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;
  86
  87         if (EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY)
  88                 return 0;
  89
  90         grp = ext4_get_group_info(sb, block_group);
  91
  92         if (buffer_verified(bh))
  93                 return 0;
  94         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
  95                 return -EFSCORRUPTED;
  96
  97         ext4_lock_group(sb, block_group);
  98         if (buffer_verified(bh))
  99                 goto verified;
 100         blk = ext4_inode_bitmap(sb, desc);
 101         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
 102                                            EXT4_INODES_PER_GROUP(sb) / 8) ||
 103             ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) {
 104                 ext4_unlock_group(sb, block_group);
 105                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 106                            "inode_bitmap = %llu", block_group, blk);
 107                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 108                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 109                 return -EFSBADCRC;
 110         }
 111         set_buffer_verified(bh);
 112 verified:
 113         ext4_unlock_group(sb, block_group);
 114         return 0;
 115 }
 116
 117 /*
 118  * Read the inode allocation bitmap for a given block_group, reading
 119  * into the specified slot in the superblock's bitmap cache.
 120  *
 121  * Return buffer_head of bitmap on success, or an ERR_PTR on error.
 122  */
 123 static struct buffer_head *
 124 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 125 {
 126         struct ext4_group_desc *desc;
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         struct buffer_head *bh = NULL;
 129         ext4_fsblk_t bitmap_blk;
 130         int err;
 131
 132         desc = ext4_get_group_desc(sb, block_group, NULL);
 133         if (!desc)
 134                 return ERR_PTR(-EFSCORRUPTED);
 135
 136         bitmap_blk = ext4_inode_bitmap(sb, desc);
 137         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 138             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 139                 ext4_error(sb, "Invalid inode bitmap blk %llu in "
 140                            "block_group %u", bitmap_blk, block_group);
 141                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 142                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 143                 return ERR_PTR(-EFSCORRUPTED);
 144         }
 145         bh = sb_getblk(sb, bitmap_blk);
 146         if (unlikely(!bh)) {
 147                 ext4_warning(sb, "Cannot read inode bitmap - "
 148                              "block_group = %u, inode_bitmap = %llu",
 149                              block_group, bitmap_blk);
 150                 return ERR_PTR(-ENOMEM);
 151         }
 152         if (bitmap_uptodate(bh))
 153                 goto verify;
 154
 155         lock_buffer(bh);
 156         if (bitmap_uptodate(bh)) {
 157                 unlock_buffer(bh);
 158                 goto verify;
 159         }
 160
 161         ext4_lock_group(sb, block_group);
 162         if (ext4_has_group_desc_csum(sb) &&
 163             (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
 164                 if (block_group == 0) {
 165                         ext4_unlock_group(sb, block_group);
 166                         unlock_buffer(bh);
 167                         ext4_error(sb, "Inode bitmap for bg 0 marked "
 168                                    "uninitialized");
 169                         err = -EFSCORRUPTED;
 170                         goto out;
 171                 }
 172                 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 173                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 174                                      sb->s_blocksize * 8, bh->b_data);
 175                 set_bitmap_uptodate(bh);
 176                 set_buffer_uptodate(bh);
 177                 set_buffer_verified(bh);
 178                 ext4_unlock_group(sb, block_group);
 179                 unlock_buffer(bh);
 180                 return bh;
 181         }
 182         ext4_unlock_group(sb, block_group);
 183
 184         if (buffer_uptodate(bh)) {
 185                 /*
 186                  * if not uninit if bh is uptodate,
 187                  * bitmap is also uptodate
 188                  */
 189                 set_bitmap_uptodate(bh);
 190                 unlock_buffer(bh);
 191                 goto verify;
 192         }
 193         /*
 194          * submit the buffer_head for reading
 195          */
 196         trace_ext4_load_inode_bitmap(sb, block_group);
 197         ext4_read_bh(bh, REQ_META | REQ_PRIO, ext4_end_bitmap_read);
 198         ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO);
 199         if (!buffer_uptodate(bh)) {
 200                 put_bh(bh);
 201                 ext4_error_err(sb, EIO, "Cannot read inode bitmap - "
 202                                "block_group = %u, inode_bitmap = %llu",
 203                                block_group, bitmap_blk);
 204                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 205                                 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 206                 return ERR_PTR(-EIO);
 207         }
 208
 209 verify:
 210         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 211         if (err)
 212                 goto out;
 213         return bh;
 214 out:
 215         put_bh(bh);
 216         return ERR_PTR(err);
 217 }
 218
 219 /*
 220  * NOTE! When we get the inode, we're the only people
 221  * that have access to it, and as such there are no
 222  * race conditions we have to worry about. The inode
 223  * is not on the hash-lists, and it cannot be reached
 224  * through the filesystem because the directory entry
 225  * has been deleted earlier.
 226  *
 227  * HOWEVER: we must make sure that we get no aliases,
 228  * which means that we have to call "clear_inode()"
 229  * _before_ we mark the inode not in use in the inode
 230  * bitmaps. Otherwise a newly created file might use
 231  * the same inode number (not actually the same pointer
 232  * though), and then we'd have two inodes sharing the
 233  * same inode number and space on the harddisk.
