fs/ext4/ialloc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/ialloc.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  BSD ufs-inspired inode and directory allocation by
  11  *  Stephen Tweedie (sct@redhat.com), 1993
  12  *  Big-endian to little-endian byte-swapping/bitmaps by
  13  *        David S. Miller (davem@caip.rutgers.edu), 1995
  14  */
  15
  16 #include <linux/time.h>
  17 #include <linux/fs.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 <linux/cred.h>
  26
  27 #include <asm/byteorder.h>
  28
  29 #include "ext4.h"
  30 #include "ext4_jbd2.h"
  31 #include "xattr.h"
  32 #include "acl.h"
  33
  34 #include <trace/events/ext4.h>
  35
  36 /*
  37  * ialloc.c contains the inodes allocation and deallocation routines
  38  */
  39
  40 /*
  41  * The free inodes are managed by bitmaps.  A file system contains several
  42  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  43  * block for inodes, N blocks for the inode table and data blocks.
  44  *
  45  * The file system contains group descriptors which are located after the
  46  * super block.  Each descriptor contains the number of the bitmap block and
  47  * the free blocks count in the block.
  48  */
  49
  50 /*
  51  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  52  * need to use it within a single byte (to ensure we get endianness right).
  53  * We can use memset for the rest of the bitmap as there are no other users.
  54  */
  55 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  56 {
  57         int i;
  58
  59         if (start_bit >= end_bit)
  60                 return;
  61
  62         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  63         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  64                 ext4_set_bit(i, bitmap);
  65         if (i < end_bit)
  66                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  67 }
  68
  69 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
  70 {
  71         if (uptodate) {
  72                 set_buffer_uptodate(bh);
  73                 set_bitmap_uptodate(bh);
  74         }
  75         unlock_buffer(bh);
  76         put_bh(bh);
  77 }
  78
  79 static int ext4_validate_inode_bitmap(struct super_block *sb,
  80                                       struct ext4_group_desc *desc,
  81                                       ext4_group_t block_group,
  82                                       struct buffer_head *bh)
  83 {
  84         ext4_fsblk_t    blk;
  85         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
  86
  87         if (buffer_verified(bh))
  88                 return 0;
  89         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
  90                 return -EFSCORRUPTED;
  91
  92         ext4_lock_group(sb, block_group);
  93         blk = ext4_inode_bitmap(sb, desc);
  94         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
  95                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
  96                 ext4_unlock_group(sb, block_group);
  97                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
  98                            "inode_bitmap = %llu", block_group, blk);
  99                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 100                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 101                 return -EFSBADCRC;
 102         }
 103         set_buffer_verified(bh);
 104         ext4_unlock_group(sb, block_group);
 105         return 0;
 106 }
 107
 108 /*
 109  * Read the inode allocation bitmap for a given block_group, reading
 110  * into the specified slot in the superblock's bitmap cache.
 111  *
 112  * Return buffer_head of bitmap on success or NULL.
 113  */
 114 static struct buffer_head *
 115 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 116 {
 117         struct ext4_group_desc *desc;
 118         struct ext4_sb_info *sbi = EXT4_SB(sb);
 119         struct buffer_head *bh = NULL;
 120         ext4_fsblk_t bitmap_blk;
 121         int err;
 122
 123         desc = ext4_get_group_desc(sb, block_group, NULL);
 124         if (!desc)
 125                 return ERR_PTR(-EFSCORRUPTED);
 126
 127         bitmap_blk = ext4_inode_bitmap(sb, desc);
 128         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 129             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 130                 ext4_error(sb, "Invalid inode bitmap blk %llu in "
 131                            "block_group %u", bitmap_blk, block_group);
 132                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 133                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 134                 return ERR_PTR(-EFSCORRUPTED);
 135         }
 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 ERR_PTR(-ENOMEM);
 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                 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 155                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 156                                      sb->s_blocksize * 8, bh->b_data);
 157                 set_bitmap_uptodate(bh);
 158                 set_buffer_uptodate(bh);
 159                 set_buffer_verified(bh);
 160                 ext4_unlock_group(sb, block_group);
 161                 unlock_buffer(bh);
 162                 return bh;
 163         }
 164         ext4_unlock_group(sb, block_group);
 165
 166         if (buffer_uptodate(bh)) {
 167                 /*
 168                  * if not uninit if bh is uptodate,
 169                  * bitmap is also uptodate
 170                  */
 171                 set_bitmap_uptodate(bh);
 172                 unlock_buffer(bh);
 173                 goto verify;
 174         }
 175         /*
 176          * submit the buffer_head for reading
 177          */
 178         trace_ext4_load_inode_bitmap(sb, block_group);
 179         bh->b_end_io = ext4_end_bitmap_read;
 180         get_bh(bh);
 181         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 182         wait_on_buffer(bh);
 183         if (!buffer_uptodate(bh)) {
 184                 put_bh(bh);
 185                 ext4_error(sb, "Cannot read inode bitmap - "
 186                            "block_group = %u, inode_bitmap = %llu",
 187                            block_group, bitmap_blk);
 188                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 189                                 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 190                 return ERR_PTR(-EIO);
 191         }
 192
 193 verify:
 194         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 195         if (err)
 196                 goto out;
 197         return bh;
 198 out:
 199         put_bh(bh);
 200         return ERR_PTR(err);
 201 }
 202
 203 /*
 204  * NOTE! When we get the inode, we're the only people
 205  * that have access to it, and as such there are no
 206  * race conditions we have to worry about. The inode
 207  * is not on the hash-lists, and it cannot be reached
 208  * through the filesystem because the directory entry
 209  * has been deleted earlier.
