fs/ext4/ialloc.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/ext4/ialloc.c
   4  *
   5  * Copyright (C) 1992, 1993, 1994, 1995
   6  * Remy Card (card@masi.ibp.fr)
   7  * Laboratoire MASI - Institut Blaise Pascal
   8  * Universite Pierre et Marie Curie (Paris VI)
   9  *
  10  *  BSD ufs-inspired inode and directory allocation by
  11  *  Stephen Tweedie (sct@redhat.com), 1993
  12  *  Big-endian to little-endian byte-swapping/bitmaps by
  13  *        David S. Miller (davem@caip.rutgers.edu), 1995
  14  */
  15
  16 #include <linux/time.h>
  17 #include <linux/fs.h>
  18 #include <linux/stat.h>
  19 #include <linux/string.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/random.h>
  23 #include <linux/bitops.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/cred.h>
  26
  27 #include <asm/byteorder.h>
  28
  29 #include "ext4.h"
  30 #include "ext4_jbd2.h"
  31 #include "xattr.h"
  32 #include "acl.h"
  33
  34 #include <trace/events/ext4.h>
  35
  36 /*
  37  * ialloc.c contains the inodes allocation and deallocation routines
  38  */
  39
  40 /*
  41  * The free inodes are managed by bitmaps.  A file system contains several
  42  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  43  * block for inodes, N blocks for the inode table and data blocks.
  44  *
  45  * The file system contains group descriptors which are located after the
  46  * super block.  Each descriptor contains the number of the bitmap block and
  47  * the free blocks count in the block.
  48  */
  49
  50 /*
  51  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  52  * need to use it within a single byte (to ensure we get endianness right).
  53  * We can use memset for the rest of the bitmap as there are no other users.
  54  */
  55 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  56 {
  57         int i;
  58
  59         if (start_bit >= end_bit)
  60                 return;
  61
  62         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  63         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  64                 ext4_set_bit(i, bitmap);
  65         if (i < end_bit)
  66                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  67 }
  68
  69 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
  70 {
  71         if (uptodate) {
  72                 set_buffer_uptodate(bh);
  73                 set_bitmap_uptodate(bh);
  74         }
  75         unlock_buffer(bh);
  76         put_bh(bh);
  77 }
  78
  79 static int ext4_validate_inode_bitmap(struct super_block *sb,
  80                                       struct ext4_group_desc *desc,
  81                                       ext4_group_t block_group,
  82                                       struct buffer_head *bh)
  83 {
  84         ext4_fsblk_t    blk;
  85         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
  86
  87         if (buffer_verified(bh))
  88                 return 0;
  89         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
  90                 return -EFSCORRUPTED;
  91
  92         ext4_lock_group(sb, block_group);
  93         if (buffer_verified(bh))
  94                 goto verified;
  95         blk = ext4_inode_bitmap(sb, desc);
  96         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
  97                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
  98                 ext4_unlock_group(sb, block_group);
  99                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 100                            "inode_bitmap = %llu", block_group, blk);
 101                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 102                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 103                 return -EFSBADCRC;
 104         }
 105         set_buffer_verified(bh);
 106 verified:
 107         ext4_unlock_group(sb, block_group);
 108         return 0;
 109 }
 110
 111 /*
 112  * Read the inode allocation bitmap for a given block_group, reading
 113  * into the specified slot in the superblock's bitmap cache.
 114  *
 115  * Return buffer_head of bitmap on success or NULL.
 116  */
 117 static struct buffer_head *
 118 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 119 {
 120         struct ext4_group_desc *desc;
 121         struct ext4_sb_info *sbi = EXT4_SB(sb);
 122         struct buffer_head *bh = NULL;
 123         ext4_fsblk_t bitmap_blk;
 124         int err;
 125
 126         desc = ext4_get_group_desc(sb, block_group, NULL);
 127         if (!desc)
 128                 return ERR_PTR(-EFSCORRUPTED);
 129
 130         bitmap_blk = ext4_inode_bitmap(sb, desc);
 131         if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) ||
 132             (bitmap_blk >= ext4_blocks_count(sbi->s_es))) {
 133                 ext4_error(sb, "Invalid inode bitmap blk %llu in "
 134                            "block_group %u", bitmap_blk, block_group);
 135                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 136                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 137                 return ERR_PTR(-EFSCORRUPTED);
 138         }
 139         bh = sb_getblk(sb, bitmap_blk);
 140         if (unlikely(!bh)) {
 141                 ext4_warning(sb, "Cannot read inode bitmap - "
 142                              "block_group = %u, inode_bitmap = %llu",
 143                              block_group, bitmap_blk);
 144                 return ERR_PTR(-ENOMEM);
 145         }
 146         if (bitmap_uptodate(bh))
 147                 goto verify;
 148
 149         lock_buffer(bh);
 150         if (bitmap_uptodate(bh)) {
 151                 unlock_buffer(bh);
 152                 goto verify;
 153         }
 154
 155         ext4_lock_group(sb, block_group);
 156         if (ext4_has_group_desc_csum(sb) &&
 157             (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) {
 158                 if (block_group == 0) {
 159                         ext4_unlock_group(sb, block_group);
 160                         unlock_buffer(bh);
 161                         ext4_error(sb, "Inode bitmap for bg 0 marked "
 162                                    "uninitialized");
 163                         err = -EFSCORRUPTED;
 164                         goto out;
 165                 }
 166                 memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
 167                 ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb),
 168                                      sb->s_blocksize * 8, bh->b_data);
 169                 set_bitmap_uptodate(bh);
 170                 set_buffer_uptodate(bh);
 171                 set_buffer_verified(bh);
 172                 ext4_unlock_group(sb, block_group);
 173                 unlock_buffer(bh);
 174                 return bh;
 175         }
 176         ext4_unlock_group(sb, block_group);
 177
 178         if (buffer_uptodate(bh)) {
 179                 /*
 180                  * if not uninit if bh is uptodate,
 181                  * bitmap is also uptodate
 182                  */
 183                 set_bitmap_uptodate(bh);
 184                 unlock_buffer(bh);
 185                 goto verify;
 186         }
 187         /*
 188          * submit the buffer_head for reading
 189          */
 190         trace_ext4_load_inode_bitmap(sb, block_group);
 191         bh->b_end_io = ext4_end_bitmap_read;
 192         get_bh(bh);
 193         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 194         wait_on_buffer(bh);
 195         if (!buffer_uptodate(bh)) {
 196                 put_bh(bh);
 197                 ext4_error(sb, "Cannot read inode bitmap - "
 198                            "block_group = %u, inode_bitmap = %llu",
 199                            block_group, bitmap_blk);
 200                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 201                                 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 202                 return ERR_PTR(-EIO);
 203         }
 204
 205 verify:
 206         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 207         if (err)
 208                 goto out;
 209         return bh;
 210 out:
 211         put_bh(bh);
 212         return ERR_PTR(err);
 213 }
 214
 215 /*
 216  * NOTE! When we get the inode, we're the only people
 217  * that have access to it, and as such there are no
 218  * race conditions we have to worry about. The inode
 219  * is not on the hash-lists, and it cannot be reached
 220  * through the filesystem because the directory entry
 221  * has been deleted earlier.
