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                 struct flex_groups *fg;
 335
 336                 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
 337                                          ext4_flex_group(sbi, block_group));
 338                 atomic_inc(&fg->free_inodes);
 339                 if (is_directory)
 340                         atomic_dec(&fg->used_dirs);
 341         }
 342         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 343         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 344 out:
 345         if (cleared) {
 346                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 347                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 348                 if (!fatal)
 349                         fatal = err;
 350         } else {
 351                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 352                 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 353                         int count;
 354                         count = ext4_free_inodes_count(sb, gdp);
 355                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 356                                            count);
 357                 }
 358                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 359         }
 360
 361 error_return:
 362         brelse(bitmap_bh);
 363         ext4_std_error(sb, fatal);
 364 }
 365
 366 struct orlov_stats {
 367         __u64 free_clusters;
 368         __u32 free_inodes;
 369         __u32 used_dirs;
 370 };
 371
 372 /*
 373  * Helper function for Orlov's allocator; returns critical information
 374  * for a particular block group or flex_bg.  If flex_size is 1, then g
 375  * is a block group number; otherwise it is flex_bg number.
 376  */
 377 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 378                             int flex_size, struct orlov_stats *stats)
 379 {
 380         struct ext4_group_desc *desc;
 381
 382         if (flex_size > 1) {
 383                 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
 384                                                              s_flex_groups, g);
 385                 stats->free_inodes = atomic_read(&fg->free_inodes);
 386                 stats->free_clusters = atomic64_read(&fg->free_clusters);
 387                 stats->used_dirs = atomic_read(&fg->used_dirs);
 388                 return;
 389         }
 390
 391         desc = ext4_get_group_desc(sb, g, NULL);
 392         if (desc) {
 393                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 394                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 395                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 396         } else {
 397                 stats->free_inodes = 0;
 398                 stats->free_clusters = 0;
 399                 stats->used_dirs = 0;
 400         }
 401 }
 402
 403 /*
 404  * Orlov's allocator for directories.
 405  *
 406  * We always try to spread first-level directories.
 407  *
 408  * If there are blockgroups with both free inodes and free clusters counts
 409  * not worse than average we return one with smallest directory count.
 410  * Otherwise we simply return a random group.
 411  *
 412  * For the rest rules look so:
 413  *
 414  * It's OK to put directory into a group unless
 415  * it has too many directories already (max_dirs) or
 416  * it has too few free inodes left (min_inodes) or
 417  * it has too few free clusters left (min_clusters) or
 418  * Parent's group is preferred, if it doesn't satisfy these
 419  * conditions we search cyclically through the rest. If none
 420  * of the groups look good we just look for a group with more
 421  * free inodes than average (starting at parent's group).
 422  */
 423
 424 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 425                             ext4_group_t *group, umode_t mode,
 426                             const struct qstr *qstr)
 427 {
 428         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 429         struct ext4_sb_info *sbi = EXT4_SB(sb);
 430         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 431         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 432         unsigned int freei, avefreei, grp_free;
 433         ext4_fsblk_t freec, avefreec;
 434         unsigned int ndirs;
 435         int max_dirs, min_inodes;
 436         ext4_grpblk_t min_clusters;
 437         ext4_group_t i, grp, g, ngroups;
 438         struct ext4_group_desc *desc;
 439         struct orlov_stats stats;
 440         int flex_size = ext4_flex_bg_size(sbi);
 441         struct dx_hash_info hinfo;
 442
 443         ngroups = real_ngroups;
 444         if (flex_size > 1) {
 445                 ngroups = (real_ngroups + flex_size - 1) >>
 446                         sbi->s_log_groups_per_flex;
 447                 parent_group >>= sbi->s_log_groups_per_flex;
 448         }
 449
 450         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 451         avefreei = freei / ngroups;
 452         freec = percpu_counter_read_positive(&sbi->s_freeclusters_counter);
 453         avefreec = freec;
 454         do_div(avefreec, ngroups);
 455         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 456
 457         if (S_ISDIR(mode) &&
 458             ((parent == d_inode(sb->s_root)) ||
 459              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 460                 int best_ndir = inodes_per_group;
 461                 int ret = -1;
 462
 463                 if (qstr) {
 464                         hinfo.hash_version = DX_HASH_HALF_MD4;
 465                         hinfo.seed = sbi->s_hash_seed;
 466                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 467                         grp = hinfo.hash;
 468                 } else
 469                         grp = prandom_u32();
 470                 parent_group = (unsigned)grp % ngroups;
 471                 for (i = 0; i < ngroups; i++) {
 472                         g = (parent_group + i) % ngroups;
 473                         get_orlov_stats(sb, g, flex_size, &stats);
 474                         if (!stats.free_inodes)
 475                                 continue;
 476                         if (stats.used_dirs >= best_ndir)
 477                                 continue;
 478                         if (stats.free_inodes < avefreei)
 479                                 continue;
 480                         if (stats.free_clusters < avefreec)
 481                                 continue;
 482                         grp = g;
 483                         ret = 0;
 484                         best_ndir = stats.used_dirs;
 485                 }
 486                 if (ret)
 487                         goto fallback;
 488         found_flex_bg:
 489                 if (flex_size == 1) {
 490                         *group = grp;
 491                         return 0;
 492                 }
 493
 494                 /*
 495                  * We pack inodes at the beginning of the flexgroup's
 496                  * inode tables.  Block allocation decisions will do
 497                  * something similar, although regular files will
 498                  * start at 2nd block group of the flexgroup.  See
 499                  * ext4_ext_find_goal() and ext4_find_near().
