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         if (buffer_verified(bh))
  94                 goto verified;
  95         blk = ext4_inode_bitmap(sb, desc);
  96         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
  97                                            EXT4_INODES_PER_GROUP(sb) / 8) ||
  98             ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
  99                 ext4_unlock_group(sb, block_group);
 100                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 101                            "inode_bitmap = %llu", block_group, blk);
 102                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 103                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 104                 return -EFSBADCRC;
 105         }
 106         set_buffer_verified(bh);
 107 verified:
 108         ext4_unlock_group(sb, block_group);
 109         return 0;
 110 }
 111
 112 /*
 113  * Read the inode allocation bitmap for a given block_group, reading
 114  * into the specified slot in the superblock's bitmap cache.
 115  *
 116  * Return buffer_head of bitmap on success or NULL.
 117  */
 118 static struct buffer_head *
 119 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 120 {
 121         struct ext4_group_desc *desc;
 122         struct ext4_sb_info *sbi = EXT4_SB(sb);
 123         struct buffer_head *bh = NULL;
 124         ext4_fsblk_t bitmap_blk;
 125         int err;
 126
 127         desc = ext4_get_group_desc(sb, block_group, NULL);
 128         if (!desc)
 129                 return ERR_PTR(-EFSCORRUPTED);
 130
 131         bitmap_blk = ext4_inode_bitmap(sb, desc);
 132         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 133             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 134                 ext4_error(sb, "Invalid inode bitmap blk %llu in "
 135                            "block_group %u", bitmap_blk, block_group);
 136                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 137                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 138                 return ERR_PTR(-EFSCORRUPTED);
 139         }
 140         bh = sb_getblk(sb, bitmap_blk);
 141         if (unlikely(!bh)) {
 142                 ext4_warning(sb, "Cannot read inode bitmap - "
 143                              "block_group = %u, inode_bitmap = %llu",
 144                              block_group, bitmap_blk);
 145                 return ERR_PTR(-ENOMEM);
 146         }
 147         if (bitmap_uptodate(bh))
 148                 goto verify;
 149
 150         lock_buffer(bh);
 151         if (bitmap_uptodate(bh)) {
 152                 unlock_buffer(bh);
 153                 goto verify;
 154         }
 155
 156         ext4_lock_group(sb, block_group);
 157         if (ext4_has_group_desc_csum(sb) &&
 158             (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
 159                 if (block_group == 0) {
 160                         ext4_unlock_group(sb, block_group);
 161                         unlock_buffer(bh);
 162                         ext4_error(sb, "Inode bitmap for bg 0 marked "
 163                                    "uninitialized");
 164                         err = -EFSCORRUPTED;
 165                         goto out;
 166                 }
 167                 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 168                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 169                                      sb->s_blocksize * 8, bh->b_data);
 170                 set_bitmap_uptodate(bh);
 171                 set_buffer_uptodate(bh);
 172                 set_buffer_verified(bh);
 173                 ext4_unlock_group(sb, block_group);
 174                 unlock_buffer(bh);
 175                 return bh;
 176         }
 177         ext4_unlock_group(sb, block_group);
 178
 179         if (buffer_uptodate(bh)) {
 180                 /*
 181                  * if not uninit if bh is uptodate,
 182                  * bitmap is also uptodate
 183                  */
 184                 set_bitmap_uptodate(bh);
 185                 unlock_buffer(bh);
 186                 goto verify;
 187         }
 188         /*
 189          * submit the buffer_head for reading
 190          */
 191         trace_ext4_load_inode_bitmap(sb, block_group);
 192         bh->b_end_io = ext4_end_bitmap_read;
 193         get_bh(bh);
 194         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 195         wait_on_buffer(bh);
 196         ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
 197         if (!buffer_uptodate(bh)) {
 198                 put_bh(bh);
 199                 ext4_set_errno(sb, EIO);
 200                 ext4_error(sb, "Cannot read inode bitmap - "
 201                            "block_group = %u, inode_bitmap = %llu",
 202                            block_group, bitmap_blk);
 203                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 204                                 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 205                 return ERR_PTR(-EIO);
 206         }
 207
 208 verify:
 209         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 210         if (err)
 211                 goto out;
 212         return bh;
 213 out:
 214         put_bh(bh);
 215         return ERR_PTR(err);
 216 }
 217
 218 /*
 219  * NOTE! When we get the inode, we're the only people
 220  * that have access to it, and as such there are no
 221  * race conditions we have to worry about. The inode
 222  * is not on the hash-lists, and it cannot be reached
 223  * through the filesystem because the directory entry
 224  * has been deleted earlier.
 225  *
 226  * HOWEVER: we must make sure that we get no aliases,
 227  * which means that we have to call "clear_inode()"
 228  * _before_ we mark the inode not in use in the inode
 229  * bitmaps. Otherwise a newly created file might use
 230  * the same inode number (not actually the same pointer
 231  * though), and then we'd have two inodes sharing the
 232  * same inode number and space on the harddisk.