 222  *
 223  * HOWEVER: we must make sure that we get no aliases,
 224  * which means that we have to call "clear_inode()"
 225  * _before_ we mark the inode not in use in the inode
 226  * bitmaps. Otherwise a newly created file might use
 227  * the same inode number (not actually the same pointer
 228  * though), and then we'd have two inodes sharing the
 229  * same inode number and space on the harddisk.
 230  */
 231 void ext4_free_inode(handle_t *handle, struct inode *inode)
 232 {
 233         struct super_block *sb = inode->i_sb;
 234         int is_directory;
 235         unsigned long ino;
 236         struct buffer_head *bitmap_bh = NULL;
 237         struct buffer_head *bh2;
 238         ext4_group_t block_group;
 239         unsigned long bit;
 240         struct ext4_group_desc *gdp;
 241         struct ext4_super_block *es;
 242         struct ext4_sb_info *sbi;
 243         int fatal = 0, err, count, cleared;
 244         struct ext4_group_info *grp;
 245
 246         if (!sb) {
 247                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 248                        "nonexistent device\n", __func__, __LINE__);
 249                 return;
 250         }
 251         if (atomic_read(&inode->i_count) > 1) {
 252                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 253                          __func__, __LINE__, inode->i_ino,
 254                          atomic_read(&inode->i_count));
 255                 return;
 256         }
 257         if (inode->i_nlink) {
 258                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 259                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 260                 return;
 261         }
 262         sbi = EXT4_SB(sb);
 263
 264         ino = inode->i_ino;
 265         ext4_debug("freeing inode %lu\n", ino);
 266         trace_ext4_free_inode(inode);
 267
 268         /*
 269          * Note: we must free any quota before locking the superblock,
 270          * as writing the quota to disk may need the lock as well.
 271          */
 272         dquot_initialize(inode);
 273         dquot_free_inode(inode);
 274         dquot_drop(inode);
 275
 276         is_directory = S_ISDIR(inode->i_mode);
 277
 278         /* Do this BEFORE marking the inode not in use or returning an error */
 279         ext4_clear_inode(inode);
 280
 281         es = sbi->s_es;
 282         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 283                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 284                 goto error_return;
 285         }
 286         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 287         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 288         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 289         /* Don't bother if the inode bitmap is corrupt. */
 290         grp = ext4_get_group_info(sb, block_group);
 291         if (IS_ERR(bitmap_bh)) {
 292                 fatal = PTR_ERR(bitmap_bh);
 293                 bitmap_bh = NULL;
 294                 goto error_return;
 295         }
 296         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 297                 fatal = -EFSCORRUPTED;
 298                 goto error_return;
 299         }
 300
 301         BUFFER_TRACE(bitmap_bh, "get_write_access");
 302         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 303         if (fatal)
 304                 goto error_return;
 305
 306         fatal = -ESRCH;
 307         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 308         if (gdp) {
 309                 BUFFER_TRACE(bh2, "get_write_access");
 310                 fatal = ext4_journal_get_write_access(handle, bh2);
 311         }
 312         ext4_lock_group(sb, block_group);
 313         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 314         if (fatal || !cleared) {
 315                 ext4_unlock_group(sb, block_group);
 316                 goto out;
 317         }
 318
 319         count = ext4_free_inodes_count(sb, gdp) + 1;
 320         ext4_free_inodes_set(sb, gdp, count);
 321         if (is_directory) {
 322                 count = ext4_used_dirs_count(sb, gdp) - 1;
 323                 ext4_used_dirs_set(sb, gdp, count);
 324                 percpu_counter_dec(&sbi->s_dirs_counter);
 325         }
 326         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 327                                    EXT4_INODES_PER_GROUP(sb) / 8);
 328         ext4_group_desc_csum_set(sb, block_group, gdp);
 329         ext4_unlock_group(sb, block_group);
 330
 331         percpu_counter_inc(&sbi->s_freeinodes_counter);
 332         if (sbi->s_log_groups_per_flex) {
 333                 struct flex_groups *fg;
 334
 335                 fg = sbi_array_rcu_deref(sbi, s_flex_groups,
 336                                          ext4_flex_group(sbi, block_group));
 337                 atomic_inc(&fg->free_inodes);
 338                 if (is_directory)
 339                         atomic_dec(&fg->used_dirs);
 340         }
 341         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 342         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 343 out:
 344         if (cleared) {
 345                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 346                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 347                 if (!fatal)
 348                         fatal = err;
 349         } else {
 350                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 351                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 352                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 353         }
 354
 355 error_return:
 356         brelse(bitmap_bh);
 357         ext4_std_error(sb, fatal);
 358 }
 359
 360 struct orlov_stats {
 361         __u64 free_clusters;
 362         __u32 free_inodes;
 363         __u32 used_dirs;
 364 };
 365
 366 /*
 367  * Helper function for Orlov's allocator; returns critical information
 368  * for a particular block group or flex_bg.  If flex_size is 1, then g
 369  * is a block group number; otherwise it is flex_bg number.
