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 /* Initializes an uninitialized inode bitmap */
  67 static unsigned ext4_init_inode_bitmap(struct super_block *sb,
  68                                        struct buffer_head *bh,
  69                                        ext4_group_t block_group,
  70                                        struct ext4_group_desc *gdp)
  71 {
  72         struct ext4_group_info *grp;
  73         struct ext4_sb_info *sbi = EXT4_SB(sb);
  74         J_ASSERT_BH(bh, buffer_locked(bh));
  75
  76         /* If checksum is bad mark all blocks and inodes use to prevent
  77          * allocation, essentially implementing a per-group read-only flag. */
  78         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
  79                 ext4_error(sb, "Checksum bad for group %u", block_group);
  80                 grp = ext4_get_group_info(sb, block_group);
  81                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
  82                         percpu_counter_sub(&sbi->s_freeclusters_counter,
  83                                            grp->bb_free);
  84                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
  85                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
  86                         int count;
  87                         count = ext4_free_inodes_count(sb, gdp);
  88                         percpu_counter_sub(&sbi->s_freeinodes_counter,
  89                                            count);
  90                 }
  91                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
  92                 return 0;
  93         }
  94
  95         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  96         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  97                         bh->b_data);
  98         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
  99                                    EXT4_INODES_PER_GROUP(sb) / 8);
 100         ext4_group_desc_csum_set(sb, block_group, gdp);
 101
 102         return EXT4_INODES_PER_GROUP(sb);
 103 }
 104
 105 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
 106 {
 107         if (uptodate) {
 108                 set_buffer_uptodate(bh);
 109                 set_bitmap_uptodate(bh);
 110         }
 111         unlock_buffer(bh);
 112         put_bh(bh);
 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 buffer_head *bh = NULL;
 126         ext4_fsblk_t bitmap_blk;
 127         struct ext4_group_info *grp;
 128         struct ext4_sb_info *sbi = EXT4_SB(sb);
 129
 130         desc = ext4_get_group_desc(sb, block_group, NULL);
 131         if (!desc)
 132                 return NULL;
 133
 134         bitmap_blk = ext4_inode_bitmap(sb, desc);
 135         bh = sb_getblk(sb, bitmap_blk);
 136         if (unlikely(!bh)) {
 137                 ext4_error(sb, "Cannot read inode bitmap - "
 138                             "block_group = %u, inode_bitmap = %llu",
 139                             block_group, bitmap_blk);
 140                 return NULL;
 141         }
 142         if (bitmap_uptodate(bh))
 143                 goto verify;
 144
 145         lock_buffer(bh);
 146         if (bitmap_uptodate(bh)) {
 147                 unlock_buffer(bh);
 148                 goto verify;
 149         }
 150
 151         ext4_lock_group(sb, block_group);
 152         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 153                 ext4_init_inode_bitmap(sb, bh, block_group, desc);
 154                 set_bitmap_uptodate(bh);
 155                 set_buffer_uptodate(bh);
 156                 set_buffer_verified(bh);
 157                 ext4_unlock_group(sb, block_group);
 158                 unlock_buffer(bh);
 159                 return bh;
 160         }
 161         ext4_unlock_group(sb, block_group);
 162
 163         if (buffer_uptodate(bh)) {
 164                 /*
 165                  * if not uninit if bh is uptodate,
 166                  * bitmap is also uptodate
 167                  */
 168                 set_bitmap_uptodate(bh);
 169                 unlock_buffer(bh);
 170                 goto verify;
 171         }
 172         /*
 173          * submit the buffer_head for reading
 174          */
 175         trace_ext4_load_inode_bitmap(sb, block_group);
 176         bh->b_end_io = ext4_end_bitmap_read;
 177         get_bh(bh);
 178         submit_bh(READ | REQ_META | REQ_PRIO, bh);
 179         wait_on_buffer(bh);
 180         if (!buffer_uptodate(bh)) {
 181                 put_bh(bh);
 182                 ext4_error(sb, "Cannot read inode bitmap - "
 183                            "block_group = %u, inode_bitmap = %llu",
 184                            block_group, bitmap_blk);
 185                 return NULL;
 186         }
 187
 188 verify:
 189         ext4_lock_group(sb, block_group);
 190         if (!buffer_verified(bh) &&
 191             !ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
 192                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
 193                 ext4_unlock_group(sb, block_group);
 194                 put_bh(bh);
 195                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 196                            "inode_bitmap = %llu", block_group, bitmap_blk);
 197                 grp = ext4_get_group_info(sb, block_group);
 198                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 199                         int count;
 200                         count = ext4_free_inodes_count(sb, desc);
 201                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 202                                            count);
 203                 }
 204                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 205                 return NULL;
 206         }
 207         ext4_unlock_group(sb, block_group);
 208         set_buffer_verified(bh);
 209         return bh;
 210 }
 211
 212 /*
 213  * NOTE! When we get the inode, we're the only people
 214  * that have access to it, and as such there are no
 215  * race conditions we have to worry about. The inode
 216  * is not on the hash-lists, and it cannot be reached
 217  * through the filesystem because the directory entry
 218  * has been deleted earlier.
 219  *
 220  * HOWEVER: we must make sure that we get no aliases,
 221  * which means that we have to call "clear_inode()"
 222  * _before_ we mark the inode not in use in the inode
 223  * bitmaps. Otherwise a newly created file might use
 224  * the same inode number (not actually the same pointer
 225  * though), and then we'd have two inodes sharing the
 226  * same inode number and space on the harddisk.