 234  */
 235 void ext4_free_inode(handle_t *handle, struct inode *inode)
 236 {
 237         struct super_block *sb = inode->i_sb;
 238         int is_directory;
 239         unsigned long ino;
 240         struct buffer_head *bitmap_bh = NULL;
 241         struct buffer_head *bh2;
 242         ext4_group_t block_group;
 243         unsigned long bit;
 244         struct ext4_group_desc *gdp;
 245         struct ext4_super_block *es;
 246         struct ext4_sb_info *sbi;
 247         int fatal = 0, err, count, cleared;
 248         struct ext4_group_info *grp;
 249
 250         if (!sb) {
 251                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 252                        "nonexistent device\n", __func__, __LINE__);
 253                 return;
 254         }
 255         if (atomic_read(&inode->i_count) > 1) {
 256                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 257                          __func__, __LINE__, inode->i_ino,
 258                          atomic_read(&inode->i_count));
 259                 return;
 260         }
 261         if (inode->i_nlink) {
 262                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 263                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 264                 return;
 265         }
 266         sbi = EXT4_SB(sb);
 267
 268         ino = inode->i_ino;
 269         ext4_debug("freeing inode %lu\n", ino);
 270         trace_ext4_free_inode(inode);
 271
 272         dquot_initialize(inode);
 273         dquot_free_inode(inode);
 274
 275         is_directory = S_ISDIR(inode->i_mode);
 276
 277         /* Do this BEFORE marking the inode not in use or returning an error */
 278         ext4_clear_inode(inode);
 279
 280         es = sbi->s_es;
 281         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 282                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 283                 goto error_return;
 284         }
 285         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 286         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 287         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 288         /* Don't bother if the inode bitmap is corrupt. */
 289         if (IS_ERR(bitmap_bh)) {
 290                 fatal = PTR_ERR(bitmap_bh);
 291                 bitmap_bh = NULL;
 292                 goto error_return;
 293         }
 294         if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
 295                 grp = ext4_get_group_info(sb, block_group);
 296                 if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 297                         fatal = -EFSCORRUPTED;
 298                         goto error_return;
 299                 }
 300         }
 301
 302         BUFFER_TRACE(bitmap_bh, "get_write_access");
 303         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 304         if (fatal)
 305                 goto error_return;
 306
 307         fatal = -ESRCH;
 308         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 309         if (gdp) {
 310                 BUFFER_TRACE(bh2, "get_write_access");
 311                 fatal = ext4_journal_get_write_access(handle, bh2);
 312         }
 313         ext4_lock_group(sb, block_group);
 314         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 315         if (fatal || !cleared) {
 316                 ext4_unlock_group(sb, block_group);
 317                 goto out;
 318         }
 319
 320         count = ext4_free_inodes_count(sb, gdp) + 1;
 321         ext4_free_inodes_set(sb, gdp, count);
 322         if (is_directory) {
 323                 count = ext4_used_dirs_count(sb, gdp) - 1;
 324                 ext4_used_dirs_set(sb, gdp, count);
 325                 percpu_counter_dec(&sbi->s_dirs_counter);
 326         }
 327         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 328                                    EXT4_INODES_PER_GROUP(sb) / 8);
 329         ext4_group_desc_csum_set(sb, block_group, gdp);
 330         ext4_unlock_group(sb, block_group);
 331
 332         percpu_counter_inc(&sbi->s_freeinodes_counter);
 333         if (sbi->s_log_groups_per_flex) {
 334                 struct flex_groups *fg;
 335
 336                 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
 337                                          ext4_flex_group(sbi, block_group));
 338                 atomic_inc(&fg->free_inodes);
 339                 if (is_directory)
 340                         atomic_dec(&fg->used_dirs);
 341         }
 342         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 343         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 344 out:
 345         if (cleared) {
 346                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 347                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 348                 if (!fatal)
 349                         fatal = err;
 350         } else {
 351                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 352                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 353                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 354         }
 355
 356 error_return:
 357         brelse(bitmap_bh);
 358         ext4_std_error(sb, fatal);
 359 }
 360
 361 struct orlov_stats {
 362         __u64 free_clusters;
 363         __u32 free_inodes;
 364         __u32 used_dirs;
 365 };
 366
 367 /*
 368  * Helper function for Orlov's allocator; returns critical information
 369  * for a particular block group or flex_bg.  If flex_size is 1, then g
 370  * is a block group number; otherwise it is flex_bg number.
 371  */
 372 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 373                             int flex_size, struct orlov_stats *stats)
 374 {
 375         struct ext4_group_desc *desc;
 376
 377         if (flex_size > 1) {
 378                 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
 379                                                              s_flex_groups, g);
 380                 stats->free_inodes = atomic_read(&fg->free_inodes);
 381                 stats->free_clusters = atomic64_read(&fg->free_clusters);
 382                 stats->used_dirs = atomic_read(&fg->used_dirs);
 383                 return;
 384         }
 385
 386         desc = ext4_get_group_desc(sb, g, NULL);
 387         if (desc) {
 388                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 389                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 390                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 391         } else {
 392                 stats->free_inodes = 0;
 393                 stats->free_clusters = 0;
 394                 stats->used_dirs = 0;
 395         }
 396 }
 397
 398 /*
 399  * Orlov's allocator for directories.
 400  *
 401  * We always try to spread first-level directories.
 402  *
 403  * If there are blockgroups with both free inodes and free blocks counts
 404  * not worse than average we return one with smallest directory count.
 405  * Otherwise we simply return a random group.
 406  *
 407  * For the rest rules look so:
 408  *
 409  * It's OK to put directory into a group unless
 410  * it has too many directories already (max_dirs) or
 411  * it has too few free inodes left (min_inodes) or
 412  * it has too few free blocks left (min_blocks) or
 413  * Parent's group is preferred, if it doesn't satisfy these
 414  * conditions we search cyclically through the rest. If none
 415  * of the groups look good we just look for a group with more
 416  * free inodes than average (starting at parent's group).
 417  */
 418
 419 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 420                             ext4_group_t *group, umode_t mode,
 421                             const struct qstr *qstr)
 422 {
 423         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 424         struct ext4_sb_info *sbi = EXT4_SB(sb);
 425         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 426         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 427         unsigned int freei, avefreei, grp_free;
 428         ext4_fsblk_t freeb, avefreec;
 429         unsigned int ndirs;
 430         int max_dirs, min_inodes;
 431         ext4_grpblk_t min_clusters;
 432         ext4_group_t i, grp, g, ngroups;
 433         struct ext4_group_desc *desc;
 434         struct orlov_stats stats;
 435         int flex_size = ext4_flex_bg_size(sbi);
 436         struct dx_hash_info hinfo;
 437
 438         ngroups = real_ngroups;
 439         if (flex_size > 1) {
 440                 ngroups = (real_ngroups + flex_size - 1) >>
 441                         sbi->s_log_groups_per_flex;
 442                 parent_group >>= sbi->s_log_groups_per_flex;
 443         }
 444
 445         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 446         avefreei = freei / ngroups;
 447         freeb = EXT4_C2B(sbi,
 448                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 449         avefreec = freeb;
 450         do_div(avefreec, ngroups);
 451         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 452
 453         if (S_ISDIR(mode) &&
 454             ((parent == d_inode(sb->s_root)) ||
 455              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 456                 int best_ndir = inodes_per_group;
 457                 int ret = -1;
 458
 459                 if (qstr) {
 460                         hinfo.hash_version = DX_HASH_HALF_MD4;
 461                         hinfo.seed = sbi->s_hash_seed;
 462                         ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
 463                         grp = hinfo.hash;
 464                 } else
 465                         grp = prandom_u32();
 466                 parent_group = (unsigned)grp % ngroups;
 467                 for (i = 0; i < ngroups; i++) {
 468                         g = (parent_group + i) % ngroups;
 469                         get_orlov_stats(sb, g, flex_size, &stats);
 470                         if (!stats.free_inodes)
 471                                 continue;
 472                         if (stats.used_dirs >= best_ndir)
 473                                 continue;
 474                         if (stats.free_inodes < avefreei)
 475                                 continue;
 476                         if (stats.free_clusters < avefreec)
 477                                 continue;
 478                         grp = g;
 479                         ret = 0;
 480                         best_ndir = stats.used_dirs;
 481                 }
 482                 if (ret)
 483                         goto fallback;
 484         found_flex_bg:
 485                 if (flex_size == 1) {
 486                         *group = grp;
 487                         return 0;
 488                 }
 489
 490                 /*
 491                  * We pack inodes at the beginning of the flexgroup's
 492                  * inode tables.  Block allocation decisions will do
 493                  * something similar, although regular files will
 494                  * start at 2nd block group of the flexgroup.  See
 495                  * ext4_ext_find_goal() and ext4_find_near().
