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_group_clusters_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         trace_ext4_load_inode_bitmap(sb, block_group);
 156         set_bitmap_uptodate(bh);
 157         if (bh_submit_read(bh) < 0) {
 158                 put_bh(bh);
 159                 ext4_error(sb, "Cannot read inode bitmap - "
 160                             "block_group = %u, inode_bitmap = %llu",
 161                             block_group, bitmap_blk);
 162                 return NULL;
 163         }
 164         return bh;
 165 }
 166
 167 /*
 168  * NOTE! When we get the inode, we're the only people
 169  * that have access to it, and as such there are no
 170  * race conditions we have to worry about. The inode
 171  * is not on the hash-lists, and it cannot be reached
 172  * through the filesystem because the directory entry
 173  * has been deleted earlier.
 174  *
 175  * HOWEVER: we must make sure that we get no aliases,
 176  * which means that we have to call "clear_inode()"
 177  * _before_ we mark the inode not in use in the inode
 178  * bitmaps. Otherwise a newly created file might use
 179  * the same inode number (not actually the same pointer
 180  * though), and then we'd have two inodes sharing the
 181  * same inode number and space on the harddisk.
 182  */
 183 void ext4_free_inode(handle_t *handle, struct inode *inode)
 184 {
 185         struct super_block *sb = inode->i_sb;
 186         int is_directory;
 187         unsigned long ino;
 188         struct buffer_head *bitmap_bh = NULL;
 189         struct buffer_head *bh2;
 190         ext4_group_t block_group;
 191         unsigned long bit;
 192         struct ext4_group_desc *gdp;
 193         struct ext4_super_block *es;
 194         struct ext4_sb_info *sbi;
 195         int fatal = 0, err, count, cleared;
 196
 197         if (atomic_read(&inode->i_count) > 1) {
 198                 printk(KERN_ERR "ext4_free_inode: inode has count=%d\n",
 199                        atomic_read(&inode->i_count));
 200                 return;
 201         }
 202         if (inode->i_nlink) {
 203                 printk(KERN_ERR "ext4_free_inode: inode has nlink=%d\n",
 204                        inode->i_nlink);
 205                 return;
 206         }
 207         if (!sb) {
 208                 printk(KERN_ERR "ext4_free_inode: inode on "
 209                        "nonexistent device\n");
 210                 return;
 211         }
 212         sbi = EXT4_SB(sb);
 213
 214         ino = inode->i_ino;
 215         ext4_debug("freeing inode %lu\n", ino);
 216         trace_ext4_free_inode(inode);
 217
 218         /*
 219          * Note: we must free any quota before locking the superblock,
 220          * as writing the quota to disk may need the lock as well.
 221          */
 222         dquot_initialize(inode);
 223         ext4_xattr_delete_inode(handle, inode);
 224         dquot_free_inode(inode);
 225         dquot_drop(inode);
 226
 227         is_directory = S_ISDIR(inode->i_mode);
 228
 229         /* Do this BEFORE marking the inode not in use or returning an error */
 230         ext4_clear_inode(inode);
 231
 232         es = EXT4_SB(sb)->s_es;
 233         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 234                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 235                 goto error_return;
 236         }
 237         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 238         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 239         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 240         if (!bitmap_bh)
 241                 goto error_return;
 242
 243         BUFFER_TRACE(bitmap_bh, "get_write_access");
 244         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 245         if (fatal)
 246                 goto error_return;
 247
 248         fatal = -ESRCH;
 249         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 250         if (gdp) {
 251                 BUFFER_TRACE(bh2, "get_write_access");
 252                 fatal = ext4_journal_get_write_access(handle, bh2);
 253         }
 254         ext4_lock_group(sb, block_group);
 255         cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
 256         if (fatal || !cleared) {
 257                 ext4_unlock_group(sb, block_group);
 258                 goto out;
 259         }
 260
 261         count = ext4_free_inodes_count(sb, gdp) + 1;
 262         ext4_free_inodes_set(sb, gdp, count);
 263         if (is_directory) {
 264                 count = ext4_used_dirs_count(sb, gdp) - 1;
 265                 ext4_used_dirs_set(sb, gdp, count);
 266                 percpu_counter_dec(&sbi->s_dirs_counter);
 267         }
 268         gdp->bg_checksum = ext4_group_desc_csum(sbi, block_group, gdp);
 269         ext4_unlock_group(sb, block_group);
 270
 271         percpu_counter_inc(&sbi->s_freeinodes_counter);
 272         if (sbi->s_log_groups_per_flex) {
 273                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 274
 275                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 276                 if (is_directory)
 277                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 278         }
 279         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 280         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 281 out:
 282         if (cleared) {
 283                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 284                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 285                 if (!fatal)
 286                         fatal = err;
 287                 ext4_mark_super_dirty(sb);
 288         } else
 289                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 290
 291 error_return:
 292         brelse(bitmap_bh);
 293         ext4_std_error(sb, fatal);
 294 }
 295
 296 /*
 297  * There are two policies for allocating an inode.  If the new inode is
 298  * a directory, then a forward search is made for a block group with both
 299  * free space and a low directory-to-inode ratio; if that fails, then of
 300  * the groups with above-average free space, that group with the fewest
 301  * directories already is chosen.
