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 ([email protected])
   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 ([email protected]), 1993
  12  *  Big-endian to little-endian byte-swapping/bitmaps by
  13  *        David S. Miller ([email protected]), 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 an ERR_PTR on error.
 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_error_err(sb, EIO, "Cannot read inode bitmap - "
 200                                "block_group = %u, inode_bitmap = %llu",
 201                                block_group, bitmap_blk);
 202                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 203                                 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 204                 return ERR_PTR(-EIO);
 205         }
 206
 207 verify:
 208         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 209         if (err)
 210                 goto out;
 211         return bh;
 212 out:
 213         put_bh(bh);
 214         return ERR_PTR(err);
 215 }
 216
 217 /*
 218  * NOTE! When we get the inode, we're the only people
 219  * that have access to it, and as such there are no
 220  * race conditions we have to worry about. The inode
 221  * is not on the hash-lists, and it cannot be reached
 222  * through the filesystem because the directory entry
 223  * has been deleted earlier.
 224  *
 225  * HOWEVER: we must make sure that we get no aliases,
 226  * which means that we have to call "clear_inode()"
 227  * _before_ we mark the inode not in use in the inode
 228  * bitmaps. Otherwise a newly created file might use
 229  * the same inode number (not actually the same pointer
 230  * though), and then we'd have two inodes sharing the
 231  * same inode number and space on the harddisk.
 232  */
 233 void ext4_free_inode(handle_t *handle, struct inode *inode)
 234 {
 235         struct super_block *sb = inode->i_sb;
 236         int is_directory;
 237         unsigned long ino;
 238         struct buffer_head *bitmap_bh = NULL;
 239         struct buffer_head *bh2;
 240         ext4_group_t block_group;
 241         unsigned long bit;
 242         struct ext4_group_desc *gdp;
 243         struct ext4_super_block *es;
 244         struct ext4_sb_info *sbi;
 245         int fatal = 0, err, count, cleared;
 246         struct ext4_group_info *grp;
 247
 248         if (!sb) {
 249                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 250                        "nonexistent device\n", __func__, __LINE__);
 251                 return;
 252         }
 253         if (atomic_read(&inode->i_count) > 1) {
 254                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 255                          __func__, __LINE__, inode->i_ino,
 256                          atomic_read(&inode->i_count));
 257                 return;
 258         }
 259         if (inode->i_nlink) {
 260                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 261                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 262                 return;
 263         }
 264         sbi = EXT4_SB(sb);
 265
 266         ino = inode->i_ino;
 267         ext4_debug("freeing inode %lu\n", ino);
 268         trace_ext4_free_inode(inode);
 269
 270         dquot_initialize(inode);
 271         dquot_free_inode(inode);
 272
 273         is_directory = S_ISDIR(inode->i_mode);
 274
 275         /* Do this BEFORE marking the inode not in use or returning an error */
 276         ext4_clear_inode(inode);
 277
 278         es = sbi->s_es;
 279         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 280                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 281                 goto error_return;
 282         }
 283         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 284         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 285         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 286         /* Don't bother if the inode bitmap is corrupt. */
 287         grp = ext4_get_group_info(sb, block_group);
 288         if (IS_ERR(bitmap_bh)) {
 289                 fatal = PTR_ERR(bitmap_bh);
 290                 bitmap_bh = NULL;
 291                 goto error_return;
 292         }
 293         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 294                 fatal = -EFSCORRUPTED;
 295                 goto error_return;
 296         }
 297
 298         BUFFER_TRACE(bitmap_bh, "get_write_access");
 299         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 300         if (fatal)
 301                 goto error_return;
 302
 303         fatal = -ESRCH;
 304         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 305         if (gdp) {
 306                 BUFFER_TRACE(bh2, "get_write_access");
 307                 fatal = ext4_journal_get_write_access(handle, bh2);
 308         }
 309         ext4_lock_group(sb, block_group);
 310         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 311         if (fatal || !cleared) {
 312                 ext4_unlock_group(sb, block_group);
 313                 goto out;
 314         }
 315
 316         count = ext4_free_inodes_count(sb, gdp) + 1;
 317         ext4_free_inodes_set(sb, gdp, count);
 318         if (is_directory) {
 319                 count = ext4_used_dirs_count(sb, gdp) - 1;
 320                 ext4_used_dirs_set(sb, gdp, count);
 321                 percpu_counter_dec(&sbi->s_dirs_counter);
 322         }
 323         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 324                                    EXT4_INODES_PER_GROUP(sb) / 8);
 325         ext4_group_desc_csum_set(sb, block_group, gdp);
 326         ext4_unlock_group(sb, block_group);
 327
 328         percpu_counter_inc(&sbi->s_freeinodes_counter);
 329         if (sbi->s_log_groups_per_flex) {
 330                 struct flex_groups *fg;
 331
 332                 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
 333                                          ext4_flex_group(sbi, block_group));
 334                 atomic_inc(&fg->free_inodes);
 335                 if (is_directory)
 336                         atomic_dec(&fg->used_dirs);
 337         }
 338         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 339         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 340 out:
 341         if (cleared) {
 342                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 343                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 344                 if (!fatal)
 345                         fatal = err;
 346         } else {
 347                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 348                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 349                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 350         }
 351
 352 error_return:
 353         brelse(bitmap_bh);
 354         ext4_std_error(sb, fatal);
 355 }
 356
 357 struct orlov_stats {
 358         __u64 free_clusters;
 359         __u32 free_inodes;
 360         __u32 used_dirs;
 361 };
 362
 363 /*
 364  * Helper function for Orlov's allocator; returns critical information
 365  * for a particular block group or flex_bg.  If flex_size is 1, then g
 366  * is a block group number; otherwise it is flex_bg number.