 496                  */
 497                 grp *= flex_size;
 498                 for (i = 0; i < flex_size; i++) {
 499                         if (grp+i >= real_ngroups)
 500                                 break;
 501                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 502                         if (desc && ext4_free_inodes_count(sb, desc)) {
 503                                 *group = grp+i;
 504                                 return 0;
 505                         }
 506                 }
 507                 goto fallback;
 508         }
 509
 510         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 511         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 512         if (min_inodes < 1)
 513                 min_inodes = 1;
 514         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 515
 516         /*
 517          * Start looking in the flex group where we last allocated an
 518          * inode for this parent directory
 519          */
 520         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 521                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 522                 if (flex_size > 1)
 523                         parent_group >>= sbi->s_log_groups_per_flex;
 524         }
 525
 526         for (i = 0; i < ngroups; i++) {
 527                 grp = (parent_group + i) % ngroups;
 528                 get_orlov_stats(sb, grp, flex_size, &stats);
 529                 if (stats.used_dirs >= max_dirs)
 530                         continue;
 531                 if (stats.free_inodes < min_inodes)
 532                         continue;
 533                 if (stats.free_clusters < min_clusters)
 534                         continue;
 535                 goto found_flex_bg;
 536         }
 537
 538 fallback:
 539         ngroups = real_ngroups;
 540         avefreei = freei / ngroups;
 541 fallback_retry:
 542         parent_group = EXT4_I(parent)->i_block_group;
 543         for (i = 0; i < ngroups; i++) {
 544                 grp = (parent_group + i) % ngroups;
 545                 desc = ext4_get_group_desc(sb, grp, NULL);
 546                 if (desc) {
 547                         grp_free = ext4_free_inodes_count(sb, desc);
 548                         if (grp_free && grp_free >= avefreei) {
 549                                 *group = grp;
 550                                 return 0;
 551                         }
 552                 }
 553         }
 554
 555         if (avefreei) {
 556                 /*
 557                  * The free-inodes counter is approximate, and for really small
 558                  * filesystems the above test can fail to find any blockgroups
 559                  */
 560                 avefreei = 0;
 561                 goto fallback_retry;
 562         }
 563
 564         return -1;
 565 }
 566
 567 static int find_group_other(struct super_block *sb, struct inode *parent,
 568                             ext4_group_t *group, umode_t mode)
 569 {
 570         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 571         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 572         struct ext4_group_desc *desc;
 573         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 574
 575         /*
 576          * Try to place the inode is the same flex group as its
 577          * parent.  If we can't find space, use the Orlov algorithm to
 578          * find another flex group, and store that information in the
 579          * parent directory's inode information so that use that flex
 580          * group for future allocations.
 581          */
 582         if (flex_size > 1) {
 583                 int retry = 0;
 584
 585         try_again:
 586                 parent_group &= ~(flex_size-1);
 587                 last = parent_group + flex_size;
 588                 if (last > ngroups)
 589                         last = ngroups;
 590                 for  (i = parent_group; i < last; i++) {
 591                         desc = ext4_get_group_desc(sb, i, NULL);
 592                         if (desc && ext4_free_inodes_count(sb, desc)) {
 593                                 *group = i;
 594                                 return 0;
 595                         }
 596                 }
 597                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 598                         retry = 1;
 599                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 600                         goto try_again;
 601                 }
 602                 /*
 603                  * If this didn't work, use the Orlov search algorithm
 604                  * to find a new flex group; we pass in the mode to
 605                  * avoid the topdir algorithms.
 606                  */
 607                 *group = parent_group + flex_size;
 608                 if (*group > ngroups)
 609                         *group = 0;
 610                 return find_group_orlov(sb, parent, group, mode, NULL);
 611         }
 612
 613         /*
 614          * Try to place the inode in its parent directory
 615          */
 616         *group = parent_group;
 617         desc = ext4_get_group_desc(sb, *group, NULL);
 618         if (desc && ext4_free_inodes_count(sb, desc) &&
 619             ext4_free_group_clusters(sb, desc))
 620                 return 0;
 621
 622         /*
 623          * We're going to place this inode in a different blockgroup from its
 624          * parent.  We want to cause files in a common directory to all land in
 625          * the same blockgroup.  But we want files which are in a different
 626          * directory which shares a blockgroup with our parent to land in a
 627          * different blockgroup.
 628          *
 629          * So add our directory's i_ino into the starting point for the hash.
 630          */
 631         *group = (*group + parent->i_ino) % ngroups;
 632
 633         /*
 634          * Use a quadratic hash to find a group with a free inode and some free
 635          * blocks.
 636          */
 637         for (i = 1; i < ngroups; i <<= 1) {
 638                 *group += i;
 639                 if (*group >= ngroups)
 640                         *group -= ngroups;
 641                 desc = ext4_get_group_desc(sb, *group, NULL);
 642                 if (desc && ext4_free_inodes_count(sb, desc) &&
 643                     ext4_free_group_clusters(sb, desc))
 644                         return 0;
 645         }
 646
 647         /*
 648          * That failed: try linear search for a free inode, even if that group
 649          * has no free blocks.