 210  *
 211  * HOWEVER: we must make sure that we get no aliases,
 212  * which means that we have to call "clear_inode()"
 213  * _before_ we mark the inode not in use in the inode
 214  * bitmaps. Otherwise a newly created file might use
 215  * the same inode number (not actually the same pointer
 216  * though), and then we'd have two inodes sharing the
 217  * same inode number and space on the harddisk.
 218  */
 219 void ext4_free_inode(handle_t *handle, struct inode *inode)
 220 {
 221         struct super_block *sb = inode->i_sb;
 222         int is_directory;
 223         unsigned long ino;
 224         struct buffer_head *bitmap_bh = NULL;
 225         struct buffer_head *bh2;
 226         ext4_group_t block_group;
 227         unsigned long bit;
 228         struct ext4_group_desc *gdp;
 229         struct ext4_super_block *es;
 230         struct ext4_sb_info *sbi;
 231         int fatal = 0, err, count, cleared;
 232         struct ext4_group_info *grp;
 233
 234         if (!sb) {
 235                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 236                        "nonexistent device\n", __func__, __LINE__);
 237                 return;
 238         }
 239         if (atomic_read(&inode->i_count) > 1) {
 240                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 241                          __func__, __LINE__, inode->i_ino,
 242                          atomic_read(&inode->i_count));
 243                 return;
 244         }
 245         if (inode->i_nlink) {
 246                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 247                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 248                 return;
 249         }
 250         sbi = EXT4_SB(sb);
 251
 252         ino = inode->i_ino;
 253         ext4_debug("freeing inode %lu\n", ino);
 254         trace_ext4_free_inode(inode);
 255
 256         /*
 257          * Note: we must free any quota before locking the superblock,
 258          * as writing the quota to disk may need the lock as well.
 259          */
 260         dquot_initialize(inode);
 261         dquot_free_inode(inode);
 262         dquot_drop(inode);
 263
 264         is_directory = S_ISDIR(inode->i_mode);
 265
 266         /* Do this BEFORE marking the inode not in use or returning an error */
 267         ext4_clear_inode(inode);
 268
 269         es = sbi->s_es;
 270         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 271                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 272                 goto error_return;
 273         }
 274         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 275         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 276         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 277         /* Don't bother if the inode bitmap is corrupt. */
 278         grp = ext4_get_group_info(sb, block_group);
 279         if (IS_ERR(bitmap_bh)) {
 280                 fatal = PTR_ERR(bitmap_bh);
 281                 bitmap_bh = NULL;
 282                 goto error_return;
 283         }
 284         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 285                 fatal = -EFSCORRUPTED;
 286                 goto error_return;
 287         }
 288
 289         BUFFER_TRACE(bitmap_bh, "get_write_access");
 290         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 291         if (fatal)
 292                 goto error_return;
 293
 294         fatal = -ESRCH;
 295         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 296         if (gdp) {
 297                 BUFFER_TRACE(bh2, "get_write_access");
 298                 fatal = ext4_journal_get_write_access(handle, bh2);
 299         }
 300         ext4_lock_group(sb, block_group);
 301         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 302         if (fatal || !cleared) {
 303                 ext4_unlock_group(sb, block_group);
 304                 goto out;
 305         }
 306
 307         count = ext4_free_inodes_count(sb, gdp) + 1;
 308         ext4_free_inodes_set(sb, gdp, count);
 309         if (is_directory) {
 310                 count = ext4_used_dirs_count(sb, gdp) - 1;
 311                 ext4_used_dirs_set(sb, gdp, count);
 312                 percpu_counter_dec(&sbi->s_dirs_counter);
 313         }
 314         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 315                                    EXT4_INODES_PER_GROUP(sb) / 8);
 316         ext4_group_desc_csum_set(sb, block_group, gdp);
 317         ext4_unlock_group(sb, block_group);
 318
 319         percpu_counter_inc(&sbi->s_freeinodes_counter);
 320         if (sbi->s_log_groups_per_flex) {
 321                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 322
 323                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 324                 if (is_directory)
 325                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 326         }
 327         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 328         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 329 out:
 330         if (cleared) {
 331                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 332                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 333                 if (!fatal)
 334                         fatal = err;
 335         } else {
 336                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 337                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 338                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 339         }
 340
 341 error_return:
 342         brelse(bitmap_bh);
 343         ext4_std_error(sb, fatal);
 344 }
 345
 346 struct orlov_stats {
 347         __u64 free_clusters;
 348         __u32 free_inodes;
 349         __u32 used_dirs;
 350 };
 351
 352 /*
 353  * Helper function for Orlov's allocator; returns critical information
 354  * for a particular block group or flex_bg.  If flex_size is 1, then g
 355  * is a block group number; otherwise it is flex_bg number.
 356  */
 357 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 358                             int flex_size, struct orlov_stats *stats)
 359 {
 360         struct ext4_group_desc *desc;
 361         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 362
 363         if (flex_size > 1) {
 364                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 365                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 366                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 367                 return;
 368         }
 369
 370         desc = ext4_get_group_desc(sb, g, NULL);
 371         if (desc) {
 372                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 373                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 374                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 375         } else {
 376                 stats->free_inodes = 0;
 377                 stats->free_clusters = 0;
 378                 stats->used_dirs = 0;
 379         }
 380 }
 381
 382 /*
 383  * Orlov's allocator for directories.
 384  *
 385  * We always try to spread first-level directories.
 386  *
 387  * If there are blockgroups with both free inodes and free blocks counts
 388  * not worse than average we return one with smallest directory count.
 389  * Otherwise we simply return a random group.
