fs/ext4/ialloc.c

   1 /*
   2  *  linux/fs/ext4/ialloc.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  BSD ufs-inspired inode and directory allocation by
  10  *  Stephen Tweedie (sct@redhat.com), 1993
  11  *  Big-endian to little-endian byte-swapping/bitmaps by
  12  *        David S. Miller (davem@caip.rutgers.edu), 1995
  13  */
  14
  15 #include <linux/time.h>
  16 #include <linux/fs.h>
  17 #include <linux/jbd2.h>
  18 #include <linux/stat.h>
  19 #include <linux/string.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/random.h>
  23 #include <linux/bitops.h>
  24 #include <linux/blkdev.h>
  25 #include <asm/byteorder.h>
  26
  27 #include "ext4.h"
  28 #include "ext4_jbd2.h"
  29 #include "xattr.h"
  30 #include "acl.h"
  31
  32 #include <trace/events/ext4.h>
  33
  34 /*
  35  * ialloc.c contains the inodes allocation and deallocation routines
  36  */
  37
  38 /*
  39  * The free inodes are managed by bitmaps.  A file system contains several
  40  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  41  * block for inodes, N blocks for the inode table and data blocks.
  42  *
  43  * The file system contains group descriptors which are located after the
  44  * super block.  Each descriptor contains the number of the bitmap block and
  45  * the free blocks count in the block.
  46  */
  47
  48 /*
  49  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  50  * need to use it within a single byte (to ensure we get endianness right).
  51  * We can use memset for the rest of the bitmap as there are no other users.
  52  */
  53 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  54 {
  55         int i;
  56
  57         if (start_bit >= end_bit)
  58                 return;
  59
  60         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  61         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  62                 ext4_set_bit(i, bitmap);
  63         if (i < end_bit)
  64                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  65 }
  66
  67 /* Initializes an uninitialized inode bitmap */
  68 static unsigned ext4_init_inode_bitmap(struct super_block *sb,
  69                                        struct buffer_head *bh,
  70                                        ext4_group_t block_group,
  71                                        struct ext4_group_desc *gdp)
  72 {
  73         struct ext4_sb_info *sbi = EXT4_SB(sb);
  74
  75         J_ASSERT_BH(bh, buffer_locked(bh));
  76
  77         /* If checksum is bad mark all blocks and inodes use to prevent
  78          * allocation, essentially implementing a per-group read-only flag. */
  79         if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
  80                 ext4_error(sb, "Checksum bad for group %u", block_group);
  81                 ext4_free_blks_set(sb, gdp, 0);
  82                 ext4_free_inodes_set(sb, gdp, 0);
  83                 ext4_itable_unused_set(sb, gdp, 0);
  84                 memset(bh->b_data, 0xff, sb->s_blocksize);
  85                 return 0;
  86         }
  87
  88         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  89         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  90                         bh->b_data);
  91
  92         return EXT4_INODES_PER_GROUP(sb);
  93 }
  94
  95 /*
  96  * Read the inode allocation bitmap for a given block_group, reading
  97  * into the specified slot in the superblock's bitmap cache.
  98  *
  99  * Return buffer_head of bitmap on success or NULL.
 100  */
 101 static struct buffer_head *
 102 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 103 {
 104         struct ext4_group_desc *desc;
 105         struct buffer_head *bh = NULL;
 106         ext4_fsblk_t bitmap_blk;
 107
 108         desc = ext4_get_group_desc(sb, block_group, NULL);
 109         if (!desc)
 110                 return NULL;
 111
 112         bitmap_blk = ext4_inode_bitmap(sb, desc);
 113         bh = sb_getblk(sb, bitmap_blk);
 114         if (unlikely(!bh)) {
 115                 ext4_error(sb, "Cannot read inode bitmap - "
 116                             "block_group = %u, inode_bitmap = %llu",
 117                             block_group, bitmap_blk);
 118                 return NULL;
 119         }
 120         if (bitmap_uptodate(bh))
 121                 return bh;
 122
 123         lock_buffer(bh);
 124         if (bitmap_uptodate(bh)) {
 125                 unlock_buffer(bh);
 126                 return bh;
 127         }
 128
 129         ext4_lock_group(sb, block_group);
 130         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 131                 ext4_init_inode_bitmap(sb, bh, block_group, desc);
 132                 set_bitmap_uptodate(bh);
 133                 set_buffer_uptodate(bh);
 134                 ext4_unlock_group(sb, block_group);
 135                 unlock_buffer(bh);
 136                 return bh;
 137         }
 138         ext4_unlock_group(sb, block_group);
 139
 140         if (buffer_uptodate(bh)) {
 141                 /*
 142                  * if not uninit if bh is uptodate,
 143                  * bitmap is also uptodate
 144                  */
 145                 set_bitmap_uptodate(bh);
 146                 unlock_buffer(bh);
 147                 return bh;
 148         }
 149         /*
 150          * submit the buffer_head for read. We can
 151          * safely mark the bitmap as uptodate now.
 152          * We do it here so the bitmap uptodate bit
 153          * get set with buffer lock held.
 154          */
 155         set_bitmap_uptodate(bh);
 156         if (bh_submit_read(bh) < 0) {
 157                 put_bh(bh);
 158                 ext4_error(sb, "Cannot read inode bitmap - "
 159                             "block_group = %u, inode_bitmap = %llu",
 160                             block_group, bitmap_blk);
 161                 return NULL;
 162         }
 163         return bh;
 164 }
 165
 166 /*
 167  * NOTE! When we get the inode, we're the only people
 168  * that have access to it, and as such there are no
 169  * race conditions we have to worry about. The inode
 170  * is not on the hash-lists, and it cannot be reached
 171  * through the filesystem because the directory entry
 172  * has been deleted earlier.
