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/stat.h>
  18 #include <linux/string.h>
  19 #include <linux/quotaops.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/random.h>
  22 #include <linux/bitops.h>
  23 #include <linux/blkdev.h>
  24 #include <asm/byteorder.h>
  25
  26 #include "ext4.h"
  27 #include "ext4_jbd2.h"
  28 #include "xattr.h"
  29 #include "acl.h"
  30
  31 #include <trace/events/ext4.h>
  32
  33 /*
  34  * ialloc.c contains the inodes allocation and deallocation routines
  35  */
  36
  37 /*
  38  * The free inodes are managed by bitmaps.  A file system contains several
  39  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  40  * block for inodes, N blocks for the inode table and data blocks.
  41  *
  42  * The file system contains group descriptors which are located after the
  43  * super block.  Each descriptor contains the number of the bitmap block and
  44  * the free blocks count in the block.
  45  */
  46
  47 /*
  48  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  49  * need to use it within a single byte (to ensure we get endianness right).
  50  * We can use memset for the rest of the bitmap as there are no other users.
  51  */
  52 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  53 {
  54         int i;
  55
  56         if (start_bit >= end_bit)
  57                 return;
  58
  59         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  60         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  61                 ext4_set_bit(i, bitmap);
  62         if (i < end_bit)
  63                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  64 }
  65
  66 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
  67 {
  68         if (uptodate) {
  69                 set_buffer_uptodate(bh);
  70                 set_bitmap_uptodate(bh);
  71         }
  72         unlock_buffer(bh);
  73         put_bh(bh);
  74 }
  75
  76 static int ext4_validate_inode_bitmap(struct super_block *sb,
  77                                       struct ext4_group_desc *desc,
  78                                       ext4_group_t block_group,
  79                                       struct buffer_head *bh)
  80 {
  81         ext4_fsblk_t    blk;
  82         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
  83         struct ext4_sb_info *sbi = EXT4_SB(sb);
  84
  85         if (buffer_verified(bh))
  86                 return 0;
  87         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
  88                 return -EFSCORRUPTED;
  89
  90         ext4_lock_group(sb, block_group);
  91         if (buffer_verified(bh))
  92                 goto verified;
  93         blk = ext4_inode_bitmap(sb, desc);
  94         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
  95                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
  96                 ext4_unlock_group(sb, block_group);
  97                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
  98                            "inode_bitmap = %llu", block_group, blk);
  99                 grp = ext4_get_group_info(sb, block_group);
 100                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 101                         int count;
 102                         count = ext4_free_inodes_count(sb, desc);
 103                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 104                                            count);
 105                 }
 106                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 107                 return -EFSBADCRC;
 108         }
 109         set_buffer_verified(bh);
 110 verified:
 111         ext4_unlock_group(sb, block_group);
 112         return 0;
 113 }
 114
 115 /*
 116  * Read the inode allocation bitmap for a given block_group, reading
 117  * into the specified slot in the superblock's bitmap cache.
 118  *
 119  * Return buffer_head of bitmap on success or NULL.
 120  */
 121 static struct buffer_head *
 122 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 123 {
 124         struct ext4_group_desc *desc;
 125         struct ext4_sb_info *sbi = EXT4_SB(sb);
 126         struct buffer_head *bh = NULL;
 127         ext4_fsblk_t bitmap_blk;
 128         int err;
 129
 130         desc = ext4_get_group_desc(sb, block_group, NULL);
 131         if (!desc)
 132                 return ERR_PTR(-EFSCORRUPTED);
 133
 134         bitmap_blk = ext4_inode_bitmap(sb, desc);
 135         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 136             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 137                 ext4_error(sb, "Invalid inode bitmap blk %llu in "
 138                            "block_group %u", bitmap_blk, block_group);
 139                 return ERR_PTR(-EFSCORRUPTED);
 140         }
 141         bh = sb_getblk(sb, bitmap_blk);
 142         if (unlikely(!bh)) {
 143                 ext4_error(sb, "Cannot read inode bitmap - "
 144                             "block_group = %u, inode_bitmap = %llu",
 145                             block_group, bitmap_blk);
 146                 return ERR_PTR(-EIO);
 147         }
 148         if (bitmap_uptodate(bh))
 149                 goto verify;
 150
 151         lock_buffer(bh);
 152         if (bitmap_uptodate(bh)) {
 153                 unlock_buffer(bh);
 154                 goto verify;
 155         }
 156
 157         ext4_lock_group(sb, block_group);
 158         if (ext4_has_group_desc_csum(sb) &&
 159             (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
 160                 if (block_group == 0) {
 161                         ext4_unlock_group(sb, block_group);
 162                         unlock_buffer(bh);
 163                         ext4_error(sb, "Inode bitmap for bg 0 marked "
 164                                    "uninitialized");
 165                         err = -EFSCORRUPTED;
 166                         goto out;
 167                 }
 168                 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 169                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 170                                      sb->s_blocksize * 8, bh->b_data);
 171                 set_bitmap_uptodate(bh);
 172                 set_buffer_uptodate(bh);
 173                 set_buffer_verified(bh);
 174                 ext4_unlock_group(sb, block_group);
 175                 unlock_buffer(bh);
 176                 return bh;
 177         }
 178         ext4_unlock_group(sb, block_group);
 179
 180         if (buffer_uptodate(bh)) {
 181                 /*
 182                  * if not uninit if bh is uptodate,
 183                  * bitmap is also uptodate
 184                  */
 185                 set_bitmap_uptodate(bh);
 186                 unlock_buffer(bh);
 187                 goto verify;
 188         }
 189         /*
 190          * submit the buffer_head for reading
 191          */
 192         trace_ext4_load_inode_bitmap(sb, block_group);
 193         bh->b_end_io = ext4_end_bitmap_read;
 194         get_bh(bh);
 195         submit_bh(READ | REQ_META | REQ_PRIO, bh);
 196         wait_on_buffer(bh);
 197         if (!buffer_uptodate(bh)) {
 198                 put_bh(bh);
 199                 ext4_error(sb, "Cannot read inode bitmap - "
 200                            "block_group = %u, inode_bitmap = %llu",
 201                            block_group, bitmap_blk);
 202                 return ERR_PTR(-EIO);
 203         }
 204
 205 verify:
 206         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 207         if (err)
 208                 goto out;
 209         return bh;
 210 out:
 211         put_bh(bh);
 212         return ERR_PTR(err);
 213 }
 214
 215 /*
 216  * NOTE! When we get the inode, we're the only people
 217  * that have access to it, and as such there are no
 218  * race conditions we have to worry about. The inode
 219  * is not on the hash-lists, and it cannot be reached
 220  * through the filesystem because the directory entry
 221  * has been deleted earlier.