 390  *
 391  * For the rest rules look so:
 392  *
 393  * It's OK to put directory into a group unless
 394  * it has too many directories already (max_dirs) or
 395  * it has too few free inodes left (min_inodes) or
 396  * it has too few free blocks left (min_blocks) or
 397  * Parent's group is preferred, if it doesn't satisfy these
 398  * conditions we search cyclically through the rest. If none
 399  * of the groups look good we just look for a group with more
 400  * free inodes than average (starting at parent's group).
 401  */
 402
 403 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 404                             ext4_group_t *group, umode_t mode,
 405                             const struct qstr *qstr)
 406 {
 407         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 408         struct ext4_sb_info *sbi = EXT4_SB(sb);
 409         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 410         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 411         unsigned int freei, avefreei, grp_free;
 412         ext4_fsblk_t freeb, avefreec;
 413         unsigned int ndirs;
 414         int max_dirs, min_inodes;
 415         ext4_grpblk_t min_clusters;
 416         ext4_group_t i, grp, g, ngroups;
 417         struct ext4_group_desc *desc;
 418         struct orlov_stats stats;
 419         int flex_size = ext4_flex_bg_size(sbi);
 420         struct dx_hash_info hinfo;
 421
 422         ngroups = real_ngroups;
 423         if (flex_size > 1) {
 424                 ngroups = (real_ngroups + flex_size - 1) >>
 425                         sbi->s_log_groups_per_flex;
 426                 parent_group >>= sbi->s_log_groups_per_flex;
 427         }
 428
 429         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 430         avefreei = freei / ngroups;
 431         freeb = EXT4_C2B(sbi,
 432                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 433         avefreec = freeb;
 434         do_div(avefreec, ngroups);
 435         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 436
 437         if (S_ISDIR(mode) &&
 438             ((parent == d_inode(sb->s_root)) ||
 439              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 440                 int best_ndir = inodes_per_group;
 441                 int ret = -1;
 442
 443                 if (qstr) {
 444                         hinfo.hash_version = DX_HASH_HALF_MD4;
 445                         hinfo.seed = sbi->s_hash_seed;
 446                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 447                         grp = hinfo.hash;
 448                 } else
 449                         grp = prandom_u32();
 450                 parent_group = (unsigned)grp % ngroups;
 451                 for (i = 0; i < ngroups; i++) {
 452                         g = (parent_group + i) % ngroups;
 453                         get_orlov_stats(sb, g, flex_size, &stats);
 454                         if (!stats.free_inodes)
 455                                 continue;
 456                         if (stats.used_dirs >= best_ndir)
 457                                 continue;
 458                         if (stats.free_inodes < avefreei)
 459                                 continue;
 460                         if (stats.free_clusters < avefreec)
 461                                 continue;
 462                         grp = g;
 463                         ret = 0;
 464                         best_ndir = stats.used_dirs;
 465                 }
 466                 if (ret)
 467                         goto fallback;
 468         found_flex_bg:
 469                 if (flex_size == 1) {
 470                         *group = grp;
 471                         return 0;
 472                 }
 473
 474                 /*
 475                  * We pack inodes at the beginning of the flexgroup's
 476                  * inode tables.  Block allocation decisions will do
 477                  * something similar, although regular files will
 478                  * start at 2nd block group of the flexgroup.  See
 479                  * ext4_ext_find_goal() and ext4_find_near().
 480                  */
 481                 grp *= flex_size;
 482                 for (i = 0; i < flex_size; i++) {
 483                         if (grp+i >= real_ngroups)
 484                                 break;
 485                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 486                         if (desc && ext4_free_inodes_count(sb, desc)) {
 487                                 *group = grp+i;
 488                                 return 0;
 489                         }
 490                 }
 491                 goto fallback;
 492         }
 493
 494         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 495         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 496         if (min_inodes < 1)
 497                 min_inodes = 1;
 498         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 499
 500         /*
 501          * Start looking in the flex group where we last allocated an
 502          * inode for this parent directory
 503          */
 504         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 505                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 506                 if (flex_size > 1)
 507                         parent_group >>= sbi->s_log_groups_per_flex;
 508         }
 509
 510         for (i = 0; i < ngroups; i++) {
 511                 grp = (parent_group + i) % ngroups;
 512                 get_orlov_stats(sb, grp, flex_size, &stats);
 513                 if (stats.used_dirs >= max_dirs)
 514                         continue;
 515                 if (stats.free_inodes < min_inodes)
 516                         continue;
 517                 if (stats.free_clusters < min_clusters)
 518                         continue;
 519                 goto found_flex_bg;
 520         }
 521
 522 fallback:
 523         ngroups = real_ngroups;
 524         avefreei = freei / ngroups;
 525 fallback_retry:
 526         parent_group = EXT4_I(parent)->i_block_group;
 527         for (i = 0; i < ngroups; i++) {
 528                 grp = (parent_group + i) % ngroups;
 529                 desc = ext4_get_group_desc(sb, grp, NULL);
 530                 if (desc) {
 531                         grp_free = ext4_free_inodes_count(sb, desc);
 532                         if (grp_free && grp_free >= avefreei) {
 533                                 *group = grp;
 534                                 return 0;
 535                         }
 536                 }
 537         }
 538
 539         if (avefreei) {
 540                 /*
 541                  * The free-inodes counter is approximate, and for really small
 542                  * filesystems the above test can fail to find any blockgroups
 543                  */
 544                 avefreei = 0;
 545                 goto fallback_retry;
 546         }
 547
 548         return -1;
 549 }
 550
 551 static int find_group_other(struct super_block *sb, struct inode *parent,
 552                             ext4_group_t *group, umode_t mode)
 553 {
 554         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 555         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 556         struct ext4_group_desc *desc;
 557         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 558
 559         /*
 560          * Try to place the inode is the same flex group as its
 561          * parent.  If we can't find space, use the Orlov algorithm to
 562          * find another flex group, and store that information in the
 563          * parent directory's inode information so that use that flex
 564          * group for future allocations.