 367  */
 368 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 369                             int flex_size, struct orlov_stats *stats)
 370 {
 371         struct ext4_group_desc *desc;
 372
 373         if (flex_size > 1) {
 374                 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
 375                                                              s_flex_groups, g);
 376                 stats->free_inodes = atomic_read(&fg->free_inodes);
 377                 stats->free_clusters = atomic64_read(&fg->free_clusters);
 378                 stats->used_dirs = atomic_read(&fg->used_dirs);
 379                 return;
 380         }
 381
 382         desc = ext4_get_group_desc(sb, g, NULL);
 383         if (desc) {
 384                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 385                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 386                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 387         } else {
 388                 stats->free_inodes = 0;
 389                 stats->free_clusters = 0;
 390                 stats->used_dirs = 0;
 391         }
 392 }
 393
 394 /*
 395  * Orlov's allocator for directories.
 396  *
 397  * We always try to spread first-level directories.
 398  *
 399  * If there are blockgroups with both free inodes and free blocks counts
 400  * not worse than average we return one with smallest directory count.
 401  * Otherwise we simply return a random group.
 402  *
 403  * For the rest rules look so:
 404  *
 405  * It's OK to put directory into a group unless
 406  * it has too many directories already (max_dirs) or
 407  * it has too few free inodes left (min_inodes) or
 408  * it has too few free blocks left (min_blocks) or
 409  * Parent's group is preferred, if it doesn't satisfy these
 410  * conditions we search cyclically through the rest. If none
 411  * of the groups look good we just look for a group with more
 412  * free inodes than average (starting at parent's group).
 413  */
 414
 415 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 416                             ext4_group_t *group, umode_t mode,
 417                             const struct qstr *qstr)
 418 {
 419         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 420         struct ext4_sb_info *sbi = EXT4_SB(sb);
 421         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 422         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 423         unsigned int freei, avefreei, grp_free;
 424         ext4_fsblk_t freeb, avefreec;
 425         unsigned int ndirs;
 426         int max_dirs, min_inodes;
 427         ext4_grpblk_t min_clusters;
 428         ext4_group_t i, grp, g, ngroups;
 429         struct ext4_group_desc *desc;
 430         struct orlov_stats stats;
 431         int flex_size = ext4_flex_bg_size(sbi);
 432         struct dx_hash_info hinfo;
 433
 434         ngroups = real_ngroups;
 435         if (flex_size > 1) {
 436                 ngroups = (real_ngroups + flex_size - 1) >>
 437                         sbi->s_log_groups_per_flex;
 438                 parent_group >>= sbi->s_log_groups_per_flex;
 439         }
 440
 441         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 442         avefreei = freei / ngroups;
 443         freeb = EXT4_C2B(sbi,
 444                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 445         avefreec = freeb;
 446         do_div(avefreec, ngroups);
 447         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 448
 449         if (S_ISDIR(mode) &&
 450             ((parent == d_inode(sb->s_root)) ||
 451              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 452                 int best_ndir = inodes_per_group;
 453                 int ret = -1;
 454
 455                 if (qstr) {
 456                         hinfo.hash_version = DX_HASH_HALF_MD4;
 457                         hinfo.seed = sbi->s_hash_seed;
 458                         ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
 459                         grp = hinfo.hash;
 460                 } else
 461                         grp = prandom_u32();
 462                 parent_group = (unsigned)grp % ngroups;
 463                 for (i = 0; i < ngroups; i++) {
 464                         g = (parent_group + i) % ngroups;
 465                         get_orlov_stats(sb, g, flex_size, &stats);
 466                         if (!stats.free_inodes)
 467                                 continue;
 468                         if (stats.used_dirs >= best_ndir)
 469                                 continue;
 470                         if (stats.free_inodes < avefreei)
 471                                 continue;
 472                         if (stats.free_clusters < avefreec)
 473                                 continue;
 474                         grp = g;
 475                         ret = 0;
 476                         best_ndir = stats.used_dirs;
 477                 }
 478                 if (ret)
 479                         goto fallback;
 480         found_flex_bg:
 481                 if (flex_size == 1) {
 482                         *group = grp;
 483                         return 0;
 484                 }
 485
 486                 /*
 487                  * We pack inodes at the beginning of the flexgroup's
 488                  * inode tables.  Block allocation decisions will do
 489                  * something similar, although regular files will
 490                  * start at 2nd block group of the flexgroup.  See
 491                  * ext4_ext_find_goal() and ext4_find_near().