 650          */
 651         *group = parent_group;
 652         for (i = 0; i < ngroups; i++) {
 653                 if (++*group >= ngroups)
 654                         *group = 0;
 655                 desc = ext4_get_group_desc(sb, *group, NULL);
 656                 if (desc && ext4_free_inodes_count(sb, desc))
 657                         return 0;
 658         }
 659
 660         return -1;
 661 }
 662
 663 /*
 664  * In no journal mode, if an inode has recently been deleted, we want
 665  * to avoid reusing it until we're reasonably sure the inode table
 666  * block has been written back to disk.  (Yes, these values are
 667  * somewhat arbitrary...)
 668  */
 669 #define RECENTCY_MIN    60
 670 #define RECENTCY_DIRTY  300
 671
 672 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 673 {
 674         struct ext4_group_desc  *gdp;
 675         struct ext4_inode       *raw_inode;
 676         struct buffer_head      *bh;
 677         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 678         int offset, ret = 0;
 679         int recentcy = RECENTCY_MIN;
 680         u32 dtime, now;
 681
 682         gdp = ext4_get_group_desc(sb, group, NULL);
 683         if (unlikely(!gdp))
 684                 return 0;
 685
 686         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 687                        (ino / inodes_per_block));
 688         if (!bh || !buffer_uptodate(bh))
 689                 /*
 690                  * If the block is not in the buffer cache, then it
 691                  * must have been written out.
 692                  */
 693                 goto out;
 694
 695         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 696         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 697
 698         /* i_dtime is only 32 bits on disk, but we only care about relative
 699          * times in the range of a few minutes (i.e. long enough to sync a
 700          * recently-deleted inode to disk), so using the low 32 bits of the
 701          * clock (a 68 year range) is enough, see time_before32() */
 702         dtime = le32_to_cpu(raw_inode->i_dtime);
 703         now = ktime_get_real_seconds();
 704         if (buffer_dirty(bh))
 705                 recentcy += RECENTCY_DIRTY;
 706
 707         if (dtime && time_before32(dtime, now) &&
 708             time_before32(now, dtime + recentcy))
 709                 ret = 1;
 710 out:
 711         brelse(bh);
 712         return ret;
 713 }
 714
 715 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 716                           struct buffer_head *bitmap, unsigned long *ino)
 717 {
 718         bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
 719         unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
 720
 721 next:
 722         *ino = ext4_find_next_zero_bit((unsigned long *)
 723                                        bitmap->b_data,
 724                                        EXT4_INODES_PER_GROUP(sb), *ino);
 725         if (*ino >= EXT4_INODES_PER_GROUP(sb))
 726                 goto not_found;
 727
 728         if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
 729                 recently_deleted_ino = *ino;
 730                 *ino = *ino + 1;
 731                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 732                         goto next;
 733                 goto not_found;
 734         }
 735         return 1;
 736 not_found:
 737         if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
 738                 return 0;
 739         /*
 740          * Not reusing recently deleted inodes is mostly a preference. We don't
 741          * want to report ENOSPC or skew allocation patterns because of that.
 742          * So return even recently deleted inode if we could find better in the
 743          * given range.
 744          */
 745         *ino = recently_deleted_ino;
 746         return 1;
 747 }
 748
 749 int ext4_mark_inode_used(struct super_block *sb, int ino)
 750 {
 751         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
 752         struct buffer_head *inode_bitmap_bh = NULL, *group_desc_bh = NULL;
 753         struct ext4_group_desc *gdp;
 754         ext4_group_t group;
 755         int bit;
 756         int err = -EFSCORRUPTED;
 757
 758         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
 759                 goto out;
 760
 761         group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 762         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 763         inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 764         if (IS_ERR(inode_bitmap_bh))
 765                 return PTR_ERR(inode_bitmap_bh);
 766
 767         if (ext4_test_bit(bit, inode_bitmap_bh->b_data)) {
 768                 err = 0;
 769                 goto out;
 770         }
 771
 772         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 773         if (!gdp || !group_desc_bh) {
 774                 err = -EINVAL;
 775                 goto out;
 776         }
 777
 778         ext4_set_bit(bit, inode_bitmap_bh->b_data);
 779
 780         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 781         err = ext4_handle_dirty_metadata(NULL, NULL, inode_bitmap_bh);
 782         if (err) {
 783                 ext4_std_error(sb, err);
 784                 goto out;
 785         }
 786         err = sync_dirty_buffer(inode_bitmap_bh);
 787         if (err) {
 788                 ext4_std_error(sb, err);
 789                 goto out;
 790         }
 791
 792         /* We may have to initialize the block bitmap if it isn't already */
 793         if (ext4_has_group_desc_csum(sb) &&
 794             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 795                 struct buffer_head *block_bitmap_bh;
 796
 797                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 798                 if (IS_ERR(block_bitmap_bh)) {
 799                         err = PTR_ERR(block_bitmap_bh);
 800                         goto out;
 801                 }
 802
 803                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
 804                 err = ext4_handle_dirty_metadata(NULL, NULL, block_bitmap_bh);
 805                 sync_dirty_buffer(block_bitmap_bh);
 806
 807                 /* recheck and clear flag under lock if we still need to */
 808                 ext4_lock_group(sb, group);
 809                 if (ext4_has_group_desc_csum(sb) &&
 810                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
 811                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 812                         ext4_free_group_clusters_set(sb, gdp,
 813                                 ext4_free_clusters_after_init(sb, group, gdp));
 814                         ext4_block_bitmap_csum_set(sb, group, gdp,
 815                                                    block_bitmap_bh);
 816                         ext4_group_desc_csum_set(sb, group, gdp);
 817                 }
 818                 ext4_unlock_group(sb, group);
 819                 brelse(block_bitmap_bh);
 820
 821                 if (err) {
 822                         ext4_std_error(sb, err);
 823                         goto out;
 824                 }
 825         }
 826
 827         /* Update the relevant bg descriptor fields */
 828         if (ext4_has_group_desc_csum(sb)) {
 829                 int free;
 830
 831                 ext4_lock_group(sb, group); /* while we modify the bg desc */
 832                 free = EXT4_INODES_PER_GROUP(sb) -
 833                         ext4_itable_unused_count(sb, gdp);
 834                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 835                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 836                         free = 0;