 565          */
 566         if (flex_size > 1) {
 567                 int retry = 0;
 568
 569         try_again:
 570                 parent_group &= ~(flex_size-1);
 571                 last = parent_group + flex_size;
 572                 if (last > ngroups)
 573                         last = ngroups;
 574                 for  (i = parent_group; i < last; i++) {
 575                         desc = ext4_get_group_desc(sb, i, NULL);
 576                         if (desc && ext4_free_inodes_count(sb, desc)) {
 577                                 *group = i;
 578                                 return 0;
 579                         }
 580                 }
 581                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 582                         retry = 1;
 583                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 584                         goto try_again;
 585                 }
 586                 /*
 587                  * If this didn't work, use the Orlov search algorithm
 588                  * to find a new flex group; we pass in the mode to
 589                  * avoid the topdir algorithms.
 590                  */
 591                 *group = parent_group + flex_size;
 592                 if (*group > ngroups)
 593                         *group = 0;
 594                 return find_group_orlov(sb, parent, group, mode, NULL);
 595         }
 596
 597         /*
 598          * Try to place the inode in its parent directory
 599          */
 600         *group = parent_group;
 601         desc = ext4_get_group_desc(sb, *group, NULL);
 602         if (desc && ext4_free_inodes_count(sb, desc) &&
 603             ext4_free_group_clusters(sb, desc))
 604                 return 0;
 605
 606         /*
 607          * We're going to place this inode in a different blockgroup from its
 608          * parent.  We want to cause files in a common directory to all land in
 609          * the same blockgroup.  But we want files which are in a different
 610          * directory which shares a blockgroup with our parent to land in a
 611          * different blockgroup.
 612          *
 613          * So add our directory's i_ino into the starting point for the hash.
 614          */
 615         *group = (*group + parent->i_ino) % ngroups;
 616
 617         /*
 618          * Use a quadratic hash to find a group with a free inode and some free
 619          * blocks.
 620          */
 621         for (i = 1; i < ngroups; i <<= 1) {
 622                 *group += i;
 623                 if (*group >= ngroups)
 624                         *group -= ngroups;
 625                 desc = ext4_get_group_desc(sb, *group, NULL);
 626                 if (desc && ext4_free_inodes_count(sb, desc) &&
 627                     ext4_free_group_clusters(sb, desc))
 628                         return 0;
 629         }
 630
 631         /*
 632          * That failed: try linear search for a free inode, even if that group
 633          * has no free blocks.
 634          */
 635         *group = parent_group;
 636         for (i = 0; i < ngroups; i++) {
 637                 if (++*group >= ngroups)
 638                         *group = 0;
 639                 desc = ext4_get_group_desc(sb, *group, NULL);
 640                 if (desc && ext4_free_inodes_count(sb, desc))
 641                         return 0;
 642         }
 643
 644         return -1;
 645 }
 646
 647 /*
 648  * In no journal mode, if an inode has recently been deleted, we want
 649  * to avoid reusing it until we're reasonably sure the inode table
 650  * block has been written back to disk.  (Yes, these values are
 651  * somewhat arbitrary...)
 652  */
 653 #define RECENTCY_MIN    5
 654 #define RECENTCY_DIRTY  300
 655
 656 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 657 {
 658         struct ext4_group_desc  *gdp;
 659         struct ext4_inode       *raw_inode;
 660         struct buffer_head      *bh;
 661         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 662         int offset, ret = 0;
 663         int recentcy = RECENTCY_MIN;
 664         u32 dtime, now;
 665
 666         gdp = ext4_get_group_desc(sb, group, NULL);
 667         if (unlikely(!gdp))
 668                 return 0;
 669
 670         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 671                        (ino / inodes_per_block));
 672         if (!bh || !buffer_uptodate(bh))
 673                 /*
 674                  * If the block is not in the buffer cache, then it
 675                  * must have been written out.
 676                  */
 677                 goto out;
 678
 679         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 680         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 681
 682         /* i_dtime is only 32 bits on disk, but we only care about relative
 683          * times in the range of a few minutes (i.e. long enough to sync a
 684          * recently-deleted inode to disk), so using the low 32 bits of the
 685          * clock (a 68 year range) is enough, see time_before32() */
 686         dtime = le32_to_cpu(raw_inode->i_dtime);
 687         now = ktime_get_real_seconds();
 688         if (buffer_dirty(bh))
 689                 recentcy += RECENTCY_DIRTY;
 690
 691         if (dtime && time_before32(dtime, now) &&
 692             time_before32(now, dtime + recentcy))
 693                 ret = 1;
 694 out:
 695         brelse(bh);
 696         return ret;
 697 }
 698
 699 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 700                           struct buffer_head *bitmap, unsigned long *ino)
 701 {
 702 next:
 703         *ino = ext4_find_next_zero_bit((unsigned long *)
 704                                        bitmap->b_data,
 705                                        EXT4_INODES_PER_GROUP(sb), *ino);
 706         if (*ino >= EXT4_INODES_PER_GROUP(sb))
 707                 return 0;
 708
 709         if ((EXT4_SB(sb)->s_journal == NULL) &&
 710             recently_deleted(sb, group, *ino)) {
 711                 *ino = *ino + 1;
 712                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 713                         goto next;
 714                 return 0;
 715         }
 716
 717         return 1;
 718 }
 719
 720 /*
 721  * There are two policies for allocating an inode.  If the new inode is
 722  * a directory, then a forward search is made for a block group with both
 723  * free space and a low directory-to-inode ratio; if that fails, then of
 724  * the groups with above-average free space, that group with the fewest
 725  * directories already is chosen.