 233  */
 234 void ext4_free_inode(handle_t *handle, struct inode *inode)
 235 {
 236         struct super_block *sb = inode->i_sb;
 237         int is_directory;
 238         unsigned long ino;
 239         struct buffer_head *bitmap_bh = NULL;
 240         struct buffer_head *bh2;
 241         ext4_group_t block_group;
 242         unsigned long bit;
 243         struct ext4_group_desc *gdp;
 244         struct ext4_super_block *es;
 245         struct ext4_sb_info *sbi;
 246         int fatal = 0, err, count, cleared;
 247         struct ext4_group_info *grp;
 248
 249         if (!sb) {
 250                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 251                        "nonexistent device\n", __func__, __LINE__);
 252                 return;
 253         }
 254         if (atomic_read(&inode->i_count) > 1) {
 255                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 256                          __func__, __LINE__, inode->i_ino,
 257                          atomic_read(&inode->i_count));
 258                 return;
 259         }
 260         if (inode->i_nlink) {
 261                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 262                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 263                 return;
 264         }
 265         sbi = EXT4_SB(sb);
 266
 267         ino = inode->i_ino;
 268         ext4_debug("freeing inode %lu\n", ino);
 269         trace_ext4_free_inode(inode);
 270
 271         dquot_initialize(inode);
 272         dquot_free_inode(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 = sbi->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 (IS_ERR(bitmap_bh)) {
 290                 fatal = PTR_ERR(bitmap_bh);
 291                 bitmap_bh = NULL;
 292                 goto error_return;
 293         }
 294         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 295                 fatal = -EFSCORRUPTED;
 296                 goto error_return;
 297         }
 298
 299         BUFFER_TRACE(bitmap_bh, "get_write_access");
 300         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 301         if (fatal)
 302                 goto error_return;
 303
 304         fatal = -ESRCH;
 305         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 306         if (gdp) {
 307                 BUFFER_TRACE(bh2, "get_write_access");
 308                 fatal = ext4_journal_get_write_access(handle, bh2);
 309         }
 310         ext4_lock_group(sb, block_group);
 311         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 312         if (fatal || !cleared) {
 313                 ext4_unlock_group(sb, block_group);
 314                 goto out;
 315         }
 316
 317         count = ext4_free_inodes_count(sb, gdp) + 1;
 318         ext4_free_inodes_set(sb, gdp, count);
 319         if (is_directory) {
 320                 count = ext4_used_dirs_count(sb, gdp) - 1;
 321                 ext4_used_dirs_set(sb, gdp, count);
 322                 percpu_counter_dec(&sbi->s_dirs_counter);
 323         }
 324         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 325                                    EXT4_INODES_PER_GROUP(sb) / 8);
 326         ext4_group_desc_csum_set(sb, block_group, gdp);
 327         ext4_unlock_group(sb, block_group);
 328
 329         percpu_counter_inc(&sbi->s_freeinodes_counter);
 330         if (sbi->s_log_groups_per_flex) {
 331                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 332
 333                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 334                 if (is_directory)
 335                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 336         }
 337         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 338         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 339 out:
 340         if (cleared) {
 341                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 342                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 343                 if (!fatal)
 344                         fatal = err;
 345         } else {
 346                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 347                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 348                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 349         }
 350
 351 error_return:
 352         brelse(bitmap_bh);
 353         ext4_std_error(sb, fatal);
 354 }
 355
 356 struct orlov_stats {
 357         __u64 free_clusters;
 358         __u32 free_inodes;
 359         __u32 used_dirs;
 360 };
 361
 362 /*
 363  * Helper function for Orlov's allocator; returns critical information
 364  * for a particular block group or flex_bg.  If flex_size is 1, then g
 365  * is a block group number; otherwise it is flex_bg number.
 366  */
 367 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 368                             int flex_size, struct orlov_stats *stats)
 369 {
 370         struct ext4_group_desc *desc;
 371         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 372
 373         if (flex_size > 1) {
 374                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 375                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 376                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 377                 return;
 378         }
 379
 380         desc = ext4_get_group_desc(sb, g, NULL);
 381         if (desc) {
 382                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 383                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 384                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 385         } else {
 386                 stats->free_inodes = 0;
 387                 stats->free_clusters = 0;
 388                 stats->used_dirs = 0;
 389         }
 390 }
 391
 392 /*
 393  * Orlov's allocator for directories.
 394  *
 395  * We always try to spread first-level directories.
 396  *
 397  * If there are blockgroups with both free inodes and free blocks counts
 398  * not worse than average we return one with smallest directory count.
 399  * Otherwise we simply return a random group.
 400  *
 401  * For the rest rules look so:
 402  *
 403  * It's OK to put directory into a group unless
 404  * it has too many directories already (max_dirs) or
 405  * it has too few free inodes left (min_inodes) or
 406  * it has too few free blocks left (min_blocks) or
 407  * Parent's group is preferred, if it doesn't satisfy these
 408  * conditions we search cyclically through the rest. If none
 409  * of the groups look good we just look for a group with more
 410  * free inodes than average (starting at parent's group).