 837                 }
 838
 839                 /*
 840                  * Check the relative inode number against the last used
 841                  * relative inode number in this group. if it is greater
 842                  * we need to update the bg_itable_unused count
 843                  */
 844                 if (bit >= free)
 845                         ext4_itable_unused_set(sb, gdp,
 846                                         (EXT4_INODES_PER_GROUP(sb) - bit - 1));
 847         } else {
 848                 ext4_lock_group(sb, group);
 849         }
 850
 851         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
 852         if (ext4_has_group_desc_csum(sb)) {
 853                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
 854                                            EXT4_INODES_PER_GROUP(sb) / 8);
 855                 ext4_group_desc_csum_set(sb, group, gdp);
 856         }
 857
 858         ext4_unlock_group(sb, group);
 859         err = ext4_handle_dirty_metadata(NULL, NULL, group_desc_bh);
 860         sync_dirty_buffer(group_desc_bh);
 861 out:
 862         return err;
 863 }
 864
 865 static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
 866                                             bool encrypt)
 867 {
 868         struct super_block *sb = dir->i_sb;
 869         int nblocks = 0;
 870 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 871         struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 872
 873         if (IS_ERR(p))
 874                 return PTR_ERR(p);
 875         if (p) {
 876                 int acl_size = p->a_count * sizeof(ext4_acl_entry);
 877
 878                 nblocks += (S_ISDIR(mode) ? 2 : 1) *
 879                         __ext4_xattr_set_credits(sb, NULL /* inode */,
 880                                                  NULL /* block_bh */, acl_size,
 881                                                  true /* is_create */);
 882                 posix_acl_release(p);
 883         }
 884 #endif
 885
 886 #ifdef CONFIG_SECURITY
 887         {
 888                 int num_security_xattrs = 1;
 889
 890 #ifdef CONFIG_INTEGRITY
 891                 num_security_xattrs++;
 892 #endif
 893                 /*
 894                  * We assume that security xattrs are never more than 1k.
 895                  * In practice they are under 128 bytes.
 896                  */
 897                 nblocks += num_security_xattrs *
 898                         __ext4_xattr_set_credits(sb, NULL /* inode */,
 899                                                  NULL /* block_bh */, 1024,
 900                                                  true /* is_create */);
 901         }
 902 #endif
 903         if (encrypt)
 904                 nblocks += __ext4_xattr_set_credits(sb,
 905                                                     NULL /* inode */,
 906                                                     NULL /* block_bh */,
 907                                                     FSCRYPT_SET_CONTEXT_MAX_SIZE,
 908                                                     true /* is_create */);
 909         return nblocks;
 910 }
 911
 912 /*
 913  * There are two policies for allocating an inode.  If the new inode is
 914  * a directory, then a forward search is made for a block group with both
 915  * free space and a low directory-to-inode ratio; if that fails, then of
 916  * the groups with above-average free space, that group with the fewest
 917  * directories already is chosen.
 918  *
 919  * For other inodes, search forward from the parent directory's block
 920  * group to find a free inode.
 921  */
 922 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 923                                umode_t mode, const struct qstr *qstr,
 924                                __u32 goal, uid_t *owner, __u32 i_flags,
 925                                int handle_type, unsigned int line_no,
 926                                int nblocks)
 927 {
 928         struct super_block *sb;
 929         struct buffer_head *inode_bitmap_bh = NULL;
 930         struct buffer_head *group_desc_bh;
 931         ext4_group_t ngroups, group = 0;
 932         unsigned long ino = 0;
 933         struct inode *inode;
 934         struct ext4_group_desc *gdp = NULL;
 935         struct ext4_inode_info *ei;
 936         struct ext4_sb_info *sbi;
 937         int ret2, err;
 938         struct inode *ret;
 939         ext4_group_t i;
 940         ext4_group_t flex_group;
 941         struct ext4_group_info *grp = NULL;
 942         bool encrypt = false;
 943
 944         /* Cannot create files in a deleted directory */
 945         if (!dir || !dir->i_nlink)
 946                 return ERR_PTR(-EPERM);
 947
 948         sb = dir->i_sb;
 949         sbi = EXT4_SB(sb);
 950
 951         if (unlikely(ext4_forced_shutdown(sbi)))
 952                 return ERR_PTR(-EIO);
 953
 954         ngroups = ext4_get_groups_count(sb);
 955         trace_ext4_request_inode(dir, mode);
 956         inode = new_inode(sb);
 957         if (!inode)
 958                 return ERR_PTR(-ENOMEM);
 959         ei = EXT4_I(inode);
 960
 961         /*
 962          * Initialize owners and quota early so that we don't have to account
 963          * for quota initialization worst case in standard inode creating
 964          * transaction
 965          */
 966         if (owner) {
 967                 inode->i_mode = mode;
 968                 i_uid_write(inode, owner[0]);
 969                 i_gid_write(inode, owner[1]);
 970         } else if (test_opt(sb, GRPID)) {
 971                 inode->i_mode = mode;
 972                 inode->i_uid = current_fsuid();
 973                 inode->i_gid = dir->i_gid;
 974         } else
 975                 inode_init_owner(inode, dir, mode);
 976
 977         if (ext4_has_feature_project(sb) &&
 978             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 979                 ei->i_projid = EXT4_I(dir)->i_projid;
 980         else
 981                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 982
 983         if (!(i_flags & EXT4_EA_INODE_FL)) {
 984                 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
 985                 if (err)
 986                         goto out;
 987         }
 988
 989         err = dquot_initialize(inode);
 990         if (err)
 991                 goto out;
 992
 993         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 994                 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
 995                 if (ret2 < 0) {
 996                         err = ret2;
 997                         goto out;
 998                 }
 999                 nblocks += ret2;
1000         }
1001
1002         if (!goal)
1003                 goal = sbi->s_inode_goal;
1004
1005         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
1006                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
1007                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
1008                 ret2 = 0;
1009                 goto got_group;
1010         }
1011
1012         if (S_ISDIR(mode))
1013                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
1014         else
1015                 ret2 = find_group_other(sb, dir, &group, mode);
1016
1017 got_group:
1018         EXT4_I(dir)->i_last_alloc_group = group;
1019         err = -ENOSPC;
1020         if (ret2 == -1)
1021                 goto out;
1022
1023         /*
1024          * Normally we will only go through one pass of this loop,
1025          * unless we get unlucky and it turns out the group we selected
1026          * had its last inode grabbed by someone else.