 302  *
 303  * For other inodes, search forward from the parent directory\'s block
 304  * group to find a free inode.
 305  */
 306 static int find_group_dir(struct super_block *sb, struct inode *parent,
 307                                 ext4_group_t *best_group)
 308 {
 309         ext4_group_t ngroups = ext4_get_groups_count(sb);
 310         unsigned int freei, avefreei;
 311         struct ext4_group_desc *desc, *best_desc = NULL;
 312         ext4_group_t group;
 313         int ret = -1;
 314
 315         freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
 316         avefreei = freei / ngroups;
 317
 318         for (group = 0; group < ngroups; group++) {
 319                 desc = ext4_get_group_desc(sb, group, NULL);
 320                 if (!desc || !ext4_free_inodes_count(sb, desc))
 321                         continue;
 322                 if (ext4_free_inodes_count(sb, desc) < avefreei)
 323                         continue;
 324                 if (!best_desc ||
 325                     (ext4_free_group_clusters(sb, desc) >
 326                      ext4_free_group_clusters(sb, best_desc))) {
 327                         *best_group = group;
 328                         best_desc = desc;
 329                         ret = 0;
 330                 }
 331         }
 332         return ret;
 333 }
 334
 335 #define free_block_ratio 10
 336
 337 static int find_group_flex(struct super_block *sb, struct inode *parent,
 338                            ext4_group_t *best_group)
 339 {
 340         struct ext4_sb_info *sbi = EXT4_SB(sb);
 341         struct ext4_group_desc *desc;
 342         struct flex_groups *flex_group = sbi->s_flex_groups;
 343         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 344         ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
 345         ext4_group_t ngroups = ext4_get_groups_count(sb);
 346         int flex_size = ext4_flex_bg_size(sbi);
 347         ext4_group_t best_flex = parent_fbg_group;
 348         int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
 349         int flexbg_free_clusters;
 350         int flex_freeb_ratio;
 351         ext4_group_t n_fbg_groups;
 352         ext4_group_t i;
 353
 354         n_fbg_groups = (ngroups + flex_size - 1) >>
 355                 sbi->s_log_groups_per_flex;
 356
 357 find_close_to_parent:
 358         flexbg_free_clusters = atomic_read(&flex_group[best_flex].free_clusters);
 359         flex_freeb_ratio = EXT4_C2B(sbi, flexbg_free_clusters) * 100 /
 360                 blocks_per_flex;
 361         if (atomic_read(&flex_group[best_flex].free_inodes) &&
 362             flex_freeb_ratio > free_block_ratio)
 363                 goto found_flexbg;
 364
 365         if (best_flex && best_flex == parent_fbg_group) {
 366                 best_flex--;
 367                 goto find_close_to_parent;
 368         }
 369
 370         for (i = 0; i < n_fbg_groups; i++) {
 371                 if (i == parent_fbg_group || i == parent_fbg_group - 1)
 372                         continue;
 373
 374                 flexbg_free_clusters = atomic_read(&flex_group[i].free_clusters);
 375                 flex_freeb_ratio = EXT4_C2B(sbi, flexbg_free_clusters) * 100 /
 376                         blocks_per_flex;
 377
 378                 if (flex_freeb_ratio > free_block_ratio &&
 379                     (atomic_read(&flex_group[i].free_inodes))) {
 380                         best_flex = i;
 381                         goto found_flexbg;
 382                 }
 383
 384                 if ((atomic_read(&flex_group[best_flex].free_inodes) == 0) ||
 385                     ((atomic_read(&flex_group[i].free_clusters) >
 386                       atomic_read(&flex_group[best_flex].free_clusters)) &&
 387                      atomic_read(&flex_group[i].free_inodes)))
 388                         best_flex = i;
 389         }
 390
 391         if (!atomic_read(&flex_group[best_flex].free_inodes) ||
 392             !atomic_read(&flex_group[best_flex].free_clusters))
 393                 return -1;
 394
 395 found_flexbg:
 396         for (i = best_flex * flex_size; i < ngroups &&
 397                      i < (best_flex + 1) * flex_size; i++) {
 398                 desc = ext4_get_group_desc(sb, i, NULL);
 399                 if (ext4_free_inodes_count(sb, desc)) {
 400                         *best_group = i;
 401                         goto out;
 402                 }
 403         }
 404
 405         return -1;
 406 out:
 407         return 0;
 408 }
 409
 410 struct orlov_stats {
 411         __u32 free_inodes;
 412         __u32 free_clusters;
 413         __u32 used_dirs;
 414 };
 415
 416 /*
 417  * Helper function for Orlov's allocator; returns critical information
 418  * for a particular block group or flex_bg.  If flex_size is 1, then g
 419  * is a block group number; otherwise it is flex_bg number.