 173  *
 174  * HOWEVER: we must make sure that we get no aliases,
 175  * which means that we have to call "clear_inode()"
 176  * _before_ we mark the inode not in use in the inode
 177  * bitmaps. Otherwise a newly created file might use
 178  * the same inode number (not actually the same pointer
 179  * though), and then we'd have two inodes sharing the
 180  * same inode number and space on the harddisk.
 181  */
 182 void ext4_free_inode(handle_t *handle, struct inode *inode)
 183 {
 184         struct super_block *sb = inode->i_sb;
 185         int is_directory;
 186         unsigned long ino;
 187         struct buffer_head *bitmap_bh = NULL;
 188         struct buffer_head *bh2;
 189         ext4_group_t block_group;
 190         unsigned long bit;
 191         struct ext4_group_desc *gdp;
 192         struct ext4_super_block *es;
 193         struct ext4_sb_info *sbi;
 194         int fatal = 0, err, count, cleared;
 195
 196         if (atomic_read(&inode->i_count) > 1) {
 197                 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
 198                        atomic_read(&inode->i_count));
 199                 return;
 200         }
 201         if (inode->i_nlink) {
 202                 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
 203                        inode->i_nlink);
 204                 return;
 205         }
 206         if (!sb) {
 207                 printk(KERN_ERR "ext4_free_inode: inode on "
 208                        "nonexistent device\n");
 209                 return;
 210         }
 211         sbi = EXT4_SB(sb);
 212
 213         ino = inode->i_ino;
 214         ext4_debug("freeing inode %lu\n", ino);
 215         trace_ext4_free_inode(inode);
 216
 217         /*
 218          * Note: we must free any quota before locking the superblock,
 219          * as writing the quota to disk may need the lock as well.
 220          */
 221         dquot_initialize(inode);
 222         ext4_xattr_delete_inode(handle, inode);
 223         dquot_free_inode(inode);
 224         dquot_drop(inode);
 225
 226         is_directory = S_ISDIR(inode->i_mode);
 227
 228         /* Do this BEFORE marking the inode not in use or returning an error */
 229         ext4_clear_inode(inode);
 230
 231         es = EXT4_SB(sb)->s_es;
 232         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 233                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 234                 goto error_return;
 235         }
 236         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 237         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 238         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 239         if (!bitmap_bh)
 240                 goto error_return;
 241
 242         BUFFER_TRACE(bitmap_bh, "get_write_access");
 243         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 244         if (fatal)
 245                 goto error_return;
 246
 247         fatal = -ESRCH;
 248         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 249         if (gdp) {
 250                 BUFFER_TRACE(bh2, "get_write_access");
 251                 fatal = ext4_journal_get_write_access(handle, bh2);
 252         }
 253         ext4_lock_group(sb, block_group);
 254         cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
 255         if (fatal || !cleared) {
 256                 ext4_unlock_group(sb, block_group);
 257                 goto out;
 258         }
 259
 260         count = ext4_free_inodes_count(sb, gdp) + 1;
 261         ext4_free_inodes_set(sb, gdp, count);
 262         if (is_directory) {
 263                 count = ext4_used_dirs_count(sb, gdp) - 1;
 264                 ext4_used_dirs_set(sb, gdp, count);
 265                 percpu_counter_dec(&sbi->s_dirs_counter);
 266         }
 267         gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
 268         ext4_unlock_group(sb, block_group);
 269
 270         percpu_counter_inc(&sbi->s_freeinodes_counter);
 271         if (sbi->s_log_groups_per_flex) {
 272                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 273
 274                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 275                 if (is_directory)
 276                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 277         }
 278         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 279         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 280 out:
 281         if (cleared) {
 282                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 283                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 284                 if (!fatal)
 285                         fatal = err;
 286                 ext4_mark_super_dirty(sb);
 287         } else
 288                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 289
 290 error_return:
 291         brelse(bitmap_bh);
 292         ext4_std_error(sb, fatal);
 293 }
 294
 295 /*
 296  * There are two policies for allocating an inode.  If the new inode is
 297  * a directory, then a forward search is made for a block group with both
 298  * free space and a low directory-to-inode ratio; if that fails, then of
 299  * the groups with above-average free space, that group with the fewest
 300  * directories already is chosen.
 301  *
 302  * For other inodes, search forward from the parent directory\'s block
 303  * group to find a free inode.