1027          */
1028         for (i = 0; i < ngroups; i++, ino = 0) {
1029                 err = -EIO;
1030
1031                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1032                 if (!gdp)
1033                         goto out;
1034
1035                 /*
1036                  * Check free inodes count before loading bitmap.
1037                  */
1038                 if (ext4_free_inodes_count(sb, gdp) == 0)
1039                         goto next_group;
1040
1041                 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1042                         grp = ext4_get_group_info(sb, group);
1043                         /*
1044                          * Skip groups with already-known suspicious inode
1045                          * tables
1046                          */
1047                         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
1048                                 goto next_group;
1049                 }
1050
1051                 brelse(inode_bitmap_bh);
1052                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
1053                 /* Skip groups with suspicious inode tables */
1054                 if (((!(sbi->s_mount_state & EXT4_FC_REPLAY))
1055                      && EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) ||
1056                     IS_ERR(inode_bitmap_bh)) {
1057                         inode_bitmap_bh = NULL;
1058                         goto next_group;
1059                 }
1060
1061 repeat_in_this_group:
1062                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1063                 if (!ret2)
1064                         goto next_group;
1065
1066                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
1067                         ext4_error(sb, "reserved inode found cleared - "
1068                                    "inode=%lu", ino + 1);
1069                         ext4_mark_group_bitmap_corrupted(sb, group,
1070                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1071                         goto next_group;
1072                 }
1073
1074                 if ((!(sbi->s_mount_state & EXT4_FC_REPLAY)) && !handle) {
1075                         BUG_ON(nblocks <= 0);
1076                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
1077                                  handle_type, nblocks, 0,
1078                                  ext4_trans_default_revoke_credits(sb));
1079                         if (IS_ERR(handle)) {
1080                                 err = PTR_ERR(handle);
1081                                 ext4_std_error(sb, err);
1082                                 goto out;
1083                         }
1084                 }
1085                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
1086                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
1087                 if (err) {
1088                         ext4_std_error(sb, err);
1089                         goto out;
1090                 }
1091                 ext4_lock_group(sb, group);
1092                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
1093                 if (ret2) {
1094                         /* Someone already took the bit. Repeat the search
1095                          * with lock held.
1096                          */
1097                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
1098                         if (ret2) {
1099                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
1100                                 ret2 = 0;
1101                         } else {
1102                                 ret2 = 1; /* we didn't grab the inode */
1103                         }
1104                 }
1105                 ext4_unlock_group(sb, group);
1106                 ino++;          /* the inode bitmap is zero-based */
1107                 if (!ret2)
1108                         goto got; /* we grabbed the inode! */
1109
1110                 if (ino < EXT4_INODES_PER_GROUP(sb))
1111                         goto repeat_in_this_group;
1112 next_group:
1113                 if (++group == ngroups)
1114                         group = 0;
1115         }
1116         err = -ENOSPC;
1117         goto out;
1118
1119 got:
1120         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
1121         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
1122         if (err) {
1123                 ext4_std_error(sb, err);
1124                 goto out;
1125         }
1126
1127         BUFFER_TRACE(group_desc_bh, "get_write_access");
1128         err = ext4_journal_get_write_access(handle, group_desc_bh);
1129         if (err) {
1130                 ext4_std_error(sb, err);
1131                 goto out;
1132         }
1133
1134         /* We may have to initialize the block bitmap if it isn't already */
1135         if (ext4_has_group_desc_csum(sb) &&
1136             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1137                 struct buffer_head *block_bitmap_bh;
1138
1139                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1140                 if (IS_ERR(block_bitmap_bh)) {
1141                         err = PTR_ERR(block_bitmap_bh);
1142                         goto out;
1143                 }
1144                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1145                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1146                 if (err) {
1147                         brelse(block_bitmap_bh);
1148                         ext4_std_error(sb, err);
1149                         goto out;
1150                 }
1151
1152                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1153                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1154
1155                 /* recheck and clear flag under lock if we still need to */
1156                 ext4_lock_group(sb, group);
1157                 if (ext4_has_group_desc_csum(sb) &&
1158                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1159                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1160                         ext4_free_group_clusters_set(sb, gdp,
1161                                 ext4_free_clusters_after_init(sb, group, gdp));
1162                         ext4_block_bitmap_csum_set(sb, group, gdp,
1163                                                    block_bitmap_bh);
1164                         ext4_group_desc_csum_set(sb, group, gdp);
1165                 }
1166                 ext4_unlock_group(sb, group);
1167                 brelse(block_bitmap_bh);
1168
1169                 if (err) {
1170                         ext4_std_error(sb, err);
1171                         goto out;
1172                 }
1173         }
1174
1175         /* Update the relevant bg descriptor fields */
1176         if (ext4_has_group_desc_csum(sb)) {
1177                 int free;
1178                 struct ext4_group_info *grp = NULL;
1179
1180                 if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
1181                         grp = ext4_get_group_info(sb, group);
1182                         down_read(&grp->alloc_sem); /*
1183                                                      * protect vs itable
1184                                                      * lazyinit
1185                                                      */
1186                 }
1187                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1188                 free = EXT4_INODES_PER_GROUP(sb) -
1189                         ext4_itable_unused_count(sb, gdp);
1190                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1191                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1192                         free = 0;
1193                 }
1194                 /*
1195                  * Check the relative inode number against the last used
1196                  * relative inode number in this group. if it is greater
1197                  * we need to update the bg_itable_unused count
1198                  */
1199                 if (ino > free)
1200                         ext4_itable_unused_set(sb, gdp,
1201                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1202                 if (!(sbi->s_mount_state & EXT4_FC_REPLAY))
1203                         up_read(&grp->alloc_sem);
1204         } else {
1205                 ext4_lock_group(sb, group);
1206         }
1207
1208         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1209         if (S_ISDIR(mode)) {
1210                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1211                 if (sbi->s_log_groups_per_flex) {
1212                         ext4_group_t f = ext4_flex_group(sbi, group);
1213
1214                         atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1215                                                         f)->used_dirs);
1216                 }
1217         }
1218         if (ext4_has_group_desc_csum(sb)) {
1219                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1220                                            EXT4_INODES_PER_GROUP(sb) / 8);
1221                 ext4_group_desc_csum_set(sb, group, gdp);
1222         }
1223         ext4_unlock_group(sb, group);
1224
1225         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1226         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1227         if (err) {
1228                 ext4_std_error(sb, err);
1229                 goto out;
1230         }
1231
1232         percpu_counter_dec(&sbi->s_freeinodes_counter);
1233         if (S_ISDIR(mode))
1234                 percpu_counter_inc(&sbi->s_dirs_counter);
1235
1236         if (sbi->s_log_groups_per_flex) {
1237                 flex_group = ext4_flex_group(sbi, group);
1238                 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1239                                                 flex_group)->free_inodes);
1240         }
1241
1242         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1243         /* This is the optimal IO size (for stat), not the fs block size */
1244         inode->i_blocks = 0;
1245         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1246         ei->i_crtime = inode->i_mtime;
1247
1248         memset(ei->i_data, 0, sizeof(ei->i_data));
1249         ei->i_dir_start_lookup = 0;
1250         ei->i_disksize = 0;
1251
1252         /* Don't inherit extent flag from directory, amongst others. */
1253         ei->i_flags =
1254                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1255         ei->i_flags |= i_flags;
1256         ei->i_file_acl = 0;
1257         ei->i_dtime = 0;
1258         ei->i_block_group = group;
1259         ei->i_last_alloc_group = ~0;
1260
1261         ext4_set_inode_flags(inode, true);
1262         if (IS_DIRSYNC(inode))
1263                 ext4_handle_sync(handle);
1264         if (insert_inode_locked(inode) < 0) {
1265                 /*
1266                  * Likely a bitmap corruption causing inode to be allocated
1267                  * twice.