 500                  */
 501                 grp *= flex_size;
 502                 for (i = 0; i < flex_size; i++) {
 503                         if (grp+i >= real_ngroups)
 504                                 break;
 505                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 506                         if (desc && ext4_free_inodes_count(sb, desc)) {
 507                                 *group = grp+i;
 508                                 return 0;
 509                         }
 510                 }
 511                 goto fallback;
 512         }
 513
 514         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 515         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 516         if (min_inodes < 1)
 517                 min_inodes = 1;
 518         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 519
 520         /*
 521          * Start looking in the flex group where we last allocated an
 522          * inode for this parent directory
 523          */
 524         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 525                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 526                 if (flex_size > 1)
 527                         parent_group >>= sbi->s_log_groups_per_flex;
 528         }
 529
 530         for (i = 0; i < ngroups; i++) {
 531                 grp = (parent_group + i) % ngroups;
 532                 get_orlov_stats(sb, grp, flex_size, &stats);
 533                 if (stats.used_dirs >= max_dirs)
 534                         continue;
 535                 if (stats.free_inodes < min_inodes)
 536                         continue;
 537                 if (stats.free_clusters < min_clusters)
 538                         continue;
 539                 goto found_flex_bg;
 540         }
 541
 542 fallback:
 543         ngroups = real_ngroups;
 544         avefreei = freei / ngroups;
 545 fallback_retry:
 546         parent_group = EXT4_I(parent)->i_block_group;
 547         for (i = 0; i < ngroups; i++) {
 548                 grp = (parent_group + i) % ngroups;
 549                 desc = ext4_get_group_desc(sb, grp, NULL);
 550                 if (desc) {
 551                         grp_free = ext4_free_inodes_count(sb, desc);
 552                         if (grp_free && grp_free >= avefreei) {
 553                                 *group = grp;
 554                                 return 0;
 555                         }
 556                 }
 557         }
 558
 559         if (avefreei) {
 560                 /*
 561                  * The free-inodes counter is approximate, and for really small
 562                  * filesystems the above test can fail to find any blockgroups
 563                  */
 564                 avefreei = 0;
 565                 goto fallback_retry;
 566         }
 567
 568         return -1;
 569 }
 570
 571 static int find_group_other(struct super_block *sb, struct inode *parent,
 572                             ext4_group_t *group, umode_t mode)
 573 {
 574         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 575         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 576         struct ext4_group_desc *desc;
 577         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 578
 579         /*
 580          * Try to place the inode is the same flex group as its
 581          * parent.  If we can't find space, use the Orlov algorithm to
 582          * find another flex group, and store that information in the
 583          * parent directory's inode information so that use that flex
 584          * group for future allocations.
 585          */
 586         if (flex_size > 1) {
 587                 int retry = 0;
 588
 589         try_again:
 590                 parent_group &= ~(flex_size-1);
 591                 last = parent_group + flex_size;
 592                 if (last > ngroups)
 593                         last = ngroups;
 594                 for  (i = parent_group; i < last; i++) {
 595                         desc = ext4_get_group_desc(sb, i, NULL);
 596                         if (desc && ext4_free_inodes_count(sb, desc)) {
 597                                 *group = i;
 598                                 return 0;
 599                         }
 600                 }
 601                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 602                         retry = 1;
 603                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 604                         goto try_again;
 605                 }
 606                 /*
 607                  * If this didn't work, use the Orlov search algorithm
 608                  * to find a new flex group; we pass in the mode to
 609                  * avoid the topdir algorithms.