 222  *
 223  * HOWEVER: we must make sure that we get no aliases,
 224  * which means that we have to call "clear_inode()"
 225  * _before_ we mark the inode not in use in the inode
 226  * bitmaps. Otherwise a newly created file might use
 227  * the same inode number (not actually the same pointer
 228  * though), and then we'd have two inodes sharing the
 229  * same inode number and space on the harddisk.
 230  */
 231 void ext4_free_inode(handle_t *handle, struct inode *inode)
 232 {
 233         struct super_block *sb = inode->i_sb;
 234         int is_directory;
 235         unsigned long ino;
 236         struct buffer_head *bitmap_bh = NULL;
 237         struct buffer_head *bh2;
 238         ext4_group_t block_group;
 239         unsigned long bit;
 240         struct ext4_group_desc *gdp;
 241         struct ext4_super_block *es;
 242         struct ext4_sb_info *sbi;
 243         int fatal = 0, err, count, cleared;
 244         struct ext4_group_info *grp;
 245
 246         if (!sb) {
 247                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 248                        "nonexistent device\n", __func__, __LINE__);
 249                 return;
 250         }
 251         if (atomic_read(&inode->i_count) > 1) {
 252                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 253                          __func__, __LINE__, inode->i_ino,
 254                          atomic_read(&inode->i_count));
 255                 return;
 256         }
 257         if (inode->i_nlink) {
 258                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 259                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 260                 return;
 261         }
 262         sbi = EXT4_SB(sb);
 263
 264         ino = inode->i_ino;
 265         ext4_debug("freeing inode %lu\n", ino);
 266         trace_ext4_free_inode(inode);
 267
 268         /*
 269          * Note: we must free any quota before locking the superblock,
 270          * as writing the quota to disk may need the lock as well.
 271          */
 272         dquot_initialize(inode);
 273         ext4_xattr_delete_inode(handle, inode);
 274         dquot_free_inode(inode);
 275         dquot_drop(inode);
 276
 277         is_directory = S_ISDIR(inode->i_mode);
 278
 279         /* Do this BEFORE marking the inode not in use or returning an error */
 280         ext4_clear_inode(inode);
 281
 282         es = EXT4_SB(sb)->s_es;
 283         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 284                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 285                 goto error_return;
 286         }
 287         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 288         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 289         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 290         /* Don't bother if the inode bitmap is corrupt. */
 291         grp = ext4_get_group_info(sb, block_group);
 292         if (IS_ERR(bitmap_bh)) {
 293                 fatal = PTR_ERR(bitmap_bh);
 294                 bitmap_bh = NULL;
 295                 goto error_return;
 296         }
 297         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 298                 fatal = -EFSCORRUPTED;
 299                 goto error_return;
 300         }
 301
 302         BUFFER_TRACE(bitmap_bh, "get_write_access");
 303         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 304         if (fatal)
 305                 goto error_return;
 306
 307         fatal = -ESRCH;
 308         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 309         if (gdp) {
 310                 BUFFER_TRACE(bh2, "get_write_access");
 311                 fatal = ext4_journal_get_write_access(handle, bh2);
 312         }
 313         ext4_lock_group(sb, block_group);
 314         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 315         if (fatal || !cleared) {
 316                 ext4_unlock_group(sb, block_group);
 317                 goto out;
 318         }
 319
 320         count = ext4_free_inodes_count(sb, gdp) + 1;
 321         ext4_free_inodes_set(sb, gdp, count);
 322         if (is_directory) {
 323                 count = ext4_used_dirs_count(sb, gdp) - 1;
 324                 ext4_used_dirs_set(sb, gdp, count);
 325                 percpu_counter_dec(&sbi->s_dirs_counter);
 326         }
 327         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 328                                    EXT4_INODES_PER_GROUP(sb) / 8);
 329         ext4_group_desc_csum_set(sb, block_group, gdp);
 330         ext4_unlock_group(sb, block_group);
 331
 332         percpu_counter_inc(&sbi->s_freeinodes_counter);
 333         if (sbi->s_log_groups_per_flex) {
 334                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 335
 336                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 337                 if (is_directory)
 338                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 339         }
 340         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 341         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 342 out:
 343         if (cleared) {
 344                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 345                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 346                 if (!fatal)
 347                         fatal = err;
 348         } else {
 349                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 350                 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 351                         int count;
 352                         count = ext4_free_inodes_count(sb, gdp);
 353                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 354                                            count);
 355                 }
 356                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 357         }
 358
 359 error_return:
 360         brelse(bitmap_bh);
 361         ext4_std_error(sb, fatal);
 362 }
 363
 364 struct orlov_stats {
 365         __u64 free_clusters;
 366         __u32 free_inodes;
 367         __u32 used_dirs;
 368 };
 369
 370 /*
 371  * Helper function for Orlov's allocator; returns critical information
 372  * for a particular block group or flex_bg.  If flex_size is 1, then g
 373  * is a block group number; otherwise it is flex_bg number.