1268                  */
1269                 err = -EIO;
1270                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1271                            inode->i_ino);
1272                 ext4_mark_group_bitmap_corrupted(sb, group,
1273                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1274                 goto out;
1275         }
1276         inode->i_generation = prandom_u32();
1277
1278         /* Precompute checksum seed for inode metadata */
1279         if (ext4_has_metadata_csum(sb)) {
1280                 __u32 csum;
1281                 __le32 inum = cpu_to_le32(inode->i_ino);
1282                 __le32 gen = cpu_to_le32(inode->i_generation);
1283                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1284                                    sizeof(inum));
1285                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1286                                               sizeof(gen));
1287         }
1288
1289         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1290         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1291
1292         ei->i_extra_isize = sbi->s_want_extra_isize;
1293         ei->i_inline_off = 0;
1294         if (ext4_has_feature_inline_data(sb))
1295                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1296         ret = inode;
1297         err = dquot_alloc_inode(inode);
1298         if (err)
1299                 goto fail_drop;
1300
1301         /*
1302          * Since the encryption xattr will always be unique, create it first so
1303          * that it's less likely to end up in an external xattr block and
1304          * prevent its deduplication.
1305          */
1306         if (encrypt) {
1307                 err = fscrypt_set_context(inode, handle);
1308                 if (err)
1309                         goto fail_free_drop;
1310         }
1311
1312         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1313                 err = ext4_init_acl(handle, inode, dir);
1314                 if (err)
1315                         goto fail_free_drop;
1316
1317                 err = ext4_init_security(handle, inode, dir, qstr);
1318                 if (err)
1319                         goto fail_free_drop;
1320         }
1321
1322         if (ext4_has_feature_extents(sb)) {
1323                 /* set extent flag only for directory, file and normal symlink*/
1324                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1325                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1326                         ext4_ext_tree_init(handle, inode);
1327                 }
1328         }
1329
1330         if (ext4_handle_valid(handle)) {
1331                 ei->i_sync_tid = handle->h_transaction->t_tid;
1332                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1333         }
1334
1335         err = ext4_mark_inode_dirty(handle, inode);
1336         if (err) {
1337                 ext4_std_error(sb, err);
1338                 goto fail_free_drop;
1339         }
1340
1341         ext4_debug("allocating inode %lu\n", inode->i_ino);
1342         trace_ext4_allocate_inode(inode, dir, mode);
1343         brelse(inode_bitmap_bh);
1344         return ret;
1345
1346 fail_free_drop:
1347         dquot_free_inode(inode);
1348 fail_drop:
1349         clear_nlink(inode);
1350         unlock_new_inode(inode);
1351 out:
1352         dquot_drop(inode);
1353         inode->i_flags |= S_NOQUOTA;
1354         iput(inode);
1355         brelse(inode_bitmap_bh);
1356         return ERR_PTR(err);
1357 }
1358
1359 /* Verify that we are loading a valid orphan from disk */
1360 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1361 {
1362         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1363         ext4_group_t block_group;
1364         int bit;
1365         struct buffer_head *bitmap_bh = NULL;
1366         struct inode *inode = NULL;
1367         int err = -EFSCORRUPTED;
1368
1369         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1370                 goto bad_orphan;
1371
1372         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1373         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1374         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1375         if (IS_ERR(bitmap_bh))
1376                 return ERR_CAST(bitmap_bh);
1377
1378         /* Having the inode bit set should be a 100% indicator that this
1379          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1380          * inodes that were being truncated, so we can't check i_nlink==0.
1381          */
1382         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1383                 goto bad_orphan;
1384
1385         inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1386         if (IS_ERR(inode)) {
1387                 err = PTR_ERR(inode);
1388                 ext4_error_err(sb, -err,
1389                                "couldn't read orphan inode %lu (err %d)",
1390                                ino, err);
1391                 brelse(bitmap_bh);
1392                 return inode;
1393         }
1394
1395         /*
1396          * If the orphans has i_nlinks > 0 then it should be able to
1397          * be truncated, otherwise it won't be removed from the orphan
1398          * list during processing and an infinite loop will result.
1399          * Similarly, it must not be a bad inode.