 726  *
 727  * For other inodes, search forward from the parent directory's block
 728  * group to find a free inode.
 729  */
 730 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 731                                umode_t mode, const struct qstr *qstr,
 732                                __u32 goal, uid_t *owner, __u32 i_flags,
 733                                int handle_type, unsigned int line_no,
 734                                int nblocks)
 735 {
 736         struct super_block *sb;
 737         struct buffer_head *inode_bitmap_bh = NULL;
 738         struct buffer_head *group_desc_bh;
 739         ext4_group_t ngroups, group = 0;
 740         unsigned long ino = 0;
 741         struct inode *inode;
 742         struct ext4_group_desc *gdp = NULL;
 743         struct ext4_inode_info *ei;
 744         struct ext4_sb_info *sbi;
 745         int ret2, err;
 746         struct inode *ret;
 747         ext4_group_t i;
 748         ext4_group_t flex_group;
 749         struct ext4_group_info *grp;
 750         int encrypt = 0;
 751
 752         /* Cannot create files in a deleted directory */
 753         if (!dir || !dir->i_nlink)
 754                 return ERR_PTR(-EPERM);
 755
 756         sb = dir->i_sb;
 757         sbi = EXT4_SB(sb);
 758
 759         if (unlikely(ext4_forced_shutdown(sbi)))
 760                 return ERR_PTR(-EIO);
 761
 762         if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
 763             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
 764             !(i_flags & EXT4_EA_INODE_FL)) {
 765                 err = fscrypt_get_encryption_info(dir);
 766                 if (err)
 767                         return ERR_PTR(err);
 768                 if (!fscrypt_has_encryption_key(dir))
 769                         return ERR_PTR(-ENOKEY);
 770                 encrypt = 1;
 771         }
 772
 773         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 774 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 775                 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 776
 777                 if (IS_ERR(p))
 778                         return ERR_CAST(p);
 779                 if (p) {
 780                         int acl_size = p->a_count * sizeof(ext4_acl_entry);
 781
 782                         nblocks += (S_ISDIR(mode) ? 2 : 1) *
 783                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 784                                         NULL /* block_bh */, acl_size,
 785                                         true /* is_create */);
 786                         posix_acl_release(p);
 787                 }
 788 #endif
 789
 790 #ifdef CONFIG_SECURITY
 791                 {
 792                         int num_security_xattrs = 1;
 793
 794 #ifdef CONFIG_INTEGRITY
 795                         num_security_xattrs++;
 796 #endif
 797                         /*
 798                          * We assume that security xattrs are never
 799                          * more than 1k.  In practice they are under
 800                          * 128 bytes.
 801                          */
 802                         nblocks += num_security_xattrs *
 803                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 804                                         NULL /* block_bh */, 1024,
 805                                         true /* is_create */);
 806                 }
 807 #endif
 808                 if (encrypt)
 809                         nblocks += __ext4_xattr_set_credits(sb,
 810                                         NULL /* inode */, NULL /* block_bh */,
 811                                         FSCRYPT_SET_CONTEXT_MAX_SIZE,
 812                                         true /* is_create */);
 813         }
 814
 815         ngroups = ext4_get_groups_count(sb);
 816         trace_ext4_request_inode(dir, mode);
 817         inode = new_inode(sb);
 818         if (!inode)
 819                 return ERR_PTR(-ENOMEM);
 820         ei = EXT4_I(inode);
 821
 822         /*
 823          * Initialize owners and quota early so that we don't have to account
 824          * for quota initialization worst case in standard inode creating
 825          * transaction
 826          */
 827         if (owner) {
 828                 inode->i_mode = mode;
 829                 i_uid_write(inode, owner[0]);
 830                 i_gid_write(inode, owner[1]);
 831         } else if (test_opt(sb, GRPID)) {
 832                 inode->i_mode = mode;
 833                 inode->i_uid = current_fsuid();
 834                 inode->i_gid = dir->i_gid;
 835         } else
 836                 inode_init_owner(inode, dir, mode);
 837
 838         if (ext4_has_feature_project(sb) &&
 839             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 840                 ei->i_projid = EXT4_I(dir)->i_projid;
 841         else
 842                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 843
 844         err = dquot_initialize(inode);
 845         if (err)
 846                 goto out;
 847
 848         if (!goal)
 849                 goal = sbi->s_inode_goal;
 850
 851         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 852                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 853                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 854                 ret2 = 0;
 855                 goto got_group;
 856         }
 857
 858         if (S_ISDIR(mode))
 859                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 860         else
 861                 ret2 = find_group_other(sb, dir, &group, mode);
 862
 863 got_group:
 864         EXT4_I(dir)->i_last_alloc_group = group;
 865         err = -ENOSPC;
 866         if (ret2 == -1)
 867                 goto out;
 868
 869         /*
 870          * Normally we will only go through one pass of this loop,
 871          * unless we get unlucky and it turns out the group we selected
 872          * had its last inode grabbed by someone else.
 873          */
 874         for (i = 0; i < ngroups; i++, ino = 0) {
 875                 err = -EIO;
 876
 877                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 878                 if (!gdp)
 879                         goto out;
 880
 881                 /*
 882                  * Check free inodes count before loading bitmap.