 374  */
 375 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 376                             int flex_size, struct orlov_stats *stats)
 377 {
 378         struct ext4_group_desc *desc;
 379         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 380
 381         if (flex_size > 1) {
 382                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 383                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 384                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 385                 return;
 386         }
 387
 388         desc = ext4_get_group_desc(sb, g, NULL);
 389         if (desc) {
 390                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 391                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 392                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 393         } else {
 394                 stats->free_inodes = 0;
 395                 stats->free_clusters = 0;
 396                 stats->used_dirs = 0;
 397         }
 398 }
 399
 400 /*
 401  * Orlov's allocator for directories.
 402  *
 403  * We always try to spread first-level directories.
 404  *
 405  * If there are blockgroups with both free inodes and free blocks counts
 406  * not worse than average we return one with smallest directory count.
 407  * Otherwise we simply return a random group.
 408  *
 409  * For the rest rules look so:
 410  *
 411  * It's OK to put directory into a group unless
 412  * it has too many directories already (max_dirs) or
 413  * it has too few free inodes left (min_inodes) or
 414  * it has too few free blocks left (min_blocks) or
 415  * Parent's group is preferred, if it doesn't satisfy these
 416  * conditions we search cyclically through the rest. If none
 417  * of the groups look good we just look for a group with more
 418  * free inodes than average (starting at parent's group).
 419  */
 420
 421 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 422                             ext4_group_t *group, umode_t mode,
 423                             const struct qstr *qstr)
 424 {
 425         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 426         struct ext4_sb_info *sbi = EXT4_SB(sb);
 427         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 428         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 429         unsigned int freei, avefreei, grp_free;
 430         ext4_fsblk_t freeb, avefreec;
 431         unsigned int ndirs;
 432         int max_dirs, min_inodes;
 433         ext4_grpblk_t min_clusters;
 434         ext4_group_t i, grp, g, ngroups;
 435         struct ext4_group_desc *desc;
 436         struct orlov_stats stats;
 437         int flex_size = ext4_flex_bg_size(sbi);
 438         struct dx_hash_info hinfo;
 439
 440         ngroups = real_ngroups;
 441         if (flex_size > 1) {
 442                 ngroups = (real_ngroups + flex_size - 1) >>
 443                         sbi->s_log_groups_per_flex;
 444                 parent_group >>= sbi->s_log_groups_per_flex;
 445         }
 446
 447         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 448         avefreei = freei / ngroups;
 449         freeb = EXT4_C2B(sbi,
 450                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 451         avefreec = freeb;
 452         do_div(avefreec, ngroups);
 453         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 454
 455         if (S_ISDIR(mode) &&
 456             ((parent == d_inode(sb->s_root)) ||
 457              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 458                 int best_ndir = inodes_per_group;
 459                 int ret = -1;
 460
 461                 if (qstr) {
 462                         hinfo.hash_version = DX_HASH_HALF_MD4;
 463                         hinfo.seed = sbi->s_hash_seed;
 464                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 465                         grp = hinfo.hash;
 466                 } else
 467                         grp = prandom_u32();
 468                 parent_group = (unsigned)grp % ngroups;
 469                 for (i = 0; i < ngroups; i++) {
 470                         g = (parent_group + i) % ngroups;
 471                         get_orlov_stats(sb, g, flex_size, &stats);
 472                         if (!stats.free_inodes)
 473                                 continue;
 474                         if (stats.used_dirs >= best_ndir)
 475                                 continue;
 476                         if (stats.free_inodes < avefreei)
 477                                 continue;
 478                         if (stats.free_clusters < avefreec)
 479                                 continue;
 480                         grp = g;
 481                         ret = 0;
 482                         best_ndir = stats.used_dirs;
 483                 }
 484                 if (ret)
 485                         goto fallback;
 486         found_flex_bg:
 487                 if (flex_size == 1) {
 488                         *group = grp;
 489                         return 0;
 490                 }
 491
 492                 /*
 493                  * We pack inodes at the beginning of the flexgroup's
 494                  * inode tables.  Block allocation decisions will do
 495                  * something similar, although regular files will
 496                  * start at 2nd block group of the flexgroup.  See
 497                  * ext4_ext_find_goal() and ext4_find_near().