 227  */
 228 void ext4_free_inode(handle_t *handle, struct inode *inode)
 229 {
 230         struct super_block *sb = inode->i_sb;
 231         int is_directory;
 232         unsigned long ino;
 233         struct buffer_head *bitmap_bh = NULL;
 234         struct buffer_head *bh2;
 235         ext4_group_t block_group;
 236         unsigned long bit;
 237         struct ext4_group_desc *gdp;
 238         struct ext4_super_block *es;
 239         struct ext4_sb_info *sbi;
 240         int fatal = 0, err, count, cleared;
 241         struct ext4_group_info *grp;
 242
 243         if (!sb) {
 244                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 245                        "nonexistent device\n", __func__, __LINE__);
 246                 return;
 247         }
 248         if (atomic_read(&inode->i_count) > 1) {
 249                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 250                          __func__, __LINE__, inode->i_ino,
 251                          atomic_read(&inode->i_count));
 252                 return;
 253         }
 254         if (inode->i_nlink) {
 255                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 256                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 257                 return;
 258         }
 259         sbi = EXT4_SB(sb);
 260
 261         ino = inode->i_ino;
 262         ext4_debug("freeing inode %lu\n", ino);
 263         trace_ext4_free_inode(inode);
 264
 265         /*
 266          * Note: we must free any quota before locking the superblock,
 267          * as writing the quota to disk may need the lock as well.
 268          */
 269         dquot_initialize(inode);
 270         ext4_xattr_delete_inode(handle, inode);
 271         dquot_free_inode(inode);
 272         dquot_drop(inode);
 273
 274         is_directory = S_ISDIR(inode->i_mode);
 275
 276         /* Do this BEFORE marking the inode not in use or returning an error */
 277         ext4_clear_inode(inode);
 278
 279         es = EXT4_SB(sb)->s_es;
 280         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 281                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 282                 goto error_return;
 283         }
 284         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 285         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 286         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 287         /* Don't bother if the inode bitmap is corrupt. */
 288         grp = ext4_get_group_info(sb, block_group);
 289         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) || !bitmap_bh)
 290                 goto error_return;
 291
 292         BUFFER_TRACE(bitmap_bh, "get_write_access");
 293         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 294         if (fatal)
 295                 goto error_return;
 296
 297         fatal = -ESRCH;
 298         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 299         if (gdp) {
 300                 BUFFER_TRACE(bh2, "get_write_access");
 301                 fatal = ext4_journal_get_write_access(handle, bh2);
 302         }
 303         ext4_lock_group(sb, block_group);
 304         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 305         if (fatal || !cleared) {
 306                 ext4_unlock_group(sb, block_group);
 307                 goto out;
 308         }
 309
 310         count = ext4_free_inodes_count(sb, gdp) + 1;
 311         ext4_free_inodes_set(sb, gdp, count);
 312         if (is_directory) {
 313                 count = ext4_used_dirs_count(sb, gdp) - 1;
 314                 ext4_used_dirs_set(sb, gdp, count);
 315                 percpu_counter_dec(&sbi->s_dirs_counter);
 316         }
 317         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 318                                    EXT4_INODES_PER_GROUP(sb) / 8);
 319         ext4_group_desc_csum_set(sb, block_group, gdp);
 320         ext4_unlock_group(sb, block_group);
 321
 322         percpu_counter_inc(&sbi->s_freeinodes_counter);
 323         if (sbi->s_log_groups_per_flex) {
 324                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 325
 326                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 327                 if (is_directory)
 328                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 329         }
 330         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 331         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 332 out:
 333         if (cleared) {
 334                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 335                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 336                 if (!fatal)
 337                         fatal = err;
 338         } else {
 339                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 340                 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 341                         int count;
 342                         count = ext4_free_inodes_count(sb, gdp);
 343                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 344                                            count);
 345                 }
 346                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 347         }
 348
 349 error_return:
 350         brelse(bitmap_bh);
 351         ext4_std_error(sb, fatal);
 352 }
 353
 354 struct orlov_stats {
 355         __u64 free_clusters;
 356         __u32 free_inodes;
 357         __u32 used_dirs;
 358 };
 359
 360 /*
 361  * Helper function for Orlov's allocator; returns critical information
 362  * for a particular block group or flex_bg.  If flex_size is 1, then g
 363  * is a block group number; otherwise it is flex_bg number.
 364  */
 365 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 366                             int flex_size, struct orlov_stats *stats)
 367 {
 368         struct ext4_group_desc *desc;
 369         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 370
 371         if (flex_size > 1) {
 372                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 373                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 374                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 375                 return;
 376         }
 377
 378         desc = ext4_get_group_desc(sb, g, NULL);
 379         if (desc) {
 380                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 381                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 382                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 383         } else {
 384                 stats->free_inodes = 0;
 385                 stats->free_clusters = 0;
 386                 stats->used_dirs = 0;
 387         }
 388 }
 389
 390 /*
 391  * Orlov's allocator for directories.
 392  *
 393  * We always try to spread first-level directories.
 394  *
 395  * If there are blockgroups with both free inodes and free blocks counts
 396  * not worse than average we return one with smallest directory count.
 397  * Otherwise we simply return a random group.