 610                  */
 611                 *group = parent_group + flex_size;
 612                 if (*group > ngroups)
 613                         *group = 0;
 614                 return find_group_orlov(sb, parent, group, mode, NULL);
 615         }
 616
 617         /*
 618          * Try to place the inode in its parent directory
 619          */
 620         *group = parent_group;
 621         desc = ext4_get_group_desc(sb, *group, NULL);
 622         if (desc && ext4_free_inodes_count(sb, desc) &&
 623             ext4_free_group_clusters(sb, desc))
 624                 return 0;
 625
 626         /*
 627          * We're going to place this inode in a different blockgroup from its
 628          * parent.  We want to cause files in a common directory to all land in
 629          * the same blockgroup.  But we want files which are in a different
 630          * directory which shares a blockgroup with our parent to land in a
 631          * different blockgroup.
 632          *
 633          * So add our directory's i_ino into the starting point for the hash.
 634          */
 635         *group = (*group + parent->i_ino) % ngroups;
 636
 637         /*
 638          * Use a quadratic hash to find a group with a free inode and some free
 639          * blocks.
 640          */
 641         for (i = 1; i < ngroups; i <<= 1) {
 642                 *group += i;
 643                 if (*group >= ngroups)
 644                         *group -= ngroups;
 645                 desc = ext4_get_group_desc(sb, *group, NULL);
 646                 if (desc && ext4_free_inodes_count(sb, desc) &&
 647                     ext4_free_group_clusters(sb, desc))
 648                         return 0;
 649         }
 650
 651         /*
 652          * That failed: try linear search for a free inode, even if that group
 653          * has no free blocks.
 654          */
 655         *group = parent_group;
 656         for (i = 0; i < ngroups; i++) {
 657                 if (++*group >= ngroups)
 658                         *group = 0;
 659                 desc = ext4_get_group_desc(sb, *group, NULL);
 660                 if (desc && ext4_free_inodes_count(sb, desc))
 661                         return 0;
 662         }
 663
 664         return -1;
 665 }
 666
 667 /*
 668  * In no journal mode, if an inode has recently been deleted, we want
 669  * to avoid reusing it until we're reasonably sure the inode table
 670  * block has been written back to disk.  (Yes, these values are
 671  * somewhat arbitrary...)
 672  */
 673 #define RECENTCY_MIN    5
 674 #define RECENTCY_DIRTY  30
 675
 676 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 677 {
 678         struct ext4_group_desc  *gdp;
 679         struct ext4_inode       *raw_inode;
 680         struct buffer_head      *bh;
 681         unsigned long           dtime, now;
 682         int     inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 683         int     offset, ret = 0, recentcy = RECENTCY_MIN;
 684
 685         gdp = ext4_get_group_desc(sb, group, NULL);
 686         if (unlikely(!gdp))
 687                 return 0;
 688
 689         bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
 690                        (ino / inodes_per_block));
 691         if (unlikely(!bh) || !buffer_uptodate(bh))
 692                 /*
 693                  * If the block is not in the buffer cache, then it
 694                  * must have been written out.
 695                  */
 696                 goto out;
 697
 698         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 699         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 700         dtime = le32_to_cpu(raw_inode->i_dtime);
 701         now = get_seconds();
 702         if (buffer_dirty(bh))
 703                 recentcy += RECENTCY_DIRTY;
 704
 705         if (dtime && (dtime < now) && (now < dtime + recentcy))
 706                 ret = 1;
 707 out:
 708         brelse(bh);
 709         return ret;
 710 }
 711
 712 /*
 713  * There are two policies for allocating an inode.  If the new inode is
 714  * a directory, then a forward search is made for a block group with both
 715  * free space and a low directory-to-inode ratio; if that fails, then of
 716  * the groups with above-average free space, that group with the fewest
 717  * directories already is chosen.
 718  *
 719  * For other inodes, search forward from the parent directory's block
 720  * group to find a free inode.