 883                  */
 884                 if (ext4_free_inodes_count(sb, gdp) == 0)
 885                         goto next_group;
 886
 887                 grp = ext4_get_group_info(sb, group);
 888                 /* Skip groups with already-known suspicious inode tables */
 889                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 890                         goto next_group;
 891
 892                 brelse(inode_bitmap_bh);
 893                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 894                 /* Skip groups with suspicious inode tables */
 895                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 896                     IS_ERR(inode_bitmap_bh)) {
 897                         inode_bitmap_bh = NULL;
 898                         goto next_group;
 899                 }
 900
 901 repeat_in_this_group:
 902                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 903                 if (!ret2)
 904                         goto next_group;
 905
 906                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 907                         ext4_error(sb, "reserved inode found cleared - "
 908                                    "inode=%lu", ino + 1);
 909                         ext4_mark_group_bitmap_corrupted(sb, group,
 910                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 911                         goto next_group;
 912                 }
 913
 914                 if (!handle) {
 915                         BUG_ON(nblocks <= 0);
 916                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 917                                                          handle_type, nblocks,
 918                                                          0);
 919                         if (IS_ERR(handle)) {
 920                                 err = PTR_ERR(handle);
 921                                 ext4_std_error(sb, err);
 922                                 goto out;
 923                         }
 924                 }
 925                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 926                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 927                 if (err) {
 928                         ext4_std_error(sb, err);
 929                         goto out;
 930                 }
 931                 ext4_lock_group(sb, group);
 932                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 933                 if (ret2) {
 934                         /* Someone already took the bit. Repeat the search
 935                          * with lock held.
 936                          */
 937                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 938                         if (ret2) {
 939                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
 940                                 ret2 = 0;
 941                         } else {
 942                                 ret2 = 1; /* we didn't grab the inode */
 943                         }
 944                 }
 945                 ext4_unlock_group(sb, group);
 946                 ino++;          /* the inode bitmap is zero-based */
 947                 if (!ret2)
 948                         goto got; /* we grabbed the inode! */
 949
 950                 if (ino < EXT4_INODES_PER_GROUP(sb))
 951                         goto repeat_in_this_group;
 952 next_group:
 953                 if (++group == ngroups)
 954                         group = 0;
 955         }
 956         err = -ENOSPC;
 957         goto out;
 958
 959 got:
 960         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 961         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 962         if (err) {
 963                 ext4_std_error(sb, err);
 964                 goto out;
 965         }
 966
 967         BUFFER_TRACE(group_desc_bh, "get_write_access");
 968         err = ext4_journal_get_write_access(handle, group_desc_bh);
 969         if (err) {
 970                 ext4_std_error(sb, err);
 971                 goto out;
 972         }
 973
 974         /* We may have to initialize the block bitmap if it isn't already */
 975         if (ext4_has_group_desc_csum(sb) &&
 976             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 977                 struct buffer_head *block_bitmap_bh;
 978
 979                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 980                 if (IS_ERR(block_bitmap_bh)) {
 981                         err = PTR_ERR(block_bitmap_bh);
 982                         goto out;
 983                 }
 984                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 985                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 986                 if (err) {
 987                         brelse(block_bitmap_bh);
 988                         ext4_std_error(sb, err);
 989                         goto out;
 990                 }
 991
 992                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 993                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
 994
 995                 /* recheck and clear flag under lock if we still need to */
 996                 ext4_lock_group(sb, group);
 997                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 998                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 999                         ext4_free_group_clusters_set(sb, gdp,
1000                                 ext4_free_clusters_after_init(sb, group, gdp));
1001                         ext4_block_bitmap_csum_set(sb, group, gdp,
1002                                                    block_bitmap_bh);
1003                         ext4_group_desc_csum_set(sb, group, gdp);
1004                 }
1005                 ext4_unlock_group(sb, group);
1006                 brelse(block_bitmap_bh);
1007
1008                 if (err) {
1009                         ext4_std_error(sb, err);
1010                         goto out;
1011                 }
1012         }
1013
1014         /* Update the relevant bg descriptor fields */
1015         if (ext4_has_group_desc_csum(sb)) {
1016                 int free;
1017                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1018
1019                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1020                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1021                 free = EXT4_INODES_PER_GROUP(sb) -
1022                         ext4_itable_unused_count(sb, gdp);
1023                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1024                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1025                         free = 0;
1026                 }
1027                 /*
1028                  * Check the relative inode number against the last used
1029                  * relative inode number in this group. if it is greater
1030                  * we need to update the bg_itable_unused count
1031                  */
1032                 if (ino > free)
1033                         ext4_itable_unused_set(sb, gdp,
1034                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1035                 up_read(&grp->alloc_sem);
1036         } else {
1037                 ext4_lock_group(sb, group);
1038         }
1039
1040         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1041         if (S_ISDIR(mode)) {
1042                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1043                 if (sbi->s_log_groups_per_flex) {
1044                         ext4_group_t f = ext4_flex_group(sbi, group);
1045
1046                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1047                 }
1048         }
1049         if (ext4_has_group_desc_csum(sb)) {
1050                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1051                                            EXT4_INODES_PER_GROUP(sb) / 8);
1052                 ext4_group_desc_csum_set(sb, group, gdp);
1053         }
1054         ext4_unlock_group(sb, group);
1055
1056         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1057         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1058         if (err) {
1059                 ext4_std_error(sb, err);
1060                 goto out;
1061         }
1062
1063         percpu_counter_dec(&sbi->s_freeinodes_counter);
1064         if (S_ISDIR(mode))
1065                 percpu_counter_inc(&sbi->s_dirs_counter);
1066
1067         if (sbi->s_log_groups_per_flex) {
1068                 flex_group = ext4_flex_group(sbi, group);
1069                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1070         }
1071
1072         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1073         /* This is the optimal IO size (for stat), not the fs block size */
1074         inode->i_blocks = 0;
1075         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1076         ei->i_crtime = timespec64_to_timespec(inode->i_mtime);
1077
1078         memset(ei->i_data, 0, sizeof(ei->i_data));
1079         ei->i_dir_start_lookup = 0;
1080         ei->i_disksize = 0;
1081
1082         /* Don't inherit extent flag from directory, amongst others. */
1083         ei->i_flags =
1084                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1085         ei->i_flags |= i_flags;
1086         ei->i_file_acl = 0;
1087         ei->i_dtime = 0;
1088         ei->i_block_group = group;
1089         ei->i_last_alloc_group = ~0;
1090
1091         ext4_set_inode_flags(inode);
1092         if (IS_DIRSYNC(inode))
1093                 ext4_handle_sync(handle);
1094         if (insert_inode_locked(inode) < 0) {
1095                 /*
1096                  * Likely a bitmap corruption causing inode to be allocated
1097                  * twice.