 498                  */
 499                 grp *= flex_size;
 500                 for (i = 0; i < flex_size; i++) {
 501                         if (grp+i >= real_ngroups)
 502                                 break;
 503                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 504                         if (desc && ext4_free_inodes_count(sb, desc)) {
 505                                 *group = grp+i;
 506                                 return 0;
 507                         }
 508                 }
 509                 goto fallback;
 510         }
 511
 512         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 513         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 514         if (min_inodes < 1)
 515                 min_inodes = 1;
 516         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 517
 518         /*
 519          * Start looking in the flex group where we last allocated an
 520          * inode for this parent directory
 521          */
 522         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 523                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 524                 if (flex_size > 1)
 525                         parent_group >>= sbi->s_log_groups_per_flex;
 526         }
 527
 528         for (i = 0; i < ngroups; i++) {
 529                 grp = (parent_group + i) % ngroups;
 530                 get_orlov_stats(sb, grp, flex_size, &stats);
 531                 if (stats.used_dirs >= max_dirs)
 532                         continue;
 533                 if (stats.free_inodes < min_inodes)
 534                         continue;
 535                 if (stats.free_clusters < min_clusters)
 536                         continue;
 537                 goto found_flex_bg;
 538         }
 539
 540 fallback:
 541         ngroups = real_ngroups;
 542         avefreei = freei / ngroups;
 543 fallback_retry:
 544         parent_group = EXT4_I(parent)->i_block_group;
 545         for (i = 0; i < ngroups; i++) {
 546                 grp = (parent_group + i) % ngroups;
 547                 desc = ext4_get_group_desc(sb, grp, NULL);
 548                 if (desc) {
 549                         grp_free = ext4_free_inodes_count(sb, desc);
 550                         if (grp_free && grp_free >= avefreei) {
 551                                 *group = grp;
 552                                 return 0;
 553                         }
 554                 }
 555         }
 556
 557         if (avefreei) {
 558                 /*
 559                  * The free-inodes counter is approximate, and for really small
 560                  * filesystems the above test can fail to find any blockgroups
 561                  */
 562                 avefreei = 0;
 563                 goto fallback_retry;
 564         }
 565
 566         return -1;
 567 }
 568
 569 static int find_group_other(struct super_block *sb, struct inode *parent,
 570                             ext4_group_t *group, umode_t mode)
 571 {
 572         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 573         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 574         struct ext4_group_desc *desc;
 575         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 576
 577         /*
 578          * Try to place the inode is the same flex group as its
 579          * parent.  If we can't find space, use the Orlov algorithm to
 580          * find another flex group, and store that information in the
 581          * parent directory's inode information so that use that flex
 582          * group for future allocations.
 583          */
 584         if (flex_size > 1) {
 585                 int retry = 0;
 586
 587         try_again:
 588                 parent_group &= ~(flex_size-1);
 589                 last = parent_group + flex_size;
 590                 if (last > ngroups)
 591                         last = ngroups;
 592                 for  (i = parent_group; i < last; i++) {
 593                         desc = ext4_get_group_desc(sb, i, NULL);
 594                         if (desc && ext4_free_inodes_count(sb, desc)) {
 595                                 *group = i;
 596                                 return 0;
 597                         }
 598                 }
 599                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 600                         retry = 1;
 601                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 602                         goto try_again;
 603                 }
 604                 /*
 605                  * If this didn't work, use the Orlov search algorithm
 606                  * to find a new flex group; we pass in the mode to
 607                  * avoid the topdir algorithms.
 608                  */
 609                 *group = parent_group + flex_size;
 610                 if (*group > ngroups)
 611                         *group = 0;
 612                 return find_group_orlov(sb, parent, group, mode, NULL);
 613         }
 614
 615         /*
 616          * Try to place the inode in its parent directory
 617          */
 618         *group = parent_group;
 619         desc = ext4_get_group_desc(sb, *group, NULL);
 620         if (desc && ext4_free_inodes_count(sb, desc) &&
 621             ext4_free_group_clusters(sb, desc))
 622                 return 0;
 623
 624         /*
 625          * We're going to place this inode in a different blockgroup from its
 626          * parent.  We want to cause files in a common directory to all land in
 627          * the same blockgroup.  But we want files which are in a different
 628          * directory which shares a blockgroup with our parent to land in a
 629          * different blockgroup.
 630          *
 631          * So add our directory's i_ino into the starting point for the hash.
 632          */
 633         *group = (*group + parent->i_ino) % ngroups;
 634
 635         /*
 636          * Use a quadratic hash to find a group with a free inode and some free
 637          * blocks.
 638          */
 639         for (i = 1; i < ngroups; i <<= 1) {
 640                 *group += i;
 641                 if (*group >= ngroups)
 642                         *group -= ngroups;
 643                 desc = ext4_get_group_desc(sb, *group, NULL);
 644                 if (desc && ext4_free_inodes_count(sb, desc) &&
 645                     ext4_free_group_clusters(sb, desc))
 646                         return 0;
 647         }
 648
 649         /*
 650          * That failed: try linear search for a free inode, even if that group
 651          * has no free blocks.
 652          */
 653         *group = parent_group;
 654         for (i = 0; i < ngroups; i++) {
 655                 if (++*group >= ngroups)
 656                         *group = 0;
 657                 desc = ext4_get_group_desc(sb, *group, NULL);
 658                 if (desc && ext4_free_inodes_count(sb, desc))
 659                         return 0;
 660         }
 661
 662         return -1;
 663 }
 664
 665 /*
 666  * In no journal mode, if an inode has recently been deleted, we want
 667  * to avoid reusing it until we're reasonably sure the inode table
 668  * block has been written back to disk.  (Yes, these values are
 669  * somewhat arbitrary...)
 670  */
 671 #define RECENTCY_MIN    5
 672 #define RECENTCY_DIRTY  30
 673
 674 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 675 {
 676         struct ext4_group_desc  *gdp;
 677         struct ext4_inode       *raw_inode;
 678         struct buffer_head      *bh;
 679         unsigned long           dtime, now;
 680         int     inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 681         int     offset, ret = 0, recentcy = RECENTCY_MIN;
 682
 683         gdp = ext4_get_group_desc(sb, group, NULL);
 684         if (unlikely(!gdp))
 685                 return 0;
 686
 687         bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
 688                        (ino / inodes_per_block));
 689         if (unlikely(!bh) || !buffer_uptodate(bh))
 690                 /*
 691                  * If the block is not in the buffer cache, then it
 692                  * must have been written out.