 398  *
 399  * For the rest rules look so:
 400  *
 401  * It's OK to put directory into a group unless
 402  * it has too many directories already (max_dirs) or
 403  * it has too few free inodes left (min_inodes) or
 404  * it has too few free blocks left (min_blocks) or
 405  * Parent's group is preferred, if it doesn't satisfy these
 406  * conditions we search cyclically through the rest. If none
 407  * of the groups look good we just look for a group with more
 408  * free inodes than average (starting at parent's group).
 409  */
 410
 411 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 412                             ext4_group_t *group, umode_t mode,
 413                             const struct qstr *qstr)
 414 {
 415         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 416         struct ext4_sb_info *sbi = EXT4_SB(sb);
 417         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 418         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 419         unsigned int freei, avefreei, grp_free;
 420         ext4_fsblk_t freeb, avefreec;
 421         unsigned int ndirs;
 422         int max_dirs, min_inodes;
 423         ext4_grpblk_t min_clusters;
 424         ext4_group_t i, grp, g, ngroups;
 425         struct ext4_group_desc *desc;
 426         struct orlov_stats stats;
 427         int flex_size = ext4_flex_bg_size(sbi);
 428         struct dx_hash_info hinfo;
 429
 430         ngroups = real_ngroups;
 431         if (flex_size > 1) {
 432                 ngroups = (real_ngroups + flex_size - 1) >>
 433                         sbi->s_log_groups_per_flex;
 434                 parent_group >>= sbi->s_log_groups_per_flex;
 435         }
 436
 437         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 438         avefreei = freei / ngroups;
 439         freeb = EXT4_C2B(sbi,
 440                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 441         avefreec = freeb;
 442         do_div(avefreec, ngroups);
 443         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 444
 445         if (S_ISDIR(mode) &&
 446             ((parent == d_inode(sb->s_root)) ||
 447              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 448                 int best_ndir = inodes_per_group;
 449                 int ret = -1;
 450
 451                 if (qstr) {
 452                         hinfo.hash_version = DX_HASH_HALF_MD4;
 453                         hinfo.seed = sbi->s_hash_seed;
 454                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 455                         grp = hinfo.hash;
 456                 } else
 457                         grp = prandom_u32();
 458                 parent_group = (unsigned)grp % ngroups;
 459                 for (i = 0; i < ngroups; i++) {
 460                         g = (parent_group + i) % ngroups;
 461                         get_orlov_stats(sb, g, flex_size, &stats);
 462                         if (!stats.free_inodes)
 463                                 continue;
 464                         if (stats.used_dirs >= best_ndir)
 465                                 continue;
 466                         if (stats.free_inodes < avefreei)
 467                                 continue;
 468                         if (stats.free_clusters < avefreec)
 469                                 continue;
 470                         grp = g;
 471                         ret = 0;
 472                         best_ndir = stats.used_dirs;
 473                 }
 474                 if (ret)
 475                         goto fallback;
 476         found_flex_bg:
 477                 if (flex_size == 1) {
 478                         *group = grp;
 479                         return 0;
 480                 }
 481
 482                 /*
 483                  * We pack inodes at the beginning of the flexgroup's
 484                  * inode tables.  Block allocation decisions will do
 485                  * something similar, although regular files will
 486                  * start at 2nd block group of the flexgroup.  See
 487                  * ext4_ext_find_goal() and ext4_find_near().
 488                  */
 489                 grp *= flex_size;
 490                 for (i = 0; i < flex_size; i++) {
 491                         if (grp+i >= real_ngroups)
 492                                 break;
 493                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 494                         if (desc && ext4_free_inodes_count(sb, desc)) {
 495                                 *group = grp+i;
 496                                 return 0;
 497                         }
 498                 }
 499                 goto fallback;
 500         }
 501
 502         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 503         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 504         if (min_inodes < 1)
 505                 min_inodes = 1;
 506         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 507
 508         /*
 509          * Start looking in the flex group where we last allocated an
 510          * inode for this parent directory
 511          */
 512         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 513                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 514                 if (flex_size > 1)
 515                         parent_group >>= sbi->s_log_groups_per_flex;
 516         }
 517
 518         for (i = 0; i < ngroups; i++) {
 519                 grp = (parent_group + i) % ngroups;
 520                 get_orlov_stats(sb, grp, flex_size, &stats);
 521                 if (stats.used_dirs >= max_dirs)
 522                         continue;
 523                 if (stats.free_inodes < min_inodes)
 524                         continue;
 525                 if (stats.free_clusters < min_clusters)
 526                         continue;
 527                 goto found_flex_bg;
 528         }
 529
 530 fallback:
 531         ngroups = real_ngroups;
 532         avefreei = freei / ngroups;
 533 fallback_retry:
 534         parent_group = EXT4_I(parent)->i_block_group;
 535         for (i = 0; i < ngroups; i++) {
 536                 grp = (parent_group + i) % ngroups;
 537                 desc = ext4_get_group_desc(sb, grp, NULL);
 538                 if (desc) {
 539                         grp_free = ext4_free_inodes_count(sb, desc);
 540                         if (grp_free && grp_free >= avefreei) {
 541                                 *group = grp;
 542                                 return 0;
 543                         }
 544                 }
 545         }
 546
 547         if (avefreei) {
 548                 /*
 549                  * The free-inodes counter is approximate, and for really small
 550                  * filesystems the above test can fail to find any blockgroups
 551                  */
 552                 avefreei = 0;
 553                 goto fallback_retry;
 554         }
 555
 556         return -1;
 557 }
 558
 559 static int find_group_other(struct super_block *sb, struct inode *parent,
 560                             ext4_group_t *group, umode_t mode)
 561 {
 562         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 563         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 564         struct ext4_group_desc *desc;
 565         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 566
 567         /*
 568          * Try to place the inode is the same flex group as its
 569          * parent.  If we can't find space, use the Orlov algorithm to
 570          * find another flex group, and store that information in the
 571          * parent directory's inode information so that use that flex
 572          * group for future allocations.