 370  */
 371 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 372                             int flex_size, struct orlov_stats *stats)
 373 {
 374         struct ext4_group_desc *desc;
 375
 376         if (flex_size > 1) {
 377                 struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb),
 378                                                              s_flex_groups, g);
 379                 stats->free_inodes = atomic_read(&fg->free_inodes);
 380                 stats->free_clusters = atomic64_read(&fg->free_clusters);
 381                 stats->used_dirs = atomic_read(&fg->used_dirs);
 382                 return;
 383         }
 384
 385         desc = ext4_get_group_desc(sb, g, NULL);
 386         if (desc) {
 387                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 388                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 389                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 390         } else {
 391                 stats->free_inodes = 0;
 392                 stats->free_clusters = 0;
 393                 stats->used_dirs = 0;
 394         }
 395 }
 396
 397 /*
 398  * Orlov's allocator for directories.
 399  *
 400  * We always try to spread first-level directories.
 401  *
 402  * If there are blockgroups with both free inodes and free blocks counts
 403  * not worse than average we return one with smallest directory count.
 404  * Otherwise we simply return a random group.
 405  *
 406  * For the rest rules look so:
 407  *
 408  * It's OK to put directory into a group unless
 409  * it has too many directories already (max_dirs) or
 410  * it has too few free inodes left (min_inodes) or
 411  * it has too few free blocks left (min_blocks) or
 412  * Parent's group is preferred, if it doesn't satisfy these
 413  * conditions we search cyclically through the rest. If none
 414  * of the groups look good we just look for a group with more
 415  * free inodes than average (starting at parent's group).
 416  */
 417
 418 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 419                             ext4_group_t *group, umode_t mode,
 420                             const struct qstr *qstr)
 421 {
 422         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 423         struct ext4_sb_info *sbi = EXT4_SB(sb);
 424         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 425         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 426         unsigned int freei, avefreei, grp_free;
 427         ext4_fsblk_t freeb, avefreec;
 428         unsigned int ndirs;
 429         int max_dirs, min_inodes;
 430         ext4_grpblk_t min_clusters;
 431         ext4_group_t i, grp, g, ngroups;
 432         struct ext4_group_desc *desc;
 433         struct orlov_stats stats;
 434         int flex_size = ext4_flex_bg_size(sbi);
 435         struct dx_hash_info hinfo;
 436
 437         ngroups = real_ngroups;
 438         if (flex_size > 1) {
 439                 ngroups = (real_ngroups + flex_size - 1) >>
 440                         sbi->s_log_groups_per_flex;
 441                 parent_group >>= sbi->s_log_groups_per_flex;
 442         }
 443
 444         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 445         avefreei = freei / ngroups;
 446         freeb = EXT4_C2B(sbi,
 447                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 448         avefreec = freeb;
 449         do_div(avefreec, ngroups);
 450         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 451
 452         if (S_ISDIR(mode) &&
 453             ((parent == d_inode(sb->s_root)) ||
 454              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 455                 int best_ndir = inodes_per_group;
 456                 int ret = -1;
 457
 458                 if (qstr) {
 459                         hinfo.hash_version = DX_HASH_HALF_MD4;
 460                         hinfo.seed = sbi->s_hash_seed;
 461                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 462                         grp = hinfo.hash;
 463                 } else
 464                         grp = prandom_u32();
 465                 parent_group = (unsigned)grp % ngroups;
 466                 for (i = 0; i < ngroups; i++) {
 467                         g = (parent_group + i) % ngroups;
 468                         get_orlov_stats(sb, g, flex_size, &stats);
 469                         if (!stats.free_inodes)
 470                                 continue;
 471                         if (stats.used_dirs >= best_ndir)
 472                                 continue;
 473                         if (stats.free_inodes < avefreei)
 474                                 continue;
 475                         if (stats.free_clusters < avefreec)
 476                                 continue;
 477                         grp = g;
 478                         ret = 0;
 479                         best_ndir = stats.used_dirs;
 480                 }
 481                 if (ret)
 482                         goto fallback;
 483         found_flex_bg:
 484                 if (flex_size == 1) {
 485                         *group = grp;
 486                         return 0;
 487                 }
 488
 489                 /*
 490                  * We pack inodes at the beginning of the flexgroup's
 491                  * inode tables.  Block allocation decisions will do
 492                  * something similar, although regular files will
 493                  * start at 2nd block group of the flexgroup.  See
 494                  * ext4_ext_find_goal() and ext4_find_near().