 693                  */
 694                 goto out;
 695
 696         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 697         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 698         dtime = le32_to_cpu(raw_inode->i_dtime);
 699         now = get_seconds();
 700         if (buffer_dirty(bh))
 701                 recentcy += RECENTCY_DIRTY;
 702
 703         if (dtime && (dtime < now) && (now < dtime + recentcy))
 704                 ret = 1;
 705 out:
 706         brelse(bh);
 707         return ret;
 708 }
 709
 710 /*
 711  * There are two policies for allocating an inode.  If the new inode is
 712  * a directory, then a forward search is made for a block group with both
 713  * free space and a low directory-to-inode ratio; if that fails, then of
 714  * the groups with above-average free space, that group with the fewest
 715  * directories already is chosen.
 716  *
 717  * For other inodes, search forward from the parent directory's block
 718  * group to find a free inode.
 719  */
 720 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 721                                umode_t mode, const struct qstr *qstr,
 722                                __u32 goal, uid_t *owner, int handle_type,
 723                                unsigned int line_no, int nblocks)
 724 {
 725         struct super_block *sb;
 726         struct buffer_head *inode_bitmap_bh = NULL;
 727         struct buffer_head *group_desc_bh;
 728         ext4_group_t ngroups, group = 0;
 729         unsigned long ino = 0;
 730         struct inode *inode;
 731         struct ext4_group_desc *gdp = NULL;
 732         struct ext4_inode_info *ei;
 733         struct ext4_sb_info *sbi;
 734         int ret2, err;
 735         struct inode *ret;
 736         ext4_group_t i;
 737         ext4_group_t flex_group;
 738         struct ext4_group_info *grp;
 739         int encrypt = 0;
 740
 741         /* Cannot create files in a deleted directory */
 742         if (!dir || !dir->i_nlink)
 743                 return ERR_PTR(-EPERM);
 744
 745         if ((ext4_encrypted_inode(dir) ||
 746              DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
 747             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
 748                 err = ext4_get_encryption_info(dir);
 749                 if (err)
 750                         return ERR_PTR(err);
 751                 if (ext4_encryption_info(dir) == NULL)
 752                         return ERR_PTR(-EPERM);
 753                 if (!handle)
 754                         nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
 755                 encrypt = 1;
 756         }
 757
 758         sb = dir->i_sb;
 759         ngroups = ext4_get_groups_count(sb);
 760         trace_ext4_request_inode(dir, mode);
 761         inode = new_inode(sb);
 762         if (!inode)
 763                 return ERR_PTR(-ENOMEM);
 764         ei = EXT4_I(inode);
 765         sbi = EXT4_SB(sb);
 766
 767         /*
 768          * Initalize owners and quota early so that we don't have to account
 769          * for quota initialization worst case in standard inode creating
 770          * transaction
 771          */
 772         if (owner) {
 773                 inode->i_mode = mode;
 774                 i_uid_write(inode, owner[0]);
 775                 i_gid_write(inode, owner[1]);
 776         } else if (test_opt(sb, GRPID)) {
 777                 inode->i_mode = mode;
 778                 inode->i_uid = current_fsuid();
 779                 inode->i_gid = dir->i_gid;
 780         } else
 781                 inode_init_owner(inode, dir, mode);
 782         err = dquot_initialize(inode);
 783         if (err)
 784                 goto out;
 785
 786         if (!goal)
 787                 goal = sbi->s_inode_goal;
 788
 789         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 790                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 791                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 792                 ret2 = 0;
 793                 goto got_group;
 794         }
 795
 796         if (S_ISDIR(mode))
 797                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 798         else
 799                 ret2 = find_group_other(sb, dir, &group, mode);
 800
 801 got_group:
 802         EXT4_I(dir)->i_last_alloc_group = group;
 803         err = -ENOSPC;
 804         if (ret2 == -1)
 805                 goto out;
 806
 807         /*
 808          * Normally we will only go through one pass of this loop,
 809          * unless we get unlucky and it turns out the group we selected
 810          * had its last inode grabbed by someone else.
 811          */
 812         for (i = 0; i < ngroups; i++, ino = 0) {
 813                 err = -EIO;
 814
 815                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 816                 if (!gdp)
 817                         goto out;
 818
 819                 /*
 820                  * Check free inodes count before loading bitmap.