 573          */
 574         if (flex_size > 1) {
 575                 int retry = 0;
 576
 577         try_again:
 578                 parent_group &= ~(flex_size-1);
 579                 last = parent_group + flex_size;
 580                 if (last > ngroups)
 581                         last = ngroups;
 582                 for  (i = parent_group; i < last; i++) {
 583                         desc = ext4_get_group_desc(sb, i, NULL);
 584                         if (desc && ext4_free_inodes_count(sb, desc)) {
 585                                 *group = i;
 586                                 return 0;
 587                         }
 588                 }
 589                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 590                         retry = 1;
 591                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 592                         goto try_again;
 593                 }
 594                 /*
 595                  * If this didn't work, use the Orlov search algorithm
 596                  * to find a new flex group; we pass in the mode to
 597                  * avoid the topdir algorithms.
 598                  */
 599                 *group = parent_group + flex_size;
 600                 if (*group > ngroups)
 601                         *group = 0;
 602                 return find_group_orlov(sb, parent, group, mode, NULL);
 603         }
 604
 605         /*
 606          * Try to place the inode in its parent directory
 607          */
 608         *group = parent_group;
 609         desc = ext4_get_group_desc(sb, *group, NULL);
 610         if (desc && ext4_free_inodes_count(sb, desc) &&
 611             ext4_free_group_clusters(sb, desc))
 612                 return 0;
 613
 614         /*
 615          * We're going to place this inode in a different blockgroup from its
 616          * parent.  We want to cause files in a common directory to all land in
 617          * the same blockgroup.  But we want files which are in a different
 618          * directory which shares a blockgroup with our parent to land in a
 619          * different blockgroup.
 620          *
 621          * So add our directory's i_ino into the starting point for the hash.
 622          */
 623         *group = (*group + parent->i_ino) % ngroups;
 624
 625         /*
 626          * Use a quadratic hash to find a group with a free inode and some free
 627          * blocks.
 628          */
 629         for (i = 1; i < ngroups; i <<= 1) {
 630                 *group += i;
 631                 if (*group >= ngroups)
 632                         *group -= ngroups;
 633                 desc = ext4_get_group_desc(sb, *group, NULL);
 634                 if (desc && ext4_free_inodes_count(sb, desc) &&
 635                     ext4_free_group_clusters(sb, desc))
 636                         return 0;
 637         }
 638
 639         /*
 640          * That failed: try linear search for a free inode, even if that group
 641          * has no free blocks.
 642          */
 643         *group = parent_group;
 644         for (i = 0; i < ngroups; i++) {
 645                 if (++*group >= ngroups)
 646                         *group = 0;
 647                 desc = ext4_get_group_desc(sb, *group, NULL);
 648                 if (desc && ext4_free_inodes_count(sb, desc))
 649                         return 0;
 650         }
 651
 652         return -1;
 653 }
 654
 655 /*
 656  * In no journal mode, if an inode has recently been deleted, we want
 657  * to avoid reusing it until we're reasonably sure the inode table
 658  * block has been written back to disk.  (Yes, these values are
 659  * somewhat arbitrary...)
 660  */
 661 #define RECENTCY_MIN    5
 662 #define RECENTCY_DIRTY  30
 663
 664 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 665 {
 666         struct ext4_group_desc  *gdp;
 667         struct ext4_inode       *raw_inode;
 668         struct buffer_head      *bh;
 669         unsigned long           dtime, now;
 670         int     inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 671         int     offset, ret = 0, recentcy = RECENTCY_MIN;
 672
 673         gdp = ext4_get_group_desc(sb, group, NULL);
 674         if (unlikely(!gdp))
 675                 return 0;
 676
 677         bh = sb_getblk(sb, ext4_inode_table(sb, gdp) +
 678                        (ino / inodes_per_block));
 679         if (unlikely(!bh) || !buffer_uptodate(bh))
 680                 /*
 681                  * If the block is not in the buffer cache, then it
 682                  * must have been written out.
 683                  */
 684                 goto out;
 685
 686         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 687         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 688         dtime = le32_to_cpu(raw_inode->i_dtime);
 689         now = get_seconds();
 690         if (buffer_dirty(bh))
 691                 recentcy += RECENTCY_DIRTY;
 692
 693         if (dtime && (dtime < now) && (now < dtime + recentcy))
 694                 ret = 1;
 695 out:
 696         brelse(bh);
 697         return ret;
 698 }
 699
 700 /*
 701  * There are two policies for allocating an inode.  If the new inode is
 702  * a directory, then a forward search is made for a block group with both
 703  * free space and a low directory-to-inode ratio; if that fails, then of
 704  * the groups with above-average free space, that group with the fewest
 705  * directories already is chosen.
 706  *
 707  * For other inodes, search forward from the parent directory's block
 708  * group to find a free inode.