 420  */
 421 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 422                             int flex_size, struct orlov_stats *stats)
 423 {
 424         struct ext4_group_desc *desc;
 425         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 426
 427         if (flex_size > 1) {
 428                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 429                 stats->free_clusters = atomic_read(&flex_group[g].free_clusters);
 430                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 431                 return;
 432         }
 433
 434         desc = ext4_get_group_desc(sb, g, NULL);
 435         if (desc) {
 436                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 437                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 438                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 439         } else {
 440                 stats->free_inodes = 0;
 441                 stats->free_clusters = 0;
 442                 stats->used_dirs = 0;
 443         }
 444 }
 445
 446 /*
 447  * Orlov's allocator for directories.
 448  *
 449  * We always try to spread first-level directories.
 450  *
 451  * If there are blockgroups with both free inodes and free blocks counts
 452  * not worse than average we return one with smallest directory count.
 453  * Otherwise we simply return a random group.
 454  *
 455  * For the rest rules look so:
 456  *
 457  * It's OK to put directory into a group unless
 458  * it has too many directories already (max_dirs) or
 459  * it has too few free inodes left (min_inodes) or
 460  * it has too few free blocks left (min_blocks) or
 461  * Parent's group is preferred, if it doesn't satisfy these
 462  * conditions we search cyclically through the rest. If none
 463  * of the groups look good we just look for a group with more
 464  * free inodes than average (starting at parent's group).
 465  */
 466
 467 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 468                             ext4_group_t *group, int mode,
 469                             const struct qstr *qstr)
 470 {
 471         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 472         struct ext4_sb_info *sbi = EXT4_SB(sb);
 473         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 474         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 475         unsigned int freei, avefreei;
 476         ext4_fsblk_t freeb, avefreec;
 477         unsigned int ndirs;
 478         int max_dirs, min_inodes;
 479         ext4_grpblk_t min_clusters;
 480         ext4_group_t i, grp, g, ngroups;
 481         struct ext4_group_desc *desc;
 482         struct orlov_stats stats;
 483         int flex_size = ext4_flex_bg_size(sbi);
 484         struct dx_hash_info hinfo;
 485
 486         ngroups = real_ngroups;
 487         if (flex_size > 1) {
 488                 ngroups = (real_ngroups + flex_size - 1) >>
 489                         sbi->s_log_groups_per_flex;
 490                 parent_group >>= sbi->s_log_groups_per_flex;
 491         }
 492
 493         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 494         avefreei = freei / ngroups;
 495         freeb = EXT4_C2B(sbi,
 496                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 497         avefreec = freeb;
 498         do_div(avefreec, ngroups);
 499         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 500
 501         if (S_ISDIR(mode) &&
 502             ((parent == sb->s_root->d_inode) ||
 503              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 504                 int best_ndir = inodes_per_group;
 505                 int ret = -1;
 506
 507                 if (qstr) {
 508                         hinfo.hash_version = DX_HASH_HALF_MD4;
 509                         hinfo.seed = sbi->s_hash_seed;
 510                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 511                         grp = hinfo.hash;
 512                 } else
 513                         get_random_bytes(&grp, sizeof(grp));
 514                 parent_group = (unsigned)grp % ngroups;
 515                 for (i = 0; i < ngroups; i++) {
 516                         g = (parent_group + i) % ngroups;
 517                         get_orlov_stats(sb, g, flex_size, &stats);
 518                         if (!stats.free_inodes)
 519                                 continue;
 520                         if (stats.used_dirs >= best_ndir)
 521                                 continue;
 522                         if (stats.free_inodes < avefreei)
 523                                 continue;
 524                         if (stats.free_clusters < avefreec)
 525                                 continue;
 526                         grp = g;
 527                         ret = 0;
 528                         best_ndir = stats.used_dirs;
 529                 }
 530                 if (ret)
 531                         goto fallback;
 532         found_flex_bg:
 533                 if (flex_size == 1) {
 534                         *group = grp;
 535                         return 0;
 536                 }
 537
 538                 /*
 539                  * We pack inodes at the beginning of the flexgroup's
 540                  * inode tables.  Block allocation decisions will do
 541                  * something similar, although regular files will
 542                  * start at 2nd block group of the flexgroup.  See
 543                  * ext4_ext_find_goal() and ext4_find_near().