 495                  */
 496                 grp *= flex_size;
 497                 for (i = 0; i < flex_size; i++) {
 498                         if (grp+i >= real_ngroups)
 499                                 break;
 500                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 501                         if (desc && ext4_free_inodes_count(sb, desc)) {
 502                                 *group = grp+i;
 503                                 return 0;
 504                         }
 505                 }
 506                 goto fallback;
 507         }
 508
 509         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 510         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 511         if (min_inodes < 1)
 512                 min_inodes = 1;
 513         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 514
 515         /*
 516          * Start looking in the flex group where we last allocated an
 517          * inode for this parent directory
 518          */
 519         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 520                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 521                 if (flex_size > 1)
 522                         parent_group >>= sbi->s_log_groups_per_flex;
 523         }
 524
 525         for (i = 0; i < ngroups; i++) {
 526                 grp = (parent_group + i) % ngroups;
 527                 get_orlov_stats(sb, grp, flex_size, &stats);
 528                 if (stats.used_dirs >= max_dirs)
 529                         continue;
 530                 if (stats.free_inodes < min_inodes)
 531                         continue;
 532                 if (stats.free_clusters < min_clusters)
 533                         continue;
 534                 goto found_flex_bg;
 535         }
 536
 537 fallback:
 538         ngroups = real_ngroups;
 539         avefreei = freei / ngroups;
 540 fallback_retry:
 541         parent_group = EXT4_I(parent)->i_block_group;
 542         for (i = 0; i < ngroups; i++) {
 543                 grp = (parent_group + i) % ngroups;
 544                 desc = ext4_get_group_desc(sb, grp, NULL);
 545                 if (desc) {
 546                         grp_free = ext4_free_inodes_count(sb, desc);
 547                         if (grp_free && grp_free >= avefreei) {
 548                                 *group = grp;
 549                                 return 0;
 550                         }
 551                 }
 552         }
 553
 554         if (avefreei) {
 555                 /*
 556                  * The free-inodes counter is approximate, and for really small
 557                  * filesystems the above test can fail to find any blockgroups
 558                  */
 559                 avefreei = 0;
 560                 goto fallback_retry;
 561         }
 562
 563         return -1;
 564 }
 565
 566 static int find_group_other(struct super_block *sb, struct inode *parent,
 567                             ext4_group_t *group, umode_t mode)
 568 {
 569         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 570         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 571         struct ext4_group_desc *desc;
 572         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 573
 574         /*
 575          * Try to place the inode is the same flex group as its
 576          * parent.  If we can't find space, use the Orlov algorithm to
 577          * find another flex group, and store that information in the
 578          * parent directory's inode information so that use that flex
 579          * group for future allocations.
 580          */
 581         if (flex_size > 1) {
 582                 int retry = 0;
 583
 584         try_again:
 585                 parent_group &= ~(flex_size-1);
 586                 last = parent_group + flex_size;
 587                 if (last > ngroups)
 588                         last = ngroups;
 589                 for  (i = parent_group; i < last; i++) {
 590                         desc = ext4_get_group_desc(sb, i, NULL);
 591                         if (desc && ext4_free_inodes_count(sb, desc)) {
 592                                 *group = i;
 593                                 return 0;
 594                         }
 595                 }
 596                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 597                         retry = 1;
 598                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 599                         goto try_again;
 600                 }
 601                 /*
 602                  * If this didn't work, use the Orlov search algorithm
 603                  * to find a new flex group; we pass in the mode to
 604                  * avoid the topdir algorithms.
 605                  */
 606                 *group = parent_group + flex_size;
 607                 if (*group > ngroups)
 608                         *group = 0;
 609                 return find_group_orlov(sb, parent, group, mode, NULL);
 610         }
 611
 612         /*
 613          * Try to place the inode in its parent directory
 614          */
 615         *group = parent_group;
 616         desc = ext4_get_group_desc(sb, *group, NULL);
 617         if (desc && ext4_free_inodes_count(sb, desc) &&
 618             ext4_free_group_clusters(sb, desc))
 619                 return 0;
 620
 621         /*
 622          * We're going to place this inode in a different blockgroup from its
 623          * parent.  We want to cause files in a common directory to all land in
 624          * the same blockgroup.  But we want files which are in a different
 625          * directory which shares a blockgroup with our parent to land in a
 626          * different blockgroup.
 627          *
 628          * So add our directory's i_ino into the starting point for the hash.
 629          */
 630         *group = (*group + parent->i_ino) % ngroups;
 631
 632         /*
 633          * Use a quadratic hash to find a group with a free inode and some free
 634          * blocks.
 635          */
 636         for (i = 1; i < ngroups; i <<= 1) {
 637                 *group += i;
 638                 if (*group >= ngroups)
 639                         *group -= ngroups;
 640                 desc = ext4_get_group_desc(sb, *group, NULL);
 641                 if (desc && ext4_free_inodes_count(sb, desc) &&
 642                     ext4_free_group_clusters(sb, desc))
 643                         return 0;
 644         }
 645
 646         /*
 647          * That failed: try linear search for a free inode, even if that group
 648          * has no free blocks.
 649          */
 650         *group = parent_group;
 651         for (i = 0; i < ngroups; i++) {
 652                 if (++*group >= ngroups)
 653                         *group = 0;
 654                 desc = ext4_get_group_desc(sb, *group, NULL);
 655                 if (desc && ext4_free_inodes_count(sb, desc))
 656                         return 0;
 657         }
 658
 659         return -1;
 660 }
 661
 662 /*
 663  * In no journal mode, if an inode has recently been deleted, we want
 664  * to avoid reusing it until we're reasonably sure the inode table
 665  * block has been written back to disk.  (Yes, these values are
 666  * somewhat arbitrary...)
 667  */
 668 #define RECENTCY_MIN    60
 669 #define RECENTCY_DIRTY  300
 670
 671 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 672 {
 673         struct ext4_group_desc  *gdp;
 674         struct ext4_inode       *raw_inode;
 675         struct buffer_head      *bh;
 676         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 677         int offset, ret = 0;
 678         int recentcy = RECENTCY_MIN;
 679         u32 dtime, now;
 680
 681         gdp = ext4_get_group_desc(sb, group, NULL);
 682         if (unlikely(!gdp))
 683                 return 0;
 684
 685         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 686                        (ino / inodes_per_block));
 687         if (!bh || !buffer_uptodate(bh))
 688                 /*
 689                  * If the block is not in the buffer cache, then it
 690                  * must have been written out.