 411  */
 412
 413 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 414                             ext4_group_t *group, umode_t mode,
 415                             const struct qstr *qstr)
 416 {
 417         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 418         struct ext4_sb_info *sbi = EXT4_SB(sb);
 419         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 420         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 421         unsigned int freei, avefreei, grp_free;
 422         ext4_fsblk_t freeb, avefreec;
 423         unsigned int ndirs;
 424         int max_dirs, min_inodes;
 425         ext4_grpblk_t min_clusters;
 426         ext4_group_t i, grp, g, ngroups;
 427         struct ext4_group_desc *desc;
 428         struct orlov_stats stats;
 429         int flex_size = ext4_flex_bg_size(sbi);
 430         struct dx_hash_info hinfo;
 431
 432         ngroups = real_ngroups;
 433         if (flex_size > 1) {
 434                 ngroups = (real_ngroups + flex_size - 1) >>
 435                         sbi->s_log_groups_per_flex;
 436                 parent_group >>= sbi->s_log_groups_per_flex;
 437         }
 438
 439         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 440         avefreei = freei / ngroups;
 441         freeb = EXT4_C2B(sbi,
 442                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 443         avefreec = freeb;
 444         do_div(avefreec, ngroups);
 445         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 446
 447         if (S_ISDIR(mode) &&
 448             ((parent == d_inode(sb->s_root)) ||
 449              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 450                 int best_ndir = inodes_per_group;
 451                 int ret = -1;
 452
 453                 if (qstr) {
 454                         hinfo.hash_version = DX_HASH_HALF_MD4;
 455                         hinfo.seed = sbi->s_hash_seed;
 456                         ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
 457                         grp = hinfo.hash;
 458                 } else
 459                         grp = prandom_u32();
 460                 parent_group = (unsigned)grp % ngroups;
 461                 for (i = 0; i < ngroups; i++) {
 462                         g = (parent_group + i) % ngroups;
 463                         get_orlov_stats(sb, g, flex_size, &stats);
 464                         if (!stats.free_inodes)
 465                                 continue;
 466                         if (stats.used_dirs >= best_ndir)
 467                                 continue;
 468                         if (stats.free_inodes < avefreei)
 469                                 continue;
 470                         if (stats.free_clusters < avefreec)
 471                                 continue;
 472                         grp = g;
 473                         ret = 0;
 474                         best_ndir = stats.used_dirs;
 475                 }
 476                 if (ret)
 477                         goto fallback;
 478         found_flex_bg:
 479                 if (flex_size == 1) {
 480                         *group = grp;
 481                         return 0;
 482                 }
 483
 484                 /*
 485                  * We pack inodes at the beginning of the flexgroup's
 486                  * inode tables.  Block allocation decisions will do
 487                  * something similar, although regular files will
 488                  * start at 2nd block group of the flexgroup.  See
 489                  * ext4_ext_find_goal() and ext4_find_near().
 490                  */
 491                 grp *= flex_size;
 492                 for (i = 0; i < flex_size; i++) {
 493                         if (grp+i >= real_ngroups)
 494                                 break;
 495                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 496                         if (desc && ext4_free_inodes_count(sb, desc)) {
 497                                 *group = grp+i;
 498                                 return 0;
 499                         }
 500                 }
 501                 goto fallback;
 502         }
 503
 504         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 505         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 506         if (min_inodes < 1)
 507                 min_inodes = 1;
 508         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 509
 510         /*
 511          * Start looking in the flex group where we last allocated an
 512          * inode for this parent directory
 513          */
 514         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 515                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 516                 if (flex_size > 1)
 517                         parent_group >>= sbi->s_log_groups_per_flex;
 518         }
 519
 520         for (i = 0; i < ngroups; i++) {
 521                 grp = (parent_group + i) % ngroups;
 522                 get_orlov_stats(sb, grp, flex_size, &stats);
 523                 if (stats.used_dirs >= max_dirs)
 524                         continue;
 525                 if (stats.free_inodes < min_inodes)
 526                         continue;
 527                 if (stats.free_clusters < min_clusters)
 528                         continue;
 529                 goto found_flex_bg;
 530         }
 531
 532 fallback:
 533         ngroups = real_ngroups;
 534         avefreei = freei / ngroups;
 535 fallback_retry:
 536         parent_group = EXT4_I(parent)->i_block_group;
 537         for (i = 0; i < ngroups; i++) {
 538                 grp = (parent_group + i) % ngroups;
 539                 desc = ext4_get_group_desc(sb, grp, NULL);
 540                 if (desc) {
 541                         grp_free = ext4_free_inodes_count(sb, desc);
 542                         if (grp_free && grp_free >= avefreei) {
 543                                 *group = grp;
 544                                 return 0;
 545                         }
 546                 }
 547         }
 548
 549         if (avefreei) {
 550                 /*
 551                  * The free-inodes counter is approximate, and for really small
 552                  * filesystems the above test can fail to find any blockgroups
 553                  */
 554                 avefreei = 0;
 555                 goto fallback_retry;
 556         }
 557
 558         return -1;
 559 }
 560
 561 static int find_group_other(struct super_block *sb, struct inode *parent,
 562                             ext4_group_t *group, umode_t mode)
 563 {
 564         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 565         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 566         struct ext4_group_desc *desc;
 567         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 568
 569         /*
 570          * Try to place the inode is the same flex group as its
 571          * parent.  If we can't find space, use the Orlov algorithm to
 572          * find another flex group, and store that information in the
 573          * parent directory's inode information so that use that flex
 574          * group for future allocations.
 575          */
 576         if (flex_size > 1) {
 577                 int retry = 0;
 578
 579         try_again:
 580                 parent_group &= ~(flex_size-1);
 581                 last = parent_group + flex_size;
 582                 if (last > ngroups)
 583                         last = ngroups;
 584                 for  (i = parent_group; i < last; i++) {
 585                         desc = ext4_get_group_desc(sb, i, NULL);
 586                         if (desc && ext4_free_inodes_count(sb, desc)) {
 587                                 *group = i;
 588                                 return 0;
 589                         }
 590                 }
 591                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 592                         retry = 1;
 593                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 594                         goto try_again;
 595                 }
 596                 /*
 597                  * If this didn't work, use the Orlov search algorithm
 598                  * to find a new flex group; we pass in the mode to
 599                  * avoid the topdir algorithms.