 492                  */
 493                 grp *= flex_size;
 494                 for (i = 0; i < flex_size; i++) {
 495                         if (grp+i >= real_ngroups)
 496                                 break;
 497                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 498                         if (desc && ext4_free_inodes_count(sb, desc)) {
 499                                 *group = grp+i;
 500                                 return 0;
 501                         }
 502                 }
 503                 goto fallback;
 504         }
 505
 506         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 507         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 508         if (min_inodes < 1)
 509                 min_inodes = 1;
 510         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 511
 512         /*
 513          * Start looking in the flex group where we last allocated an
 514          * inode for this parent directory
 515          */
 516         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 517                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 518                 if (flex_size > 1)
 519                         parent_group >>= sbi->s_log_groups_per_flex;
 520         }
 521
 522         for (i = 0; i < ngroups; i++) {
 523                 grp = (parent_group + i) % ngroups;
 524                 get_orlov_stats(sb, grp, flex_size, &stats);
 525                 if (stats.used_dirs >= max_dirs)
 526                         continue;
 527                 if (stats.free_inodes < min_inodes)
 528                         continue;
 529                 if (stats.free_clusters < min_clusters)
 530                         continue;
 531                 goto found_flex_bg;
 532         }
 533
 534 fallback:
 535         ngroups = real_ngroups;
 536         avefreei = freei / ngroups;
 537 fallback_retry:
 538         parent_group = EXT4_I(parent)->i_block_group;
 539         for (i = 0; i < ngroups; i++) {
 540                 grp = (parent_group + i) % ngroups;
 541                 desc = ext4_get_group_desc(sb, grp, NULL);
 542                 if (desc) {
 543                         grp_free = ext4_free_inodes_count(sb, desc);
 544                         if (grp_free && grp_free >= avefreei) {
 545                                 *group = grp;
 546                                 return 0;
 547                         }
 548                 }
 549         }
 550
 551         if (avefreei) {
 552                 /*
 553                  * The free-inodes counter is approximate, and for really small
 554                  * filesystems the above test can fail to find any blockgroups
 555                  */
 556                 avefreei = 0;
 557                 goto fallback_retry;
 558         }
 559
 560         return -1;
 561 }
 562
 563 static int find_group_other(struct super_block *sb, struct inode *parent,
 564                             ext4_group_t *group, umode_t mode)
 565 {
 566         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 567         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 568         struct ext4_group_desc *desc;
 569         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 570
 571         /*
 572          * Try to place the inode is the same flex group as its
 573          * parent.  If we can't find space, use the Orlov algorithm to
 574          * find another flex group, and store that information in the
 575          * parent directory's inode information so that use that flex
 576          * group for future allocations.
 577          */
 578         if (flex_size > 1) {
 579                 int retry = 0;
 580
 581         try_again:
 582                 parent_group &= ~(flex_size-1);
 583                 last = parent_group + flex_size;
 584                 if (last > ngroups)
 585                         last = ngroups;
 586                 for  (i = parent_group; i < last; i++) {
 587                         desc = ext4_get_group_desc(sb, i, NULL);
 588                         if (desc && ext4_free_inodes_count(sb, desc)) {
 589                                 *group = i;
 590                                 return 0;
 591                         }
 592                 }
 593                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 594                         retry = 1;
 595                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 596                         goto try_again;
 597                 }
 598                 /*
 599                  * If this didn't work, use the Orlov search algorithm
 600                  * to find a new flex group; we pass in the mode to
 601                  * avoid the topdir algorithms.
 602                  */
 603                 *group = parent_group + flex_size;
 604                 if (*group > ngroups)
 605                         *group = 0;
 606                 return find_group_orlov(sb, parent, group, mode, NULL);
 607         }
 608
 609         /*
 610          * Try to place the inode in its parent directory
 611          */
 612         *group = parent_group;
 613         desc = ext4_get_group_desc(sb, *group, NULL);
 614         if (desc && ext4_free_inodes_count(sb, desc) &&
 615             ext4_free_group_clusters(sb, desc))
 616                 return 0;
 617
 618         /*
 619          * We're going to place this inode in a different blockgroup from its
 620          * parent.  We want to cause files in a common directory to all land in
 621          * the same blockgroup.  But we want files which are in a different
 622          * directory which shares a blockgroup with our parent to land in a
 623          * different blockgroup.
 624          *
 625          * So add our directory's i_ino into the starting point for the hash.
 626          */
 627         *group = (*group + parent->i_ino) % ngroups;
 628
 629         /*
 630          * Use a quadratic hash to find a group with a free inode and some free
 631          * blocks.
 632          */
 633         for (i = 1; i < ngroups; i <<= 1) {
 634                 *group += i;
 635                 if (*group >= ngroups)
 636                         *group -= ngroups;
 637                 desc = ext4_get_group_desc(sb, *group, NULL);
 638                 if (desc && ext4_free_inodes_count(sb, desc) &&
 639                     ext4_free_group_clusters(sb, desc))
 640                         return 0;
 641         }
 642
 643         /*
 644          * That failed: try linear search for a free inode, even if that group
 645          * has no free blocks.
 646          */
 647         *group = parent_group;
 648         for (i = 0; i < ngroups; i++) {
 649                 if (++*group >= ngroups)
 650                         *group = 0;
 651                 desc = ext4_get_group_desc(sb, *group, NULL);
 652                 if (desc && ext4_free_inodes_count(sb, desc))
 653                         return 0;
 654         }
 655
 656         return -1;
 657 }
 658
 659 /*
 660  * In no journal mode, if an inode has recently been deleted, we want
 661  * to avoid reusing it until we're reasonably sure the inode table
 662  * block has been written back to disk.  (Yes, these values are
 663  * somewhat arbitrary...)