 691                  */
 692                 goto out;
 693
 694         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 695         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 696
 697         /* i_dtime is only 32 bits on disk, but we only care about relative
 698          * times in the range of a few minutes (i.e. long enough to sync a
 699          * recently-deleted inode to disk), so using the low 32 bits of the
 700          * clock (a 68 year range) is enough, see time_before32() */
 701         dtime = le32_to_cpu(raw_inode->i_dtime);
 702         now = ktime_get_real_seconds();
 703         if (buffer_dirty(bh))
 704                 recentcy += RECENTCY_DIRTY;
 705
 706         if (dtime && time_before32(dtime, now) &&
 707             time_before32(now, dtime + recentcy))
 708                 ret = 1;
 709 out:
 710         brelse(bh);
 711         return ret;
 712 }
 713
 714 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 715                           struct buffer_head *bitmap, unsigned long *ino)
 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                 return 0;
 723
 724         if ((EXT4_SB(sb)->s_journal == NULL) &&
 725             recently_deleted(sb, group, *ino)) {
 726                 *ino = *ino + 1;
 727                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 728                         goto next;
 729                 return 0;
 730         }
 731
 732         return 1;
 733 }
 734
 735 /*
 736  * There are two policies for allocating an inode.  If the new inode is
 737  * a directory, then a forward search is made for a block group with both
 738  * free space and a low directory-to-inode ratio; if that fails, then of
 739  * the groups with above-average free space, that group with the fewest
 740  * directories already is chosen.
 741  *
 742  * For other inodes, search forward from the parent directory's block
 743  * group to find a free inode.
 744  */
 745 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 746                                umode_t mode, const struct qstr *qstr,
 747                                __u32 goal, uid_t *owner, __u32 i_flags,
 748                                int handle_type, unsigned int line_no,
 749                                int nblocks)
 750 {
 751         struct super_block *sb;
 752         struct buffer_head *inode_bitmap_bh = NULL;
 753         struct buffer_head *group_desc_bh;
 754         ext4_group_t ngroups, group = 0;
 755         unsigned long ino = 0;
 756         struct inode *inode;
 757         struct ext4_group_desc *gdp = NULL;
 758         struct ext4_inode_info *ei;
 759         struct ext4_sb_info *sbi;
 760         int ret2, err;
 761         struct inode *ret;
 762         ext4_group_t i;
 763         ext4_group_t flex_group;
 764         struct ext4_group_info *grp;
 765         int encrypt = 0;
 766
 767         /* Cannot create files in a deleted directory */
 768         if (!dir || !dir->i_nlink)
 769                 return ERR_PTR(-EPERM);
 770
 771         sb = dir->i_sb;
 772         sbi = EXT4_SB(sb);
 773
 774         if (unlikely(ext4_forced_shutdown(sbi)))
 775                 return ERR_PTR(-EIO);
 776
 777         if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
 778             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
 779             !(i_flags & EXT4_EA_INODE_FL)) {
 780                 err = fscrypt_get_encryption_info(dir);
 781                 if (err)
 782                         return ERR_PTR(err);
 783                 if (!fscrypt_has_encryption_key(dir))
 784                         return ERR_PTR(-ENOKEY);
 785                 encrypt = 1;
 786         }
 787
 788         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 789 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 790                 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 791
 792                 if (IS_ERR(p))
 793                         return ERR_CAST(p);
 794                 if (p) {
 795                         int acl_size = p->a_count * sizeof(ext4_acl_entry);
 796
 797                         nblocks += (S_ISDIR(mode) ? 2 : 1) *
 798                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 799                                         NULL /* block_bh */, acl_size,
 800                                         true /* is_create */);
 801                         posix_acl_release(p);
 802                 }
 803 #endif
 804
 805 #ifdef CONFIG_SECURITY
 806                 {
 807                         int num_security_xattrs = 1;
 808
 809 #ifdef CONFIG_INTEGRITY
 810                         num_security_xattrs++;
 811 #endif
 812                         /*
 813                          * We assume that security xattrs are never
 814                          * more than 1k.  In practice they are under
 815                          * 128 bytes.
 816                          */
 817                         nblocks += num_security_xattrs *
 818                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 819                                         NULL /* block_bh */, 1024,
 820                                         true /* is_create */);
 821                 }
 822 #endif
 823                 if (encrypt)
 824                         nblocks += __ext4_xattr_set_credits(sb,
 825                                         NULL /* inode */, NULL /* block_bh */,
 826                                         FSCRYPT_SET_CONTEXT_MAX_SIZE,
 827                                         true /* is_create */);
 828         }
 829
 830         ngroups = ext4_get_groups_count(sb);
 831         trace_ext4_request_inode(dir, mode);
 832         inode = new_inode(sb);
 833         if (!inode)
 834                 return ERR_PTR(-ENOMEM);
 835         ei = EXT4_I(inode);
 836
 837         /*
 838          * Initialize owners and quota early so that we don't have to account
 839          * for quota initialization worst case in standard inode creating
 840          * transaction
 841          */
 842         if (owner) {
 843                 inode->i_mode = mode;
 844                 i_uid_write(inode, owner[0]);
 845                 i_gid_write(inode, owner[1]);
 846         } else if (test_opt(sb, GRPID)) {
 847                 inode->i_mode = mode;
 848                 inode->i_uid = current_fsuid();
 849                 inode->i_gid = dir->i_gid;
 850         } else
 851                 inode_init_owner(inode, dir, mode);
 852
 853         if (ext4_has_feature_project(sb) &&
 854             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 855                 ei->i_projid = EXT4_I(dir)->i_projid;
 856         else
 857                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 858
 859         err = dquot_initialize(inode);
 860         if (err)
 861                 goto out;
 862
 863         if (!goal)
 864                 goal = sbi->s_inode_goal;
 865
 866         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 867                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 868                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 869                 ret2 = 0;
 870                 goto got_group;
 871         }
 872
 873         if (S_ISDIR(mode))
 874                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 875         else
 876                 ret2 = find_group_other(sb, dir, &group, mode);
 877
 878 got_group:
 879         EXT4_I(dir)->i_last_alloc_group = group;
 880         err = -ENOSPC;
 881         if (ret2 == -1)
 882                 goto out;
 883
 884         /*
 885          * Normally we will only go through one pass of this loop,
 886          * unless we get unlucky and it turns out the group we selected
 887          * had its last inode grabbed by someone else.