 600                  */
 601                 *group = parent_group + flex_size;
 602                 if (*group > ngroups)
 603                         *group = 0;
 604                 return find_group_orlov(sb, parent, group, mode, NULL);
 605         }
 606
 607         /*
 608          * Try to place the inode in its parent directory
 609          */
 610         *group = parent_group;
 611         desc = ext4_get_group_desc(sb, *group, NULL);
 612         if (desc && ext4_free_inodes_count(sb, desc) &&
 613             ext4_free_group_clusters(sb, desc))
 614                 return 0;
 615
 616         /*
 617          * We're going to place this inode in a different blockgroup from its
 618          * parent.  We want to cause files in a common directory to all land in
 619          * the same blockgroup.  But we want files which are in a different
 620          * directory which shares a blockgroup with our parent to land in a
 621          * different blockgroup.
 622          *
 623          * So add our directory's i_ino into the starting point for the hash.
 624          */
 625         *group = (*group + parent->i_ino) % ngroups;
 626
 627         /*
 628          * Use a quadratic hash to find a group with a free inode and some free
 629          * blocks.
 630          */
 631         for (i = 1; i < ngroups; i <<= 1) {
 632                 *group += i;
 633                 if (*group >= ngroups)
 634                         *group -= ngroups;
 635                 desc = ext4_get_group_desc(sb, *group, NULL);
 636                 if (desc && ext4_free_inodes_count(sb, desc) &&
 637                     ext4_free_group_clusters(sb, desc))
 638                         return 0;
 639         }
 640
 641         /*
 642          * That failed: try linear search for a free inode, even if that group
 643          * has no free blocks.
 644          */
 645         *group = parent_group;
 646         for (i = 0; i < ngroups; i++) {
 647                 if (++*group >= ngroups)
 648                         *group = 0;
 649                 desc = ext4_get_group_desc(sb, *group, NULL);
 650                 if (desc && ext4_free_inodes_count(sb, desc))
 651                         return 0;
 652         }
 653
 654         return -1;
 655 }
 656
 657 /*
 658  * In no journal mode, if an inode has recently been deleted, we want
 659  * to avoid reusing it until we're reasonably sure the inode table
 660  * block has been written back to disk.  (Yes, these values are
 661  * somewhat arbitrary...)
 662  */
 663 #define RECENTCY_MIN    5
 664 #define RECENTCY_DIRTY  300
 665
 666 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 667 {
 668         struct ext4_group_desc  *gdp;
 669         struct ext4_inode       *raw_inode;
 670         struct buffer_head      *bh;
 671         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 672         int offset, ret = 0;
 673         int recentcy = RECENTCY_MIN;
 674         u32 dtime, now;
 675
 676         gdp = ext4_get_group_desc(sb, group, NULL);
 677         if (unlikely(!gdp))
 678                 return 0;
 679
 680         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 681                        (ino / inodes_per_block));
 682         if (!bh || !buffer_uptodate(bh))
 683                 /*
 684                  * If the block is not in the buffer cache, then it
 685                  * must have been written out.
 686                  */
 687                 goto out;
 688
 689         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 690         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 691
 692         /* i_dtime is only 32 bits on disk, but we only care about relative
 693          * times in the range of a few minutes (i.e. long enough to sync a
 694          * recently-deleted inode to disk), so using the low 32 bits of the
 695          * clock (a 68 year range) is enough, see time_before32() */
 696         dtime = le32_to_cpu(raw_inode->i_dtime);
 697         now = ktime_get_real_seconds();
 698         if (buffer_dirty(bh))
 699                 recentcy += RECENTCY_DIRTY;
 700
 701         if (dtime && time_before32(dtime, now) &&
 702             time_before32(now, dtime + recentcy))
 703                 ret = 1;
 704 out:
 705         brelse(bh);
 706         return ret;
 707 }
 708
 709 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 710                           struct buffer_head *bitmap, unsigned long *ino)
 711 {
 712 next:
 713         *ino = ext4_find_next_zero_bit((unsigned long *)
 714                                        bitmap->b_data,
 715                                        EXT4_INODES_PER_GROUP(sb), *ino);
 716         if (*ino >= EXT4_INODES_PER_GROUP(sb))
 717                 return 0;
 718
 719         if ((EXT4_SB(sb)->s_journal == NULL) &&
 720             recently_deleted(sb, group, *ino)) {
 721                 *ino = *ino + 1;
 722                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 723                         goto next;
 724                 return 0;
 725         }
 726
 727         return 1;
 728 }
 729
 730 /*
 731  * There are two policies for allocating an inode.  If the new inode is
 732  * a directory, then a forward search is made for a block group with both
 733  * free space and a low directory-to-inode ratio; if that fails, then of
 734  * the groups with above-average free space, that group with the fewest
 735  * directories already is chosen.
 736  *
 737  * For other inodes, search forward from the parent directory's block
 738  * group to find a free inode.
