fs/ext4/ialloc.c

   1 /*
   2  *  linux/fs/ext4/ialloc.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  BSD ufs-inspired inode and directory allocation by
  10  *  Stephen Tweedie (sct@redhat.com), 1993
  11  *  Big-endian to little-endian byte-swapping/bitmaps by
  12  *        David S. Miller (davem@caip.rutgers.edu), 1995
  13  */
  14
  15 #include <linux/time.h>
  16 #include <linux/fs.h>
  17 #include <linux/jbd2.h>
  18 #include <linux/stat.h>
  19 #include <linux/string.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/random.h>
  23 #include <linux/bitops.h>
  24 #include <linux/blkdev.h>
  25 #include <asm/byteorder.h>
  26
  27 #include "ext4.h"
  28 #include "ext4_jbd2.h"
  29 #include "xattr.h"
  30 #include "acl.h"
  31
  32 #include <trace/events/ext4.h>
  33
  34 /*
  35  * ialloc.c contains the inodes allocation and deallocation routines
  36  */
  37
  38 /*
  39  * The free inodes are managed by bitmaps.  A file system contains several
  40  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  41  * block for inodes, N blocks for the inode table and data blocks.
  42  *
  43  * The file system contains group descriptors which are located after the
  44  * super block.  Each descriptor contains the number of the bitmap block and
  45  * the free blocks count in the block.
  46  */
  47
  48 /*
  49  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  50  * need to use it within a single byte (to ensure we get endianness right).
  51  * We can use memset for the rest of the bitmap as there are no other users.
  52  */
  53 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  54 {
  55         int i;
  56
  57         if (start_bit >= end_bit)
  58                 return;
  59
  60         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  61         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  62                 ext4_set_bit(i, bitmap);
  63         if (i < end_bit)
  64                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  65 }
  66
  67 /* Initializes an uninitialized inode bitmap */
  68 static unsigned ext4_init_inode_bitmap(struct super_block *sb,
  69                                        struct buffer_head *bh,
  70                                        ext4_group_t block_group,
  71                                        struct ext4_group_desc *gdp)
  72 {
  73         struct ext4_group_info *grp;
  74         struct ext4_sb_info *sbi = EXT4_SB(sb);
  75         J_ASSERT_BH(bh, buffer_locked(bh));
  76
  77         /* If checksum is bad mark all blocks and inodes use to prevent
  78          * allocation, essentially implementing a per-group read-only flag. */
  79         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
  80                 ext4_error(sb, "Checksum bad for group %u", block_group);
  81                 grp = ext4_get_group_info(sb, block_group);
  82                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
  83                         percpu_counter_sub(&sbi->s_freeclusters_counter,
  84                                            grp->bb_free);
  85                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
  86                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
  87                         int count;
  88                         count = ext4_free_inodes_count(sb, gdp);
  89                         percpu_counter_sub(&sbi->s_freeinodes_counter,
  90                                            count);
  91                 }
  92                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
  93                 return 0;
  94         }
  95
  96         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  97         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  98                         bh->b_data);
  99         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
 100                                    EXT4_INODES_PER_GROUP(sb) / 8);
 101         ext4_group_desc_csum_set(sb, block_group, gdp);
 102
 103         return EXT4_INODES_PER_GROUP(sb);
 104 }
 105
 106 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
 107 {
 108         if (uptodate) {
 109                 set_buffer_uptodate(bh);
 110                 set_bitmap_uptodate(bh);
 111         }
 112         unlock_buffer(bh);
 113         put_bh(bh);
 114 }
 115
 116 /*
 117  * Read the inode allocation bitmap for a given block_group, reading
 118  * into the specified slot in the superblock's bitmap cache.
 119  *
 120  * Return buffer_head of bitmap on success or NULL.
 121  */
 122 static struct buffer_head *
 123 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 124 {
 125         struct ext4_group_desc *desc;
 126         struct buffer_head *bh = NULL;
 127         ext4_fsblk_t bitmap_blk;
 128         struct ext4_group_info *grp;
 129         struct ext4_sb_info *sbi = EXT4_SB(sb);
 130
 131         desc = ext4_get_group_desc(sb, block_group, NULL);
 132         if (!desc)
 133                 return NULL;
 134
 135         bitmap_blk = ext4_inode_bitmap(sb, desc);
 136         bh = sb_getblk(sb, bitmap_blk);
 137         if (unlikely(!bh)) {
 138                 ext4_error(sb, "Cannot read inode bitmap - "
 139                             "block_group = %u, inode_bitmap = %llu",
 140                             block_group, bitmap_blk);
 141                 return NULL;
 142         }
 143         if (bitmap_uptodate(bh))
 144                 goto verify;
 145
 146         lock_buffer(bh);
 147         if (bitmap_uptodate(bh)) {
 148                 unlock_buffer(bh);
 149                 goto verify;
 150         }
 151
 152         ext4_lock_group(sb, block_group);
 153         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 154                 ext4_init_inode_bitmap(sb, bh, block_group, desc);
 155                 set_bitmap_uptodate(bh);
 156                 set_buffer_uptodate(bh);
 157                 set_buffer_verified(bh);
 158                 ext4_unlock_group(sb, block_group);
 159                 unlock_buffer(bh);
 160                 return bh;
 161         }
 162         ext4_unlock_group(sb, block_group);
 163
 164         if (buffer_uptodate(bh)) {
 165                 /*
 166                  * if not uninit if bh is uptodate,
 167                  * bitmap is also uptodate
 168                  */
 169                 set_bitmap_uptodate(bh);
 170                 unlock_buffer(bh);
 171                 goto verify;
 172         }
 173         /*
 174          * submit the buffer_head for reading
 175          */
 176         trace_ext4_load_inode_bitmap(sb, block_group);
 177         bh->b_end_io = ext4_end_bitmap_read;
 178         get_bh(bh);
 179         submit_bh(READ | REQ_META | REQ_PRIO, bh);
 180         wait_on_buffer(bh);
 181         if (!buffer_uptodate(bh)) {
 182                 put_bh(bh);
 183                 ext4_error(sb, "Cannot read inode bitmap - "
 184                            "block_group = %u, inode_bitmap = %llu",
 185                            block_group, bitmap_blk);
 186                 return NULL;
 187         }
 188
 189 verify:
 190         ext4_lock_group(sb, block_group);
 191         if (!buffer_verified(bh) &&
 192             !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
 193                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