1098                  */
1099                 err = -EIO;
1100                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1101                            inode->i_ino);
1102                 ext4_mark_group_bitmap_corrupted(sb, group,
1103                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1104                 goto out;
1105         }
1106         inode->i_generation = prandom_u32();
1107
1108         /* Precompute checksum seed for inode metadata */
1109         if (ext4_has_metadata_csum(sb)) {
1110                 __u32 csum;
1111                 __le32 inum = cpu_to_le32(inode->i_ino);
1112                 __le32 gen = cpu_to_le32(inode->i_generation);
1113                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1114                                    sizeof(inum));
1115                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1116                                               sizeof(gen));
1117         }
1118
1119         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1120         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1121
1122         ei->i_extra_isize = sbi->s_want_extra_isize;
1123         ei->i_inline_off = 0;
1124         if (ext4_has_feature_inline_data(sb))
1125                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1126         ret = inode;
1127         err = dquot_alloc_inode(inode);
1128         if (err)
1129                 goto fail_drop;
1130
1131         /*
1132          * Since the encryption xattr will always be unique, create it first so
1133          * that it's less likely to end up in an external xattr block and
1134          * prevent its deduplication.
1135          */
1136         if (encrypt) {
1137                 err = fscrypt_inherit_context(dir, inode, handle, true);
1138                 if (err)
1139                         goto fail_free_drop;
1140         }
1141
1142         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1143                 err = ext4_init_acl(handle, inode, dir);
1144                 if (err)
1145                         goto fail_free_drop;
1146
1147                 err = ext4_init_security(handle, inode, dir, qstr);
1148                 if (err)
1149                         goto fail_free_drop;
1150         }
1151
1152         if (ext4_has_feature_extents(sb)) {
1153                 /* set extent flag only for directory, file and normal symlink*/
1154                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1155                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1156                         ext4_ext_tree_init(handle, inode);
1157                 }
1158         }
1159
1160         if (ext4_handle_valid(handle)) {
1161                 ei->i_sync_tid = handle->h_transaction->t_tid;
1162                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1163         }
1164
1165         err = ext4_mark_inode_dirty(handle, inode);
1166         if (err) {
1167                 ext4_std_error(sb, err);
1168                 goto fail_free_drop;
1169         }
1170
1171         ext4_debug("allocating inode %lu\n", inode->i_ino);
1172         trace_ext4_allocate_inode(inode, dir, mode);
1173         brelse(inode_bitmap_bh);
1174         return ret;
1175
1176 fail_free_drop:
1177         dquot_free_inode(inode);
1178 fail_drop:
1179         clear_nlink(inode);
1180         unlock_new_inode(inode);
1181 out:
1182         dquot_drop(inode);
1183         inode->i_flags |= S_NOQUOTA;
1184         iput(inode);
1185         brelse(inode_bitmap_bh);
1186         return ERR_PTR(err);
1187 }
1188
1189 /* Verify that we are loading a valid orphan from disk */
1190 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1191 {
1192         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1193         ext4_group_t block_group;
1194         int bit;
1195         struct buffer_head *bitmap_bh = NULL;
1196         struct inode *inode = NULL;
1197         int err = -EFSCORRUPTED;
1198
1199         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1200                 goto bad_orphan;
1201
1202         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1203         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1204         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1205         if (IS_ERR(bitmap_bh))
1206                 return (struct inode *) bitmap_bh;
1207
1208         /* Having the inode bit set should be a 100% indicator that this
1209          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1210          * inodes that were being truncated, so we can't check i_nlink==0.
1211          */
1212         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1213                 goto bad_orphan;
1214
1215         inode = ext4_iget(sb, ino);
1216         if (IS_ERR(inode)) {
1217                 err = PTR_ERR(inode);
1218                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1219                            ino, err);
1220                 return inode;
1221         }
1222
1223         /*
1224          * If the orphans has i_nlinks > 0 then it should be able to
1225          * be truncated, otherwise it won't be removed from the orphan
1226          * list during processing and an infinite loop will result.
1227          * Similarly, it must not be a bad inode.