 709  */
 710 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 711                                umode_t mode, const struct qstr *qstr,
 712                                __u32 goal, uid_t *owner, int handle_type,
 713                                unsigned int line_no, int nblocks)
 714 {
 715         struct super_block *sb;
 716         struct buffer_head *inode_bitmap_bh = NULL;
 717         struct buffer_head *group_desc_bh;
 718         ext4_group_t ngroups, group = 0;
 719         unsigned long ino = 0;
 720         struct inode *inode;
 721         struct ext4_group_desc *gdp = NULL;
 722         struct ext4_inode_info *ei;
 723         struct ext4_sb_info *sbi;
 724         int ret2, err = 0;
 725         struct inode *ret;
 726         ext4_group_t i;
 727         ext4_group_t flex_group;
 728         struct ext4_group_info *grp;
 729         int encrypt = 0;
 730
 731         /* Cannot create files in a deleted directory */
 732         if (!dir || !dir->i_nlink)
 733                 return ERR_PTR(-EPERM);
 734
 735         if ((ext4_encrypted_inode(dir) ||
 736              DUMMY_ENCRYPTION_ENABLED(EXT4_SB(dir->i_sb))) &&
 737             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
 738                 err = ext4_get_encryption_info(dir);
 739                 if (err)
 740                         return ERR_PTR(err);
 741                 if (ext4_encryption_info(dir) == NULL)
 742                         return ERR_PTR(-EPERM);
 743                 if (!handle)
 744                         nblocks += EXT4_DATA_TRANS_BLOCKS(dir->i_sb);
 745                 encrypt = 1;
 746         }
 747
 748         sb = dir->i_sb;
 749         ngroups = ext4_get_groups_count(sb);
 750         trace_ext4_request_inode(dir, mode);
 751         inode = new_inode(sb);
 752         if (!inode)
 753                 return ERR_PTR(-ENOMEM);
 754         ei = EXT4_I(inode);
 755         sbi = EXT4_SB(sb);
 756
 757         /*
 758          * Initalize owners and quota early so that we don't have to account
 759          * for quota initialization worst case in standard inode creating
 760          * transaction
 761          */
 762         if (owner) {
 763                 inode->i_mode = mode;
 764                 i_uid_write(inode, owner[0]);
 765                 i_gid_write(inode, owner[1]);
 766         } else if (test_opt(sb, GRPID)) {
 767                 inode->i_mode = mode;
 768                 inode->i_uid = current_fsuid();
 769                 inode->i_gid = dir->i_gid;
 770         } else
 771                 inode_init_owner(inode, dir, mode);
 772         dquot_initialize(inode);
 773
 774         if (!goal)
 775                 goal = sbi->s_inode_goal;
 776
 777         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 778                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 779                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 780                 ret2 = 0;
 781                 goto got_group;
 782         }
 783
 784         if (S_ISDIR(mode))
 785                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 786         else
 787                 ret2 = find_group_other(sb, dir, &group, mode);
 788
 789 got_group:
 790         EXT4_I(dir)->i_last_alloc_group = group;
 791         err = -ENOSPC;
 792         if (ret2 == -1)
 793                 goto out;
 794
 795         /*
 796          * Normally we will only go through one pass of this loop,
 797          * unless we get unlucky and it turns out the group we selected
 798          * had its last inode grabbed by someone else.
 799          */
 800         for (i = 0; i < ngroups; i++, ino = 0) {
 801                 err = -EIO;
 802
 803                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 804                 if (!gdp)
 805                         goto out;
 806
 807                 /*
 808                  * Check free inodes count before loading bitmap.
 809                  */
 810                 if (ext4_free_inodes_count(sb, gdp) == 0) {
 811                         if (++group == ngroups)
 812                                 group = 0;
 813                         continue;
 814                 }
 815
 816                 grp = ext4_get_group_info(sb, group);
 817                 /* Skip groups with already-known suspicious inode tables */
 818                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 819                         if (++group == ngroups)
 820                                 group = 0;
 821                         continue;
 822                 }
 823
 824                 brelse(inode_bitmap_bh);
 825                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 826                 /* Skip groups with suspicious inode tables */
 827                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || !inode_bitmap_bh) {
 828                         if (++group == ngroups)
 829                                 group = 0;
 830                         continue;
 831                 }
 832
 833 repeat_in_this_group:
 834                 ino = ext4_find_next_zero_bit((unsigned long *)
 835                                               inode_bitmap_bh->b_data,
 836                                               EXT4_INODES_PER_GROUP(sb), ino);
 837                 if (ino >= EXT4_INODES_PER_GROUP(sb))
 838                         goto next_group;
 839                 if (group == 0 && (ino+1) < EXT4_FIRST_INO(sb)) {
 840                         ext4_error(sb, "reserved inode found cleared - "
 841                                    "inode=%lu", ino + 1);
 842                         continue;
 843                 }
 844                 if ((EXT4_SB(sb)->s_journal == NULL) &&
 845                     recently_deleted(sb, group, ino)) {
 846                         ino++;
 847                         goto next_inode;
 848                 }
 849                 if (!handle) {
 850                         BUG_ON(nblocks <= 0);
 851                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 852                                                          handle_type, nblocks,
 853                                                          0);
 854                         if (IS_ERR(handle)) {
 855                                 err = PTR_ERR(handle);
 856                                 ext4_std_error(sb, err);
 857                                 goto out;
 858                         }
 859                 }
 860                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 861                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 862                 if (err) {
 863                         ext4_std_error(sb, err);
 864                         goto out;
 865                 }
 866                 ext4_lock_group(sb, group);