 888          */
 889         for (i = 0; i < ngroups; i++, ino = 0) {
 890                 err = -EIO;
 891
 892                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 893                 if (!gdp)
 894                         goto out;
 895
 896                 /*
 897                  * Check free inodes count before loading bitmap.
 898                  */
 899                 if (ext4_free_inodes_count(sb, gdp) == 0)
 900                         goto next_group;
 901
 902                 grp = ext4_get_group_info(sb, group);
 903                 /* Skip groups with already-known suspicious inode tables */
 904                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 905                         goto next_group;
 906
 907                 brelse(inode_bitmap_bh);
 908                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 909                 /* Skip groups with suspicious inode tables */
 910                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 911                     IS_ERR(inode_bitmap_bh)) {
 912                         inode_bitmap_bh = NULL;
 913                         goto next_group;
 914                 }
 915
 916 repeat_in_this_group:
 917                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 918                 if (!ret2)
 919                         goto next_group;
 920
 921                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 922                         ext4_error(sb, "reserved inode found cleared - "
 923                                    "inode=%lu", ino + 1);
 924                         ext4_mark_group_bitmap_corrupted(sb, group,
 925                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 926                         goto next_group;
 927                 }
 928
 929                 if (!handle) {
 930                         BUG_ON(nblocks <= 0);
 931                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 932                                                          handle_type, nblocks,
 933                                                          0);
 934                         if (IS_ERR(handle)) {
 935                                 err = PTR_ERR(handle);
 936                                 ext4_std_error(sb, err);
 937                                 goto out;
 938                         }
 939                 }
 940                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 941                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 942                 if (err) {
 943                         ext4_std_error(sb, err);
 944                         goto out;
 945                 }
 946                 ext4_lock_group(sb, group);
 947                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 948                 if (ret2) {
 949                         /* Someone already took the bit. Repeat the search
 950                          * with lock held.
 951                          */
 952                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 953                         if (ret2) {
 954                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
 955                                 ret2 = 0;
 956                         } else {
 957                                 ret2 = 1; /* we didn't grab the inode */
 958                         }
 959                 }
 960                 ext4_unlock_group(sb, group);
 961                 ino++;          /* the inode bitmap is zero-based */
 962                 if (!ret2)
 963                         goto got; /* we grabbed the inode! */
 964
 965                 if (ino < EXT4_INODES_PER_GROUP(sb))
 966                         goto repeat_in_this_group;
 967 next_group:
 968                 if (++group == ngroups)
 969                         group = 0;
 970         }
 971         err = -ENOSPC;
 972         goto out;
 973
 974 got:
 975         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 976         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 977         if (err) {
 978                 ext4_std_error(sb, err);
 979                 goto out;
 980         }
 981
 982         BUFFER_TRACE(group_desc_bh, "get_write_access");
 983         err = ext4_journal_get_write_access(handle, group_desc_bh);
 984         if (err) {
 985                 ext4_std_error(sb, err);
 986                 goto out;
 987         }
 988
 989         /* We may have to initialize the block bitmap if it isn't already */
 990         if (ext4_has_group_desc_csum(sb) &&
 991             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 992                 struct buffer_head *block_bitmap_bh;
 993
 994                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 995                 if (IS_ERR(block_bitmap_bh)) {
 996                         err = PTR_ERR(block_bitmap_bh);
 997                         goto out;
 998                 }
 999                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1000                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1001                 if (err) {
1002                         brelse(block_bitmap_bh);
1003                         ext4_std_error(sb, err);
1004                         goto out;
1005                 }
1006
1007                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1008                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1009
1010                 /* recheck and clear flag under lock if we still need to */
1011                 ext4_lock_group(sb, group);
1012                 if (ext4_has_group_desc_csum(sb) &&
1013                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1014                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1015                         ext4_free_group_clusters_set(sb, gdp,
1016                                 ext4_free_clusters_after_init(sb, group, gdp));
1017                         ext4_block_bitmap_csum_set(sb, group, gdp,
1018                                                    block_bitmap_bh);
1019                         ext4_group_desc_csum_set(sb, group, gdp);
1020                 }
1021                 ext4_unlock_group(sb, group);
1022                 brelse(block_bitmap_bh);
1023
1024                 if (err) {
1025                         ext4_std_error(sb, err);
1026                         goto out;
1027                 }
1028         }
1029
1030         /* Update the relevant bg descriptor fields */
1031         if (ext4_has_group_desc_csum(sb)) {
1032                 int free;
1033                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1034
1035                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1036                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1037                 free = EXT4_INODES_PER_GROUP(sb) -
1038                         ext4_itable_unused_count(sb, gdp);
1039                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1040                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1041                         free = 0;
1042                 }
1043                 /*
1044                  * Check the relative inode number against the last used
1045                  * relative inode number in this group. if it is greater
1046                  * we need to update the bg_itable_unused count
1047                  */
1048                 if (ino > free)
1049                         ext4_itable_unused_set(sb, gdp,
1050                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1051                 up_read(&grp->alloc_sem);
1052         } else {
1053                 ext4_lock_group(sb, group);
1054         }
1055
1056         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1057         if (S_ISDIR(mode)) {
1058                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1059                 if (sbi->s_log_groups_per_flex) {
1060                         ext4_group_t f = ext4_flex_group(sbi, group);
1061
1062                         atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups,
1063                                                         f)->used_dirs);
1064                 }
1065         }
1066         if (ext4_has_group_desc_csum(sb)) {
1067                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1068                                            EXT4_INODES_PER_GROUP(sb) / 8);
1069                 ext4_group_desc_csum_set(sb, group, gdp);
1070         }
1071         ext4_unlock_group(sb, group);
1072
1073         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1074         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1075         if (err) {
1076                 ext4_std_error(sb, err);
1077                 goto out;
1078         }
1079
1080         percpu_counter_dec(&sbi->s_freeinodes_counter);
1081         if (S_ISDIR(mode))
1082                 percpu_counter_inc(&sbi->s_dirs_counter);
1083
1084         if (sbi->s_log_groups_per_flex) {
1085                 flex_group = ext4_flex_group(sbi, group);
1086                 atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups,
1087                                                 flex_group)->free_inodes);
1088         }
1089
1090         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1091         /* This is the optimal IO size (for stat), not the fs block size */
1092         inode->i_blocks = 0;
1093         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1094         ei->i_crtime = inode->i_mtime;
1095
1096         memset(ei->i_data, 0, sizeof(ei->i_data));
1097         ei->i_dir_start_lookup = 0;
1098         ei->i_disksize = 0;
1099
1100         /* Don't inherit extent flag from directory, amongst others. */
1101         ei->i_flags =
1102                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1103         ei->i_flags |= i_flags;
1104         ei->i_file_acl = 0;
1105         ei->i_dtime = 0;
1106         ei->i_block_group = group;
1107         ei->i_last_alloc_group = ~0;
1108
1109         ext4_set_inode_flags(inode);
1110         if (IS_DIRSYNC(inode))
1111                 ext4_handle_sync(handle);
1112         if (insert_inode_locked(inode) < 0) {
1113                 /*
1114                  * Likely a bitmap corruption causing inode to be allocated
1115                  * twice.