 664  */
 665 #define RECENTCY_MIN    60
 666 #define RECENTCY_DIRTY  300
 667
 668 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 669 {
 670         struct ext4_group_desc  *gdp;
 671         struct ext4_inode       *raw_inode;
 672         struct buffer_head      *bh;
 673         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 674         int offset, ret = 0;
 675         int recentcy = RECENTCY_MIN;
 676         u32 dtime, now;
 677
 678         gdp = ext4_get_group_desc(sb, group, NULL);
 679         if (unlikely(!gdp))
 680                 return 0;
 681
 682         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 683                        (ino / inodes_per_block));
 684         if (!bh || !buffer_uptodate(bh))
 685                 /*
 686                  * If the block is not in the buffer cache, then it
 687                  * must have been written out.
 688                  */
 689                 goto out;
 690
 691         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 692         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 693
 694         /* i_dtime is only 32 bits on disk, but we only care about relative
 695          * times in the range of a few minutes (i.e. long enough to sync a
 696          * recently-deleted inode to disk), so using the low 32 bits of the
 697          * clock (a 68 year range) is enough, see time_before32() */
 698         dtime = le32_to_cpu(raw_inode->i_dtime);
 699         now = ktime_get_real_seconds();
 700         if (buffer_dirty(bh))
 701                 recentcy += RECENTCY_DIRTY;
 702
 703         if (dtime && time_before32(dtime, now) &&
 704             time_before32(now, dtime + recentcy))
 705                 ret = 1;
 706 out:
 707         brelse(bh);
 708         return ret;
 709 }
 710
 711 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 712                           struct buffer_head *bitmap, unsigned long *ino)
 713 {
 714         bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL;
 715         unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb);
 716
 717 next:
 718         *ino = ext4_find_next_zero_bit((unsigned long *)
 719                                        bitmap->b_data,
 720                                        EXT4_INODES_PER_GROUP(sb), *ino);
 721         if (*ino >= EXT4_INODES_PER_GROUP(sb))
 722                 goto not_found;
 723
 724         if (check_recently_deleted && recently_deleted(sb, group, *ino)) {
 725                 recently_deleted_ino = *ino;
 726                 *ino = *ino + 1;
 727                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 728                         goto next;
 729                 goto not_found;
 730         }
 731         return 1;
 732 not_found:
 733         if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb))
 734                 return 0;
 735         /*
 736          * Not reusing recently deleted inodes is mostly a preference. We don't
 737          * want to report ENOSPC or skew allocation patterns because of that.
 738          * So return even recently deleted inode if we could find better in the
 739          * given range.
 740          */
 741         *ino = recently_deleted_ino;
 742         return 1;
 743 }
 744
 745 static int ext4_xattr_credits_for_new_inode(struct inode *dir, mode_t mode,
 746                                             bool encrypt)
 747 {
 748         struct super_block *sb = dir->i_sb;
 749         int nblocks = 0;
 750 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 751         struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 752
 753         if (IS_ERR(p))
 754                 return PTR_ERR(p);
 755         if (p) {
 756                 int acl_size = p->a_count * sizeof(ext4_acl_entry);
 757
 758                 nblocks += (S_ISDIR(mode) ? 2 : 1) *
 759                         __ext4_xattr_set_credits(sb, NULL /* inode */,
 760                                                  NULL /* block_bh */, acl_size,
 761                                                  true /* is_create */);
 762                 posix_acl_release(p);
 763         }
 764 #endif
 765
 766 #ifdef CONFIG_SECURITY
 767         {
 768                 int num_security_xattrs = 1;
 769
 770 #ifdef CONFIG_INTEGRITY
 771                 num_security_xattrs++;
 772 #endif
 773                 /*
 774                  * We assume that security xattrs are never more than 1k.
 775                  * In practice they are under 128 bytes.
 776                  */
 777                 nblocks += num_security_xattrs *
 778                         __ext4_xattr_set_credits(sb, NULL /* inode */,
 779                                                  NULL /* block_bh */, 1024,
 780                                                  true /* is_create */);
 781         }
 782 #endif
 783         if (encrypt)
 784                 nblocks += __ext4_xattr_set_credits(sb,
 785                                                     NULL /* inode */,
 786                                                     NULL /* block_bh */,
 787                                                     FSCRYPT_SET_CONTEXT_MAX_SIZE,
 788                                                     true /* is_create */);
 789         return nblocks;
 790 }
 791
 792 /*
 793  * There are two policies for allocating an inode.  If the new inode is
 794  * a directory, then a forward search is made for a block group with both
 795  * free space and a low directory-to-inode ratio; if that fails, then of
 796  * the groups with above-average free space, that group with the fewest
 797  * directories already is chosen.
 798  *
 799  * For other inodes, search forward from the parent directory's block
 800  * group to find a free inode.