 304  */
 305 static int find_group_dir(struct super_block *sb, struct inode *parent,
 306                                 ext4_group_t *best_group)
 307 {
 308         ext4_group_t ngroups = ext4_get_groups_count(sb);
 309         unsigned int freei, avefreei;
 310         struct ext4_group_desc *desc, *best_desc = NULL;
 311         ext4_group_t group;
 312         int ret = -1;
 313
 314         freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
 315         avefreei = freei / ngroups;
 316
 317         for (group = 0; group < ngroups; group++) {
 318                 desc = ext4_get_group_desc(sb, group, NULL);
 319                 if (!desc || !ext4_free_inodes_count(sb, desc))
 320                         continue;
 321                 if (ext4_free_inodes_count(sb, desc) < avefreei)
 322                         continue;
 323                 if (!best_desc ||
 324                     (ext4_free_blks_count(sb, desc) >
 325                      ext4_free_blks_count(sb, best_desc))) {
 326                         *best_group = group;
 327                         best_desc = desc;
 328                         ret = 0;
 329                 }
 330         }
 331         return ret;
 332 }
 333
 334 #define free_block_ratio 10
 335
 336 static int find_group_flex(struct super_block *sb, struct inode *parent,
 337                            ext4_group_t *best_group)
 338 {
 339         struct ext4_sb_info *sbi = EXT4_SB(sb);
 340         struct ext4_group_desc *desc;
 341         struct flex_groups *flex_group = sbi->s_flex_groups;
 342         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 343         ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
 344         ext4_group_t ngroups = ext4_get_groups_count(sb);
 345         int flex_size = ext4_flex_bg_size(sbi);
 346         ext4_group_t best_flex = parent_fbg_group;
 347         int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
 348         int flexbg_free_blocks;
 349         int flex_freeb_ratio;
 350         ext4_group_t n_fbg_groups;
 351         ext4_group_t i;
 352
 353         n_fbg_groups = (ngroups + flex_size - 1) >>
 354                 sbi->s_log_groups_per_flex;
 355
 356 find_close_to_parent:
 357         flexbg_free_blocks = atomic_read(&flex_group[best_flex].free_blocks);
 358         flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 359         if (atomic_read(&flex_group[best_flex].free_inodes) &&
 360             flex_freeb_ratio > free_block_ratio)
 361                 goto found_flexbg;
 362
 363         if (best_flex && best_flex == parent_fbg_group) {
 364                 best_flex--;
 365                 goto find_close_to_parent;
 366         }
 367
 368         for (i = 0; i < n_fbg_groups; i++) {
 369                 if (i == parent_fbg_group || i == parent_fbg_group - 1)
 370                         continue;
 371
 372                 flexbg_free_blocks = atomic_read(&flex_group[i].free_blocks);
 373                 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 374
 375                 if (flex_freeb_ratio > free_block_ratio &&
 376                     (atomic_read(&flex_group[i].free_inodes))) {
 377                         best_flex = i;
 378                         goto found_flexbg;
 379                 }
 380
 381                 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
 382                     ((atomic_read(&flex_group[i].free_blocks) >
 383                       atomic_read(&flex_group[best_flex].free_blocks)) &&
 384                      atomic_read(&flex_group[i].free_inodes)))
 385                         best_flex = i;
 386         }
 387
 388         if (!atomic_read(&flex_group[best_flex].free_inodes) ||
 389             !atomic_read(&flex_group[best_flex].free_blocks))
 390                 return -1;
 391
 392 found_flexbg:
 393         for (i = best_flex * flex_size; i < ngroups &&
 394                      i < (best_flex + 1) * flex_size; i++) {
 395                 desc = ext4_get_group_desc(sb, i, NULL);
 396                 if (ext4_free_inodes_count(sb, desc)) {
 397                         *best_group = i;
 398                         goto out;
 399                 }
 400         }
 401
 402         return -1;
 403 out:
 404         return 0;
 405 }
 406
 407 struct orlov_stats {
 408         __u32 free_inodes;
 409         __u32 free_blocks;
 410         __u32 used_dirs;
 411 };
 412
 413 /*
 414  * Helper function for Orlov's allocator; returns critical information
 415  * for a particular block group or flex_bg.  If flex_size is 1, then g
 416  * is a block group number; otherwise it is flex_bg number.
 417  */
 418 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 419                             int flex_size, struct orlov_stats *stats)
 420 {
 421         struct ext4_group_desc *desc;
 422         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 423
 424         if (flex_size > 1) {
 425                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 426                 stats->free_blocks = atomic_read(&flex_group[g].free_blocks);
 427                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 428                 return;
 429         }
 430
 431         desc = ext4_get_group_desc(sb, g, NULL);
 432         if (desc) {
 433                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 434                 stats->free_blocks = ext4_free_blks_count(sb, desc);
 435                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 436         } else {
 437                 stats->free_inodes = 0;
 438                 stats->free_blocks = 0;
 439                 stats->used_dirs = 0;
 440         }
 441 }
 442
 443 /*
 444  * Orlov's allocator for directories.
 445  *
 446  * We always try to spread first-level directories.
 447  *
 448  * If there are blockgroups with both free inodes and free blocks counts
 449  * not worse than average we return one with smallest directory count.
 450  * Otherwise we simply return a random group.
 451  *
 452  * For the rest rules look so:
 453  *
 454  * It's OK to put directory into a group unless
 455  * it has too many directories already (max_dirs) or
 456  * it has too few free inodes left (min_inodes) or
 457  * it has too few free blocks left (min_blocks) or
 458  * Parent's group is preferred, if it doesn't satisfy these
 459  * conditions we search cyclically through the rest. If none
 460  * of the groups look good we just look for a group with more
 461  * free inodes than average (starting at parent's group).