1116                  */
1117                 err = -EIO;
1118                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1119                            inode->i_ino);
1120                 ext4_mark_group_bitmap_corrupted(sb, group,
1121                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1122                 goto out;
1123         }
1124         inode->i_generation = prandom_u32();
1125
1126         /* Precompute checksum seed for inode metadata */
1127         if (ext4_has_metadata_csum(sb)) {
1128                 __u32 csum;
1129                 __le32 inum = cpu_to_le32(inode->i_ino);
1130                 __le32 gen = cpu_to_le32(inode->i_generation);
1131                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1132                                    sizeof(inum));
1133                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1134                                               sizeof(gen));
1135         }
1136
1137         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1138         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1139
1140         ei->i_extra_isize = sbi->s_want_extra_isize;
1141         ei->i_inline_off = 0;
1142         if (ext4_has_feature_inline_data(sb))
1143                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1144         ret = inode;
1145         err = dquot_alloc_inode(inode);
1146         if (err)
1147                 goto fail_drop;
1148
1149         /*
1150          * Since the encryption xattr will always be unique, create it first so
1151          * that it's less likely to end up in an external xattr block and
1152          * prevent its deduplication.
1153          */
1154         if (encrypt) {
1155                 err = fscrypt_inherit_context(dir, inode, handle, true);
1156                 if (err)
1157                         goto fail_free_drop;
1158         }
1159
1160         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1161                 err = ext4_init_acl(handle, inode, dir);
1162                 if (err)
1163                         goto fail_free_drop;
1164
1165                 err = ext4_init_security(handle, inode, dir, qstr);
1166                 if (err)
1167                         goto fail_free_drop;
1168         }
1169
1170         if (ext4_has_feature_extents(sb)) {
1171                 /* set extent flag only for directory, file and normal symlink*/
1172                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1173                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1174                         ext4_ext_tree_init(handle, inode);
1175                 }
1176         }
1177
1178         if (ext4_handle_valid(handle)) {
1179                 ei->i_sync_tid = handle->h_transaction->t_tid;
1180                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1181         }
1182
1183         err = ext4_mark_inode_dirty(handle, inode);
1184         if (err) {
1185                 ext4_std_error(sb, err);
1186                 goto fail_free_drop;
1187         }
1188
1189         ext4_debug("allocating inode %lu\n", inode->i_ino);
1190         trace_ext4_allocate_inode(inode, dir, mode);
1191         brelse(inode_bitmap_bh);
1192         return ret;
1193
1194 fail_free_drop:
1195         dquot_free_inode(inode);
1196 fail_drop:
1197         clear_nlink(inode);
1198         unlock_new_inode(inode);
1199 out:
1200         dquot_drop(inode);
1201         inode->i_flags |= S_NOQUOTA;
1202         iput(inode);
1203         brelse(inode_bitmap_bh);
1204         return ERR_PTR(err);
1205 }
1206
1207 /* Verify that we are loading a valid orphan from disk */
1208 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1209 {
1210         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1211         ext4_group_t block_group;
1212         int bit;
1213         struct buffer_head *bitmap_bh = NULL;
1214         struct inode *inode = NULL;
1215         int err = -EFSCORRUPTED;
1216
1217         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1218                 goto bad_orphan;
1219
1220         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1221         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1222         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1223         if (IS_ERR(bitmap_bh))
1224                 return (struct inode *) bitmap_bh;
1225
1226         /* Having the inode bit set should be a 100% indicator that this
1227          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1228          * inodes that were being truncated, so we can't check i_nlink==0.
1229          */
1230         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1231                 goto bad_orphan;
1232
1233         inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1234         if (IS_ERR(inode)) {
1235                 err = PTR_ERR(inode);
1236                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1237                            ino, err);
1238                 return inode;
1239         }
1240
1241         /*
1242          * If the orphans has i_nlinks > 0 then it should be able to
1243          * be truncated, otherwise it won't be removed from the orphan
1244          * list during processing and an infinite loop will result.
1245          * Similarly, it must not be a bad inode.