1400          */
1401         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1402             is_bad_inode(inode))
1403                 goto bad_orphan;
1404
1405         if (NEXT_ORPHAN(inode) > max_ino)
1406                 goto bad_orphan;
1407         brelse(bitmap_bh);
1408         return inode;
1409
1410 bad_orphan:
1411         ext4_error(sb, "bad orphan inode %lu", ino);
1412         if (bitmap_bh)
1413                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1414                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1415                        ext4_test_bit(bit, bitmap_bh->b_data));
1416         if (inode) {
1417                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1418                        is_bad_inode(inode));
1419                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1420                        NEXT_ORPHAN(inode));
1421                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1422                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1423                 /* Avoid freeing blocks if we got a bad deleted inode */
1424                 if (inode->i_nlink == 0)
1425                         inode->i_blocks = 0;
1426                 iput(inode);
1427         }
1428         brelse(bitmap_bh);
1429         return ERR_PTR(err);
1430 }
1431
1432 unsigned long ext4_count_free_inodes(struct super_block *sb)
1433 {
1434         unsigned long desc_count;
1435         struct ext4_group_desc *gdp;
1436         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1437 #ifdef EXT4FS_DEBUG
1438         struct ext4_super_block *es;
1439         unsigned long bitmap_count, x;
1440         struct buffer_head *bitmap_bh = NULL;
1441
1442         es = EXT4_SB(sb)->s_es;
1443         desc_count = 0;
1444         bitmap_count = 0;
1445         gdp = NULL;
1446         for (i = 0; i < ngroups; i++) {
1447                 gdp = ext4_get_group_desc(sb, i, NULL);
1448                 if (!gdp)
1449                         continue;
1450                 desc_count += ext4_free_inodes_count(sb, gdp);
1451                 brelse(bitmap_bh);
1452                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1453                 if (IS_ERR(bitmap_bh)) {
1454                         bitmap_bh = NULL;
1455                         continue;
1456                 }
1457
1458                 x = ext4_count_free(bitmap_bh->b_data,
1459                                     EXT4_INODES_PER_GROUP(sb) / 8);
1460                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1461                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1462                 bitmap_count += x;
1463         }
1464         brelse(bitmap_bh);
1465         printk(KERN_DEBUG "ext4_count_free_inodes: "
1466                "stored = %u, computed = %lu, %lu\n",
1467                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1468         return desc_count;
1469 #else
1470         desc_count = 0;
1471         for (i = 0; i < ngroups; i++) {
1472                 gdp = ext4_get_group_desc(sb, i, NULL);
1473                 if (!gdp)
1474                         continue;
1475                 desc_count += ext4_free_inodes_count(sb, gdp);
1476                 cond_resched();
1477         }
1478         return desc_count;
1479 #endif
1480 }
1481
1482 /* Called at mount-time, super-block is locked */
1483 unsigned long ext4_count_dirs(struct super_block * sb)
1484 {
1485         unsigned long count = 0;
1486         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1487
1488         for (i = 0; i < ngroups; i++) {
1489                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1490                 if (!gdp)
1491                         continue;
1492                 count += ext4_used_dirs_count(sb, gdp);
1493         }
1494         return count;
1495 }
1496
1497 /*
1498  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1499  * inode table. Must be called without any spinlock held. The only place
1500  * where it is called from on active part of filesystem is ext4lazyinit
1501  * thread, so we do not need any special locks, however we have to prevent
1502  * inode allocation from the current group, so we take alloc_sem lock, to
1503  * block ext4_new_inode() until we are finished.
1504  */
1505 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1506                                  int barrier)
1507 {
1508         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1509         struct ext4_sb_info *sbi = EXT4_SB(sb);
1510         struct ext4_group_desc *gdp = NULL;
1511         struct buffer_head *group_desc_bh;
1512         handle_t *handle;
1513         ext4_fsblk_t blk;
1514         int num, ret = 0, used_blks = 0;
1515
1516         /* This should not happen, but just to be sure check this */
1517         if (sb_rdonly(sb)) {
1518                 ret = 1;
1519                 goto out;
1520         }
1521
1522         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1523         if (!gdp)
1524                 goto out;
1525
1526         /*
1527          * We do not need to lock this, because we are the only one
1528          * handling this flag.
1529          */
1530         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1531                 goto out;
1532
1533         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1534         if (IS_ERR(handle)) {
1535                 ret = PTR_ERR(handle);
1536                 goto out;
1537         }
1538
1539         down_write(&grp->alloc_sem);
1540         /*
1541          * If inode bitmap was already initialized there may be some
1542          * used inodes so we need to skip blocks with used inodes in
1543          * inode table.
1544          */
1545         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1546                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1547                             ext4_itable_unused_count(sb, gdp)),
1548                             sbi->s_inodes_per_block);
1549
1550         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1551             ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1552                                ext4_itable_unused_count(sb, gdp)) <
1553                               EXT4_FIRST_INO(sb)))) {
1554                 ext4_error(sb, "Something is wrong with group %u: "
1555                            "used itable blocks: %d; "
1556                            "itable unused count: %u",
1557                            group, used_blks,
1558                            ext4_itable_unused_count(sb, gdp));
1559                 ret = 1;
1560                 goto err_out;
1561         }
1562
1563         blk = ext4_inode_table(sb, gdp) + used_blks;
1564         num = sbi->s_itb_per_group - used_blks;
1565
1566         BUFFER_TRACE(group_desc_bh, "get_write_access");
1567         ret = ext4_journal_get_write_access(handle,
1568                                             group_desc_bh);
1569         if (ret)
1570                 goto err_out;
1571
1572         /*
1573          * Skip zeroout if the inode table is full. But we set the ZEROED
1574          * flag anyway, because obviously, when it is full it does not need
1575          * further zeroing.
1576          */
1577         if (unlikely(num == 0))
1578                 goto skip_zeroout;
1579
1580         ext4_debug("going to zero out inode table in group %d\n",
1581                    group);
1582         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1583         if (ret < 0)
1584                 goto err_out;
1585         if (barrier)
1586                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS);
1587
1588 skip_zeroout:
1589         ext4_lock_group(sb, group);
1590         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1591         ext4_group_desc_csum_set(sb, group, gdp);
1592         ext4_unlock_group(sb, group);
1593
1594         BUFFER_TRACE(group_desc_bh,
1595                      "call ext4_handle_dirty_metadata");
1596         ret = ext4_handle_dirty_metadata(handle, NULL,
1597                                          group_desc_bh);
1598
1599 err_out:
1600         up_write(&grp->alloc_sem);
1601         ext4_journal_stop(handle);
1602 out:
1603         return ret;
1604 }