 801  */
 802 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 803                                umode_t mode, const struct qstr *qstr,
 804                                __u32 goal, uid_t *owner, __u32 i_flags,
 805                                int handle_type, unsigned int line_no,
 806                                int nblocks)
 807 {
 808         struct super_block *sb;
 809         struct buffer_head *inode_bitmap_bh = NULL;
 810         struct buffer_head *group_desc_bh;
 811         ext4_group_t ngroups, group = 0;
 812         unsigned long ino = 0;
 813         struct inode *inode;
 814         struct ext4_group_desc *gdp = NULL;
 815         struct ext4_inode_info *ei;
 816         struct ext4_sb_info *sbi;
 817         int ret2, err;
 818         struct inode *ret;
 819         ext4_group_t i;
 820         ext4_group_t flex_group;
 821         struct ext4_group_info *grp;
 822         bool encrypt = false;
 823
 824         /* Cannot create files in a deleted directory */
 825         if (!dir || !dir->i_nlink)
 826                 return ERR_PTR(-EPERM);
 827
 828         sb = dir->i_sb;
 829         sbi = EXT4_SB(sb);
 830
 831         if (unlikely(ext4_forced_shutdown(sbi)))
 832                 return ERR_PTR(-EIO);
 833
 834         ngroups = ext4_get_groups_count(sb);
 835         trace_ext4_request_inode(dir, mode);
 836         inode = new_inode(sb);
 837         if (!inode)
 838                 return ERR_PTR(-ENOMEM);
 839         ei = EXT4_I(inode);
 840
 841         /*
 842          * Initialize owners and quota early so that we don't have to account
 843          * for quota initialization worst case in standard inode creating
 844          * transaction
 845          */
 846         if (owner) {
 847                 inode->i_mode = mode;
 848                 i_uid_write(inode, owner[0]);
 849                 i_gid_write(inode, owner[1]);
 850         } else if (test_opt(sb, GRPID)) {
 851                 inode->i_mode = mode;
 852                 inode->i_uid = current_fsuid();
 853                 inode->i_gid = dir->i_gid;
 854         } else
 855                 inode_init_owner(inode, dir, mode);
 856
 857         if (ext4_has_feature_project(sb) &&
 858             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 859                 ei->i_projid = EXT4_I(dir)->i_projid;
 860         else
 861                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 862
 863         if (!(i_flags & EXT4_EA_INODE_FL)) {
 864                 err = fscrypt_prepare_new_inode(dir, inode, &encrypt);
 865                 if (err)
 866                         goto out;
 867         }
 868
 869         err = dquot_initialize(inode);
 870         if (err)
 871                 goto out;
 872
 873         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 874                 ret2 = ext4_xattr_credits_for_new_inode(dir, mode, encrypt);
 875                 if (ret2 < 0) {
 876                         err = ret2;
 877                         goto out;
 878                 }
 879                 nblocks += ret2;
 880         }
 881
 882         if (!goal)
 883                 goal = sbi->s_inode_goal;
 884
 885         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 886                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 887                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 888                 ret2 = 0;
 889                 goto got_group;
 890         }
 891
 892         if (S_ISDIR(mode))
 893                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 894         else
 895                 ret2 = find_group_other(sb, dir, &group, mode);
 896
 897 got_group:
 898         EXT4_I(dir)->i_last_alloc_group = group;
 899         err = -ENOSPC;
 900         if (ret2 == -1)
 901                 goto out;
 902
 903         /*
 904          * Normally we will only go through one pass of this loop,
 905          * unless we get unlucky and it turns out the group we selected
 906          * had its last inode grabbed by someone else.
 907          */
 908         for (i = 0; i < ngroups; i++, ino = 0) {
 909                 err = -EIO;
 910
 911                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 912                 if (!gdp)
 913                         goto out;
 914
 915                 /*
 916                  * Check free inodes count before loading bitmap.
 917                  */
 918                 if (ext4_free_inodes_count(sb, gdp) == 0)
 919                         goto next_group;
 920
 921                 grp = ext4_get_group_info(sb, group);
 922                 /* Skip groups with already-known suspicious inode tables */
 923                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 924                         goto next_group;
 925
 926                 brelse(inode_bitmap_bh);
 927                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 928                 /* Skip groups with suspicious inode tables */
 929                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 930                     IS_ERR(inode_bitmap_bh)) {
 931                         inode_bitmap_bh = NULL;
 932                         goto next_group;
 933                 }
 934
 935 repeat_in_this_group:
 936                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 937                 if (!ret2)
 938                         goto next_group;
 939
 940                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 941                         ext4_error(sb, "reserved inode found cleared - "
 942                                    "inode=%lu", ino + 1);
 943                         ext4_mark_group_bitmap_corrupted(sb, group,
 944                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 945                         goto next_group;
 946                 }
 947
 948                 if (!handle) {
 949                         BUG_ON(nblocks <= 0);
 950                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 951                                  handle_type, nblocks, 0,
 952                                  ext4_trans_default_revoke_credits(sb));
 953                         if (IS_ERR(handle)) {
 954                                 err = PTR_ERR(handle);
 955                                 ext4_std_error(sb, err);
 956                                 goto out;
 957                         }
 958                 }
 959                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 960                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 961                 if (err) {
 962                         ext4_std_error(sb, err);
 963                         goto out;
 964                 }
 965                 ext4_lock_group(sb, group);
 966                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 967                 if (ret2) {
 968                         /* Someone already took the bit. Repeat the search
 969                          * with lock held.