 721  */
 722 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 723                                umode_t mode, const struct qstr *qstr,
 724                                __u32 goal, uid_t *owner, int handle_type,
 725                                unsigned int line_no, int nblocks)
 726 {
 727         struct super_block *sb;
 728         struct buffer_head *inode_bitmap_bh = NULL;
 729         struct buffer_head *group_desc_bh;
 730         ext4_group_t ngroups, group = 0;
 731         unsigned long ino = 0;
 732         struct inode *inode;
 733         struct ext4_group_desc *gdp = NULL;
 734         struct ext4_inode_info *ei;
 735         struct ext4_sb_info *sbi;
 736         int ret2, err;
 737         struct inode *ret;
 738         ext4_group_t i;
 739         ext4_group_t flex_group;
 740         struct ext4_group_info *grp;
 741         int encrypt = 0;
 742
 743         /* Cannot create files in a deleted directory */
 744         if (!dir || !dir->i_nlink)
 745                 return ERR_PTR(-EPERM);
 746
 747         if ((ext4_encrypted_inode(dir) ||
 748              DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
 749             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
 750                 err = ext4_get_encryption_info(dir);
 751                 if (err)
 752                         return ERR_PTR(err);
 753                 if (ext4_encryption_info(dir) == NULL)
 754                         return ERR_PTR(-EPERM);
 755                 if (!handle)
 756                         nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
 757                 encrypt = 1;
 758         }
 759
 760         sb = dir->i_sb;
 761         ngroups = ext4_get_groups_count(sb);
 762         trace_ext4_request_inode(dir, mode);
 763         inode = new_inode(sb);
 764         if (!inode)
 765                 return ERR_PTR(-ENOMEM);
 766         ei = EXT4_I(inode);
 767         sbi = EXT4_SB(sb);
 768
 769         /*
 770          * Initalize owners and quota early so that we don't have to account
 771          * for quota initialization worst case in standard inode creating
 772          * transaction
 773          */
 774         if (owner) {
 775                 inode->i_mode = mode;
 776                 i_uid_write(inode, owner[0]);
 777                 i_gid_write(inode, owner[1]);
 778         } else if (test_opt(sb, GRPID)) {
 779                 inode->i_mode = mode;
 780                 inode->i_uid = current_fsuid();
 781                 inode->i_gid = dir->i_gid;
 782         } else
 783                 inode_init_owner(inode, dir, mode);
 784         err = dquot_initialize(inode);
 785         if (err)
 786                 goto out;
 787
 788         if (!goal)
 789                 goal = sbi->s_inode_goal;
 790
 791         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 792                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 793                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 794                 ret2 = 0;
 795                 goto got_group;
 796         }
 797
 798         if (S_ISDIR(mode))
 799                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 800         else
 801                 ret2 = find_group_other(sb, dir, &group, mode);
 802
 803 got_group:
 804         EXT4_I(dir)->i_last_alloc_group = group;
 805         err = -ENOSPC;
 806         if (ret2 == -1)
 807                 goto out;
 808
 809         /*
 810          * Normally we will only go through one pass of this loop,
 811          * unless we get unlucky and it turns out the group we selected
 812          * had its last inode grabbed by someone else.
 813          */
 814         for (i = 0; i < ngroups; i++, ino = 0) {
 815                 err = -EIO;
 816
 817                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 818                 if (!gdp)
 819                         goto out;
 820
 821                 /*
 822                  * Check free inodes count before loading bitmap.