 821                  */
 822                 if (ext4_free_inodes_count(sb, gdp) == 0) {
 823                         if (++group == ngroups)
 824                                 group = 0;
 825                         continue;
 826                 }
 827
 828                 grp = ext4_get_group_info(sb, group);
 829                 /* Skip groups with already-known suspicious inode tables */
 830                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 831                         if (++group == ngroups)
 832                                 group = 0;
 833                         continue;
 834                 }
 835
 836                 brelse(inode_bitmap_bh);
 837                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 838                 /* Skip groups with suspicious inode tables */
 839                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 840                     IS_ERR(inode_bitmap_bh)) {
 841                         inode_bitmap_bh = NULL;
 842                         if (++group == ngroups)
 843                                 group = 0;
 844                         continue;
 845                 }
 846
 847 repeat_in_this_group:
 848                 ino = ext4_find_next_zero_bit((unsigned long *)
 849                                               inode_bitmap_bh->b_data,
 850                                               EXT4_INODES_PER_GROUP(sb), ino);
 851                 if (ino >= EXT4_INODES_PER_GROUP(sb))
 852                         goto next_group;
 853                 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
 854                         ext4_error(sb, "reserved inode found cleared - "
 855                                    "inode=%lu", ino + 1);
 856                         continue;
 857                 }
 858                 if ((EXT4_SB(sb)->s_journal == NULL) &&
 859                     recently_deleted(sb, group, ino)) {
 860                         ino++;
 861                         goto next_inode;
 862                 }
 863                 if (!handle) {
 864                         BUG_ON(nblocks <= 0);
 865                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 866                                                          handle_type, nblocks,
 867                                                          0);
 868                         if (IS_ERR(handle)) {
 869                                 err = PTR_ERR(handle);
 870                                 ext4_std_error(sb, err);
 871                                 goto out;
 872                         }
 873                 }
 874                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 875                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 876                 if (err) {
 877                         ext4_std_error(sb, err);
 878                         goto out;
 879                 }
 880                 ext4_lock_group(sb, group);
 881                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 882                 ext4_unlock_group(sb, group);
 883                 ino++;          /* the inode bitmap is zero-based */
 884                 if (!ret2)
 885                         goto got; /* we grabbed the inode! */
 886 next_inode:
 887                 if (ino < EXT4_INODES_PER_GROUP(sb))
 888                         goto repeat_in_this_group;
 889 next_group:
 890                 if (++group == ngroups)
 891                         group = 0;
 892         }
 893         err = -ENOSPC;
 894         goto out;
 895
 896 got:
 897         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 898         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 899         if (err) {
 900                 ext4_std_error(sb, err);
 901                 goto out;
 902         }
 903
 904         BUFFER_TRACE(group_desc_bh, "get_write_access");
 905         err = ext4_journal_get_write_access(handle, group_desc_bh);
 906         if (err) {
 907                 ext4_std_error(sb, err);
 908                 goto out;
 909         }
 910
 911         /* We may have to initialize the block bitmap if it isn't already */
 912         if (ext4_has_group_desc_csum(sb) &&
 913             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 914                 struct buffer_head *block_bitmap_bh;
 915
 916                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 917                 if (IS_ERR(block_bitmap_bh)) {
 918                         err = PTR_ERR(block_bitmap_bh);
 919                         goto out;
 920                 }
 921                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 922                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 923                 if (err) {
 924                         brelse(block_bitmap_bh);
 925                         ext4_std_error(sb, err);
 926                         goto out;
 927                 }
 928
 929                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 930                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 931
 932                 /* recheck and clear flag under lock if we still need to */
 933                 ext4_lock_group(sb, group);
 934                 if (ext4_has_group_desc_csum(sb) &&
 935                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 936                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 937                         ext4_free_group_clusters_set(sb, gdp,
 938                                 ext4_free_clusters_after_init(sb, group, gdp));
 939                         ext4_block_bitmap_csum_set(sb, group, gdp,
 940                                                    block_bitmap_bh);
 941                         ext4_group_desc_csum_set(sb, group, gdp);
 942                 }
 943                 ext4_unlock_group(sb, group);
 944                 brelse(block_bitmap_bh);
 945
 946                 if (err) {
 947                         ext4_std_error(sb, err);
 948                         goto out;
 949                 }
 950         }
 951
 952         /* Update the relevant bg descriptor fields */
 953         if (ext4_has_group_desc_csum(sb)) {
 954                 int free;
 955                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 956
 957                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
 958                 ext4_lock_group(sb, group); /* while we modify the bg desc */
 959                 free = EXT4_INODES_PER_GROUP(sb) -
 960                         ext4_itable_unused_count(sb, gdp);
 961                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 962                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 963                         free = 0;
 964                 }
 965                 /*
 966                  * Check the relative inode number against the last used
 967                  * relative inode number in this group. if it is greater
 968                  * we need to update the bg_itable_unused count
 969                  */
 970                 if (ino > free)
 971                         ext4_itable_unused_set(sb, gdp,
 972                                         (EXT4_INODES_PER_GROUP(sb) - ino));
 973                 up_read(&grp->alloc_sem);
 974         } else {
 975                 ext4_lock_group(sb, group);
 976         }
 977
 978         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
 979         if (S_ISDIR(mode)) {
 980                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
 981                 if (sbi->s_log_groups_per_flex) {
 982                         ext4_group_t f = ext4_flex_group(sbi, group);
 983
 984                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
 985                 }
 986         }
 987         if (ext4_has_group_desc_csum(sb)) {
 988                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
 989                                            EXT4_INODES_PER_GROUP(sb) / 8);
 990                 ext4_group_desc_csum_set(sb, group, gdp);
 991         }
 992         ext4_unlock_group(sb, group);
 993
 994         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 995         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 996         if (err) {
 997                 ext4_std_error(sb, err);
 998                 goto out;
 999         }
1000
1001         percpu_counter_dec(&sbi->s_freeinodes_counter);
1002         if (S_ISDIR(mode))
1003                 percpu_counter_inc(&sbi->s_dirs_counter);
1004
1005         if (sbi->s_log_groups_per_flex) {
1006                 flex_group = ext4_flex_group(sbi, group);
1007                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1008         }
1009
1010         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1011         /* This is the optimal IO size (for stat), not the fs block size */
1012         inode->i_blocks = 0;
1013         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1014                                                        ext4_current_time(inode);
1015
1016         memset(ei->i_data, 0, sizeof(ei->i_data));
1017         ei->i_dir_start_lookup = 0;
1018         ei->i_disksize = 0;
1019
1020         /* Don't inherit extent flag from directory, amongst others. */
1021         ei->i_flags =
1022                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1023         ei->i_file_acl = 0;
1024         ei->i_dtime = 0;
1025         ei->i_block_group = group;
1026         ei->i_last_alloc_group = ~0;
1027
1028         ext4_set_inode_flags(inode);
1029         if (IS_DIRSYNC(inode))
1030                 ext4_handle_sync(handle);
1031         if (insert_inode_locked(inode) < 0) {
1032                 /*
1033                  * Likely a bitmap corruption causing inode to be allocated
1034                  * twice.