 544                  */
 545                 grp *= flex_size;
 546                 for (i = 0; i < flex_size; i++) {
 547                         if (grp+i >= real_ngroups)
 548                                 break;
 549                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 550                         if (desc && ext4_free_inodes_count(sb, desc)) {
 551                                 *group = grp+i;
 552                                 return 0;
 553                         }
 554                 }
 555                 goto fallback;
 556         }
 557
 558         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 559         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 560         if (min_inodes < 1)
 561                 min_inodes = 1;
 562         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 563
 564         /*
 565          * Start looking in the flex group where we last allocated an
 566          * inode for this parent directory
 567          */
 568         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 569                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 570                 if (flex_size > 1)
 571                         parent_group >>= sbi->s_log_groups_per_flex;
 572         }
 573
 574         for (i = 0; i < ngroups; i++) {
 575                 grp = (parent_group + i) % ngroups;
 576                 get_orlov_stats(sb, grp, flex_size, &stats);
 577                 if (stats.used_dirs >= max_dirs)
 578                         continue;
 579                 if (stats.free_inodes < min_inodes)
 580                         continue;
 581                 if (stats.free_clusters < min_clusters)
 582                         continue;
 583                 goto found_flex_bg;
 584         }
 585
 586 fallback:
 587         ngroups = real_ngroups;
 588         avefreei = freei / ngroups;
 589 fallback_retry:
 590         parent_group = EXT4_I(parent)->i_block_group;
 591         for (i = 0; i < ngroups; i++) {
 592                 grp = (parent_group + i) % ngroups;
 593                 desc = ext4_get_group_desc(sb, grp, NULL);
 594                 if (desc && ext4_free_inodes_count(sb, desc) &&
 595                     ext4_free_inodes_count(sb, desc) >= avefreei) {
 596                         *group = grp;
 597                         return 0;
 598                 }
 599         }
 600
 601         if (avefreei) {
 602                 /*
 603                  * The free-inodes counter is approximate, and for really small
 604                  * filesystems the above test can fail to find any blockgroups
 605                  */
 606                 avefreei = 0;
 607                 goto fallback_retry;
 608         }
 609
 610         return -1;
 611 }
 612
 613 static int find_group_other(struct super_block *sb, struct inode *parent,
 614                             ext4_group_t *group, int mode)
 615 {
 616         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 617         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 618         struct ext4_group_desc *desc;
 619         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 620
 621         /*
 622          * Try to place the inode is the same flex group as its
 623          * parent.  If we can't find space, use the Orlov algorithm to
 624          * find another flex group, and store that information in the
 625          * parent directory's inode information so that use that flex
 626          * group for future allocations.
 627          */
 628         if (flex_size > 1) {
 629                 int retry = 0;
 630
 631         try_again:
 632                 parent_group &= ~(flex_size-1);
 633                 last = parent_group + flex_size;
 634                 if (last > ngroups)
 635                         last = ngroups;
 636                 for  (i = parent_group; i < last; i++) {
 637                         desc = ext4_get_group_desc(sb, i, NULL);
 638                         if (desc && ext4_free_inodes_count(sb, desc)) {
 639                                 *group = i;
 640                                 return 0;
 641                         }
 642                 }
 643                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 644                         retry = 1;
 645                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 646                         goto try_again;
 647                 }
 648                 /*
 649                  * If this didn't work, use the Orlov search algorithm
 650                  * to find a new flex group; we pass in the mode to
 651                  * avoid the topdir algorithms.
 652                  */
 653                 *group = parent_group + flex_size;
 654                 if (*group > ngroups)
 655                         *group = 0;
 656                 return find_group_orlov(sb, parent, group, mode, NULL);
 657         }
 658
 659         /*
 660          * Try to place the inode in its parent directory
 661          */
 662         *group = parent_group;
 663         desc = ext4_get_group_desc(sb, *group, NULL);
 664         if (desc && ext4_free_inodes_count(sb, desc) &&
 665             ext4_free_group_clusters(sb, desc))
 666                 return 0;
 667
 668         /*
 669          * We're going to place this inode in a different blockgroup from its
 670          * parent.  We want to cause files in a common directory to all land in
 671          * the same blockgroup.  But we want files which are in a different
 672          * directory which shares a blockgroup with our parent to land in a
 673          * different blockgroup.
 674          *
 675          * So add our directory's i_ino into the starting point for the hash.
 676          */
 677         *group = (*group + parent->i_ino) % ngroups;
 678
 679         /*
 680          * Use a quadratic hash to find a group with a free inode and some free
 681          * blocks.
 682          */
 683         for (i = 1; i < ngroups; i <<= 1) {
 684                 *group += i;
 685                 if (*group >= ngroups)
 686                         *group -= ngroups;
 687                 desc = ext4_get_group_desc(sb, *group, NULL);
 688                 if (desc && ext4_free_inodes_count(sb, desc) &&
 689                     ext4_free_group_clusters(sb, desc))
 690                         return 0;
 691         }
 692
 693         /*
 694          * That failed: try linear search for a free inode, even if that group
 695          * has no free blocks.
 696          */
 697         *group = parent_group;
 698         for (i = 0; i < ngroups; i++) {
 699                 if (++*group >= ngroups)
 700                         *group = 0;
 701                 desc = ext4_get_group_desc(sb, *group, NULL);
 702                 if (desc && ext4_free_inodes_count(sb, desc))
 703                         return 0;
 704         }
 705
 706         return -1;
 707 }
 708
 709 /*
 710  * claim the inode from the inode bitmap. If the group
 711  * is uninit we need to take the groups's ext4_group_lock
 712  * and clear the uninit flag. The inode bitmap update
 713  * and group desc uninit flag clear should be done
 714  * after holding ext4_group_lock so that ext4_read_inode_bitmap
 715  * doesn't race with the ext4_claim_inode
 716  */
 717 static int ext4_claim_inode(struct super_block *sb,
 718                         struct buffer_head *inode_bitmap_bh,
 719                         unsigned long ino, ext4_group_t group, int mode)
 720 {
 721         int free = 0, retval = 0, count;
 722         struct ext4_sb_info *sbi = EXT4_SB(sb);
 723         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 724         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 725
 726         /*
 727          * We have to be sure that new inode allocation does not race with
 728          * inode table initialization, because otherwise we may end up
 729          * allocating and writing new inode right before sb_issue_zeroout
 730          * takes place and overwriting our new inode with zeroes. So we
 731          * take alloc_sem to prevent it.