 823                  */
 824                 if (ext4_free_inodes_count(sb, gdp) == 0) {
 825                         if (++group == ngroups)
 826                                 group = 0;
 827                         continue;
 828                 }
 829
 830                 grp = ext4_get_group_info(sb, group);
 831                 /* Skip groups with already-known suspicious inode tables */
 832                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 833                         if (++group == ngroups)
 834                                 group = 0;
 835                         continue;
 836                 }
 837
 838                 brelse(inode_bitmap_bh);
 839                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 840                 /* Skip groups with suspicious inode tables */
 841                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 842                     IS_ERR(inode_bitmap_bh)) {
 843                         inode_bitmap_bh = NULL;
 844                         if (++group == ngroups)
 845                                 group = 0;
 846                         continue;
 847                 }
 848
 849 repeat_in_this_group:
 850                 ino = ext4_find_next_zero_bit((unsigned long *)
 851                                               inode_bitmap_bh->b_data,
 852                                               EXT4_INODES_PER_GROUP(sb), ino);
 853                 if (ino >= EXT4_INODES_PER_GROUP(sb))
 854                         goto next_group;
 855                 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
 856                         ext4_error(sb, "reserved inode found cleared - "
 857                                    "inode=%lu", ino + 1);
 858                         continue;
 859                 }
 860                 if ((EXT4_SB(sb)->s_journal == NULL) &&
 861                     recently_deleted(sb, group, ino)) {
 862                         ino++;
 863                         goto next_inode;
 864                 }
 865                 if (!handle) {
 866                         BUG_ON(nblocks <= 0);
 867                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 868                                                          handle_type, nblocks,
 869                                                          0);
 870                         if (IS_ERR(handle)) {
 871                                 err = PTR_ERR(handle);
 872                                 ext4_std_error(sb, err);
 873                                 goto out;
 874                         }
 875                 }
 876                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 877                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 878                 if (err) {
 879                         ext4_std_error(sb, err);
 880                         goto out;
 881                 }
 882                 ext4_lock_group(sb, group);
 883                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 884                 ext4_unlock_group(sb, group);
 885                 ino++;          /* the inode bitmap is zero-based */
 886                 if (!ret2)
 887                         goto got; /* we grabbed the inode! */
 888 next_inode:
 889                 if (ino < EXT4_INODES_PER_GROUP(sb))
 890                         goto repeat_in_this_group;
 891 next_group:
 892                 if (++group == ngroups)
 893                         group = 0;
 894         }
 895         err = -ENOSPC;
 896         goto out;
 897
 898 got:
 899         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 900         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 901         if (err) {
 902                 ext4_std_error(sb, err);
 903                 goto out;
 904         }
 905
 906         BUFFER_TRACE(group_desc_bh, "get_write_access");
 907         err = ext4_journal_get_write_access(handle, group_desc_bh);
 908         if (err) {
 909                 ext4_std_error(sb, err);
 910                 goto out;
 911         }
 912
 913         /* We may have to initialize the block bitmap if it isn't already */
 914         if (ext4_has_group_desc_csum(sb) &&
 915             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 916                 struct buffer_head *block_bitmap_bh;
 917
 918                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 919                 if (IS_ERR(block_bitmap_bh)) {
 920                         err = PTR_ERR(block_bitmap_bh);
 921                         goto out;
 922                 }
 923                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 924                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 925                 if (err) {
 926                         brelse(block_bitmap_bh);
 927                         ext4_std_error(sb, err);
 928                         goto out;
 929                 }
 930
 931                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 932                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 933
 934                 /* recheck and clear flag under lock if we still need to */
 935                 ext4_lock_group(sb, group);
 936                 if (ext4_has_group_desc_csum(sb) &&
 937                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 938                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 939                         ext4_free_group_clusters_set(sb, gdp,
 940                                 ext4_free_clusters_after_init(sb, group, gdp));
 941                         ext4_block_bitmap_csum_set(sb, group, gdp,
 942                                                    block_bitmap_bh);
 943                         ext4_group_desc_csum_set(sb, group, gdp);
 944                 }
 945                 ext4_unlock_group(sb, group);
 946                 brelse(block_bitmap_bh);
 947
 948                 if (err) {
 949                         ext4_std_error(sb, err);
 950                         goto out;
 951                 }
 952         }
 953
 954         /* Update the relevant bg descriptor fields */
 955         if (ext4_has_group_desc_csum(sb)) {
 956                 int free;
 957                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 958
 959                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
 960                 ext4_lock_group(sb, group); /* while we modify the bg desc */
 961                 free = EXT4_INODES_PER_GROUP(sb) -
 962                         ext4_itable_unused_count(sb, gdp);
 963                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 964                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 965                         free = 0;
 966                 }
 967                 /*
 968                  * Check the relative inode number against the last used
 969                  * relative inode number in this group. if it is greater
 970                  * we need to update the bg_itable_unused count
 971                  */
 972                 if (ino > free)
 973                         ext4_itable_unused_set(sb, gdp,