1035                  */
1036                 err = -EIO;
1037                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1038                            inode->i_ino);
1039                 goto out;
1040         }
1041         spin_lock(&sbi->s_next_gen_lock);
1042         inode->i_generation = sbi->s_next_generation++;
1043         spin_unlock(&sbi->s_next_gen_lock);
1044
1045         /* Precompute checksum seed for inode metadata */
1046         if (ext4_has_metadata_csum(sb)) {
1047                 __u32 csum;
1048                 __le32 inum = cpu_to_le32(inode->i_ino);
1049                 __le32 gen = cpu_to_le32(inode->i_generation);
1050                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1051                                    sizeof(inum));
1052                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1053                                               sizeof(gen));
1054         }
1055
1056         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1057         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1058
1059         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1060         ei->i_inline_off = 0;
1061         if (ext4_has_feature_inline_data(sb))
1062                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1063         ret = inode;
1064         err = dquot_alloc_inode(inode);
1065         if (err)
1066                 goto fail_drop;
1067
1068         err = ext4_init_acl(handle, inode, dir);
1069         if (err)
1070                 goto fail_free_drop;
1071
1072         err = ext4_init_security(handle, inode, dir, qstr);
1073         if (err)
1074                 goto fail_free_drop;
1075
1076         if (ext4_has_feature_extents(sb)) {
1077                 /* set extent flag only for directory, file and normal symlink*/
1078                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1079                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1080                         ext4_ext_tree_init(handle, inode);
1081                 }
1082         }
1083
1084         if (ext4_handle_valid(handle)) {
1085                 ei->i_sync_tid = handle->h_transaction->t_tid;
1086                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1087         }
1088
1089         if (encrypt) {
1090                 err = ext4_inherit_context(dir, inode);
1091                 if (err)
1092                         goto fail_free_drop;
1093         }
1094
1095         err = ext4_mark_inode_dirty(handle, inode);
1096         if (err) {
1097                 ext4_std_error(sb, err);
1098                 goto fail_free_drop;
1099         }
1100
1101         ext4_debug("allocating inode %lu\n", inode->i_ino);
1102         trace_ext4_allocate_inode(inode, dir, mode);
1103         brelse(inode_bitmap_bh);
1104         return ret;
1105
1106 fail_free_drop:
1107         dquot_free_inode(inode);
1108 fail_drop:
1109         clear_nlink(inode);
1110         unlock_new_inode(inode);
1111 out:
1112         dquot_drop(inode);
1113         inode->i_flags |= S_NOQUOTA;
1114         iput(inode);
1115         brelse(inode_bitmap_bh);
1116         return ERR_PTR(err);
1117 }
1118
1119 /* Verify that we are loading a valid orphan from disk */
1120 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1121 {
1122         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1123         ext4_group_t block_group;
1124         int bit;
1125         struct buffer_head *bitmap_bh = NULL;
1126         struct inode *inode = NULL;
1127         int err = -EFSCORRUPTED;
1128
1129         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1130                 goto bad_orphan;
1131
1132         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1133         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1134         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1135         if (IS_ERR(bitmap_bh)) {
1136                 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1137                            ino, PTR_ERR(bitmap_bh));
1138                 return (struct inode *) bitmap_bh;
1139         }
1140
1141         /* Having the inode bit set should be a 100% indicator that this
1142          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1143          * inodes that were being truncated, so we can't check i_nlink==0.
1144          */
1145         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1146                 goto bad_orphan;
1147
1148         inode = ext4_iget(sb, ino);
1149         if (IS_ERR(inode)) {
1150                 err = PTR_ERR(inode);
1151                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1152                            ino, err);
1153                 return inode;
1154         }
1155
1156         /*
1157          * If the orphans has i_nlinks > 0 then it should be able to
1158          * be truncated, otherwise it won't be removed from the orphan
1159          * list during processing and an infinite loop will result.
1160          * Similarly, it must not be a bad inode.