 867                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 868                 ext4_unlock_group(sb, group);
 869                 ino++;          /* the inode bitmap is zero-based */
 870                 if (!ret2)
 871                         goto got; /* we grabbed the inode! */
 872 next_inode:
 873                 if (ino < EXT4_INODES_PER_GROUP(sb))
 874                         goto repeat_in_this_group;
 875 next_group:
 876                 if (++group == ngroups)
 877                         group = 0;
 878         }
 879         err = -ENOSPC;
 880         goto out;
 881
 882 got:
 883         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 884         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 885         if (err) {
 886                 ext4_std_error(sb, err);
 887                 goto out;
 888         }
 889
 890         BUFFER_TRACE(group_desc_bh, "get_write_access");
 891         err = ext4_journal_get_write_access(handle, group_desc_bh);
 892         if (err) {
 893                 ext4_std_error(sb, err);
 894                 goto out;
 895         }
 896
 897         /* We may have to initialize the block bitmap if it isn't already */
 898         if (ext4_has_group_desc_csum(sb) &&
 899             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 900                 struct buffer_head *block_bitmap_bh;
 901
 902                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 903                 if (!block_bitmap_bh) {
 904                         err = -EIO;
 905                         goto out;
 906                 }
 907                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 908                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 909                 if (err) {
 910                         brelse(block_bitmap_bh);
 911                         ext4_std_error(sb, err);
 912                         goto out;
 913                 }
 914
 915                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 916                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 917
 918                 /* recheck and clear flag under lock if we still need to */
 919                 ext4_lock_group(sb, group);
 920                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 921                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 922                         ext4_free_group_clusters_set(sb, gdp,
 923                                 ext4_free_clusters_after_init(sb, group, gdp));
 924                         ext4_block_bitmap_csum_set(sb, group, gdp,
 925                                                    block_bitmap_bh);
 926                         ext4_group_desc_csum_set(sb, group, gdp);
 927                 }
 928                 ext4_unlock_group(sb, group);
 929                 brelse(block_bitmap_bh);
 930
 931                 if (err) {
 932                         ext4_std_error(sb, err);
 933                         goto out;
 934                 }
 935         }
 936
 937         /* Update the relevant bg descriptor fields */
 938         if (ext4_has_group_desc_csum(sb)) {
 939                 int free;
 940                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
 941
 942                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
 943                 ext4_lock_group(sb, group); /* while we modify the bg desc */
 944                 free = EXT4_INODES_PER_GROUP(sb) -
 945                         ext4_itable_unused_count(sb, gdp);
 946                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 947                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 948                         free = 0;
 949                 }
 950                 /*
 951                  * Check the relative inode number against the last used
 952                  * relative inode number in this group. if it is greater
 953                  * we need to update the bg_itable_unused count
 954                  */
 955                 if (ino > free)
 956                         ext4_itable_unused_set(sb, gdp,
 957                                         (EXT4_INODES_PER_GROUP(sb) - ino));
 958                 up_read(&grp->alloc_sem);
 959         } else {
 960                 ext4_lock_group(sb, group);
 961         }
 962
 963         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
 964         if (S_ISDIR(mode)) {
 965                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
 966                 if (sbi->s_log_groups_per_flex) {
 967                         ext4_group_t f = ext4_flex_group(sbi, group);
 968
 969                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
 970                 }
 971         }
 972         if (ext4_has_group_desc_csum(sb)) {
 973                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
 974                                            EXT4_INODES_PER_GROUP(sb) / 8);
 975                 ext4_group_desc_csum_set(sb, group, gdp);
 976         }
 977         ext4_unlock_group(sb, group);
 978
 979         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
 980         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
 981         if (err) {
 982                 ext4_std_error(sb, err);
 983                 goto out;
 984         }
 985
 986         percpu_counter_dec(&sbi->s_freeinodes_counter);
 987         if (S_ISDIR(mode))
 988                 percpu_counter_inc(&sbi->s_dirs_counter);
 989
 990         if (sbi->s_log_groups_per_flex) {
 991                 flex_group = ext4_flex_group(sbi, group);
 992                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
 993         }
 994
 995         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
 996         /* This is the optimal IO size (for stat), not the fs block size */
 997         inode->i_blocks = 0;
 998         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
 999                                                        ext4_current_time(inode);
1000
1001         memset(ei->i_data, 0, sizeof(ei->i_data));
1002         ei->i_dir_start_lookup = 0;
1003         ei->i_disksize = 0;
1004
1005         /* Don't inherit extent flag from directory, amongst others. */
1006         ei->i_flags =
1007                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1008         ei->i_file_acl = 0;
1009         ei->i_dtime = 0;
1010         ei->i_block_group = group;
1011         ei->i_last_alloc_group = ~0;
1012
1013         ext4_set_inode_flags(inode);
1014         if (IS_DIRSYNC(inode))
1015                 ext4_handle_sync(handle);
1016         if (insert_inode_locked(inode) < 0) {
1017                 /*
1018                  * Likely a bitmap corruption causing inode to be allocated
1019                  * twice.