 462  */
 463
 464 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 465                             ext4_group_t *group, int mode,
 466                             const struct qstr *qstr)
 467 {
 468         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 469         struct ext4_sb_info *sbi = EXT4_SB(sb);
 470         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 471         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 472         unsigned int freei, avefreei;
 473         ext4_fsblk_t freeb, avefreeb;
 474         unsigned int ndirs;
 475         int max_dirs, min_inodes;
 476         ext4_grpblk_t min_blocks;
 477         ext4_group_t i, grp, g, ngroups;
 478         struct ext4_group_desc *desc;
 479         struct orlov_stats stats;
 480         int flex_size = ext4_flex_bg_size(sbi);
 481         struct dx_hash_info hinfo;
 482
 483         ngroups = real_ngroups;
 484         if (flex_size > 1) {
 485                 ngroups = (real_ngroups + flex_size - 1) >>
 486                         sbi->s_log_groups_per_flex;
 487                 parent_group >>= sbi->s_log_groups_per_flex;
 488         }
 489
 490         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 491         avefreei = freei / ngroups;
 492         freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
 493         avefreeb = freeb;
 494         do_div(avefreeb, ngroups);
 495         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 496
 497         if (S_ISDIR(mode) &&
 498             ((parent == sb->s_root->d_inode) ||
 499              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 500                 int best_ndir = inodes_per_group;
 501                 int ret = -1;
 502
 503                 if (qstr) {
 504                         hinfo.hash_version = DX_HASH_HALF_MD4;
 505                         hinfo.seed = sbi->s_hash_seed;
 506                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 507                         grp = hinfo.hash;
 508                 } else
 509                         get_random_bytes(&grp, sizeof(grp));
 510                 parent_group = (unsigned)grp % ngroups;
 511                 for (i = 0; i < ngroups; i++) {
 512                         g = (parent_group + i) % ngroups;
 513                         get_orlov_stats(sb, g, flex_size, &stats);
 514                         if (!stats.free_inodes)
 515                                 continue;
 516                         if (stats.used_dirs >= best_ndir)
 517                                 continue;
 518                         if (stats.free_inodes < avefreei)
 519                                 continue;
 520                         if (stats.free_blocks < avefreeb)
 521                                 continue;
 522                         grp = g;
 523                         ret = 0;
 524                         best_ndir = stats.used_dirs;
 525                 }
 526                 if (ret)
 527                         goto fallback;
 528         found_flex_bg:
 529                 if (flex_size == 1) {
 530                         *group = grp;
 531                         return 0;
 532                 }
 533
 534                 /*
 535                  * We pack inodes at the beginning of the flexgroup's
 536                  * inode tables.  Block allocation decisions will do
 537                  * something similar, although regular files will
 538                  * start at 2nd block group of the flexgroup.  See
 539                  * ext4_ext_find_goal() and ext4_find_near().
 540                  */
 541                 grp *= flex_size;
 542                 for (i = 0; i < flex_size; i++) {
 543                         if (grp+i >= real_ngroups)
 544                                 break;
 545                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 546                         if (desc && ext4_free_inodes_count(sb, desc)) {
 547                                 *group = grp+i;
 548                                 return 0;
 549                         }
 550                 }
 551                 goto fallback;
 552         }
 553
 554         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 555         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 556         if (min_inodes < 1)
 557                 min_inodes = 1;
 558         min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb)*flex_size / 4;
 559
 560         /*
 561          * Start looking in the flex group where we last allocated an
 562          * inode for this parent directory
 563          */
 564         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 565                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 566                 if (flex_size > 1)
 567                         parent_group >>= sbi->s_log_groups_per_flex;
 568         }
 569
 570         for (i = 0; i < ngroups; i++) {
 571                 grp = (parent_group + i) % ngroups;
 572                 get_orlov_stats(sb, grp, flex_size, &stats);
 573                 if (stats.used_dirs >= max_dirs)
 574                         continue;
 575                 if (stats.free_inodes < min_inodes)
 576                         continue;
 577                 if (stats.free_blocks < min_blocks)
 578                         continue;
 579                 goto found_flex_bg;
 580         }
 581
 582 fallback:
 583         ngroups = real_ngroups;
 584         avefreei = freei / ngroups;
 585 fallback_retry:
 586         parent_group = EXT4_I(parent)->i_block_group;
 587         for (i = 0; i < ngroups; i++) {
 588                 grp = (parent_group + i) % ngroups;
 589                 desc = ext4_get_group_desc(sb, grp, NULL);
 590                 if (desc && ext4_free_inodes_count(sb, desc) &&
 591                     ext4_free_inodes_count(sb, desc) >= avefreei) {
 592                         *group = grp;
 593                         return 0;
 594                 }
 595         }
 596
 597         if (avefreei) {
 598                 /*
 599                  * The free-inodes counter is approximate, and for really small
 600                  * filesystems the above test can fail to find any blockgroups
 601                  */
 602                 avefreei = 0;
 603                 goto fallback_retry;
 604         }
 605
 606         return -1;
 607 }
 608
 609 static int find_group_other(struct super_block *sb, struct inode *parent,
 610                             ext4_group_t *group, int mode)
 611 {
 612         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 613         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 614         struct ext4_group_desc *desc;
 615         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 616
 617         /*
 618          * Try to place the inode is the same flex group as its
 619          * parent.  If we can't find space, use the Orlov algorithm to
 620          * find another flex group, and store that information in the
 621          * parent directory's inode information so that use that flex
 622          * group for future allocations.
 623          */
 624         if (flex_size > 1) {
 625                 int retry = 0;
 626
 627         try_again:
 628                 parent_group &= ~(flex_size-1);
 629                 last = parent_group + flex_size;
 630                 if (last > ngroups)
 631                         last = ngroups;
 632                 for  (i = parent_group; i < last; i++) {
 633                         desc = ext4_get_group_desc(sb, i, NULL);
 634                         if (desc && ext4_free_inodes_count(sb, desc)) {
 635                                 *group = i;
 636                                 return 0;
 637                         }
 638                 }
 639                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 640                         retry = 1;
 641                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 642                         goto try_again;
 643                 }
 644                 /*
 645                  * If this didn't work, use the Orlov search algorithm
 646                  * to find a new flex group; we pass in the mode to
 647                  * avoid the topdir algorithms.
 648                  */
 649                 *group = parent_group + flex_size;
 650                 if (*group > ngroups)
 651                         *group = 0;
 652                 return find_group_orlov(sb, parent, group, mode, 0);
 653         }
 654
 655         /*
 656          * Try to place the inode in its parent directory
 657          */
 658         *group = parent_group;
 659         desc = ext4_get_group_desc(sb, *group, NULL);
 660         if (desc && ext4_free_inodes_count(sb, desc) &&
 661                         ext4_free_blks_count(sb, desc))
 662                 return 0;
 663
 664         /*
 665          * We're going to place this inode in a different blockgroup from its
 666          * parent.  We want to cause files in a common directory to all land in
 667          * the same blockgroup.  But we want files which are in a different
 668          * directory which shares a blockgroup with our parent to land in a
 669          * different blockgroup.