 739  */
 740 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 741                                umode_t mode, const struct qstr *qstr,
 742                                __u32 goal, uid_t *owner, __u32 i_flags,
 743                                int handle_type, unsigned int line_no,
 744                                int nblocks)
 745 {
 746         struct super_block *sb;
 747         struct buffer_head *inode_bitmap_bh = NULL;
 748         struct buffer_head *group_desc_bh;
 749         ext4_group_t ngroups, group = 0;
 750         unsigned long ino = 0;
 751         struct inode *inode;
 752         struct ext4_group_desc *gdp = NULL;
 753         struct ext4_inode_info *ei;
 754         struct ext4_sb_info *sbi;
 755         int ret2, err;
 756         struct inode *ret;
 757         ext4_group_t i;
 758         ext4_group_t flex_group;
 759         struct ext4_group_info *grp;
 760         int encrypt = 0;
 761
 762         /* Cannot create files in a deleted directory */
 763         if (!dir || !dir->i_nlink)
 764                 return ERR_PTR(-EPERM);
 765
 766         sb = dir->i_sb;
 767         sbi = EXT4_SB(sb);
 768
 769         if (unlikely(ext4_forced_shutdown(sbi)))
 770                 return ERR_PTR(-EIO);
 771
 772         if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
 773             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
 774             !(i_flags & EXT4_EA_INODE_FL)) {
 775                 err = fscrypt_get_encryption_info(dir);
 776                 if (err)
 777                         return ERR_PTR(err);
 778                 if (!fscrypt_has_encryption_key(dir))
 779                         return ERR_PTR(-ENOKEY);
 780                 encrypt = 1;
 781         }
 782
 783         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 784 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 785                 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 786
 787                 if (IS_ERR(p))
 788                         return ERR_CAST(p);
 789                 if (p) {
 790                         int acl_size = p->a_count * sizeof(ext4_acl_entry);
 791
 792                         nblocks += (S_ISDIR(mode) ? 2 : 1) *
 793                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 794                                         NULL /* block_bh */, acl_size,
 795                                         true /* is_create */);
 796                         posix_acl_release(p);
 797                 }
 798 #endif
 799
 800 #ifdef CONFIG_SECURITY
 801                 {
 802                         int num_security_xattrs = 1;
 803
 804 #ifdef CONFIG_INTEGRITY
 805                         num_security_xattrs++;
 806 #endif
 807                         /*
 808                          * We assume that security xattrs are never
 809                          * more than 1k.  In practice they are under
 810                          * 128 bytes.
 811                          */
 812                         nblocks += num_security_xattrs *
 813                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 814                                         NULL /* block_bh */, 1024,
 815                                         true /* is_create */);
 816                 }
 817 #endif
 818                 if (encrypt)
 819                         nblocks += __ext4_xattr_set_credits(sb,
 820                                         NULL /* inode */, NULL /* block_bh */,
 821                                         FSCRYPT_SET_CONTEXT_MAX_SIZE,
 822                                         true /* is_create */);
 823         }
 824
 825         ngroups = ext4_get_groups_count(sb);
 826         trace_ext4_request_inode(dir, mode);
 827         inode = new_inode(sb);
 828         if (!inode)
 829                 return ERR_PTR(-ENOMEM);
 830         ei = EXT4_I(inode);
 831
 832         /*
 833          * Initialize owners and quota early so that we don't have to account
 834          * for quota initialization worst case in standard inode creating
 835          * transaction
 836          */
 837         if (owner) {
 838                 inode->i_mode = mode;
 839                 i_uid_write(inode, owner[0]);
 840                 i_gid_write(inode, owner[1]);
 841         } else if (test_opt(sb, GRPID)) {
 842                 inode->i_mode = mode;
 843                 inode->i_uid = current_fsuid();
 844                 inode->i_gid = dir->i_gid;
 845         } else
 846                 inode_init_owner(inode, dir, mode);
 847
 848         if (ext4_has_feature_project(sb) &&
 849             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 850                 ei->i_projid = EXT4_I(dir)->i_projid;
 851         else
 852                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 853
 854         err = dquot_initialize(inode);
 855         if (err)
 856                 goto out;
 857
 858         if (!goal)
 859                 goal = sbi->s_inode_goal;
 860
 861         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 862                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 863                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 864                 ret2 = 0;
 865                 goto got_group;
 866         }
 867
 868         if (S_ISDIR(mode))
 869                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 870         else
 871                 ret2 = find_group_other(sb, dir, &group, mode);
 872
 873 got_group:
 874         EXT4_I(dir)->i_last_alloc_group = group;
 875         err = -ENOSPC;
 876         if (ret2 == -1)
 877                 goto out;
 878
 879         /*
 880          * Normally we will only go through one pass of this loop,
 881          * unless we get unlucky and it turns out the group we selected
 882          * had its last inode grabbed by someone else.
 883          */
 884         for (i = 0; i < ngroups; i++, ino = 0) {
 885                 err = -EIO;
 886
 887                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 888                 if (!gdp)
 889                         goto out;
 890
 891                 /*
 892                  * Check free inodes count before loading bitmap.
 893                  */
 894                 if (ext4_free_inodes_count(sb, gdp) == 0)
 895                         goto next_group;
 896
 897                 grp = ext4_get_group_info(sb, group);
 898                 /* Skip groups with already-known suspicious inode tables */
 899                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 900                         goto next_group;
 901
 902                 brelse(inode_bitmap_bh);
 903                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 904                 /* Skip groups with suspicious inode tables */
 905                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 906                     IS_ERR(inode_bitmap_bh)) {
 907                         inode_bitmap_bh = NULL;
 908                         goto next_group;
 909                 }
 910
 911 repeat_in_this_group:
 912                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 913                 if (!ret2)
 914                         goto next_group;
 915
 916                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 917                         ext4_error(sb, "reserved inode found cleared - "
 918                                    "inode=%lu", ino + 1);
 919                         ext4_mark_group_bitmap_corrupted(sb, group,
 920                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 921                         goto next_group;
 922                 }
 923
 924                 if (!handle) {
 925                         BUG_ON(nblocks <= 0);
 926                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 927                                  handle_type, nblocks, 0,
 928                                  ext4_trans_default_revoke_credits(sb));
 929                         if (IS_ERR(handle)) {
 930                                 err = PTR_ERR(handle);
 931                                 ext4_std_error(sb, err);
 932                                 goto out;
 933                         }
 934                 }
 935                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 936                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 937                 if (err) {
 938                         ext4_std_error(sb, err);
 939                         goto out;
 940                 }
 941                 ext4_lock_group(sb, group);
 942                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 943                 if (ret2) {
 944                         /* Someone already took the bit. Repeat the search
 945                          * with lock held.