1020                  */
1021                 err = -EIO;
1022                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1023                            inode->i_ino);
1024                 goto out;
1025         }
1026         spin_lock(&sbi->s_next_gen_lock);
1027         inode->i_generation = sbi->s_next_generation++;
1028         spin_unlock(&sbi->s_next_gen_lock);
1029
1030         /* Precompute checksum seed for inode metadata */
1031         if (ext4_has_metadata_csum(sb)) {
1032                 __u32 csum;
1033                 __le32 inum = cpu_to_le32(inode->i_ino);
1034                 __le32 gen = cpu_to_le32(inode->i_generation);
1035                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1036                                    sizeof(inum));
1037                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1038                                               sizeof(gen));
1039         }
1040
1041         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1042         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1043
1044         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1045         ei->i_inline_off = 0;
1046         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_INLINE_DATA))
1047                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1048         ret = inode;
1049         err = dquot_alloc_inode(inode);
1050         if (err)
1051                 goto fail_drop;
1052
1053         err = ext4_init_acl(handle, inode, dir);
1054         if (err)
1055                 goto fail_free_drop;
1056
1057         err = ext4_init_security(handle, inode, dir, qstr);
1058         if (err)
1059                 goto fail_free_drop;
1060
1061         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
1062                 /* set extent flag only for directory, file and normal symlink*/
1063                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1064                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1065                         ext4_ext_tree_init(handle, inode);
1066                 }
1067         }
1068
1069         if (ext4_handle_valid(handle)) {
1070                 ei->i_sync_tid = handle->h_transaction->t_tid;
1071                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1072         }
1073
1074         if (encrypt) {
1075                 err = ext4_inherit_context(dir, inode);
1076                 if (err)
1077                         goto fail_free_drop;
1078         }
1079
1080         err = ext4_mark_inode_dirty(handle, inode);
1081         if (err) {
1082                 ext4_std_error(sb, err);
1083                 goto fail_free_drop;
1084         }
1085
1086         ext4_debug("allocating inode %lu\n", inode->i_ino);
1087         trace_ext4_allocate_inode(inode, dir, mode);
1088         brelse(inode_bitmap_bh);
1089         return ret;
1090
1091 fail_free_drop:
1092         dquot_free_inode(inode);
1093 fail_drop:
1094         clear_nlink(inode);
1095         unlock_new_inode(inode);
1096 out:
1097         dquot_drop(inode);
1098         inode->i_flags |= S_NOQUOTA;
1099         iput(inode);
1100         brelse(inode_bitmap_bh);
1101         return ERR_PTR(err);
1102 }
1103
1104 /* Verify that we are loading a valid orphan from disk */
1105 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1106 {
1107         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1108         ext4_group_t block_group;
1109         int bit;
1110         struct buffer_head *bitmap_bh;
1111         struct inode *inode = NULL;
1112         long err = -EIO;
1113
1114         /* Error cases - e2fsck has already cleaned up for us */
1115         if (ino > max_ino) {
1116                 ext4_warning(sb, "bad orphan ino %lu!  e2fsck was run?", ino);
1117                 goto error;
1118         }
1119
1120         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1121         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1122         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1123         if (!bitmap_bh) {
1124                 ext4_warning(sb, "inode bitmap error for orphan %lu", ino);
1125                 goto error;
1126         }
1127
1128         /* Having the inode bit set should be a 100% indicator that this
1129          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1130          * inodes that were being truncated, so we can't check i_nlink==0.
1131          */
1132         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1133                 goto bad_orphan;
1134
1135         inode = ext4_iget(sb, ino);
1136         if (IS_ERR(inode))
1137                 goto iget_failed;
1138
1139         /*
1140          * If the orphans has i_nlinks > 0 then it should be able to be
1141          * truncated, otherwise it won't be removed from the orphan list
1142          * during processing and an infinite loop will result.
1143          */
1144         if (inode->i_nlink && !ext4_can_truncate(inode))
1145                 goto bad_orphan;
1146
1147         if (NEXT_ORPHAN(inode) > max_ino)
1148                 goto bad_orphan;
1149         brelse(bitmap_bh);
1150         return inode;
1151
1152 iget_failed:
1153         err = PTR_ERR(inode);
1154         inode = NULL;
1155 bad_orphan:
1156         ext4_warning(sb, "bad orphan inode %lu!  e2fsck was run?", ino);
1157         printk(KERN_WARNING "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1158                bit, (unsigned long long)bitmap_bh->b_blocknr,
1159                ext4_test_bit(bit, bitmap_bh->b_data));
1160         printk(KERN_WARNING "inode=%p\n", inode);
1161         if (inode) {
1162                 printk(KERN_WARNING "is_bad_inode(inode)=%d\n",
1163                        is_bad_inode(inode));
1164                 printk(KERN_WARNING "NEXT_ORPHAN(inode)=%u\n",
1165                        NEXT_ORPHAN(inode));
1166                 printk(KERN_WARNING "max_ino=%lu\n", max_ino);
1167                 printk(KERN_WARNING "i_nlink=%u\n", inode->i_nlink);
1168                 /* Avoid freeing blocks if we got a bad deleted inode */
1169                 if (inode->i_nlink == 0)
1170                         inode->i_blocks = 0;
1171                 iput(inode);
1172         }
1173         brelse(bitmap_bh);
1174 error:
1175         return ERR_PTR(err);
1176 }
1177
1178 unsigned long ext4_count_free_inodes(struct super_block *sb)
1179 {
1180         unsigned long desc_count;
1181         struct ext4_group_desc *gdp;