 974                                         (EXT4_INODES_PER_GROUP(sb) - ino));
 975                 up_read(&grp->alloc_sem);
 976         } else {
 977                 ext4_lock_group(sb, group);
 978         }
 979
 980         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
 981         if (S_ISDIR(mode)) {
 982                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
 983                 if (sbi->s_log_groups_per_flex) {
 984                         ext4_group_t f = ext4_flex_group(sbi, group);
 985
 986                         atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
 987                                                         f)->used_dirs);
 988                 }
 989         }
 990         if (ext4_has_group_desc_csum(sb)) {
 991                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
 992                                            EXT4_INODES_PER_GROUP(sb) / 8);
 993                 ext4_group_desc_csum_set(sb, group, gdp);
 994         }
 995         ext4_unlock_group(sb, group);
 996
 997         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 998         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 999         if (err) {
1000                 ext4_std_error(sb, err);
1001                 goto out;
1002         }
1003
1004         percpu_counter_dec(&sbi->s_freeinodes_counter);
1005         if (S_ISDIR(mode))
1006                 percpu_counter_inc(&sbi->s_dirs_counter);
1007
1008         if (sbi->s_log_groups_per_flex) {
1009                 flex_group = ext4_flex_group(sbi, group);
1010                 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1011                                                 flex_group)->free_inodes);
1012         }
1013
1014         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1015         /* This is the optimal IO size (for stat), not the fs block size */
1016         inode->i_blocks = 0;
1017         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1018                                                        ext4_current_time(inode);
1019
1020         memset(ei->i_data, 0, sizeof(ei->i_data));
1021         ei->i_dir_start_lookup = 0;
1022         ei->i_disksize = 0;
1023
1024         /* Don't inherit extent flag from directory, amongst others. */
1025         ei->i_flags =
1026                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1027         ei->i_file_acl = 0;
1028         ei->i_dtime = 0;
1029         ei->i_block_group = group;
1030         ei->i_last_alloc_group = ~0;
1031
1032         ext4_set_inode_flags(inode);
1033         if (IS_DIRSYNC(inode))
1034                 ext4_handle_sync(handle);
1035         if (insert_inode_locked(inode) < 0) {
1036                 /*
1037                  * Likely a bitmap corruption causing inode to be allocated
1038                  * twice.
1039                  */
1040                 err = -EIO;
1041                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1042                            inode->i_ino);
1043                 goto out;
1044         }
1045         spin_lock(&sbi->s_next_gen_lock);
1046         inode->i_generation = sbi->s_next_generation++;
1047         spin_unlock(&sbi->s_next_gen_lock);
1048
1049         /* Precompute checksum seed for inode metadata */
1050         if (ext4_has_metadata_csum(sb)) {
1051                 __u32 csum;
1052                 __le32 inum = cpu_to_le32(inode->i_ino);
1053                 __le32 gen = cpu_to_le32(inode->i_generation);
1054                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1055                                    sizeof(inum));
1056                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1057                                               sizeof(gen));
1058         }
1059
1060         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1061         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1062
1063         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1064         ei->i_inline_off = 0;
1065         if (ext4_has_feature_inline_data(sb))
1066                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1067         ret = inode;
1068         err = dquot_alloc_inode(inode);
1069         if (err)
1070                 goto fail_drop;
1071
1072         err = ext4_init_acl(handle, inode, dir);
1073         if (err)
1074                 goto fail_free_drop;
1075
1076         err = ext4_init_security(handle, inode, dir, qstr);
1077         if (err)
1078                 goto fail_free_drop;
1079
1080         if (ext4_has_feature_extents(sb)) {
1081                 /* set extent flag only for directory, file and normal symlink*/
1082                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1083                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1084                         ext4_ext_tree_init(handle, inode);
1085                 }
1086         }
1087
1088         if (ext4_handle_valid(handle)) {
1089                 ei->i_sync_tid = handle->h_transaction->t_tid;
1090                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1091         }
1092
1093         if (encrypt) {
1094                 err = ext4_inherit_context(dir, inode);
1095                 if (err)
1096                         goto fail_free_drop;
1097         }
1098
1099         err = ext4_mark_inode_dirty(handle, inode);
1100         if (err) {
1101                 ext4_std_error(sb, err);
1102                 goto fail_free_drop;
1103         }
1104
1105         ext4_debug("allocating inode %lu\n", inode->i_ino);
1106         trace_ext4_allocate_inode(inode, dir, mode);
1107         brelse(inode_bitmap_bh);
1108         return ret;
1109
1110 fail_free_drop:
1111         dquot_free_inode(inode);
1112 fail_drop:
1113         clear_nlink(inode);
1114         unlock_new_inode(inode);
1115 out:
1116         dquot_drop(inode);
1117         inode->i_flags |= S_NOQUOTA;
1118         iput(inode);
1119         brelse(inode_bitmap_bh);
1120         return ERR_PTR(err);
1121 }
1122
1123 /* Verify that we are loading a valid orphan from disk */
1124 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1125 {
1126         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1127         ext4_group_t block_group;
1128         int bit;
1129         struct buffer_head *bitmap_bh = NULL;
1130         struct inode *inode = NULL;
1131         int err = -EFSCORRUPTED;
1132
1133         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1134                 goto bad_orphan;
1135
1136         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1137         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1138         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1139         if (IS_ERR(bitmap_bh)) {
1140                 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1141                            ino, PTR_ERR(bitmap_bh));
1142                 return (struct inode *) bitmap_bh;
1143         }
1144
1145         /* Having the inode bit set should be a 100% indicator that this
1146          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1147          * inodes that were being truncated, so we can't check i_nlink==0.