 970                          */
 971                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 972                         if (ret2) {
 973                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
 974                                 ret2 = 0;
 975                         } else {
 976                                 ret2 = 1; /* we didn't grab the inode */
 977                         }
 978                 }
 979                 ext4_unlock_group(sb, group);
 980                 ino++;          /* the inode bitmap is zero-based */
 981                 if (!ret2)
 982                         goto got; /* we grabbed the inode! */
 983
 984                 if (ino < EXT4_INODES_PER_GROUP(sb))
 985                         goto repeat_in_this_group;
 986 next_group:
 987                 if (++group == ngroups)
 988                         group = 0;
 989         }
 990         err = -ENOSPC;
 991         goto out;
 992
 993 got:
 994         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 995         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 996         if (err) {
 997                 ext4_std_error(sb, err);
 998                 goto out;
 999         }
1000
1001         BUFFER_TRACE(group_desc_bh, "get_write_access");
1002         err = ext4_journal_get_write_access(handle, group_desc_bh);
1003         if (err) {
1004                 ext4_std_error(sb, err);
1005                 goto out;
1006         }
1007
1008         /* We may have to initialize the block bitmap if it isn't already */
1009         if (ext4_has_group_desc_csum(sb) &&
1010             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1011                 struct buffer_head *block_bitmap_bh;
1012
1013                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1014                 if (IS_ERR(block_bitmap_bh)) {
1015                         err = PTR_ERR(block_bitmap_bh);
1016                         goto out;
1017                 }
1018                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1019                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1020                 if (err) {
1021                         brelse(block_bitmap_bh);
1022                         ext4_std_error(sb, err);
1023                         goto out;
1024                 }
1025
1026                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1027                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1028
1029                 /* recheck and clear flag under lock if we still need to */
1030                 ext4_lock_group(sb, group);
1031                 if (ext4_has_group_desc_csum(sb) &&
1032                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1033                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1034                         ext4_free_group_clusters_set(sb, gdp,
1035                                 ext4_free_clusters_after_init(sb, group, gdp));
1036                         ext4_block_bitmap_csum_set(sb, group, gdp,
1037                                                    block_bitmap_bh);
1038                         ext4_group_desc_csum_set(sb, group, gdp);
1039                 }
1040                 ext4_unlock_group(sb, group);
1041                 brelse(block_bitmap_bh);
1042
1043                 if (err) {
1044                         ext4_std_error(sb, err);
1045                         goto out;
1046                 }
1047         }
1048
1049         /* Update the relevant bg descriptor fields */
1050         if (ext4_has_group_desc_csum(sb)) {
1051                 int free;
1052                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1053
1054                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1055                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1056                 free = EXT4_INODES_PER_GROUP(sb) -
1057                         ext4_itable_unused_count(sb, gdp);
1058                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1059                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1060                         free = 0;
1061                 }
1062                 /*
1063                  * Check the relative inode number against the last used
1064                  * relative inode number in this group. if it is greater
1065                  * we need to update the bg_itable_unused count
1066                  */
1067                 if (ino > free)
1068                         ext4_itable_unused_set(sb, gdp,
1069                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1070                 up_read(&grp->alloc_sem);
1071         } else {
1072                 ext4_lock_group(sb, group);
1073         }
1074
1075         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1076         if (S_ISDIR(mode)) {
1077                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1078                 if (sbi->s_log_groups_per_flex) {
1079                         ext4_group_t f = ext4_flex_group(sbi, group);
1080
1081                         atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1082                                                         f)->used_dirs);
1083                 }
1084         }
1085         if (ext4_has_group_desc_csum(sb)) {
1086                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1087                                            EXT4_INODES_PER_GROUP(sb) / 8);
1088                 ext4_group_desc_csum_set(sb, group, gdp);
1089         }
1090         ext4_unlock_group(sb, group);
1091
1092         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1093         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1094         if (err) {
1095                 ext4_std_error(sb, err);
1096                 goto out;
1097         }
1098
1099         percpu_counter_dec(&sbi->s_freeinodes_counter);
1100         if (S_ISDIR(mode))
1101                 percpu_counter_inc(&sbi->s_dirs_counter);
1102
1103         if (sbi->s_log_groups_per_flex) {
1104                 flex_group = ext4_flex_group(sbi, group);
1105                 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1106                                                 flex_group)->free_inodes);
1107         }
1108
1109         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1110         /* This is the optimal IO size (for stat), not the fs block size */
1111         inode->i_blocks = 0;
1112         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1113         ei->i_crtime = inode->i_mtime;
1114
1115         memset(ei->i_data, 0, sizeof(ei->i_data));
1116         ei->i_dir_start_lookup = 0;
1117         ei->i_disksize = 0;
1118
1119         /* Don't inherit extent flag from directory, amongst others. */
1120         ei->i_flags =
1121                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1122         ei->i_flags |= i_flags;
1123         ei->i_file_acl = 0;
1124         ei->i_dtime = 0;
1125         ei->i_block_group = group;
1126         ei->i_last_alloc_group = ~0;
1127
1128         ext4_set_inode_flags(inode, true);
1129         if (IS_DIRSYNC(inode))
1130                 ext4_handle_sync(handle);
1131         if (insert_inode_locked(inode) < 0) {
1132                 /*
1133                  * Likely a bitmap corruption causing inode to be allocated
1134                  * twice.