1182         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1183 #ifdef EXT4FS_DEBUG
1184         struct ext4_super_block *es;
1185         unsigned long bitmap_count, x;
1186         struct buffer_head *bitmap_bh = NULL;
1187
1188         es = EXT4_SB(sb)->s_es;
1189         desc_count = 0;
1190         bitmap_count = 0;
1191         gdp = NULL;
1192         for (i = 0; i < ngroups; i++) {
1193                 gdp = ext4_get_group_desc(sb, i, NULL);
1194                 if (!gdp)
1195                         continue;
1196                 desc_count += ext4_free_inodes_count(sb, gdp);
1197                 brelse(bitmap_bh);
1198                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1199                 if (!bitmap_bh)
1200                         continue;
1201
1202                 x = ext4_count_free(bitmap_bh->b_data,
1203                                     EXT4_INODES_PER_GROUP(sb) / 8);
1204                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1205                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1206                 bitmap_count += x;
1207         }
1208         brelse(bitmap_bh);
1209         printk(KERN_DEBUG "ext4_count_free_inodes: "
1210                "stored = %u, computed = %lu, %lu\n",
1211                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1212         return desc_count;
1213 #else
1214         desc_count = 0;
1215         for (i = 0; i < ngroups; i++) {
1216                 gdp = ext4_get_group_desc(sb, i, NULL);
1217                 if (!gdp)
1218                         continue;
1219                 desc_count += ext4_free_inodes_count(sb, gdp);
1220                 cond_resched();
1221         }
1222         return desc_count;
1223 #endif
1224 }
1225
1226 /* Called at mount-time, super-block is locked */
1227 unsigned long ext4_count_dirs(struct super_block * sb)
1228 {
1229         unsigned long count = 0;
1230         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1231
1232         for (i = 0; i < ngroups; i++) {
1233                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1234                 if (!gdp)
1235                         continue;
1236                 count += ext4_used_dirs_count(sb, gdp);
1237         }
1238         return count;
1239 }
1240
1241 /*
1242  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1243  * inode table. Must be called without any spinlock held. The only place
1244  * where it is called from on active part of filesystem is ext4lazyinit
1245  * thread, so we do not need any special locks, however we have to prevent
1246  * inode allocation from the current group, so we take alloc_sem lock, to
1247  * block ext4_new_inode() until we are finished.
1248  */
1249 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1250                                  int barrier)
1251 {
1252         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1253         struct ext4_sb_info *sbi = EXT4_SB(sb);
1254         struct ext4_group_desc *gdp = NULL;
1255         struct buffer_head *group_desc_bh;
1256         handle_t *handle;
1257         ext4_fsblk_t blk;
1258         int num, ret = 0, used_blks = 0;
1259
1260         /* This should not happen, but just to be sure check this */
1261         if (sb->s_flags & MS_RDONLY) {
1262                 ret = 1;
1263                 goto out;
1264         }
1265
1266         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1267         if (!gdp)
1268                 goto out;
1269
1270         /*
1271          * We do not need to lock this, because we are the only one
1272          * handling this flag.
1273          */
1274         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1275                 goto out;
1276
1277         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1278         if (IS_ERR(handle)) {
1279                 ret = PTR_ERR(handle);
1280                 goto out;
1281         }
1282
1283         down_write(&grp->alloc_sem);
1284         /*
1285          * If inode bitmap was already initialized there may be some
1286          * used inodes so we need to skip blocks with used inodes in
1287          * inode table.
1288          */
1289         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1290                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1291                             ext4_itable_unused_count(sb, gdp)),
1292                             sbi->s_inodes_per_block);
1293
1294         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1295                 ext4_error(sb, "Something is wrong with group %u: "
1296                            "used itable blocks: %d; "
1297                            "itable unused count: %u",
1298                            group, used_blks,
1299                            ext4_itable_unused_count(sb, gdp));
1300                 ret = 1;
1301                 goto err_out;
1302         }
1303
1304         blk = ext4_inode_table(sb, gdp) + used_blks;
1305         num = sbi->s_itb_per_group - used_blks;
1306
1307         BUFFER_TRACE(group_desc_bh, "get_write_access");
1308         ret = ext4_journal_get_write_access(handle,
1309                                             group_desc_bh);
1310         if (ret)
1311                 goto err_out;
1312
1313         /*
1314          * Skip zeroout if the inode table is full. But we set the ZEROED
1315          * flag anyway, because obviously, when it is full it does not need
1316          * further zeroing.
1317          */
1318         if (unlikely(num == 0))
1319                 goto skip_zeroout;
1320
1321         ext4_debug("going to zero out inode table in group %d\n",
1322                    group);
1323         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1324         if (ret < 0)
1325                 goto err_out;
1326         if (barrier)
1327                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1328
1329 skip_zeroout:
1330         ext4_lock_group(sb, group);
1331         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1332         ext4_group_desc_csum_set(sb, group, gdp);
1333         ext4_unlock_group(sb, group);
1334
1335         BUFFER_TRACE(group_desc_bh,
1336                      "call ext4_handle_dirty_metadata");
1337         ret = ext4_handle_dirty_metadata(handle, NULL,
1338                                          group_desc_bh);
1339
1340 err_out:
1341         up_write(&grp->alloc_sem);
1342         ext4_journal_stop(handle);
1343 out:
1344         return ret;
1345 }