1148          */
1149         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1150                 goto bad_orphan;
1151
1152         inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1153         if (IS_ERR(inode)) {
1154                 err = PTR_ERR(inode);
1155                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1156                            ino, err);
1157                 return inode;
1158         }
1159
1160         /*
1161          * If the orphans has i_nlinks > 0 then it should be able to
1162          * be truncated, otherwise it won't be removed from the orphan
1163          * list during processing and an infinite loop will result.
1164          * Similarly, it must not be a bad inode.
1165          */
1166         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1167             is_bad_inode(inode))
1168                 goto bad_orphan;
1169
1170         if (NEXT_ORPHAN(inode) > max_ino)
1171                 goto bad_orphan;
1172         brelse(bitmap_bh);
1173         return inode;
1174
1175 bad_orphan:
1176         ext4_error(sb, "bad orphan inode %lu", ino);
1177         if (bitmap_bh)
1178                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1179                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1180                        ext4_test_bit(bit, bitmap_bh->b_data));
1181         if (inode) {
1182                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1183                        is_bad_inode(inode));
1184                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1185                        NEXT_ORPHAN(inode));
1186                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1187                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1188                 /* Avoid freeing blocks if we got a bad deleted inode */
1189                 if (inode->i_nlink == 0)
1190                         inode->i_blocks = 0;
1191                 iput(inode);
1192         }
1193         brelse(bitmap_bh);
1194         return ERR_PTR(err);
1195 }
1196
1197 unsigned long ext4_count_free_inodes(struct super_block *sb)
1198 {
1199         unsigned long desc_count;
1200         struct ext4_group_desc *gdp;
1201         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1202 #ifdef EXT4FS_DEBUG
1203         struct ext4_super_block *es;
1204         unsigned long bitmap_count, x;
1205         struct buffer_head *bitmap_bh = NULL;
1206
1207         es = EXT4_SB(sb)->s_es;
1208         desc_count = 0;
1209         bitmap_count = 0;
1210         gdp = NULL;
1211         for (i = 0; i < ngroups; i++) {
1212                 gdp = ext4_get_group_desc(sb, i, NULL);
1213                 if (!gdp)
1214                         continue;
1215                 desc_count += ext4_free_inodes_count(sb, gdp);
1216                 brelse(bitmap_bh);
1217                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1218                 if (IS_ERR(bitmap_bh)) {
1219                         bitmap_bh = NULL;
1220                         continue;
1221                 }
1222
1223                 x = ext4_count_free(bitmap_bh->b_data,
1224                                     EXT4_INODES_PER_GROUP(sb) / 8);
1225                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1226                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1227                 bitmap_count += x;
1228         }
1229         brelse(bitmap_bh);
1230         printk(KERN_DEBUG "ext4_count_free_inodes: "
1231                "stored = %u, computed = %lu, %lu\n",
1232                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1233         return desc_count;
1234 #else
1235         desc_count = 0;
1236         for (i = 0; i < ngroups; i++) {
1237                 gdp = ext4_get_group_desc(sb, i, NULL);
1238                 if (!gdp)
1239                         continue;
1240                 desc_count += ext4_free_inodes_count(sb, gdp);
1241                 cond_resched();
1242         }
1243         return desc_count;
1244 #endif
1245 }
1246
1247 /* Called at mount-time, super-block is locked */
1248 unsigned long ext4_count_dirs(struct super_block * sb)
1249 {
1250         unsigned long count = 0;
1251         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1252
1253         for (i = 0; i < ngroups; i++) {
1254                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1255                 if (!gdp)
1256                         continue;
1257                 count += ext4_used_dirs_count(sb, gdp);
1258         }
1259         return count;
1260 }
1261
1262 /*
1263  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1264  * inode table. Must be called without any spinlock held. The only place
1265  * where it is called from on active part of filesystem is ext4lazyinit
1266  * thread, so we do not need any special locks, however we have to prevent
1267  * inode allocation from the current group, so we take alloc_sem lock, to
1268  * block ext4_new_inode() until we are finished.