 194                 ext4_unlock_group(sb, block_group);
 195                 put_bh(bh);
 196                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 197                            "inode_bitmap = %llu", block_group, bitmap_blk);
 198                 grp = ext4_get_group_info(sb, block_group);
 199                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 200                         int count;
 201                         count = ext4_free_inodes_count(sb, desc);
 202                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 203                                            count);
 204                 }
 205                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 206                 return NULL;
 207         }
 208         ext4_unlock_group(sb, block_group);
 209         set_buffer_verified(bh);
 210         return bh;
 211 }
 212
 213 /*
 214  * NOTE! When we get the inode, we're the only people
 215  * that have access to it, and as such there are no
 216  * race conditions we have to worry about. The inode
 217  * is not on the hash-lists, and it cannot be reached
 218  * through the filesystem because the directory entry
 219  * has been deleted earlier.
 220  *
 221  * HOWEVER: we must make sure that we get no aliases,
 222  * which means that we have to call "clear_inode()"
 223  * _before_ we mark the inode not in use in the inode
 224  * bitmaps. Otherwise a newly created file might use
 225  * the same inode number (not actually the same pointer
 226  * though), and then we'd have two inodes sharing the
 227  * same inode number and space on the harddisk.
 228  */
 229 void ext4_free_inode(handle_t *handle, struct inode *inode)
 230 {
 231         struct super_block *sb = inode->i_sb;
 232         int is_directory;
 233         unsigned long ino;
 234         struct buffer_head *bitmap_bh = NULL;
 235         struct buffer_head *bh2;
 236         ext4_group_t block_group;
 237         unsigned long bit;
 238         struct ext4_group_desc *gdp;
 239         struct ext4_super_block *es;
 240         struct ext4_sb_info *sbi;
 241         int fatal = 0, err, count, cleared;
 242         struct ext4_group_info *grp;
 243
 244         if (!sb) {
 245                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 246                        "nonexistent device\n", __func__, __LINE__);
 247                 return;
 248         }
 249         if (atomic_read(&inode->i_count) > 1) {
 250                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 251                          __func__, __LINE__, inode->i_ino,
 252                          atomic_read(&inode->i_count));
 253                 return;
 254         }
 255         if (inode->i_nlink) {
 256                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 257                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 258                 return;
 259         }
 260         sbi = EXT4_SB(sb);
 261
 262         ino = inode->i_ino;
 263         ext4_debug("freeing inode %lu\n", ino);
 264         trace_ext4_free_inode(inode);
 265
 266         /*
 267          * Note: we must free any quota before locking the superblock,
 268          * as writing the quota to disk may need the lock as well.
 269          */
 270         dquot_initialize(inode);
 271         ext4_xattr_delete_inode(handle, inode);
 272         dquot_free_inode(inode);
 273         dquot_drop(inode);
 274
 275         is_directory = S_ISDIR(inode->i_mode);
 276
 277         /* Do this BEFORE marking the inode not in use or returning an error */
 278         ext4_clear_inode(inode);
 279
 280         es = EXT4_SB(sb)->s_es;
 281         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 282                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 283                 goto error_return;
 284         }
 285         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 286         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 287         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 288         /* Don't bother if the inode bitmap is corrupt. */
 289         grp = ext4_get_group_info(sb, block_group);
 290         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh)
 291                 goto error_return;
 292
 293         BUFFER_TRACE(bitmap_bh, "get_write_access");
 294         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 295         if (fatal)
 296                 goto error_return;
 297
 298         fatal = -ESRCH;
 299         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 300         if (gdp) {
 301                 BUFFER_TRACE(bh2, "get_write_access");
 302                 fatal = ext4_journal_get_write_access(handle, bh2);
 303         }
 304         ext4_lock_group(sb, block_group);
 305         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 306         if (fatal || !cleared) {
 307                 ext4_unlock_group(sb, block_group);
 308                 goto out;
 309         }
 310
 311         count = ext4_free_inodes_count(sb, gdp) + 1;
 312         ext4_free_inodes_set(sb, gdp, count);
 313         if (is_directory) {
 314                 count = ext4_used_dirs_count(sb, gdp) - 1;
 315                 ext4_used_dirs_set(sb, gdp, count);
 316                 percpu_counter_dec(&sbi->s_dirs_counter);
 317         }
 318         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 319                                    EXT4_INODES_PER_GROUP(sb) / 8);
 320         ext4_group_desc_csum_set(sb, block_group, gdp);
 321         ext4_unlock_group(sb, block_group);
 322
 323         percpu_counter_inc(&sbi->s_freeinodes_counter);
 324         if (sbi->s_log_groups_per_flex) {
 325                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 326
 327                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 328                 if (is_directory)
 329                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 330         }
 331         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 332         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 333 out:
 334         if (cleared) {
 335                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 336                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 337                 if (!fatal)
 338                         fatal = err;
 339         } else {
 340                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 341                 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 342                         int count;
 343                         count = ext4_free_inodes_count(sb, gdp);
 344                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 345                                            count);
 346                 }
 347                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 348         }
 349
 350 error_return:
 351         brelse(bitmap_bh);
 352         ext4_std_error(sb, fatal);
 353 }
 354
 355 struct orlov_stats {
 356         __u64 free_clusters;
 357         __u32 free_inodes;
 358         __u32 used_dirs;
 359 };
 360
 361 /*
 362  * Helper function for Orlov's allocator; returns critical information
 363  * for a particular block group or flex_bg.  If flex_size is 1, then g
 364  * is a block group number; otherwise it is flex_bg number.