 732          */
 733         down_read(&grp->alloc_sem);
 734         ext4_lock_group(sb, group);
 735         if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
 736                 /* not a free inode */
 737                 retval = 1;
 738                 goto err_ret;
 739         }
 740         ino++;
 741         if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
 742                         ino > EXT4_INODES_PER_GROUP(sb)) {
 743                 ext4_unlock_group(sb, group);
 744                 up_read(&grp->alloc_sem);
 745                 ext4_error(sb, "reserved inode or inode > inodes count - "
 746                            "block_group = %u, inode=%lu", group,
 747                            ino + group * EXT4_INODES_PER_GROUP(sb));
 748                 return 1;
 749         }
 750         /* If we didn't allocate from within the initialized part of the inode
 751          * table then we need to initialize up to this inode. */
 752         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
 753
 754                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 755                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 756                         /* When marking the block group with
 757                          * ~EXT4_BG_INODE_UNINIT we don't want to depend
 758                          * on the value of bg_itable_unused even though
 759                          * mke2fs could have initialized the same for us.
 760                          * Instead we calculated the value below
 761                          */
 762
 763                         free = 0;
 764                 } else {
 765                         free = EXT4_INODES_PER_GROUP(sb) -
 766                                 ext4_itable_unused_count(sb, gdp);
 767                 }
 768
 769                 /*
 770                  * Check the relative inode number against the last used
 771                  * relative inode number in this group. if it is greater
 772                  * we need to  update the bg_itable_unused count
 773                  *
 774                  */
 775                 if (ino > free)
 776                         ext4_itable_unused_set(sb, gdp,
 777                                         (EXT4_INODES_PER_GROUP(sb) - ino));
 778         }
 779         count = ext4_free_inodes_count(sb, gdp) - 1;
 780         ext4_free_inodes_set(sb, gdp, count);
 781         if (S_ISDIR(mode)) {
 782                 count = ext4_used_dirs_count(sb, gdp) + 1;
 783                 ext4_used_dirs_set(sb, gdp, count);
 784                 if (sbi->s_log_groups_per_flex) {
 785                         ext4_group_t f = ext4_flex_group(sbi, group);
 786
 787                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
 788                 }
 789         }
 790         gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
 791 err_ret:
 792         ext4_unlock_group(sb, group);
 793         up_read(&grp->alloc_sem);
 794         return retval;
 795 }
 796
 797 /*
 798  * There are two policies for allocating an inode.  If the new inode is
 799  * a directory, then a forward search is made for a block group with both
 800  * free space and a low directory-to-inode ratio; if that fails, then of
 801  * the groups with above-average free space, that group with the fewest
 802  * directories already is chosen.
 803  *
 804  * For other inodes, search forward from the parent directory's block
 805  * group to find a free inode.
 806  */
 807 struct inode *ext4_new_inode(handle_t *handle, struct inode *dir, int mode,
 808                              const struct qstr *qstr, __u32 goal)
 809 {
 810         struct super_block *sb;
 811         struct buffer_head *inode_bitmap_bh = NULL;
 812         struct buffer_head *group_desc_bh;
 813         ext4_group_t ngroups, group = 0;
 814         unsigned long ino = 0;
 815         struct inode *inode;
 816         struct ext4_group_desc *gdp = NULL;
 817         struct ext4_inode_info *ei;
 818         struct ext4_sb_info *sbi;
 819         int ret2, err = 0;
 820         struct inode *ret;
 821         ext4_group_t i;
 822         static int once = 1;
 823         ext4_group_t flex_group;
 824
 825         /* Cannot create files in a deleted directory */
 826         if (!dir || !dir->i_nlink)
 827                 return ERR_PTR(-EPERM);
 828
 829         sb = dir->i_sb;
 830         ngroups = ext4_get_groups_count(sb);
 831         trace_ext4_request_inode(dir, mode);
 832         inode = new_inode(sb);
 833         if (!inode)
 834                 return ERR_PTR(-ENOMEM);
 835         ei = EXT4_I(inode);
 836         sbi = EXT4_SB(sb);
 837
 838         if (!goal)
 839                 goal = sbi->s_inode_goal;
 840
 841         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 842                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 843                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 844                 ret2 = 0;
 845                 goto got_group;
 846         }
 847
 848         if (sbi->s_log_groups_per_flex && test_opt(sb, OLDALLOC)) {
 849                 ret2 = find_group_flex(sb, dir, &group);
 850                 if (ret2 == -1) {
 851                         ret2 = find_group_other(sb, dir, &group, mode);
 852                         if (ret2 == 0 && once) {
 853                                 once = 0;
 854                                 printk(KERN_NOTICE "ext4: find_group_flex "
 855                                        "failed, fallback succeeded dir %lu\n",
 856                                        dir->i_ino);
 857                         }
 858                 }
 859                 goto got_group;
