 }