 670          *
 671          * So add our directory's i_ino into the starting point for the hash.
 672          */
 673         *group = (*group + parent->i_ino) % ngroups;
 674
 675         /*
 676          * Use a quadratic hash to find a group with a free inode and some free
 677          * blocks.
 678          */
 679         for (i = 1; i < ngroups; i <<= 1) {
 680                 *group += i;
 681                 if (*group >= ngroups)
 682                         *group -= ngroups;
 683                 desc = ext4_get_group_desc(sb, *group, NULL);
 684                 if (desc && ext4_free_inodes_count(sb, desc) &&
 685                                 ext4_free_blks_count(sb, desc))
 686                         return 0;
 687         }
 688
 689         /*
 690          * That failed: try linear search for a free inode, even if that group
 691          * has no free blocks.
 692          */
 693         *group = parent_group;
 694         for (i = 0; i < ngroups; i++) {
 695                 if (++*group >= ngroups)
 696                         *group = 0;
 697                 desc = ext4_get_group_desc(sb, *group, NULL);
 698                 if (desc && ext4_free_inodes_count(sb, desc))
 699                         return 0;
 700         }
 701
 702         return -1;
 703 }
 704
 705 /*
 706  * claim the inode from the inode bitmap. If the group
 707  * is uninit we need to take the groups's ext4_group_lock
 708  * and clear the uninit flag. The inode bitmap update
 709  * and group desc uninit flag clear should be done
 710  * after holding ext4_group_lock so that ext4_read_inode_bitmap
 711  * doesn't race with the ext4_claim_inode
 712  */
 713 static int ext4_claim_inode(struct super_block *sb,
 714                         struct buffer_head *inode_bitmap_bh,
 715                         unsigned long ino, ext4_group_t group, int mode)
 716 {
 717         int free = 0, retval = 0, count;
 718         struct ext4_sb_info *sbi = EXT4_SB(sb);
 719         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 720         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 721
 722         /*
 723          * We have to be sure that new inode allocation does not race with
 724          * inode table initialization, because otherwise we may end up
 725          * allocating and writing new inode right before sb_issue_zeroout
 726          * takes place and overwriting our new inode with zeroes. So we
 727          * take alloc_sem to prevent it.
 728          */
 729         down_read(&grp->alloc_sem);
 730         ext4_lock_group(sb, group);
 731         if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
 732                 /* not a free inode */
 733                 retval = 1;
 734                 goto err_ret;
 735         }
 736         ino++;
 737         if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
 738                         ino > EXT4_INODES_PER_GROUP(sb)) {
 739                 ext4_unlock_group(sb, group);
 740                 up_read(&grp->alloc_sem);
 741                 ext4_error(sb, "reserved inode or inode > inodes count - "
 742                            "block_group = %u, inode=%lu", group,
 743                            ino + group * EXT4_INODES_PER_GROUP(sb));
 744                 return 1;
 745         }
 746         /* If we didn't allocate from within the initialized part of the inode
 747          * table then we need to initialize up to this inode. */
 748         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
 749
 750                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 751                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 752                         /* When marking the block group with
 753                          * ~EXT4_BG_INODE_UNINIT we don't want to depend
 754                          * on the value of bg_itable_unused even though
 755                          * mke2fs could have initialized the same for us.
 756                          * Instead we calculated the value below
 757                          */
 758
 759                         free = 0;
 760                 } else {
 761                         free = EXT4_INODES_PER_GROUP(sb) -
 762                                 ext4_itable_unused_count(sb, gdp);
 763                 }
 764
 765                 /*
 766                  * Check the relative inode number against the last used
 767                  * relative inode number in this group. if it is greater
 768                  * we need to  update the bg_itable_unused count
 769                  *
 770                  */
 771                 if (ino > free)
 772                         ext4_itable_unused_set(sb, gdp,
 773                                         (EXT4_INODES_PER_GROUP(sb) - ino));
 774         }
 775         count = ext4_free_inodes_count(sb, gdp) - 1;
 776         ext4_free_inodes_set(sb, gdp, count);
 777         if (S_ISDIR(mode)) {
 778                 count = ext4_used_dirs_count(sb, gdp) + 1;
 779                 ext4_used_dirs_set(sb, gdp, count);
 780                 if (sbi->s_log_groups_per_flex) {
 781                         ext4_group_t f = ext4_flex_group(sbi, group);
 782
 783                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
 784                 }
 785         }
 786         gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
 787 err_ret:
 788         ext4_unlock_group(sb, group);
 789         up_read(&grp->alloc_sem);
 790         return retval;
 791 }
 792
 793 /*
 794  * There are two policies for allocating an inode.  If the new inode is
 795  * a directory, then a forward search is made for a block group with both
 796  * free space and a low directory-to-inode ratio; if that fails, then of
 797  * the groups with above-average free space, that group with the fewest
 798  * directories already is chosen.
 799  *
 800  * For other inodes, search forward from the parent directory's block
 801  * group to find a free inode.