 860         }
 861
 862         if (S_ISDIR(mode)) {
 863                 if (test_opt(sb, OLDALLOC))
 864                         ret2 = find_group_dir(sb, dir, &group);
 865                 else
 866                         ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 867         } else
 868                 ret2 = find_group_other(sb, dir, &group, mode);
 869
 870 got_group:
 871         EXT4_I(dir)->i_last_alloc_group = group;
 872         err = -ENOSPC;
 873         if (ret2 == -1)
 874                 goto out;
 875
 876         for (i = 0; i < ngroups; i++, ino = 0) {
 877                 err = -EIO;
 878
 879                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 880                 if (!gdp)
 881                         goto fail;
 882
 883                 brelse(inode_bitmap_bh);
 884                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 885                 if (!inode_bitmap_bh)
 886                         goto fail;
 887
 888 repeat_in_this_group:
 889                 ino = ext4_find_next_zero_bit((unsigned long *)
 890                                               inode_bitmap_bh->b_data,
 891                                               EXT4_INODES_PER_GROUP(sb), ino);
 892
 893                 if (ino < EXT4_INODES_PER_GROUP(sb)) {
 894
 895                         BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 896                         err = ext4_journal_get_write_access(handle,
 897                                                             inode_bitmap_bh);
 898                         if (err)
 899                                 goto fail;
 900
 901                         BUFFER_TRACE(group_desc_bh, "get_write_access");
 902                         err = ext4_journal_get_write_access(handle,
 903                                                                 group_desc_bh);
 904                         if (err)
 905                                 goto fail;
 906                         if (!ext4_claim_inode(sb, inode_bitmap_bh,
 907                                                 ino, group, mode)) {
 908                                 /* we won it */
 909                                 BUFFER_TRACE(inode_bitmap_bh,
 910                                         "call ext4_handle_dirty_metadata");
 911                                 err = ext4_handle_dirty_metadata(handle,
 912                                                                  NULL,
 913                                                         inode_bitmap_bh);
 914                                 if (err)
 915                                         goto fail;
 916                                 /* zero bit is inode number 1*/
 917                                 ino++;
 918                                 goto got;
 919                         }
 920                         /* we lost it */
 921                         ext4_handle_release_buffer(handle, inode_bitmap_bh);
 922                         ext4_handle_release_buffer(handle, group_desc_bh);
 923
 924                         if (++ino < EXT4_INODES_PER_GROUP(sb))
 925                                 goto repeat_in_this_group;
 926                 }
 927
 928                 /*
 929                  * This case is possible in concurrent environment.  It is very
 930                  * rare.  We cannot repeat the find_group_xxx() call because
 931                  * that will simply return the same blockgroup, because the
 932                  * group descriptor metadata has not yet been updated.
 933                  * So we just go onto the next blockgroup.
 934                  */
 935                 if (++group == ngroups)
 936                         group = 0;
 937         }
 938         err = -ENOSPC;
 939         goto out;
 940
 941 got:
 942         /* We may have to initialize the block bitmap if it isn't already */
 943         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
 944             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 945                 struct buffer_head *block_bitmap_bh;
 946
 947                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 948                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 949                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 950                 if (err) {
 951                         brelse(block_bitmap_bh);
 952                         goto fail;
 953                 }
 954
 955                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 956                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 957                 brelse(block_bitmap_bh);
 958
 959                 /* recheck and clear flag under lock if we still need to */
 960                 ext4_lock_group(sb, group);
 961                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 962                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 963                         ext4_free_group_clusters_set(sb, gdp,
 964                                 ext4_free_blocks_after_init(sb, group, gdp));
 965                         gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
 966                                                                 gdp);
 967                 }
 968                 ext4_unlock_group(sb, group);
 969
 970                 if (err)
 971                         goto fail;
 972         }
 973         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 974         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 975         if (err)
 976                 goto fail;
 977
 978         percpu_counter_dec(&sbi->s_freeinodes_counter);
 979         if (S_ISDIR(mode))
 980                 percpu_counter_inc(&sbi->s_dirs_counter);