 365  */
 366 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 367                             int flex_size, struct orlov_stats *stats)
 368 {
 369         struct ext4_group_desc *desc;
 370         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 371
 372         if (flex_size > 1) {
 373                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 374                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 375                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 376                 return;
 377         }
 378
 379         desc = ext4_get_group_desc(sb, g, NULL);
 380         if (desc) {
 381                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 382                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 383                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 384         } else {
 385                 stats->free_inodes = 0;
 386                 stats->free_clusters = 0;
 387                 stats->used_dirs = 0;
 388         }
 389 }
 390
 391 /*
 392  * Orlov's allocator for directories.
 393  *
 394  * We always try to spread first-level directories.
 395  *
 396  * If there are blockgroups with both free inodes and free blocks counts
 397  * not worse than average we return one with smallest directory count.
 398  * Otherwise we simply return a random group.
 399  *
 400  * For the rest rules look so:
 401  *
 402  * It's OK to put directory into a group unless
 403  * it has too many directories already (max_dirs) or
 404  * it has too few free inodes left (min_inodes) or
 405  * it has too few free blocks left (min_blocks) or
 406  * Parent's group is preferred, if it doesn't satisfy these
 407  * conditions we search cyclically through the rest. If none
 408  * of the groups look good we just look for a group with more
 409  * free inodes than average (starting at parent's group).
 410  */
 411
 412 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 413                             ext4_group_t *group, umode_t mode,
 414                             const struct qstr *qstr)
 415 {
 416         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 417         struct ext4_sb_info *sbi = EXT4_SB(sb);
 418         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 419         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 420         unsigned int freei, avefreei, grp_free;
 421         ext4_fsblk_t freeb, avefreec;
 422         unsigned int ndirs;
 423         int max_dirs, min_inodes;
 424         ext4_grpblk_t min_clusters;
 425         ext4_group_t i, grp, g, ngroups;
 426         struct ext4_group_desc *desc;
 427         struct orlov_stats stats;
 428         int flex_size = ext4_flex_bg_size(sbi);
 429         struct dx_hash_info hinfo;
 430
 431         ngroups = real_ngroups;
 432         if (flex_size > 1) {
 433                 ngroups = (real_ngroups + flex_size - 1) >>
 434                         sbi->s_log_groups_per_flex;
 435                 parent_group >>= sbi->s_log_groups_per_flex;
 436         }
 437
 438         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 439         avefreei = freei / ngroups;
 440         freeb = EXT4_C2B(sbi,
 441                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 442         avefreec = freeb;
 443         do_div(avefreec, ngroups);
 444         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 445
 446         if (S_ISDIR(mode) &&
 447             ((parent == sb->s_root->d_inode) ||
 448              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 449                 int best_ndir = inodes_per_group;
 450                 int ret = -1;
 451
 452                 if (qstr) {
 453                         hinfo.hash_version = DX_HASH_HALF_MD4;
 454                         hinfo.seed = sbi->s_hash_seed;
 455                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 456                         grp = hinfo.hash;
 457                 } else
 458                         grp = prandom_u32();
 459                 parent_group = (unsigned)grp % ngroups;
 460                 for (i = 0; i < ngroups; i++) {
 461                         g = (parent_group + i) % ngroups;
 462                         get_orlov_stats(sb, g, flex_size, &stats);
 463                         if (!stats.free_inodes)
 464                                 continue;
 465                         if (stats.used_dirs >= best_ndir)
 466                                 continue;
 467                         if (stats.free_inodes < avefreei)
 468                                 continue;
 469                         if (stats.free_clusters < avefreec)
 470                                 continue;
 471                         grp = g;
 472                         ret = 0;
 473                         best_ndir = stats.used_dirs;
 474                 }
 475                 if (ret)
 476                         goto fallback;
 477         found_flex_bg:
 478                 if (flex_size == 1) {
 479                         *group = grp;
 480                         return 0;
 481                 }
 482
 483                 /*
 484                  * We pack inodes at the beginning of the flexgroup's
 485                  * inode tables.  Block allocation decisions will do
 486                  * something similar, although regular files will
 487                  * start at 2nd block group of the flexgroup.  See
 488                  * ext4_ext_find_goal() and ext4_find_near().