1228          */
1229         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1230             is_bad_inode(inode))
1231                 goto bad_orphan;
1232
1233         if (NEXT_ORPHAN(inode) > max_ino)
1234                 goto bad_orphan;
1235         brelse(bitmap_bh);
1236         return inode;
1237
1238 bad_orphan:
1239         ext4_error(sb, "bad orphan inode %lu", ino);
1240         if (bitmap_bh)
1241                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1242                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1243                        ext4_test_bit(bit, bitmap_bh->b_data));
1244         if (inode) {
1245                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1246                        is_bad_inode(inode));
1247                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1248                        NEXT_ORPHAN(inode));
1249                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1250                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1251                 /* Avoid freeing blocks if we got a bad deleted inode */
1252                 if (inode->i_nlink == 0)
1253                         inode->i_blocks = 0;
1254                 iput(inode);
1255         }
1256         brelse(bitmap_bh);
1257         return ERR_PTR(err);
1258 }
1259
1260 unsigned long ext4_count_free_inodes(struct super_block *sb)
1261 {
1262         unsigned long desc_count;
1263         struct ext4_group_desc *gdp;
1264         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1265 #ifdef EXT4FS_DEBUG
1266         struct ext4_super_block *es;
1267         unsigned long bitmap_count, x;
1268         struct buffer_head *bitmap_bh = NULL;
1269
1270         es = EXT4_SB(sb)->s_es;
1271         desc_count = 0;
1272         bitmap_count = 0;
1273         gdp = NULL;
1274         for (i = 0; i < ngroups; i++) {
1275                 gdp = ext4_get_group_desc(sb, i, NULL);
1276                 if (!gdp)
1277                         continue;
1278                 desc_count += ext4_free_inodes_count(sb, gdp);
1279                 brelse(bitmap_bh);
1280                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1281                 if (IS_ERR(bitmap_bh)) {
1282                         bitmap_bh = NULL;
1283                         continue;
1284                 }
1285
1286                 x = ext4_count_free(bitmap_bh->b_data,
1287                                     EXT4_INODES_PER_GROUP(sb) / 8);
1288                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1289                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1290                 bitmap_count += x;
1291         }
1292         brelse(bitmap_bh);
1293         printk(KERN_DEBUG "ext4_count_free_inodes: "
1294                "stored = %u, computed = %lu, %lu\n",
1295                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1296         return desc_count;
1297 #else
1298         desc_count = 0;
1299         for (i = 0; i < ngroups; i++) {
1300                 gdp = ext4_get_group_desc(sb, i, NULL);
1301                 if (!gdp)
1302                         continue;
1303                 desc_count += ext4_free_inodes_count(sb, gdp);
1304                 cond_resched();
1305         }
1306         return desc_count;
1307 #endif
1308 }
1309
1310 /* Called at mount-time, super-block is locked */
1311 unsigned long ext4_count_dirs(struct super_block * sb)
1312 {
1313         unsigned long count = 0;
1314         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1315
1316         for (i = 0; i < ngroups; i++) {
1317                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1318                 if (!gdp)
1319                         continue;
1320                 count += ext4_used_dirs_count(sb, gdp);
1321         }
1322         return count;
1323 }
1324
1325 /*
1326  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1327  * inode table. Must be called without any spinlock held. The only place
1328  * where it is called from on active part of filesystem is ext4lazyinit
1329  * thread, so we do not need any special locks, however we have to prevent
1330  * inode allocation from the current group, so we take alloc_sem lock, to
1331  * block ext4_new_inode() until we are finished.
1332  */
1333 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1334                                  int barrier)
1335 {
1336         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1337         struct ext4_sb_info *sbi = EXT4_SB(sb);
1338         struct ext4_group_desc *gdp = NULL;
1339         struct buffer_head *group_desc_bh;
1340         handle_t *handle;
1341         ext4_fsblk_t blk;
1342         int num, ret = 0, used_blks = 0;
1343
1344         /* This should not happen, but just to be sure check this */
1345         if (sb_rdonly(sb)) {
1346                 ret = 1;
1347                 goto out;
1348         }
1349
1350         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1351         if (!gdp)
1352                 goto out;
1353
1354         /*
1355          * We do not need to lock this, because we are the only one
1356          * handling this flag.
1357          */
1358         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1359                 goto out;
1360
1361         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1362         if (IS_ERR(handle)) {
1363                 ret = PTR_ERR(handle);
1364                 goto out;
1365         }
1366
1367         down_write(&grp->alloc_sem);
1368         /*
1369          * If inode bitmap was already initialized there may be some
1370          * used inodes so we need to skip blocks with used inodes in
1371          * inode table.
1372          */
1373         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1374                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1375                             ext4_itable_unused_count(sb, gdp)),
1376                             sbi->s_inodes_per_block);
1377
1378         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1379                 ext4_error(sb, "Something is wrong with group %u: "
1380                            "used itable blocks: %d; "
1381                            "itable unused count: %u",
1382                            group, used_blks,
1383                            ext4_itable_unused_count(sb, gdp));
1384                 ret = 1;
1385                 goto err_out;
1386         }
1387
1388         blk = ext4_inode_table(sb, gdp) + used_blks;
1389         num = sbi->s_itb_per_group - used_blks;
1390
1391         BUFFER_TRACE(group_desc_bh, "get_write_access");
1392         ret = ext4_journal_get_write_access(handle,
1393                                             group_desc_bh);
1394         if (ret)
1395                 goto err_out;
1396
1397         /*
1398          * Skip zeroout if the inode table is full. But we set the ZEROED
1399          * flag anyway, because obviously, when it is full it does not need
1400          * further zeroing.
1401          */
1402         if (unlikely(num == 0))
1403                 goto skip_zeroout;
1404
1405         ext4_debug("going to zero out inode table in group %d\n",
1406                    group);
1407         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1408         if (ret < 0)
1409                 goto err_out;
1410         if (barrier)
1411                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1412
1413 skip_zeroout:
1414         ext4_lock_group(sb, group);
1415         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1416         ext4_group_desc_csum_set(sb, group, gdp);
1417         ext4_unlock_group(sb, group);
1418
1419         BUFFER_TRACE(group_desc_bh,
1420                      "call ext4_handle_dirty_metadata");
1421         ret = ext4_handle_dirty_metadata(handle, NULL,
1422                                          group_desc_bh);
1423
1424 err_out:
1425         up_write(&grp->alloc_sem);
1426         ext4_journal_stop(handle);
1427 out:
1428         return ret;
1429 }