1135                  */
1136                 err = -EIO;
1137                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1138                            inode->i_ino);
1139                 ext4_mark_group_bitmap_corrupted(sb, group,
1140                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1141                 goto out;
1142         }
1143         inode->i_generation = prandom_u32();
1144
1145         /* Precompute checksum seed for inode metadata */
1146         if (ext4_has_metadata_csum(sb)) {
1147                 __u32 csum;
1148                 __le32 inum = cpu_to_le32(inode->i_ino);
1149                 __le32 gen = cpu_to_le32(inode->i_generation);
1150                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1151                                    sizeof(inum));
1152                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1153                                               sizeof(gen));
1154         }
1155
1156         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1157         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1158
1159         ei->i_extra_isize = sbi->s_want_extra_isize;
1160         ei->i_inline_off = 0;
1161         if (ext4_has_feature_inline_data(sb))
1162                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1163         ret = inode;
1164         err = dquot_alloc_inode(inode);
1165         if (err)
1166                 goto fail_drop;
1167
1168         /*
1169          * Since the encryption xattr will always be unique, create it first so
1170          * that it's less likely to end up in an external xattr block and
1171          * prevent its deduplication.
1172          */
1173         if (encrypt) {
1174                 err = fscrypt_set_context(inode, handle);
1175                 if (err)
1176                         goto fail_free_drop;
1177         }
1178
1179         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1180                 err = ext4_init_acl(handle, inode, dir);
1181                 if (err)
1182                         goto fail_free_drop;
1183
1184                 err = ext4_init_security(handle, inode, dir, qstr);
1185                 if (err)
1186                         goto fail_free_drop;
1187         }
1188
1189         if (ext4_has_feature_extents(sb)) {
1190                 /* set extent flag only for directory, file and normal symlink*/
1191                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1192                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1193                         ext4_ext_tree_init(handle, inode);
1194                 }
1195         }
1196
1197         if (ext4_handle_valid(handle)) {
1198                 ei->i_sync_tid = handle->h_transaction->t_tid;
1199                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1200         }
1201
1202         err = ext4_mark_inode_dirty(handle, inode);
1203         if (err) {
1204                 ext4_std_error(sb, err);
1205                 goto fail_free_drop;
1206         }
1207
1208         ext4_debug("allocating inode %lu\n", inode->i_ino);
1209         trace_ext4_allocate_inode(inode, dir, mode);
1210         brelse(inode_bitmap_bh);
1211         return ret;
1212
1213 fail_free_drop:
1214         dquot_free_inode(inode);
1215 fail_drop:
1216         clear_nlink(inode);
1217         unlock_new_inode(inode);
1218 out:
1219         dquot_drop(inode);
1220         inode->i_flags |= S_NOQUOTA;
1221         iput(inode);
1222         brelse(inode_bitmap_bh);
1223         return ERR_PTR(err);
1224 }
1225
1226 /* Verify that we are loading a valid orphan from disk */
1227 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1228 {
1229         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1230         ext4_group_t block_group;
1231         int bit;
1232         struct buffer_head *bitmap_bh = NULL;
1233         struct inode *inode = NULL;
1234         int err = -EFSCORRUPTED;
1235
1236         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1237                 goto bad_orphan;
1238
1239         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1240         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1241         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1242         if (IS_ERR(bitmap_bh))
1243                 return ERR_CAST(bitmap_bh);
1244
1245         /* Having the inode bit set should be a 100% indicator that this
1246          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1247          * inodes that were being truncated, so we can't check i_nlink==0.
1248          */
1249         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1250                 goto bad_orphan;
1251
1252         inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1253         if (IS_ERR(inode)) {
1254                 err = PTR_ERR(inode);
1255                 ext4_error_err(sb, -err,
1256                                "couldn't read orphan inode %lu (err %d)",
1257                                ino, err);
1258                 brelse(bitmap_bh);
1259                 return inode;
1260         }
1261
1262         /*
1263          * If the orphans has i_nlinks > 0 then it should be able to
1264          * be truncated, otherwise it won't be removed from the orphan
1265          * list during processing and an infinite loop will result.
1266          * Similarly, it must not be a bad inode.