1269  */
1270 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1271                                  int barrier)
1272 {
1273         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1274         struct ext4_sb_info *sbi = EXT4_SB(sb);
1275         struct ext4_group_desc *gdp = NULL;
1276         struct buffer_head *group_desc_bh;
1277         handle_t *handle;
1278         ext4_fsblk_t blk;
1279         int num, ret = 0, used_blks = 0;
1280         unsigned long used_inos = 0;
1281
1282         /* This should not happen, but just to be sure check this */
1283         if (sb->s_flags & MS_RDONLY) {
1284                 ret = 1;
1285                 goto out;
1286         }
1287
1288         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1289         if (!gdp)
1290                 goto out;
1291
1292         /*
1293          * We do not need to lock this, because we are the only one
1294          * handling this flag.
1295          */
1296         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1297                 goto out;
1298
1299         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1300         if (IS_ERR(handle)) {
1301                 ret = PTR_ERR(handle);
1302                 goto out;
1303         }
1304
1305         down_write(&grp->alloc_sem);
1306         /*
1307          * If inode bitmap was already initialized there may be some
1308          * used inodes so we need to skip blocks with used inodes in
1309          * inode table.
1310          */
1311         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
1312                 used_inos = EXT4_INODES_PER_GROUP(sb) -
1313                             ext4_itable_unused_count(sb, gdp);
1314                 used_blks = DIV_ROUND_UP(used_inos, sbi->s_inodes_per_block);
1315
1316                 /* Bogus inode unused count? */
1317                 if (used_blks < 0 || used_blks > sbi->s_itb_per_group) {
1318                         ext4_error(sb, "Something is wrong with group %u: "
1319                                    "used itable blocks: %d; "
1320                                    "itable unused count: %u",
1321                                    group, used_blks,
1322                                    ext4_itable_unused_count(sb, gdp));
1323                         ret = 1;
1324                         goto err_out;
1325                 }
1326
1327                 used_inos += group * EXT4_INODES_PER_GROUP(sb);
1328                 /*
1329                  * Are there some uninitialized inodes in the inode table
1330                  * before the first normal inode?
1331                  */
1332                 if ((used_blks != sbi->s_itb_per_group) &&
1333                      (used_inos < EXT4_FIRST_INO(sb))) {
1334                         ext4_error(sb, "Something is wrong with group %u: "
1335                                    "itable unused count: %u; "
1336                                    "itables initialized count: %ld",
1337                                    group, ext4_itable_unused_count(sb, gdp),
1338                                    used_inos);
1339                         ret = 1;
1340                         goto err_out;
1341                 }
1342         }
1343
1344         blk = ext4_inode_table(sb, gdp) + used_blks;
1345         num = sbi->s_itb_per_group - used_blks;
1346
1347         BUFFER_TRACE(group_desc_bh, "get_write_access");
1348         ret = ext4_journal_get_write_access(handle,
1349                                             group_desc_bh);
1350         if (ret)
1351                 goto err_out;
1352
1353         /*
1354          * Skip zeroout if the inode table is full. But we set the ZEROED
1355          * flag anyway, because obviously, when it is full it does not need
1356          * further zeroing.
1357          */
1358         if (unlikely(num == 0))
1359                 goto skip_zeroout;
1360
1361         ext4_debug("going to zero out inode table in group %d\n",
1362                    group);
1363         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1364         if (ret < 0)
1365                 goto err_out;
1366         if (barrier)
1367                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1368
1369 skip_zeroout:
1370         ext4_lock_group(sb, group);
1371         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1372         ext4_group_desc_csum_set(sb, group, gdp);
1373         ext4_unlock_group(sb, group);
1374
1375         BUFFER_TRACE(group_desc_bh,
1376                      "call ext4_handle_dirty_metadata");
1377         ret = ext4_handle_dirty_metadata(handle, NULL,
1378                                          group_desc_bh);
1379
1380 err_out:
1381         up_write(&grp->alloc_sem);
1382         ext4_journal_stop(handle);
1383 out:
1384         return ret;
1385 }