 802  */
 803 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
 804                              const struct qstr *qstr, __u32 goal)
 805 {
 806         struct super_block *sb;
 807         struct buffer_head *inode_bitmap_bh = NULL;
 808         struct buffer_head *group_desc_bh;
 809         ext4_group_t ngroups, group = 0;
 810         unsigned long ino = 0;
 811         struct inode *inode;
 812         struct ext4_group_desc *gdp = NULL;
 813         struct ext4_inode_info *ei;
 814         struct ext4_sb_info *sbi;
 815         int ret2, err = 0;
 816         struct inode *ret;
 817         ext4_group_t i;
 818         int free = 0;
 819         static int once = 1;
 820         ext4_group_t flex_group;
 821
 822         /* Cannot create files in a deleted directory */
 823         if (!dir || !dir->i_nlink)
 824                 return ERR_PTR(-EPERM);
 825
 826         sb = dir->i_sb;
 827         ngroups = ext4_get_groups_count(sb);
 828         trace_ext4_request_inode(dir, mode);
 829         inode = new_inode(sb);
 830         if (!inode)
 831                 return ERR_PTR(-ENOMEM);
 832         ei = EXT4_I(inode);
 833         sbi = EXT4_SB(sb);
 834
 835         if (!goal)
 836                 goal = sbi->s_inode_goal;
 837
 838         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 839                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 840                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 841                 ret2 = 0;
 842                 goto got_group;
 843         }
 844
 845         if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
 846                 ret2 = find_group_flex(sb, dir, &group);
 847                 if (ret2 == -1) {
 848                         ret2 = find_group_other(sb, dir, &group, mode);
 849                         if (ret2 == 0 && once) {
 850                                 once = 0;
 851                                 printk(KERN_NOTICE "ext4: find_group_flex "
 852                                        "failed, fallback succeeded dir %lu\n",
 853                                        dir->i_ino);
 854                         }
 855                 }
 856                 goto got_group;
 857         }
 858
 859         if (S_ISDIR(mode)) {
 860                 if (test_opt(sb, OLDALLOC))
 861                         ret2 = find_group_dir(sb, dir, &group);
 862                 else
 863                         ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 864         } else
 865                 ret2 = find_group_other(sb, dir, &group, mode);
 866
 867 got_group:
 868         EXT4_I(dir)->i_last_alloc_group = group;
 869         err = -ENOSPC;
 870         if (ret2 == -1)
 871                 goto out;
 872
 873         for (i = 0; i < ngroups; i++, ino = 0) {
 874                 err = -EIO;
 875
 876                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 877                 if (!gdp)
 878                         goto fail;
 879
 880                 brelse(inode_bitmap_bh);
 881                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 882                 if (!inode_bitmap_bh)
 883                         goto fail;
 884
 885 repeat_in_this_group:
 886                 ino = ext4_find_next_zero_bit((unsigned long *)
 887                                               inode_bitmap_bh->b_data,
 888                                               EXT4_INODES_PER_GROUP(sb), ino);
 889
 890                 if (ino < EXT4_INODES_PER_GROUP(sb)) {
 891
 892                         BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 893                         err = ext4_journal_get_write_access(handle,
 894                                                             inode_bitmap_bh);
 895                         if (err)
 896                                 goto fail;
 897
 898                         BUFFER_TRACE(group_desc_bh, "get_write_access");
 899                         err = ext4_journal_get_write_access(handle,
 900                                                                 group_desc_bh);
 901                         if (err)
 902                                 goto fail;
 903                         if (!ext4_claim_inode(sb, inode_bitmap_bh,
 904                                                 ino, group, mode)) {
 905                                 /* we won it */
 906                                 BUFFER_TRACE(inode_bitmap_bh,
 907                                         "call ext4_handle_dirty_metadata");
 908                                 err = ext4_handle_dirty_metadata(handle,
 909                                                                  NULL,
 910                                                         inode_bitmap_bh);
 911                                 if (err)
 912                                         goto fail;
 913                                 /* zero bit is inode number 1*/
 914                                 ino++;
 915                                 goto got;
 916                         }
 917                         /* we lost it */
 918                         ext4_handle_release_buffer(handle, inode_bitmap_bh);
 919                         ext4_handle_release_buffer(handle, group_desc_bh);
 920
 921                         if (++ino < EXT4_INODES_PER_GROUP(sb))
 922                                 goto repeat_in_this_group;
 923                 }
 924
 925                 /*
 926                  * This case is possible in concurrent environment.  It is very
 927                  * rare.  We cannot repeat the find_group_xxx() call because
 928                  * that will simply return the same blockgroup, because the
 929                  * group descriptor metadata has not yet been updated.
 930                  * So we just go onto the next blockgroup.