1161          */
1162         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1163             is_bad_inode(inode))
1164                 goto bad_orphan;
1165
1166         if (NEXT_ORPHAN(inode) > max_ino)
1167                 goto bad_orphan;
1168         brelse(bitmap_bh);
1169         return inode;
1170
1171 bad_orphan:
1172         ext4_error(sb, "bad orphan inode %lu", ino);
1173         if (bitmap_bh)
1174                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1175                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1176                        ext4_test_bit(bit, bitmap_bh->b_data));
1177         if (inode) {
1178                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1179                        is_bad_inode(inode));
1180                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1181                        NEXT_ORPHAN(inode));
1182                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1183                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1184                 /* Avoid freeing blocks if we got a bad deleted inode */
1185                 if (inode->i_nlink == 0)
1186                         inode->i_blocks = 0;
1187                 iput(inode);
1188         }
1189         brelse(bitmap_bh);
1190         return ERR_PTR(err);
1191 }
1192
1193 unsigned long ext4_count_free_inodes(struct super_block *sb)
1194 {
1195         unsigned long desc_count;
1196         struct ext4_group_desc *gdp;
1197         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1198 #ifdef EXT4FS_DEBUG
1199         struct ext4_super_block *es;
1200         unsigned long bitmap_count, x;
1201         struct buffer_head *bitmap_bh = NULL;
1202
1203         es = EXT4_SB(sb)->s_es;
1204         desc_count = 0;
1205         bitmap_count = 0;
1206         gdp = NULL;
1207         for (i = 0; i < ngroups; i++) {
1208                 gdp = ext4_get_group_desc(sb, i, NULL);
1209                 if (!gdp)
1210                         continue;
1211                 desc_count += ext4_free_inodes_count(sb, gdp);
1212                 brelse(bitmap_bh);
1213                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1214                 if (IS_ERR(bitmap_bh)) {
1215                         bitmap_bh = NULL;
1216                         continue;
1217                 }
1218
1219                 x = ext4_count_free(bitmap_bh->b_data,
1220                                     EXT4_INODES_PER_GROUP(sb) / 8);
1221                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1222                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1223                 bitmap_count += x;
1224         }
1225         brelse(bitmap_bh);
1226         printk(KERN_DEBUG "ext4_count_free_inodes: "
1227                "stored = %u, computed = %lu, %lu\n",
1228                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1229         return desc_count;
1230 #else
1231         desc_count = 0;
1232         for (i = 0; i < ngroups; i++) {
1233                 gdp = ext4_get_group_desc(sb, i, NULL);
1234                 if (!gdp)
1235                         continue;
1236                 desc_count += ext4_free_inodes_count(sb, gdp);
1237                 cond_resched();
1238         }
1239         return desc_count;
1240 #endif
1241 }
1242
1243 /* Called at mount-time, super-block is locked */
1244 unsigned long ext4_count_dirs(struct super_block * sb)
1245 {
1246         unsigned long count = 0;
1247         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1248
1249         for (i = 0; i < ngroups; i++) {
1250                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1251                 if (!gdp)
1252                         continue;
1253                 count += ext4_used_dirs_count(sb, gdp);
1254         }
1255         return count;
1256 }
1257
1258 /*
1259  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1260  * inode table. Must be called without any spinlock held. The only place
1261  * where it is called from on active part of filesystem is ext4lazyinit
1262  * thread, so we do not need any special locks, however we have to prevent
1263  * inode allocation from the current group, so we take alloc_sem lock, to
1264  * block ext4_new_inode() until we are finished.
1265  */
1266 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1267                                  int barrier)
1268 {
1269         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1270         struct ext4_sb_info *sbi = EXT4_SB(sb);
1271         struct ext4_group_desc *gdp = NULL;
1272         struct buffer_head *group_desc_bh;
1273         handle_t *handle;
1274         ext4_fsblk_t blk;
1275         int num, ret = 0, used_blks = 0;
1276
1277         /* This should not happen, but just to be sure check this */
1278         if (sb->s_flags & MS_RDONLY) {
1279                 ret = 1;
1280                 goto out;
1281         }
1282
1283         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1284         if (!gdp)
1285                 goto out;
1286
1287         /*
1288          * We do not need to lock this, because we are the only one
1289          * handling this flag.
1290          */
1291         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1292                 goto out;
1293
1294         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1295         if (IS_ERR(handle)) {
1296                 ret = PTR_ERR(handle);
1297                 goto out;
1298         }
1299
1300         down_write(&grp->alloc_sem);
1301         /*
1302          * If inode bitmap was already initialized there may be some
1303          * used inodes so we need to skip blocks with used inodes in
1304          * inode table.
1305          */
1306         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1307                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1308                             ext4_itable_unused_count(sb, gdp)),
1309                             sbi->s_inodes_per_block);
1310
1311         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1312             ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1313                                ext4_itable_unused_count(sb, gdp)) <
1314                               EXT4_FIRST_INO(sb)))) {
1315                 ext4_error(sb, "Something is wrong with group %u: "
1316                            "used itable blocks: %d; "
1317                            "itable unused count: %u",
1318                            group, used_blks,
1319                            ext4_itable_unused_count(sb, gdp));
1320                 ret = 1;
1321                 goto err_out;
1322         }
1323
1324         blk = ext4_inode_table(sb, gdp) + used_blks;
1325         num = sbi->s_itb_per_group - used_blks;
1326
1327         BUFFER_TRACE(group_desc_bh, "get_write_access");
1328         ret = ext4_journal_get_write_access(handle,
1329                                             group_desc_bh);
1330         if (ret)
1331                 goto err_out;
1332
1333         /*
1334          * Skip zeroout if the inode table is full. But we set the ZEROED
1335          * flag anyway, because obviously, when it is full it does not need
1336          * further zeroing.
1337          */
1338         if (unlikely(num == 0))
1339                 goto skip_zeroout;
1340
1341         ext4_debug("going to zero out inode table in group %d\n",
1342                    group);
1343         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1344         if (ret < 0)
1345                 goto err_out;
1346         if (barrier)
1347                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1348
1349 skip_zeroout:
1350         ext4_lock_group(sb, group);
1351         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1352         ext4_group_desc_csum_set(sb, group, gdp);
1353         ext4_unlock_group(sb, group);
1354
1355         BUFFER_TRACE(group_desc_bh,
1356                      "call ext4_handle_dirty_metadata");
1357         ret = ext4_handle_dirty_metadata(handle, NULL,
1358                                          group_desc_bh);
1359
1360 err_out:
1361         up_write(&grp->alloc_sem);
1362         ext4_journal_stop(handle);
1363 out:
1364         return ret;
1365 }