 489                  */
 490                 grp *= flex_size;
 491                 for (i = 0; i < flex_size; i++) {
 492                         if (grp+i >= real_ngroups)
 493                                 break;
 494                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 495                         if (desc && ext4_free_inodes_count(sb, desc)) {
 496                                 *group = grp+i;
 497                                 return 0;
 498                         }
 499                 }
 500                 goto fallback;
 501         }
 502
 503         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 504         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 505         if (min_inodes < 1)
 506                 min_inodes = 1;
 507         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 508
 509         /*
 510          * Start looking in the flex group where we last allocated an
 511          * inode for this parent directory
 512          */
 513         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 514                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 515                 if (flex_size > 1)
 516                         parent_group >>= sbi->s_log_groups_per_flex;
 517         }
 518
 519         for (i = 0; i < ngroups; i++) {
 520                 grp = (parent_group + i) % ngroups;
 521                 get_orlov_stats(sb, grp, flex_size, &stats);
 522                 if (stats.used_dirs >= max_dirs)
 523                         continue;
 524                 if (stats.free_inodes < min_inodes)
 525                         continue;
 526                 if (stats.free_clusters < min_clusters)
 527                         continue;
 528                 goto found_flex_bg;
 529         }
 530
 531 fallback:
 532         ngroups = real_ngroups;
 533         avefreei = freei / ngroups;
 534 fallback_retry:
 535         parent_group = EXT4_I(parent)->i_block_group;
 536         for (i = 0; i < ngroups; i++) {
 537                 grp = (parent_group + i) % ngroups;
 538                 desc = ext4_get_group_desc(sb, grp, NULL);
 539                 if (desc) {
 540                         grp_free = ext4_free_inodes_count(sb, desc);
 541                         if (grp_free && grp_free >= avefreei) {
 542                                 *group = grp;
 543                                 return 0;
 544                         }
 545                 }
 546         }
 547
 548         if (avefreei) {
 549                 /*
 550                  * The free-inodes counter is approximate, and for really small
 551                  * filesystems the above test can fail to find any blockgroups
 552                  */
 553                 avefreei = 0;
 554                 goto fallback_retry;
 555         }
 556
 557         return -1;
 558 }
 559
 560 static int find_group_other(struct super_block *sb, struct inode *parent,
 561                             ext4_group_t *group, umode_t mode)
 562 {
 563         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 564         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 565         struct ext4_group_desc *desc;
 566         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 567
 568         /*
 569          * Try to place the inode is the same flex group as its
 570          * parent.  If we can't find space, use the Orlov algorithm to
 571          * find another flex group, and store that information in the
 572          * parent directory's inode information so that use that flex
 573          * group for future allocations.
 574          */
 575         if (flex_size > 1) {
 576                 int retry = 0;
 577
 578         try_again:
 579                 parent_group &= ~(flex_size-1);
 580                 last = parent_group + flex_size;
 581                 if (last > ngroups)
 582                         last = ngroups;
 583                 for  (i = parent_group; i < last; i++) {
 584                         desc = ext4_get_group_desc(sb, i, NULL);
 585                         if (desc && ext4_free_inodes_count(sb, desc)) {
 586                                 *group = i;
 587                                 return 0;
 588                         }
 589                 }
 590                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 591                         retry = 1;
 592                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 593                         goto try_again;
 594                 }
 595                 /*
 596                  * If this didn't work, use the Orlov search algorithm
 597                  * to find a new flex group; we pass in the mode to
 598                  * avoid the topdir algorithms.
 599                  */
 600                 *group = parent_group + flex_size;
 601                 if (*group > ngroups)
 602                         *group = 0;
 603                 return find_group_orlov(sb, parent, group, mode, NULL);
 604         }
 605
 606         /*
 607          * Try to place the inode in its parent directory
 608          */
 609         *group = parent_group;
 610         desc = ext4_get_group_desc(sb, *group, NULL);
 611         if (desc && ext4_free_inodes_count(sb, desc) &&
 612             ext4_free_group_clusters(sb, desc))
 613                 return 0;
 614
 615         /*
 616          * We're going to place this inode in a different blockgroup from its
 617          * parent.  We want to cause files in a common directory to all land in
 618          * the same blockgroup.  But we want files which are in a different
 619          * directory which shares a blockgroup with our parent to land in a
 620          * different blockgroup.
 621          *
 622          * So add our directory's i_ino into the starting point for the hash.
 623          */
 624         *group = (*group + parent->i_ino) % ngroups;
 625
 626         /*
 627          * Use a quadratic hash to find a group with a free inode and some free
 628          * blocks.
 629          */
 630         for (i = 1; i < ngroups; i <<= 1) {
 631                 *group += i;
 632                 if (*group >= ngroups)
 633                         *group -= ngroups;
 634                 desc = ext4_get_group_desc(sb, *group, NULL);
 635                 if (desc && ext4_free_inodes_count(sb, desc) &&
 636                     ext4_free_group_clusters(sb, desc))
 637                         return 0;
 638         }
 639
 640         /*
 641          * That failed: try linear search for a free inode, even if that group
 642          * has no free blocks.
 643          */
 644         *group = parent_group;
 645         for (i = 0; i < ngroups; i++) {
 646                 if (++*group >= ngroups)
 647                         *group = 0;
 648                 desc = ext4_get_group_desc(sb, *group, NULL);
 649                 if (desc && ext4_free_inodes_count(sb, desc))
 650                         return 0;
 651         }
 652
 653         return -1;
 654 }
 655
 656 /*
 657  * In no journal mode, if an inode has recently been deleted, we want
 658  * to avoid reusing it until we're reasonably sure the inode table
 659  * block has been written back to disk.  (Yes, these values are
 660  * somewhat arbitrary...)