1246          */
1247         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1248             is_bad_inode(inode))
1249                 goto bad_orphan;
1250
1251         if (NEXT_ORPHAN(inode) > max_ino)
1252                 goto bad_orphan;
1253         brelse(bitmap_bh);
1254         return inode;
1255
1256 bad_orphan:
1257         ext4_error(sb, "bad orphan inode %lu", ino);
1258         if (bitmap_bh)
1259                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1260                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1261                        ext4_test_bit(bit, bitmap_bh->b_data));
1262         if (inode) {
1263                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1264                        is_bad_inode(inode));
1265                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1266                        NEXT_ORPHAN(inode));
1267                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1268                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1269                 /* Avoid freeing blocks if we got a bad deleted inode */
1270                 if (inode->i_nlink == 0)
1271                         inode->i_blocks = 0;
1272                 iput(inode);
1273         }
1274         brelse(bitmap_bh);
1275         return ERR_PTR(err);
1276 }
1277
1278 unsigned long ext4_count_free_inodes(struct super_block *sb)
1279 {
1280         unsigned long desc_count;
1281         struct ext4_group_desc *gdp;
1282         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1283 #ifdef EXT4FS_DEBUG
1284         struct ext4_super_block *es;
1285         unsigned long bitmap_count, x;
1286         struct buffer_head *bitmap_bh = NULL;
1287
1288         es = EXT4_SB(sb)->s_es;
1289         desc_count = 0;
1290         bitmap_count = 0;
1291         gdp = NULL;
1292         for (i = 0; i < ngroups; i++) {
1293                 gdp = ext4_get_group_desc(sb, i, NULL);
1294                 if (!gdp)
1295                         continue;
1296                 desc_count += ext4_free_inodes_count(sb, gdp);
1297                 brelse(bitmap_bh);
1298                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1299                 if (IS_ERR(bitmap_bh)) {
1300                         bitmap_bh = NULL;
1301                         continue;
1302                 }
1303
1304                 x = ext4_count_free(bitmap_bh->b_data,
1305                                     EXT4_INODES_PER_GROUP(sb) / 8);
1306                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1307                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1308                 bitmap_count += x;
1309         }
1310         brelse(bitmap_bh);
1311         printk(KERN_DEBUG "ext4_count_free_inodes: "
1312                "stored = %u, computed = %lu, %lu\n",
1313                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1314         return desc_count;
1315 #else
1316         desc_count = 0;
1317         for (i = 0; i < ngroups; i++) {
1318                 gdp = ext4_get_group_desc(sb, i, NULL);
1319                 if (!gdp)
1320                         continue;
1321                 desc_count += ext4_free_inodes_count(sb, gdp);
1322                 cond_resched();
1323         }
1324         return desc_count;
1325 #endif
1326 }
1327
1328 /* Called at mount-time, super-block is locked */
1329 unsigned long ext4_count_dirs(struct super_block * sb)
1330 {
1331         unsigned long count = 0;
1332         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1333
1334         for (i = 0; i < ngroups; i++) {
1335                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1336                 if (!gdp)
1337                         continue;
1338                 count += ext4_used_dirs_count(sb, gdp);
1339         }
1340         return count;
1341 }
1342
1343 /*
1344  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1345  * inode table. Must be called without any spinlock held. The only place
1346  * where it is called from on active part of filesystem is ext4lazyinit
1347  * thread, so we do not need any special locks, however we have to prevent
1348  * inode allocation from the current group, so we take alloc_sem lock, to
1349  * block ext4_new_inode() until we are finished.
1350  */
1351 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1352                                  int barrier)
1353 {
1354         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1355         struct ext4_sb_info *sbi = EXT4_SB(sb);
1356         struct ext4_group_desc *gdp = NULL;
1357         struct buffer_head *group_desc_bh;
1358         handle_t *handle;
1359         ext4_fsblk_t blk;
1360         int num, ret = 0, used_blks = 0;
1361
1362         /* This should not happen, but just to be sure check this */
1363         if (sb_rdonly(sb)) {
1364                 ret = 1;
1365                 goto out;
1366         }
1367
1368         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1369         if (!gdp)
1370                 goto out;
1371
1372         /*
1373          * We do not need to lock this, because we are the only one
1374          * handling this flag.
1375          */
1376         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1377                 goto out;
1378
1379         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1380         if (IS_ERR(handle)) {
1381                 ret = PTR_ERR(handle);
1382                 goto out;
1383         }
1384
1385         down_write(&grp->alloc_sem);
1386         /*
1387          * If inode bitmap was already initialized there may be some
1388          * used inodes so we need to skip blocks with used inodes in
1389          * inode table.
1390          */
1391         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1392                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1393                             ext4_itable_unused_count(sb, gdp)),
1394                             sbi->s_inodes_per_block);
1395
1396         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1397             ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1398                                ext4_itable_unused_count(sb, gdp)) <
1399                               EXT4_FIRST_INO(sb)))) {
1400                 ext4_error(sb, "Something is wrong with group %u: "
1401                            "used itable blocks: %d; "
1402                            "itable unused count: %u",
1403                            group, used_blks,
1404                            ext4_itable_unused_count(sb, gdp));
1405                 ret = 1;
1406                 goto err_out;
1407         }
1408
1409         blk = ext4_inode_table(sb, gdp) + used_blks;
1410         num = sbi->s_itb_per_group - used_blks;
1411
1412         BUFFER_TRACE(group_desc_bh, "get_write_access");
1413         ret = ext4_journal_get_write_access(handle,
1414                                             group_desc_bh);
1415         if (ret)
1416                 goto err_out;
1417
1418         /*
1419          * Skip zeroout if the inode table is full. But we set the ZEROED
1420          * flag anyway, because obviously, when it is full it does not need
1421          * further zeroing.
1422          */
1423         if (unlikely(num == 0))
1424                 goto skip_zeroout;
1425
1426         ext4_debug("going to zero out inode table in group %d\n",
1427                    group);
1428         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1429         if (ret < 0)
1430                 goto err_out;
1431         if (barrier)
1432                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1433
1434 skip_zeroout:
1435         ext4_lock_group(sb, group);
1436         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1437         ext4_group_desc_csum_set(sb, group, gdp);
1438         ext4_unlock_group(sb, group);
1439
1440         BUFFER_TRACE(group_desc_bh,
1441                      "call ext4_handle_dirty_metadata");
1442         ret = ext4_handle_dirty_metadata(handle, NULL,
1443                                          group_desc_bh);
1444
1445 err_out:
1446         up_write(&grp->alloc_sem);
1447         ext4_journal_stop(handle);
1448 out:
1449         return ret;
1450 }