 981         ext4_mark_super_dirty(sb);
 982
 983         if (sbi->s_log_groups_per_flex) {
 984                 flex_group = ext4_flex_group(sbi, group);
 985                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
 986         }
 987
 988         if (test_opt(sb, GRPID)) {
 989                 inode->i_mode = mode;
 990                 inode->i_uid = current_fsuid();
 991                 inode->i_gid = dir->i_gid;
 992         } else
 993                 inode_init_owner(inode, dir, mode);
 994
 995         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
 996         /* This is the optimal IO size (for stat), not the fs block size */
 997         inode->i_blocks = 0;
 998         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
 999                                                        ext4_current_time(inode);
1000
1001         memset(ei->i_data, 0, sizeof(ei->i_data));
1002         ei->i_dir_start_lookup = 0;
1003         ei->i_disksize = 0;
1004
1005         /*
1006          * Don't inherit extent flag from directory, amongst others. We set
1007          * extent flag on newly created directory and file only if -o extent
1008          * mount option is specified
1009          */
1010         ei->i_flags =
1011                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1012         ei->i_file_acl = 0;
1013         ei->i_dtime = 0;
1014         ei->i_block_group = group;
1015         ei->i_last_alloc_group = ~0;
1016
1017         ext4_set_inode_flags(inode);
1018         if (IS_DIRSYNC(inode))
1019                 ext4_handle_sync(handle);
1020         if (insert_inode_locked(inode) < 0) {
1021                 err = -EINVAL;
1022                 goto fail_drop;
1023         }
1024         spin_lock(&sbi->s_next_gen_lock);
1025         inode->i_generation = sbi->s_next_generation++;
1026         spin_unlock(&sbi->s_next_gen_lock);
1027
1028         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1029         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1030
1031         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1032
1033         ret = inode;
1034         dquot_initialize(inode);
1035         err = dquot_alloc_inode(inode);
1036         if (err)
1037                 goto fail_drop;
1038
1039         err = ext4_init_acl(handle, inode, dir);
1040         if (err)
1041                 goto fail_free_drop;
1042
1043         err = ext4_init_security(handle, inode, dir, qstr);
1044         if (err)
1045                 goto fail_free_drop;
1046
1047         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1048                 /* set extent flag only for directory, file and normal symlink*/
1049                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1050                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1051                         ext4_ext_tree_init(handle, inode);
1052                 }
1053         }
1054
1055         if (ext4_handle_valid(handle)) {
1056                 ei->i_sync_tid = handle->h_transaction->t_tid;
1057                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1058         }
1059
1060         err = ext4_mark_inode_dirty(handle, inode);
1061         if (err) {
1062                 ext4_std_error(sb, err);
1063                 goto fail_free_drop;
1064         }
1065
1066         ext4_debug("allocating inode %lu\n", inode->i_ino);
1067         trace_ext4_allocate_inode(inode, dir, mode);
1068         goto really_out;
1069 fail:
1070         ext4_std_error(sb, err);
1071 out:
1072         iput(inode);
1073         ret = ERR_PTR(err);
1074 really_out:
1075         brelse(inode_bitmap_bh);
1076         return ret;
1077
1078 fail_free_drop:
1079         dquot_free_inode(inode);
1080
1081 fail_drop:
1082         dquot_drop(inode);
1083         inode->i_flags |= S_NOQUOTA;
1084         inode->i_nlink = 0;
1085         unlock_new_inode(inode);
1086         iput(inode);
1087         brelse(inode_bitmap_bh);
1088         return ERR_PTR(err);
1089 }
1090
1091 /* Verify that we are loading a valid orphan from disk */
1092 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1093 {
1094         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1095         ext4_group_t block_group;
1096         int bit;
1097         struct buffer_head *bitmap_bh;
1098         struct inode *inode = NULL;
1099         long err = -EIO;
1100
1101         /* Error cases - e2fsck has already cleaned up for us */
1102         if (ino > max_ino) {
1103                 ext4_warning(sb, "bad orphan ino %lu!  e2fsck was run?", ino);
1104                 goto error;
1105         }
1106
1107         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1108         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1109         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1110         if (!bitmap_bh) {
1111                 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1112                 goto error;
1113         }
1114
1115         /* Having the inode bit set should be a 100% indicator that this
1116          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1117          * inodes that were being truncated, so we can't check i_nlink==0.
1118          */
1119         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1120                 goto bad_orphan;
1121
1122         inode = ext4_iget(sb, ino);
1123         if (IS_ERR(inode))
1124                 goto iget_failed;
1125
1126         /*
1127          * If the orphans has i_nlinks > 0 then it should be able to be
1128          * truncated, otherwise it won't be removed from the orphan list
1129          * during processing and an infinite loop will result.