 661  */
 662 #define RECENTCY_MIN    5
 663 #define RECENTCY_DIRTY  30
 664
 665 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 666 {
 667         struct ext4_group_desc  *gdp;
 668         struct ext4_inode       *raw_inode;
 669         struct buffer_head      *bh;
 670         unsigned long           dtime, now;
 671         int     inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 672         int     offset, ret = 0, recentcy = RECENTCY_MIN;
 673
 674         gdp = ext4_get_group_desc(sb, group, NULL);
 675         if (unlikely(!gdp))
 676                 return 0;
 677
 678         bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
 679                        (ino / inodes_per_block));
 680         if (unlikely(!bh) || !buffer_uptodate(bh))
 681                 /*
 682                  * If the block is not in the buffer cache, then it
 683                  * must have been written out.
 684                  */
 685                 goto out;
 686
 687         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 688         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 689         dtime = le32_to_cpu(raw_inode->i_dtime);
 690         now = get_seconds();
 691         if (buffer_dirty(bh))
 692                 recentcy += RECENTCY_DIRTY;
 693
 694         if (dtime && (dtime < now) && (now < dtime + recentcy))
 695                 ret = 1;
 696 out:
 697         brelse(bh);
 698         return ret;
 699 }
 700
 701 /*
 702  * There are two policies for allocating an inode.  If the new inode is
 703  * a directory, then a forward search is made for a block group with both
 704  * free space and a low directory-to-inode ratio; if that fails, then of
 705  * the groups with above-average free space, that group with the fewest
 706  * directories already is chosen.
 707  *
 708  * For other inodes, search forward from the parent directory's block
 709  * group to find a free inode.
 710  */
 711 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 712                                umode_t mode, const struct qstr *qstr,
 713                                __u32 goal, uid_t *owner, int handle_type,
 714                                unsigned int line_no, int nblocks)
 715 {
 716         struct super_block *sb;
 717         struct buffer_head *inode_bitmap_bh = NULL;
 718         struct buffer_head *group_desc_bh;
 719         ext4_group_t ngroups, group = 0;
 720         unsigned long ino = 0;
 721         struct inode *inode;
 722         struct ext4_group_desc *gdp = NULL;
 723         struct ext4_inode_info *ei;
 724         struct ext4_sb_info *sbi;
 725         int ret2, err = 0;
 726         struct inode *ret;
 727         ext4_group_t i;
 728         ext4_group_t flex_group;
 729         struct ext4_group_info *grp;
 730
 731         /* Cannot create files in a deleted directory */
 732         if (!dir || !dir->i_nlink)
 733                 return ERR_PTR(-EPERM);
 734
 735         sb = dir->i_sb;
 736         ngroups = ext4_get_groups_count(sb);
 737         trace_ext4_request_inode(dir, mode);
 738         inode = new_inode(sb);
 739         if (!inode)
 740                 return ERR_PTR(-ENOMEM);
 741         ei = EXT4_I(inode);
 742         sbi = EXT4_SB(sb);
 743
 744         /*
 745          * Initalize owners and quota early so that we don't have to account
 746          * for quota initialization worst case in standard inode creating
 747          * transaction
 748          */
 749         if (owner) {
 750                 inode->i_mode = mode;
 751                 i_uid_write(inode, owner[0]);
 752                 i_gid_write(inode, owner[1]);
 753         } else if (test_opt(sb, GRPID)) {
 754                 inode->i_mode = mode;
 755                 inode->i_uid = current_fsuid();
 756                 inode->i_gid = dir->i_gid;
 757         } else
 758                 inode_init_owner(inode, dir, mode);
 759         dquot_initialize(inode);
 760
 761         if (!goal)
 762                 goal = sbi->s_inode_goal;
 763
 764         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 765                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 766                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 767                 ret2 = 0;
 768                 goto got_group;
 769         }
 770
 771         if (S_ISDIR(mode))
 772                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 773         else
 774                 ret2 = find_group_other(sb, dir, &group, mode);
 775
 776 got_group:
 777         EXT4_I(dir)->i_last_alloc_group = group;
 778         err = -ENOSPC;
 779         if (ret2 == -1)
 780                 goto out;
 781
 782         /*
 783          * Normally we will only go through one pass of this loop,
 784          * unless we get unlucky and it turns out the group we selected
 785          * had its last inode grabbed by someone else.
 786          */
 787         for (i = 0; i < ngroups; i++, ino = 0) {
 788                 err = -EIO;
 789
 790                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 791                 if (!gdp)
 792                         goto out;
 793
 794                 /*
 795                  * Check free inodes count before loading bitmap.