1267          */
1268         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1269             is_bad_inode(inode))
1270                 goto bad_orphan;
1271
1272         if (NEXT_ORPHAN(inode) > max_ino)
1273                 goto bad_orphan;
1274         brelse(bitmap_bh);
1275         return inode;
1276
1277 bad_orphan:
1278         ext4_error(sb, "bad orphan inode %lu", ino);
1279         if (bitmap_bh)
1280                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1281                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1282                        ext4_test_bit(bit, bitmap_bh->b_data));
1283         if (inode) {
1284                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1285                        is_bad_inode(inode));
1286                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1287                        NEXT_ORPHAN(inode));
1288                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1289                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1290                 /* Avoid freeing blocks if we got a bad deleted inode */
1291                 if (inode->i_nlink == 0)
1292                         inode->i_blocks = 0;
1293                 iput(inode);
1294         }
1295         brelse(bitmap_bh);
1296         return ERR_PTR(err);
1297 }
1298
1299 unsigned long ext4_count_free_inodes(struct super_block *sb)
1300 {
1301         unsigned long desc_count;
1302         struct ext4_group_desc *gdp;
1303         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1304 #ifdef EXT4FS_DEBUG
1305         struct ext4_super_block *es;
1306         unsigned long bitmap_count, x;
1307         struct buffer_head *bitmap_bh = NULL;
1308
1309         es = EXT4_SB(sb)->s_es;
1310         desc_count = 0;
1311         bitmap_count = 0;
1312         gdp = NULL;
1313         for (i = 0; i < ngroups; i++) {
1314                 gdp = ext4_get_group_desc(sb, i, NULL);
1315                 if (!gdp)
1316                         continue;
1317                 desc_count += ext4_free_inodes_count(sb, gdp);
1318                 brelse(bitmap_bh);
1319                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1320                 if (IS_ERR(bitmap_bh)) {
1321                         bitmap_bh = NULL;
1322                         continue;
1323                 }
1324
1325                 x = ext4_count_free(bitmap_bh->b_data,
1326                                     EXT4_INODES_PER_GROUP(sb) / 8);
1327                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1328                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1329                 bitmap_count += x;
1330         }
1331         brelse(bitmap_bh);
1332         printk(KERN_DEBUG "ext4_count_free_inodes: "
1333                "stored = %u, computed = %lu, %lu\n",
1334                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1335         return desc_count;
1336 #else
1337         desc_count = 0;
1338         for (i = 0; i < ngroups; i++) {
1339                 gdp = ext4_get_group_desc(sb, i, NULL);
1340                 if (!gdp)
1341                         continue;
1342                 desc_count += ext4_free_inodes_count(sb, gdp);
1343                 cond_resched();
1344         }
1345         return desc_count;
1346 #endif
1347 }
1348
1349 /* Called at mount-time, super-block is locked */
1350 unsigned long ext4_count_dirs(struct super_block * sb)
1351 {
1352         unsigned long count = 0;
1353         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1354
1355         for (i = 0; i < ngroups; i++) {
1356                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1357                 if (!gdp)
1358                         continue;
1359                 count += ext4_used_dirs_count(sb, gdp);
1360         }
1361         return count;
1362 }
1363
1364 /*
1365  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1366  * inode table. Must be called without any spinlock held. The only place
1367  * where it is called from on active part of filesystem is ext4lazyinit
1368  * thread, so we do not need any special locks, however we have to prevent
1369  * inode allocation from the current group, so we take alloc_sem lock, to
1370  * block ext4_new_inode() until we are finished.
1371  */
1372 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1373                                  int barrier)
1374 {
1375         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1376         struct ext4_sb_info *sbi = EXT4_SB(sb);
1377         struct ext4_group_desc *gdp = NULL;
1378         struct buffer_head *group_desc_bh;
1379         handle_t *handle;
1380         ext4_fsblk_t blk;
1381         int num, ret = 0, used_blks = 0;
1382
1383         /* This should not happen, but just to be sure check this */
1384         if (sb_rdonly(sb)) {
1385                 ret = 1;
1386                 goto out;
1387         }
1388
1389         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1390         if (!gdp)
1391                 goto out;
1392
1393         /*
1394          * We do not need to lock this, because we are the only one
1395          * handling this flag.
1396          */
1397         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1398                 goto out;
1399
1400         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1401         if (IS_ERR(handle)) {
1402                 ret = PTR_ERR(handle);
1403                 goto out;
1404         }
1405
1406         down_write(&grp->alloc_sem);
1407         /*
1408          * If inode bitmap was already initialized there may be some
1409          * used inodes so we need to skip blocks with used inodes in
1410          * inode table.
1411          */
1412         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1413                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1414                             ext4_itable_unused_count(sb, gdp)),
1415                             sbi->s_inodes_per_block);
1416
1417         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1418             ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1419                                ext4_itable_unused_count(sb, gdp)) <
1420                               EXT4_FIRST_INO(sb)))) {
1421                 ext4_error(sb, "Something is wrong with group %u: "
1422                            "used itable blocks: %d; "
1423                            "itable unused count: %u",
1424                            group, used_blks,
1425                            ext4_itable_unused_count(sb, gdp));
1426                 ret = 1;
1427                 goto err_out;
1428         }
1429
1430         blk = ext4_inode_table(sb, gdp) + used_blks;
1431         num = sbi->s_itb_per_group - used_blks;
1432
1433         BUFFER_TRACE(group_desc_bh, "get_write_access");
1434         ret = ext4_journal_get_write_access(handle,
1435                                             group_desc_bh);
1436         if (ret)
1437                 goto err_out;
1438
1439         /*
1440          * Skip zeroout if the inode table is full. But we set the ZEROED
1441          * flag anyway, because obviously, when it is full it does not need
1442          * further zeroing.
1443          */
1444         if (unlikely(num == 0))
1445                 goto skip_zeroout;
1446
1447         ext4_debug("going to zero out inode table in group %d\n",
1448                    group);
1449         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1450         if (ret < 0)
1451                 goto err_out;
1452         if (barrier)
1453                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS);
1454
1455 skip_zeroout:
1456         ext4_lock_group(sb, group);
1457         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1458         ext4_group_desc_csum_set(sb, group, gdp);
1459         ext4_unlock_group(sb, group);
1460
1461         BUFFER_TRACE(group_desc_bh,
1462                      "call ext4_handle_dirty_metadata");
1463         ret = ext4_handle_dirty_metadata(handle, NULL,
1464                                          group_desc_bh);
1465
1466 err_out:
1467         up_write(&grp->alloc_sem);
1468         ext4_journal_stop(handle);
1469 out:
1470         return ret;
1471 }