 946                          */
 947                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 948                         if (ret2) {
 949                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
 950                                 ret2 = 0;
 951                         } else {
 952                                 ret2 = 1; /* we didn't grab the inode */
 953                         }
 954                 }
 955                 ext4_unlock_group(sb, group);
 956                 ino++;          /* the inode bitmap is zero-based */
 957                 if (!ret2)
 958                         goto got; /* we grabbed the inode! */
 959
 960                 if (ino < EXT4_INODES_PER_GROUP(sb))
 961                         goto repeat_in_this_group;
 962 next_group:
 963                 if (++group == ngroups)
 964                         group = 0;
 965         }
 966         err = -ENOSPC;
 967         goto out;
 968
 969 got:
 970         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 971         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 972         if (err) {
 973                 ext4_std_error(sb, err);
 974                 goto out;
 975         }
 976
 977         BUFFER_TRACE(group_desc_bh, "get_write_access");
 978         err = ext4_journal_get_write_access(handle, group_desc_bh);
 979         if (err) {
 980                 ext4_std_error(sb, err);
 981                 goto out;
 982         }
 983
 984         /* We may have to initialize the block bitmap if it isn't already */
 985         if (ext4_has_group_desc_csum(sb) &&
 986             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 987                 struct buffer_head *block_bitmap_bh;
 988
 989                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 990                 if (IS_ERR(block_bitmap_bh)) {
 991                         err = PTR_ERR(block_bitmap_bh);
 992                         goto out;
 993                 }
 994                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 995                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 996                 if (err) {
 997                         brelse(block_bitmap_bh);
 998                         ext4_std_error(sb, err);
 999                         goto out;
1000                 }
1001
1002                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1003                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1004
1005                 /* recheck and clear flag under lock if we still need to */
1006                 ext4_lock_group(sb, group);
1007                 if (ext4_has_group_desc_csum(sb) &&
1008                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1009                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1010                         ext4_free_group_clusters_set(sb, gdp,
1011                                 ext4_free_clusters_after_init(sb, group, gdp));
1012                         ext4_block_bitmap_csum_set(sb, group, gdp,
1013                                                    block_bitmap_bh);
1014                         ext4_group_desc_csum_set(sb, group, gdp);
1015                 }
1016                 ext4_unlock_group(sb, group);
1017                 brelse(block_bitmap_bh);
1018
1019                 if (err) {
1020                         ext4_std_error(sb, err);
1021                         goto out;
1022                 }
1023         }
1024
1025         /* Update the relevant bg descriptor fields */
1026         if (ext4_has_group_desc_csum(sb)) {
1027                 int free;
1028                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1029
1030                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1031                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1032                 free = EXT4_INODES_PER_GROUP(sb) -
1033                         ext4_itable_unused_count(sb, gdp);
1034                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1035                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1036                         free = 0;
1037                 }
1038                 /*
1039                  * Check the relative inode number against the last used
1040                  * relative inode number in this group. if it is greater
1041                  * we need to update the bg_itable_unused count
1042                  */
1043                 if (ino > free)
1044                         ext4_itable_unused_set(sb, gdp,
1045                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1046                 up_read(&grp->alloc_sem);
1047         } else {
1048                 ext4_lock_group(sb, group);
1049         }
1050
1051         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1052         if (S_ISDIR(mode)) {
1053                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1054                 if (sbi->s_log_groups_per_flex) {
1055                         ext4_group_t f = ext4_flex_group(sbi, group);
1056
1057                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1058                 }
1059         }
1060         if (ext4_has_group_desc_csum(sb)) {
1061                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1062                                            EXT4_INODES_PER_GROUP(sb) / 8);
1063                 ext4_group_desc_csum_set(sb, group, gdp);
1064         }
1065         ext4_unlock_group(sb, group);
1066
1067         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1068         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1069         if (err) {
1070                 ext4_std_error(sb, err);
1071                 goto out;
1072         }
1073
1074         percpu_counter_dec(&sbi->s_freeinodes_counter);
1075         if (S_ISDIR(mode))
1076                 percpu_counter_inc(&sbi->s_dirs_counter);
1077
1078         if (sbi->s_log_groups_per_flex) {
1079                 flex_group = ext4_flex_group(sbi, group);
1080                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1081         }
1082
1083         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1084         /* This is the optimal IO size (for stat), not the fs block size */
1085         inode->i_blocks = 0;
1086         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1087         ei->i_crtime = inode->i_mtime;
1088
1089         memset(ei->i_data, 0, sizeof(ei->i_data));
1090         ei->i_dir_start_lookup = 0;
1091         ei->i_disksize = 0;
1092
1093         /* Don't inherit extent flag from directory, amongst others. */
1094         ei->i_flags =
1095                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1096         ei->i_flags |= i_flags;
1097         ei->i_file_acl = 0;
1098         ei->i_dtime = 0;
1099         ei->i_block_group = group;
1100         ei->i_last_alloc_group = ~0;
1101
1102         ext4_set_inode_flags(inode);
1103         if (IS_DIRSYNC(inode))
1104                 ext4_handle_sync(handle);
1105         if (insert_inode_locked(inode) < 0) {
1106                 /*
1107                  * Likely a bitmap corruption causing inode to be allocated
1108                  * twice.