 796                  */
 797                 if (ext4_free_inodes_count(sb, gdp) == 0) {
 798                         if (++group == ngroups)
 799                                 group = 0;
 800                         continue;
 801                 }
 802
 803                 grp = ext4_get_group_info(sb, group);
 804                 /* Skip groups with already-known suspicious inode tables */
 805                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 806                         if (++group == ngroups)
 807                                 group = 0;
 808                         continue;
 809                 }
 810
 811                 brelse(inode_bitmap_bh);
 812                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 813                 /* Skip groups with suspicious inode tables */
 814                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) {
 815                         if (++group == ngroups)
 816                                 group = 0;
 817                         continue;
 818                 }
 819
 820 repeat_in_this_group:
 821                 ino = ext4_find_next_zero_bit((unsigned long *)
 822                                               inode_bitmap_bh->b_data,
 823                                               EXT4_INODES_PER_GROUP(sb), ino);
 824                 if (ino >= EXT4_INODES_PER_GROUP(sb))
 825                         goto next_group;
 826                 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
 827                         ext4_error(sb, "reserved inode found cleared - "
 828                                    "inode=%lu", ino + 1);
 829                         continue;
 830                 }
 831                 if ((EXT4_SB(sb)->s_journal == NULL) &&
 832                     recently_deleted(sb, group, ino)) {
 833                         ino++;
 834                         goto next_inode;
 835                 }
 836                 if (!handle) {
 837                         BUG_ON(nblocks <= 0);
 838                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 839                                                          handle_type, nblocks,
 840                                                          0);
 841                         if (IS_ERR(handle)) {
 842                                 err = PTR_ERR(handle);
 843                                 ext4_std_error(sb, err);
 844                                 goto out;
 845                         }
 846                 }
 847                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 848                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 849                 if (err) {
 850                         ext4_std_error(sb, err);
 851                         goto out;
 852                 }
 853                 ext4_lock_group(sb, group);
 854                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 855                 ext4_unlock_group(sb, group);
 856                 ino++;          /* the inode bitmap is zero-based */
 857                 if (!ret2)
 858                         goto got; /* we grabbed the inode! */
 859 next_inode:
 860                 if (ino < EXT4_INODES_PER_GROUP(sb))
 861                         goto repeat_in_this_group;
 862 next_group:
 863                 if (++group == ngroups)
 864                         group = 0;
 865         }
 866         err = -ENOSPC;
 867         goto out;
 868
 869 got:
 870         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 871         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 872         if (err) {
 873                 ext4_std_error(sb, err);
 874                 goto out;
 875         }
 876
 877         BUFFER_TRACE(group_desc_bh, "get_write_access");
 878         err = ext4_journal_get_write_access(handle, group_desc_bh);
 879         if (err) {
 880                 ext4_std_error(sb, err);
 881                 goto out;
 882         }
 883
 884         /* We may have to initialize the block bitmap if it isn't already */
 885         if (ext4_has_group_desc_csum(sb) &&
 886             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 887                 struct buffer_head *block_bitmap_bh;
 888
 889                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 890                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 891                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 892                 if (err) {
 893                         brelse(block_bitmap_bh);
 894                         ext4_std_error(sb, err);
 895                         goto out;
 896                 }
 897
 898                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 899                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 900
 901                 /* recheck and clear flag under lock if we still need to */
 902                 ext4_lock_group(sb, group);
 903                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 904                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 905                         ext4_free_group_clusters_set(sb, gdp,
 906                                 ext4_free_clusters_after_init(sb, group, gdp));
 907                         ext4_block_bitmap_csum_set(sb, group, gdp,
 908                                                    block_bitmap_bh);
 909                         ext4_group_desc_csum_set(sb, group, gdp);
 910                 }
 911                 ext4_unlock_group(sb, group);
 912                 brelse(block_bitmap_bh);
 913
 914                 if (err) {
 915                         ext4_std_error(sb, err);
 916                         goto out;
 917                 }
 918         }
 919
 920         /* Update the relevant bg descriptor fields */
 921         if (ext4_has_group_desc_csum(sb)) {
 922                 int free;
 923                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 924
 925                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
 926                 ext4_lock_group(sb, group); /* while we modify the bg desc */
 927                 free = EXT4_INODES_PER_GROUP(sb) -
 928                         ext4_itable_unused_count(sb, gdp);
 929                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 930                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 931                         free = 0;
 932                 }
 933                 /*
 934                  * Check the relative inode number against the last used
 935                  * relative inode number in this group. if it is greater
 936                  * we need to update the bg_itable_unused count
 937                  */
 938                 if (ino > free)
 939                         ext4_itable_unused_set(sb, gdp,
 940                                         (EXT4_INODES_PER_GROUP(sb) - ino));
 941                 up_read(&grp->alloc_sem);
 942         } else {
 943                 ext4_lock_group(sb, group);
 944         }
 945
 946         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
 947         if (S_ISDIR(mode)) {
 948                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
 949                 if (sbi->s_log_groups_per_flex) {
 950                         ext4_group_t f = ext4_flex_group(sbi, group);
 951
 952                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
 953                 }
 954         }
 955         if (ext4_has_group_desc_csum(sb)) {
 956                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
 957                                            EXT4_INODES_PER_GROUP(sb) / 8);
 958                 ext4_group_desc_csum_set(sb, group, gdp);
 959         }
 960         ext4_unlock_group(sb, group);
 961
 962         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 963         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 964         if (err) {
 965                 ext4_std_error(sb, err);
 966                 goto out;
 967         }
 968
 969         percpu_counter_dec(&sbi->s_freeinodes_counter);
 970         if (S_ISDIR(mode))
 971                 percpu_counter_inc(&sbi->s_dirs_counter);
 972
 973         if (sbi->s_log_groups_per_flex) {
 974                 flex_group = ext4_flex_group(sbi, group);
 975                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
 976         }
 977
 978         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
 979         /* This is the optimal IO size (for stat), not the fs block size */
 980         inode->i_blocks = 0;
 981         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
 982                                                        ext4_current_time(inode);
 983
 984         memset(ei->i_data, 0, sizeof(ei->i_data));
 985         ei->i_dir_start_lookup = 0;
 986         ei->i_disksize = 0;
 987
 988         /* Don't inherit extent flag from directory, amongst others. */
 989         ei->i_flags =
 990                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
 991         ei->i_file_acl = 0;
 992         ei->i_dtime = 0;
 993         ei->i_block_group = group;
 994         ei->i_last_alloc_group = ~0;
 995
 996         ext4_set_inode_flags(inode);
 997         if (IS_DIRSYNC(inode))
 998                 ext4_handle_sync(handle);
 999         if (insert_inode_locked(inode) < 0) {
1000                 /*
1001                  * Likely a bitmap corruption causing inode to be allocated
1002                  * twice.