 931                  */
 932                 if (++group == ngroups)
 933                         group = 0;
 934         }
 935         err = -ENOSPC;
 936         goto out;
 937
 938 got:
 939         /* We may have to initialize the block bitmap if it isn't already */
 940         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
 941             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 942                 struct buffer_head *block_bitmap_bh;
 943
 944                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 945                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 946                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 947                 if (err) {
 948                         brelse(block_bitmap_bh);
 949                         goto fail;
 950                 }
 951
 952                 free = 0;
 953                 ext4_lock_group(sb, group);
 954                 /* recheck and clear flag under lock if we still need to */
 955                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 956                         free = ext4_free_blocks_after_init(sb, group, gdp);
 957                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 958                         ext4_free_blks_set(sb, gdp, free);
 959                         gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
 960                                                                 gdp);
 961                 }
 962                 ext4_unlock_group(sb, group);
 963
 964                 /* Don't need to dirty bitmap block if we didn't change it */
 965                 if (free) {
 966                         BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 967                         err = ext4_handle_dirty_metadata(handle,
 968                                                         NULL, block_bitmap_bh);
 969                 }
 970
 971                 brelse(block_bitmap_bh);
 972                 if (err)
 973                         goto fail;
 974         }
 975         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 976         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 977         if (err)
 978                 goto fail;
 979
 980         percpu_counter_dec(&sbi->s_freeinodes_counter);
 981         if (S_ISDIR(mode))
 982                 percpu_counter_inc(&sbi->s_dirs_counter);
 983         ext4_mark_super_dirty(sb);
 984
 985         if (sbi->s_log_groups_per_flex) {
 986                 flex_group = ext4_flex_group(sbi, group);
 987                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
 988         }
 989
 990         if (test_opt(sb, GRPID)) {
 991                 inode->i_mode = mode;
 992                 inode->i_uid = current_fsuid();
 993                 inode->i_gid = dir->i_gid;
 994         } else
 995                 inode_init_owner(inode, dir, mode);
 996
 997         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
 998         /* This is the optimal IO size (for stat), not the fs block size */
 999         inode->i_blocks = 0;
1000         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1001                                                        ext4_current_time(inode);
1002
1003         memset(ei->i_data, 0, sizeof(ei->i_data));
1004         ei->i_dir_start_lookup = 0;
1005         ei->i_disksize = 0;
1006
1007         /*
1008          * Don't inherit extent flag from directory, amongst others. We set
1009          * extent flag on newly created directory and file only if -o extent
1010          * mount option is specified
1011          */
1012         ei->i_flags =
1013                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1014         ei->i_file_acl = 0;
1015         ei->i_dtime = 0;
1016         ei->i_block_group = group;
1017         ei->i_last_alloc_group = ~0;
1018
1019         ext4_set_inode_flags(inode);
1020         if (IS_DIRSYNC(inode))
1021                 ext4_handle_sync(handle);
1022         if (insert_inode_locked(inode) < 0) {
1023                 err = -EINVAL;
1024                 goto fail_drop;
1025         }
1026         spin_lock(&sbi->s_next_gen_lock);
1027         inode->i_generation = sbi->s_next_generation++;
1028         spin_unlock(&sbi->s_next_gen_lock);
1029
1030         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1031         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1032
1033         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1034
1035         ret = inode;
1036         dquot_initialize(inode);
1037         err = dquot_alloc_inode(inode);
1038         if (err)
1039                 goto fail_drop;
1040
1041         err = ext4_init_acl(handle, inode, dir);
1042         if (err)
1043                 goto fail_free_drop;
1044
1045         err = ext4_init_security(handle, inode, dir);
1046         if (err)
1047                 goto fail_free_drop;
1048
1049         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1050                 /* set extent flag only for directory, file and normal symlink*/
1051                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1052                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1053                         ext4_ext_tree_init(handle, inode);
1054                 }
1055         }
1056
1057         err = ext4_mark_inode_dirty(handle, inode);
1058         if (err) {
1059                 ext4_std_error(sb, err);
1060                 goto fail_free_drop;
1061         }
1062
1063         ext4_debug("allocating inode %lu\n", inode->i_ino);
1064         trace_ext4_allocate_inode(inode, dir, mode);
1065         goto really_out;
1066 fail:
1067         ext4_std_error(sb, err);
1068 out:
1069         iput(inode);
1070         ret = ERR_PTR(err);
1071 really_out:
1072         brelse(inode_bitmap_bh);
1073         return ret;
1074
1075 fail_free_drop:
1076         dquot_free_inode(inode);
1077
1078 fail_drop:
1079         dquot_drop(inode);
1080         inode->i_flags |= S_NOQUOTA;
1081         inode->i_nlink = 0;
1082         unlock_new_inode(inode);
1083         iput(inode);
1084         brelse(inode_bitmap_bh);
1085         return ERR_PTR(err);
1086 }
1087
1088 /* Verify that we are loading a valid orphan from disk */
1089 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1090 {
1091         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1092         ext4_group_t block_group;
1093         int bit;
1094         struct buffer_head *bitmap_bh;
1095         struct inode *inode = NULL;
1096         long err = -EIO;
1097
1098         /* Error cases - e2fsck has already cleaned up for us */
1099         if (ino > max_ino) {
1100                 ext4_warning(sb, "bad orphan ino %lu!  e2fsck was run?", ino);
1101                 goto error;
1102         }
1103
1104         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1105         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1106         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1107         if (!bitmap_bh) {
1108                 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1109                 goto error;
1110         }
1111
1112         /* Having the inode bit set should be a 100% indicator that this
1113          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1114          * inodes that were being truncated, so we can't check i_nlink==0.
1115          */
1116         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1117                 goto bad_orphan;
1118
1119         inode = ext4_iget(sb, ino);
1120         if (IS_ERR(inode))
1121                 goto iget_failed;
1122
1123         /*
1124          * If the orphans has i_nlinks > 0 then it should be able to be
1125          * truncated, otherwise it won't be removed from the orphan list
1126          * during processing and an infinite loop will result.