1109                  */
1110                 err = -EIO;
1111                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1112                            inode->i_ino);
1113                 ext4_mark_group_bitmap_corrupted(sb, group,
1114                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1115                 goto out;
1116         }
1117         inode->i_generation = prandom_u32();
1118
1119         /* Precompute checksum seed for inode metadata */
1120         if (ext4_has_metadata_csum(sb)) {
1121                 __u32 csum;
1122                 __le32 inum = cpu_to_le32(inode->i_ino);
1123                 __le32 gen = cpu_to_le32(inode->i_generation);
1124                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1125                                    sizeof(inum));
1126                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1127                                               sizeof(gen));
1128         }
1129
1130         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1131         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1132
1133         ei->i_extra_isize = sbi->s_want_extra_isize;
1134         ei->i_inline_off = 0;
1135         if (ext4_has_feature_inline_data(sb))
1136                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1137         ret = inode;
1138         err = dquot_alloc_inode(inode);
1139         if (err)
1140                 goto fail_drop;
1141
1142         /*
1143          * Since the encryption xattr will always be unique, create it first so
1144          * that it's less likely to end up in an external xattr block and
1145          * prevent its deduplication.
1146          */
1147         if (encrypt) {
1148                 err = fscrypt_inherit_context(dir, inode, handle, true);
1149                 if (err)
1150                         goto fail_free_drop;
1151         }
1152
1153         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1154                 err = ext4_init_acl(handle, inode, dir);
1155                 if (err)
1156                         goto fail_free_drop;
1157
1158                 err = ext4_init_security(handle, inode, dir, qstr);
1159                 if (err)
1160                         goto fail_free_drop;
1161         }
1162
1163         if (ext4_has_feature_extents(sb)) {
1164                 /* set extent flag only for directory, file and normal symlink*/
1165                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1166                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1167                         ext4_ext_tree_init(handle, inode);
1168                 }
1169         }
1170
1171         if (ext4_handle_valid(handle)) {
1172                 ei->i_sync_tid = handle->h_transaction->t_tid;
1173                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1174         }
1175
1176         err = ext4_mark_inode_dirty(handle, inode);
1177         if (err) {
1178                 ext4_std_error(sb, err);
1179                 goto fail_free_drop;
1180         }
1181
1182         ext4_debug("allocating inode %lu\n", inode->i_ino);
1183         trace_ext4_allocate_inode(inode, dir, mode);
1184         brelse(inode_bitmap_bh);
1185         return ret;
1186
1187 fail_free_drop:
1188         dquot_free_inode(inode);
1189 fail_drop:
1190         clear_nlink(inode);
1191         unlock_new_inode(inode);
1192 out:
1193         dquot_drop(inode);
1194         inode->i_flags |= S_NOQUOTA;
1195         iput(inode);
1196         brelse(inode_bitmap_bh);
1197         return ERR_PTR(err);
1198 }
1199
1200 /* Verify that we are loading a valid orphan from disk */
1201 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1202 {
1203         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1204         ext4_group_t block_group;
1205         int bit;
1206         struct buffer_head *bitmap_bh = NULL;
1207         struct inode *inode = NULL;
1208         int err = -EFSCORRUPTED;
1209
1210         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1211                 goto bad_orphan;
1212
1213         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1214         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1215         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1216         if (IS_ERR(bitmap_bh))
1217                 return ERR_CAST(bitmap_bh);
1218
1219         /* Having the inode bit set should be a 100% indicator that this
1220          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1221          * inodes that were being truncated, so we can't check i_nlink==0.
1222          */
1223         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1224                 goto bad_orphan;
1225
1226         inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1227         if (IS_ERR(inode)) {
1228                 err = PTR_ERR(inode);
1229                 ext4_set_errno(sb, -err);
1230                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1231                            ino, err);
1232                 return inode;
1233         }
1234
1235         /*
1236          * If the orphans has i_nlinks > 0 then it should be able to
1237          * be truncated, otherwise it won't be removed from the orphan
1238          * list during processing and an infinite loop will result.
1239          * Similarly, it must not be a bad inode.