1003                  */
1004                 err = -EIO;
1005                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1006                            inode->i_ino);
1007                 goto out;
1008         }
1009         spin_lock(&sbi->s_next_gen_lock);
1010         inode->i_generation = sbi->s_next_generation++;
1011         spin_unlock(&sbi->s_next_gen_lock);
1012
1013         /* Precompute checksum seed for inode metadata */
1014         if (ext4_has_metadata_csum(sb)) {
1015                 __u32 csum;
1016                 __le32 inum = cpu_to_le32(inode->i_ino);
1017                 __le32 gen = cpu_to_le32(inode->i_generation);
1018                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1019                                    sizeof(inum));
1020                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1021                                               sizeof(gen));
1022         }
1023
1024         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1025         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1026
1027         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1028
1029         ei->i_inline_off = 0;
1030         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
1031                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1032
1033         ret = inode;
1034         err = dquot_alloc_inode(inode);
1035         if (err)
1036                 goto fail_drop;
1037
1038         err = ext4_init_acl(handle, inode, dir);
1039         if (err)
1040                 goto fail_free_drop;
1041
1042         err = ext4_init_security(handle, inode, dir, qstr);
1043         if (err)
1044                 goto fail_free_drop;
1045
1046         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1047                 /* set extent flag only for directory, file and normal symlink*/
1048                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1049                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1050                         ext4_ext_tree_init(handle, inode);
1051                 }
1052         }
1053
1054         if (ext4_handle_valid(handle)) {
1055                 ei->i_sync_tid = handle->h_transaction->t_tid;
1056                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1057         }
1058
1059         err = ext4_mark_inode_dirty(handle, inode);
1060         if (err) {
1061                 ext4_std_error(sb, err);
1062                 goto fail_free_drop;
1063         }
1064
1065         ext4_debug("allocating inode %lu\n", inode->i_ino);
1066         trace_ext4_allocate_inode(inode, dir, mode);
1067         brelse(inode_bitmap_bh);
1068         return ret;
1069
1070 fail_free_drop:
1071         dquot_free_inode(inode);
1072 fail_drop:
1073         clear_nlink(inode);
1074         unlock_new_inode(inode);
1075 out:
1076         dquot_drop(inode);
1077         inode->i_flags |= S_NOQUOTA;
1078         iput(inode);
1079         brelse(inode_bitmap_bh);
1080         return ERR_PTR(err);
1081 }
1082
1083 /* Verify that we are loading a valid orphan from disk */
1084 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1085 {
1086         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1087         ext4_group_t block_group;
1088         int bit;
1089         struct buffer_head *bitmap_bh;
1090         struct inode *inode = NULL;
1091         long err = -EIO;
1092
1093         /* Error cases - e2fsck has already cleaned up for us */
1094         if (ino > max_ino) {
1095                 ext4_warning(sb, "bad orphan ino %lu!  e2fsck was run?", ino);
1096                 goto error;
1097         }
1098
1099         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1100         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1101         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1102         if (!bitmap_bh) {
1103                 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1104                 goto error;
1105         }
1106
1107         /* Having the inode bit set should be a 100% indicator that this
1108          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1109          * inodes that were being truncated, so we can't check i_nlink==0.
1110          */
1111         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1112                 goto bad_orphan;
1113
1114         inode = ext4_iget(sb, ino);
1115         if (IS_ERR(inode))
1116                 goto iget_failed;
1117
1118         /*
1119          * If the orphans has i_nlinks > 0 then it should be able to be
1120          * truncated, otherwise it won't be removed from the orphan list
1121          * during processing and an infinite loop will result.