1127          */
1128         if (inode->i_nlink && !ext4_can_truncate(inode))
1129                 goto bad_orphan;
1130
1131         if (NEXT_ORPHAN(inode) > max_ino)
1132                 goto bad_orphan;
1133         brelse(bitmap_bh);
1134         return inode;
1135
1136 iget_failed:
1137         err = PTR_ERR(inode);
1138         inode = NULL;
1139 bad_orphan:
1140         ext4_warning(sb, "bad orphan inode %lu!  e2fsck was run?", ino);
1141         printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1142                bit, (unsigned long long)bitmap_bh->b_blocknr,
1143                ext4_test_bit(bit, bitmap_bh->b_data));
1144         printk(KERN_NOTICE "inode=%p\n", inode);
1145         if (inode) {
1146                 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1147                        is_bad_inode(inode));
1148                 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1149                        NEXT_ORPHAN(inode));
1150                 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1151                 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1152                 /* Avoid freeing blocks if we got a bad deleted inode */
1153                 if (inode->i_nlink == 0)
1154                         inode->i_blocks = 0;
1155                 iput(inode);
1156         }
1157         brelse(bitmap_bh);
1158 error:
1159         return ERR_PTR(err);
1160 }
1161
1162 unsigned long ext4_count_free_inodes(struct super_block *sb)
1163 {
1164         unsigned long desc_count;
1165         struct ext4_group_desc *gdp;
1166         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1167 #ifdef EXT4FS_DEBUG
1168         struct ext4_super_block *es;
1169         unsigned long bitmap_count, x;
1170         struct buffer_head *bitmap_bh = NULL;
1171
1172         es = EXT4_SB(sb)->s_es;
1173         desc_count = 0;
1174         bitmap_count = 0;
1175         gdp = NULL;
1176         for (i = 0; i < ngroups; i++) {
1177                 gdp = ext4_get_group_desc(sb, i, NULL);
1178                 if (!gdp)
1179                         continue;
1180                 desc_count += ext4_free_inodes_count(sb, gdp);
1181                 brelse(bitmap_bh);
1182                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1183                 if (!bitmap_bh)
1184                         continue;
1185
1186                 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1187                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1188                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1189                 bitmap_count += x;
1190         }
1191         brelse(bitmap_bh);
1192         printk(KERN_DEBUG "ext4_count_free_inodes: "
1193                "stored = %u, computed = %lu, %lu\n",
1194                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1195         return desc_count;
1196 #else
1197         desc_count = 0;
1198         for (i = 0; i < ngroups; i++) {
1199                 gdp = ext4_get_group_desc(sb, i, NULL);
1200                 if (!gdp)
1201                         continue;
1202                 desc_count += ext4_free_inodes_count(sb, gdp);
1203                 cond_resched();
1204         }
1205         return desc_count;
1206 #endif
1207 }
1208
1209 /* Called at mount-time, super-block is locked */
1210 unsigned long ext4_count_dirs(struct super_block * sb)
1211 {
1212         unsigned long count = 0;
1213         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1214
1215         for (i = 0; i < ngroups; i++) {
1216                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1217                 if (!gdp)
1218                         continue;
1219                 count += ext4_used_dirs_count(sb, gdp);
1220         }
1221         return count;
1222 }
1223
1224 /*
1225  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1226  * inode table. Must be called without any spinlock held. The only place
1227  * where it is called from on active part of filesystem is ext4lazyinit
1228  * thread, so we do not need any special locks, however we have to prevent
1229  * inode allocation from the current group, so we take alloc_sem lock, to
1230  * block ext4_claim_inode until we are finished.
1231  */
1232 extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1233                                  int barrier)
1234 {
1235         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1236         struct ext4_sb_info *sbi = EXT4_SB(sb);
1237         struct ext4_group_desc *gdp = NULL;
1238         struct buffer_head *group_desc_bh;
1239         handle_t *handle;
1240         ext4_fsblk_t blk;
1241         int num, ret = 0, used_blks = 0;
1242
1243         /* This should not happen, but just to be sure check this */
1244         if (sb->s_flags & MS_RDONLY) {
1245                 ret = 1;
1246                 goto out;
1247         }
1248
1249         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1250         if (!gdp)
1251                 goto out;
1252
1253         /*
1254          * We do not need to lock this, because we are the only one
1255          * handling this flag.
1256          */
1257         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1258                 goto out;
1259
1260         handle = ext4_journal_start_sb(sb, 1);
1261         if (IS_ERR(handle)) {
1262                 ret = PTR_ERR(handle);
1263                 goto out;
1264         }
1265
1266         down_write(&grp->alloc_sem);
1267         /*
1268          * If inode bitmap was already initialized there may be some
1269          * used inodes so we need to skip blocks with used inodes in
1270          * inode table.
1271          */
1272         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1273                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1274                             ext4_itable_unused_count(sb, gdp)),
1275                             sbi->s_inodes_per_block);
1276
1277         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1278                 ext4_error(sb, "Something is wrong with group %u\n"
1279                            "Used itable blocks: %d"
1280                            "itable unused count: %u\n",
1281                            group, used_blks,
1282                            ext4_itable_unused_count(sb, gdp));
1283                 ret = 1;
1284                 goto out;
1285         }
1286
1287         blk = ext4_inode_table(sb, gdp) + used_blks;
1288         num = sbi->s_itb_per_group - used_blks;
1289
1290         BUFFER_TRACE(group_desc_bh, "get_write_access");
1291         ret = ext4_journal_get_write_access(handle,
1292                                             group_desc_bh);
1293         if (ret)
1294                 goto err_out;
1295
1296         /*
1297          * Skip zeroout if the inode table is full. But we set the ZEROED
1298          * flag anyway, because obviously, when it is full it does not need
1299          * further zeroing.
1300          */
1301         if (unlikely(num == 0))
1302                 goto skip_zeroout;
1303
1304         ext4_debug("going to zero out inode table in group %d\n",
1305                    group);
1306         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1307         if (ret < 0)
1308                 goto err_out;
1309         if (barrier)
1310                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1311
1312 skip_zeroout:
1313         ext4_lock_group(sb, group);
1314         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1315         gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1316         ext4_unlock_group(sb, group);
1317
1318         BUFFER_TRACE(group_desc_bh,
1319                      "call ext4_handle_dirty_metadata");
1320         ret = ext4_handle_dirty_metadata(handle, NULL,
1321                                          group_desc_bh);
1322
1323 err_out:
1324         up_write(&grp->alloc_sem);
1325         ext4_journal_stop(handle);
1326 out:
1327         return ret;
1328 }