1130          */
1131         if (inode->i_nlink && !ext4_can_truncate(inode))
1132                 goto bad_orphan;
1133
1134         if (NEXT_ORPHAN(inode) > max_ino)
1135                 goto bad_orphan;
1136         brelse(bitmap_bh);
1137         return inode;
1138
1139 iget_failed:
1140         err = PTR_ERR(inode);
1141         inode = NULL;
1142 bad_orphan:
1143         ext4_warning(sb, "bad orphan inode %lu!  e2fsck was run?", ino);
1144         printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1145                bit, (unsigned long long)bitmap_bh->b_blocknr,
1146                ext4_test_bit(bit, bitmap_bh->b_data));
1147         printk(KERN_NOTICE "inode=%p\n", inode);
1148         if (inode) {
1149                 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1150                        is_bad_inode(inode));
1151                 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1152                        NEXT_ORPHAN(inode));
1153                 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1154                 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1155                 /* Avoid freeing blocks if we got a bad deleted inode */
1156                 if (inode->i_nlink == 0)
1157                         inode->i_blocks = 0;
1158                 iput(inode);
1159         }
1160         brelse(bitmap_bh);
1161 error:
1162         return ERR_PTR(err);
1163 }
1164
1165 unsigned long ext4_count_free_inodes(struct super_block *sb)
1166 {
1167         unsigned long desc_count;
1168         struct ext4_group_desc *gdp;
1169         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1170 #ifdef EXT4FS_DEBUG
1171         struct ext4_super_block *es;
1172         unsigned long bitmap_count, x;
1173         struct buffer_head *bitmap_bh = NULL;
1174
1175         es = EXT4_SB(sb)->s_es;
1176         desc_count = 0;
1177         bitmap_count = 0;
1178         gdp = NULL;
1179         for (i = 0; i < ngroups; i++) {
1180                 gdp = ext4_get_group_desc(sb, i, NULL);
1181                 if (!gdp)
1182                         continue;
1183                 desc_count += ext4_free_inodes_count(sb, gdp);
1184                 brelse(bitmap_bh);
1185                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1186                 if (!bitmap_bh)
1187                         continue;
1188
1189                 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1190                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1191                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1192                 bitmap_count += x;
1193         }
1194         brelse(bitmap_bh);
1195         printk(KERN_DEBUG "ext4_count_free_inodes: "
1196                "stored = %u, computed = %lu, %lu\n",
1197                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1198         return desc_count;
1199 #else
1200         desc_count = 0;
1201         for (i = 0; i < ngroups; i++) {
1202                 gdp = ext4_get_group_desc(sb, i, NULL);
1203                 if (!gdp)
1204                         continue;
1205                 desc_count += ext4_free_inodes_count(sb, gdp);
1206                 cond_resched();
1207         }
1208         return desc_count;
1209 #endif
1210 }
1211
1212 /* Called at mount-time, super-block is locked */
1213 unsigned long ext4_count_dirs(struct super_block * sb)
1214 {
1215         unsigned long count = 0;
1216         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1217
1218         for (i = 0; i < ngroups; i++) {
1219                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1220                 if (!gdp)
1221                         continue;
1222                 count += ext4_used_dirs_count(sb, gdp);
1223         }
1224         return count;
1225 }
1226
1227 /*
1228  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1229  * inode table. Must be called without any spinlock held. The only place
1230  * where it is called from on active part of filesystem is ext4lazyinit
1231  * thread, so we do not need any special locks, however we have to prevent
1232  * inode allocation from the current group, so we take alloc_sem lock, to
1233  * block ext4_claim_inode until we are finished.
1234  */
1235 extern int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1236                                  int barrier)
1237 {
1238         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1239         struct ext4_sb_info *sbi = EXT4_SB(sb);
1240         struct ext4_group_desc *gdp = NULL;
1241         struct buffer_head *group_desc_bh;
1242         handle_t *handle;
1243         ext4_fsblk_t blk;
1244         int num, ret = 0, used_blks = 0;
1245
1246         /* This should not happen, but just to be sure check this */
1247         if (sb->s_flags & MS_RDONLY) {
1248                 ret = 1;
1249                 goto out;
1250         }
1251
1252         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1253         if (!gdp)
1254                 goto out;
1255
1256         /*
1257          * We do not need to lock this, because we are the only one
1258          * handling this flag.
1259          */
1260         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1261                 goto out;
1262
1263         handle = ext4_journal_start_sb(sb, 1);
1264         if (IS_ERR(handle)) {
1265                 ret = PTR_ERR(handle);
1266                 goto out;
1267         }
1268
1269         down_write(&grp->alloc_sem);
1270         /*
1271          * If inode bitmap was already initialized there may be some
1272          * used inodes so we need to skip blocks with used inodes in
1273          * inode table.
1274          */
1275         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1276                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1277                             ext4_itable_unused_count(sb, gdp)),
1278                             sbi->s_inodes_per_block);
1279
1280         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1281                 ext4_error(sb, "Something is wrong with group %u\n"
1282                            "Used itable blocks: %d"
1283                            "itable unused count: %u\n",
1284                            group, used_blks,
1285                            ext4_itable_unused_count(sb, gdp));
1286                 ret = 1;
1287                 goto err_out;
1288         }
1289
1290         blk = ext4_inode_table(sb, gdp) + used_blks;
1291         num = sbi->s_itb_per_group - used_blks;
1292
1293         BUFFER_TRACE(group_desc_bh, "get_write_access");
1294         ret = ext4_journal_get_write_access(handle,
1295                                             group_desc_bh);
1296         if (ret)
1297                 goto err_out;
1298
1299         /*
1300          * Skip zeroout if the inode table is full. But we set the ZEROED
1301          * flag anyway, because obviously, when it is full it does not need
1302          * further zeroing.
1303          */
1304         if (unlikely(num == 0))
1305                 goto skip_zeroout;
1306
1307         ext4_debug("going to zero out inode table in group %d\n",
1308                    group);
1309         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1310         if (ret < 0)
1311                 goto err_out;
1312         if (barrier)
1313                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1314
1315 skip_zeroout:
1316         ext4_lock_group(sb, group);
1317         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1318         gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
1319         ext4_unlock_group(sb, group);
1320
1321         BUFFER_TRACE(group_desc_bh,
1322                      "call ext4_handle_dirty_metadata");
1323         ret = ext4_handle_dirty_metadata(handle, NULL,
1324                                          group_desc_bh);
1325
1326 err_out:
1327         up_write(&grp->alloc_sem);
1328         ext4_journal_stop(handle);
1329 out:
1330         return ret;
1331 }