1240          */
1241         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1242             is_bad_inode(inode))
1243                 goto bad_orphan;
1244
1245         if (NEXT_ORPHAN(inode) > max_ino)
1246                 goto bad_orphan;
1247         brelse(bitmap_bh);
1248         return inode;
1249
1250 bad_orphan:
1251         ext4_error(sb, "bad orphan inode %lu", ino);
1252         if (bitmap_bh)
1253                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1254                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1255                        ext4_test_bit(bit, bitmap_bh->b_data));
1256         if (inode) {
1257                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1258                        is_bad_inode(inode));
1259                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1260                        NEXT_ORPHAN(inode));
1261                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1262                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1263                 /* Avoid freeing blocks if we got a bad deleted inode */
1264                 if (inode->i_nlink == 0)
1265                         inode->i_blocks = 0;
1266                 iput(inode);
1267         }
1268         brelse(bitmap_bh);
1269         return ERR_PTR(err);
1270 }
1271
1272 unsigned long ext4_count_free_inodes(struct super_block *sb)
1273 {
1274         unsigned long desc_count;
1275         struct ext4_group_desc *gdp;
1276         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1277 #ifdef EXT4FS_DEBUG
1278         struct ext4_super_block *es;
1279         unsigned long bitmap_count, x;
1280         struct buffer_head *bitmap_bh = NULL;
1281
1282         es = EXT4_SB(sb)->s_es;
1283         desc_count = 0;
1284         bitmap_count = 0;
1285         gdp = NULL;
1286         for (i = 0; i < ngroups; i++) {
1287                 gdp = ext4_get_group_desc(sb, i, NULL);
1288                 if (!gdp)
1289                         continue;
1290                 desc_count += ext4_free_inodes_count(sb, gdp);
1291                 brelse(bitmap_bh);
1292                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1293                 if (IS_ERR(bitmap_bh)) {
1294                         bitmap_bh = NULL;
1295                         continue;
1296                 }
1297
1298                 x = ext4_count_free(bitmap_bh->b_data,
1299                                     EXT4_INODES_PER_GROUP(sb) / 8);
1300                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1301                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1302                 bitmap_count += x;
1303         }
1304         brelse(bitmap_bh);
1305         printk(KERN_DEBUG "ext4_count_free_inodes: "
1306                "stored = %u, computed = %lu, %lu\n",
1307                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1308         return desc_count;
1309 #else
1310         desc_count = 0;
1311         for (i = 0; i < ngroups; i++) {
1312                 gdp = ext4_get_group_desc(sb, i, NULL);
1313                 if (!gdp)
1314                         continue;
1315                 desc_count += ext4_free_inodes_count(sb, gdp);
1316                 cond_resched();
1317         }
1318         return desc_count;
1319 #endif
1320 }
1321
1322 /* Called at mount-time, super-block is locked */
1323 unsigned long ext4_count_dirs(struct super_block * sb)
1324 {
1325         unsigned long count = 0;
1326         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1327
1328         for (i = 0; i < ngroups; i++) {
1329                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1330                 if (!gdp)
1331                         continue;
1332                 count += ext4_used_dirs_count(sb, gdp);
1333         }
1334         return count;
1335 }
1336
1337 /*
1338  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1339  * inode table. Must be called without any spinlock held. The only place
1340  * where it is called from on active part of filesystem is ext4lazyinit
1341  * thread, so we do not need any special locks, however we have to prevent
1342  * inode allocation from the current group, so we take alloc_sem lock, to
1343  * block ext4_new_inode() until we are finished.
1344  */
1345 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1346                                  int barrier)
1347 {
1348         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1349         struct ext4_sb_info *sbi = EXT4_SB(sb);
1350         struct ext4_group_desc *gdp = NULL;
1351         struct buffer_head *group_desc_bh;
1352         handle_t *handle;
1353         ext4_fsblk_t blk;
1354         int num, ret = 0, used_blks = 0;
1355
1356         /* This should not happen, but just to be sure check this */
1357         if (sb_rdonly(sb)) {
1358                 ret = 1;
1359                 goto out;
1360         }
1361
1362         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1363         if (!gdp)
1364                 goto out;
1365
1366         /*
1367          * We do not need to lock this, because we are the only one
1368          * handling this flag.
1369          */
1370         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1371                 goto out;
1372
1373         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1374         if (IS_ERR(handle)) {
1375                 ret = PTR_ERR(handle);
1376                 goto out;
1377         }
1378
1379         down_write(&grp->alloc_sem);
1380         /*
1381          * If inode bitmap was already initialized there may be some
1382          * used inodes so we need to skip blocks with used inodes in
1383          * inode table.
1384          */
1385         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1386                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1387                             ext4_itable_unused_count(sb, gdp)),
1388                             sbi->s_inodes_per_block);
1389
1390         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1391             ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1392                                ext4_itable_unused_count(sb, gdp)) <
1393                               EXT4_FIRST_INO(sb)))) {
1394                 ext4_error(sb, "Something is wrong with group %u: "
1395                            "used itable blocks: %d; "
1396                            "itable unused count: %u",
1397                            group, used_blks,
1398                            ext4_itable_unused_count(sb, gdp));
1399                 ret = 1;
1400                 goto err_out;
1401         }
1402
1403         blk = ext4_inode_table(sb, gdp) + used_blks;
1404         num = sbi->s_itb_per_group - used_blks;
1405
1406         BUFFER_TRACE(group_desc_bh, "get_write_access");
1407         ret = ext4_journal_get_write_access(handle,
1408                                             group_desc_bh);
1409         if (ret)
1410                 goto err_out;
1411
1412         /*
1413          * Skip zeroout if the inode table is full. But we set the ZEROED
1414          * flag anyway, because obviously, when it is full it does not need
1415          * further zeroing.
1416          */
1417         if (unlikely(num == 0))
1418                 goto skip_zeroout;
1419
1420         ext4_debug("going to zero out inode table in group %d\n",
1421                    group);
1422         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1423         if (ret < 0)
1424                 goto err_out;
1425         if (barrier)
1426                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1427
1428 skip_zeroout:
1429         ext4_lock_group(sb, group);
1430         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1431         ext4_group_desc_csum_set(sb, group, gdp);
1432         ext4_unlock_group(sb, group);
1433
1434         BUFFER_TRACE(group_desc_bh,
1435                      "call ext4_handle_dirty_metadata");
1436         ret = ext4_handle_dirty_metadata(handle, NULL,
1437                                          group_desc_bh);
1438
1439 err_out:
1440         up_write(&grp->alloc_sem);
1441         ext4_journal_stop(handle);
1442 out:
1443         return ret;
1444 }