1122          */
1123         if (inode->i_nlink && !ext4_can_truncate(inode))
1124                 goto bad_orphan;
1125
1126         if (NEXT_ORPHAN(inode) > max_ino)
1127                 goto bad_orphan;
1128         brelse(bitmap_bh);
1129         return inode;
1130
1131 iget_failed:
1132         err = PTR_ERR(inode);
1133         inode = NULL;
1134 bad_orphan:
1135         ext4_warning(sb, "bad orphan inode %lu!  e2fsck was run?", ino);
1136         printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1137                bit, (unsigned long long)bitmap_bh->b_blocknr,
1138                ext4_test_bit(bit, bitmap_bh->b_data));
1139         printk(KERN_WARNING "inode=%p\n", inode);
1140         if (inode) {
1141                 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
1142                        is_bad_inode(inode));
1143                 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
1144                        NEXT_ORPHAN(inode));
1145                 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1146                 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
1147                 /* Avoid freeing blocks if we got a bad deleted inode */
1148                 if (inode->i_nlink == 0)
1149                         inode->i_blocks = 0;
1150                 iput(inode);
1151         }
1152         brelse(bitmap_bh);
1153 error:
1154         return ERR_PTR(err);
1155 }
1156
1157 unsigned long ext4_count_free_inodes(struct super_block *sb)
1158 {
1159         unsigned long desc_count;
1160         struct ext4_group_desc *gdp;
1161         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1162 #ifdef EXT4FS_DEBUG
1163         struct ext4_super_block *es;
1164         unsigned long bitmap_count, x;
1165         struct buffer_head *bitmap_bh = NULL;
1166
1167         es = EXT4_SB(sb)->s_es;
1168         desc_count = 0;
1169         bitmap_count = 0;
1170         gdp = NULL;
1171         for (i = 0; i < ngroups; i++) {
1172                 gdp = ext4_get_group_desc(sb, i, NULL);
1173                 if (!gdp)
1174                         continue;
1175                 desc_count += ext4_free_inodes_count(sb, gdp);
1176                 brelse(bitmap_bh);
1177                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1178                 if (!bitmap_bh)
1179                         continue;
1180
1181                 x = ext4_count_free(bitmap_bh->b_data,
1182                                     EXT4_INODES_PER_GROUP(sb) / 8);
1183                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1184                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1185                 bitmap_count += x;
1186         }
1187         brelse(bitmap_bh);
1188         printk(KERN_DEBUG "ext4_count_free_inodes: "
1189                "stored = %u, computed = %lu, %lu\n",
1190                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1191         return desc_count;
1192 #else
1193         desc_count = 0;
1194         for (i = 0; i < ngroups; i++) {
1195                 gdp = ext4_get_group_desc(sb, i, NULL);
1196                 if (!gdp)
1197                         continue;
1198                 desc_count += ext4_free_inodes_count(sb, gdp);
1199                 cond_resched();
1200         }
1201         return desc_count;
1202 #endif
1203 }
1204
1205 /* Called at mount-time, super-block is locked */
1206 unsigned long ext4_count_dirs(struct super_block * sb)
1207 {
1208         unsigned long count = 0;
1209         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1210
1211         for (i = 0; i < ngroups; i++) {
1212                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1213                 if (!gdp)
1214                         continue;
1215                 count += ext4_used_dirs_count(sb, gdp);
1216         }
1217         return count;
1218 }
1219
1220 /*
1221  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1222  * inode table. Must be called without any spinlock held. The only place
1223  * where it is called from on active part of filesystem is ext4lazyinit
1224  * thread, so we do not need any special locks, however we have to prevent
1225  * inode allocation from the current group, so we take alloc_sem lock, to
1226  * block ext4_new_inode() until we are finished.
1227  */
1228 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1229                                  int barrier)
1230 {
1231         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1232         struct ext4_sb_info *sbi = EXT4_SB(sb);
1233         struct ext4_group_desc *gdp = NULL;
1234         struct buffer_head *group_desc_bh;
1235         handle_t *handle;
1236         ext4_fsblk_t blk;
1237         int num, ret = 0, used_blks = 0;
1238
1239         /* This should not happen, but just to be sure check this */
1240         if (sb->s_flags & MS_RDONLY) {
1241                 ret = 1;
1242                 goto out;
1243         }
1244
1245         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1246         if (!gdp)
1247                 goto out;
1248
1249         /*
1250          * We do not need to lock this, because we are the only one
1251          * handling this flag.
1252          */
1253         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1254                 goto out;
1255
1256         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1257         if (IS_ERR(handle)) {
1258                 ret = PTR_ERR(handle);
1259                 goto out;
1260         }
1261
1262         down_write(&grp->alloc_sem);
1263         /*
1264          * If inode bitmap was already initialized there may be some
1265          * used inodes so we need to skip blocks with used inodes in
1266          * inode table.
1267          */
1268         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1269                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1270                             ext4_itable_unused_count(sb, gdp)),
1271                             sbi->s_inodes_per_block);
1272
1273         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1274                 ext4_error(sb, "Something is wrong with group %u: "
1275                            "used itable blocks: %d; "
1276                            "itable unused count: %u",
1277                            group, used_blks,
1278                            ext4_itable_unused_count(sb, gdp));
1279                 ret = 1;
1280                 goto err_out;
1281         }
1282
1283         blk = ext4_inode_table(sb, gdp) + used_blks;
1284         num = sbi->s_itb_per_group - used_blks;
1285
1286         BUFFER_TRACE(group_desc_bh, "get_write_access");
1287         ret = ext4_journal_get_write_access(handle,
1288                                             group_desc_bh);
1289         if (ret)
1290                 goto err_out;
1291
1292         /*
1293          * Skip zeroout if the inode table is full. But we set the ZEROED
1294          * flag anyway, because obviously, when it is full it does not need
1295          * further zeroing.
1296          */
1297         if (unlikely(num == 0))
1298                 goto skip_zeroout;
1299
1300         ext4_debug("going to zero out inode table in group %d\n",
1301                    group);
1302         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1303         if (ret < 0)
1304                 goto err_out;
1305         if (barrier)
1306                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1307
1308 skip_zeroout:
1309         ext4_lock_group(sb, group);
1310         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1311         ext4_group_desc_csum_set(sb, group, gdp);
1312         ext4_unlock_group(sb, group);
1313
1314         BUFFER_TRACE(group_desc_bh,
1315                      "call ext4_handle_dirty_metadata");
1316         ret = ext4_handle_dirty_metadata(handle, NULL,
1317                                          group_desc_bh);
1318
1319 err_out:
1320         up_write(&grp->alloc_sem);
1321         ext4_journal_stop(handle);
1322 out:
1323         return ret;
1324 }