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_set_errno(sb, EIO);
 198                 ext4_error(sb, "Cannot read inode bitmap - "
 199                            "block_group = %u, inode_bitmap = %llu",
 200                            block_group, bitmap_blk);
 201                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 202                                 EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 203                 return ERR_PTR(-EIO);
 204         }
 205
 206 verify:
 207         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 208         if (err)
 209                 goto out;
 210         return bh;
 211 out:
 212         put_bh(bh);
 213         return ERR_PTR(err);
 214 }
 215
 216 /*
 217  * NOTE! When we get the inode, we're the only people
 218  * that have access to it, and as such there are no
 219  * race conditions we have to worry about. The inode
 220  * is not on the hash-lists, and it cannot be reached
 221  * through the filesystem because the directory entry
 222  * has been deleted earlier.
 223  *
 224  * HOWEVER: we must make sure that we get no aliases,
 225  * which means that we have to call "clear_inode()"
 226  * _before_ we mark the inode not in use in the inode
 227  * bitmaps. Otherwise a newly created file might use
 228  * the same inode number (not actually the same pointer
 229  * though), and then we'd have two inodes sharing the
 230  * same inode number and space on the harddisk.
 231  */
 232 void ext4_free_inode(handle_t *handle, struct inode *inode)
 233 {
 234         struct super_block *sb = inode->i_sb;
 235         int is_directory;
 236         unsigned long ino;
 237         struct buffer_head *bitmap_bh = NULL;
 238         struct buffer_head *bh2;
 239         ext4_group_t block_group;
 240         unsigned long bit;
 241         struct ext4_group_desc *gdp;
 242         struct ext4_super_block *es;
 243         struct ext4_sb_info *sbi;
 244         int fatal = 0, err, count, cleared;
 245         struct ext4_group_info *grp;
 246
 247         if (!sb) {
 248                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 249                        "nonexistent device\n", __func__, __LINE__);
 250                 return;
 251         }
 252         if (atomic_read(&inode->i_count) > 1) {
 253                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 254                          __func__, __LINE__, inode->i_ino,
 255                          atomic_read(&inode->i_count));
 256                 return;
 257         }
 258         if (inode->i_nlink) {
 259                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 260                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 261                 return;
 262         }
 263         sbi = EXT4_SB(sb);
 264
 265         ino = inode->i_ino;
 266         ext4_debug("freeing inode %lu\n", ino);
 267         trace_ext4_free_inode(inode);
 268
 269         dquot_initialize(inode);
 270         dquot_free_inode(inode);
 271
 272         is_directory = S_ISDIR(inode->i_mode);
 273
 274         /* Do this BEFORE marking the inode not in use or returning an error */
 275         ext4_clear_inode(inode);
 276
 277         es = sbi->s_es;
 278         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 279                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 280                 goto error_return;
 281         }
 282         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 283         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 284         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 285         /* Don't bother if the inode bitmap is corrupt. */
 286         grp = ext4_get_group_info(sb, block_group);
 287         if (IS_ERR(bitmap_bh)) {
 288                 fatal = PTR_ERR(bitmap_bh);
 289                 bitmap_bh = NULL;
 290                 goto error_return;
 291         }
 292         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 293                 fatal = -EFSCORRUPTED;
 294                 goto error_return;
 295         }
 296
 297         BUFFER_TRACE(bitmap_bh, "get_write_access");
 298         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 299         if (fatal)
 300                 goto error_return;
 301
 302         fatal = -ESRCH;
 303         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 304         if (gdp) {
 305                 BUFFER_TRACE(bh2, "get_write_access");
 306                 fatal = ext4_journal_get_write_access(handle, bh2);
 307         }
 308         ext4_lock_group(sb, block_group);
 309         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 310         if (fatal || !cleared) {
 311                 ext4_unlock_group(sb, block_group);
 312                 goto out;
 313         }
 314
 315         count = ext4_free_inodes_count(sb, gdp) + 1;
 316         ext4_free_inodes_set(sb, gdp, count);
 317         if (is_directory) {
 318                 count = ext4_used_dirs_count(sb, gdp) - 1;
 319                 ext4_used_dirs_set(sb, gdp, count);
 320                 percpu_counter_dec(&sbi->s_dirs_counter);
 321         }
 322         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 323                                    EXT4_INODES_PER_GROUP(sb) / 8);
 324         ext4_group_desc_csum_set(sb, block_group, gdp);
 325         ext4_unlock_group(sb, block_group);
 326
 327         percpu_counter_inc(&sbi->s_freeinodes_counter);
 328         if (sbi->s_log_groups_per_flex) {
 329                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 330
 331                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 332                 if (is_directory)
 333                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 334         }
 335         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 336         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 337 out:
 338         if (cleared) {
 339                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 340                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 341                 if (!fatal)
 342                         fatal = err;
 343         } else {
 344                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 345                 ext4_mark_group_bitmap_corrupted(sb, block_group,
 346                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 347         }
 348
 349 error_return:
 350         brelse(bitmap_bh);
 351         ext4_std_error(sb, fatal);
 352 }
 353
 354 struct orlov_stats {
 355         __u64 free_clusters;
 356         __u32 free_inodes;
 357         __u32 used_dirs;
 358 };
 359
 360 /*
 361  * Helper function for Orlov's allocator; returns critical information
 362  * for a particular block group or flex_bg.  If flex_size is 1, then g
 363  * is a block group number; otherwise it is flex_bg number.
 364  */
 365 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 366                             int flex_size, struct orlov_stats *stats)
 367 {
 368         struct ext4_group_desc *desc;
 369         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 370
 371         if (flex_size > 1) {
 372                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 373                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 374                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 375                 return;
 376         }
 377
 378         desc = ext4_get_group_desc(sb, g, NULL);
 379         if (desc) {
 380                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 381                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 382                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 383         } else {
 384                 stats->free_inodes = 0;
 385                 stats->free_clusters = 0;
 386                 stats->used_dirs = 0;
 387         }
 388 }
 389
 390 /*
 391  * Orlov's allocator for directories.
 392  *
 393  * We always try to spread first-level directories.
 394  *
 395  * If there are blockgroups with both free inodes and free blocks counts
 396  * not worse than average we return one with smallest directory count.
 397  * Otherwise we simply return a random group.
 398  *
 399  * For the rest rules look so:
 400  *
 401  * It's OK to put directory into a group unless
 402  * it has too many directories already (max_dirs) or
 403  * it has too few free inodes left (min_inodes) or
 404  * it has too few free blocks left (min_blocks) or
 405  * Parent's group is preferred, if it doesn't satisfy these
 406  * conditions we search cyclically through the rest. If none
 407  * of the groups look good we just look for a group with more
 408  * free inodes than average (starting at parent's group).
 409  */
 410
 411 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 412                             ext4_group_t *group, umode_t mode,
 413                             const struct qstr *qstr)
 414 {
 415         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 416         struct ext4_sb_info *sbi = EXT4_SB(sb);
 417         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 418         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 419         unsigned int freei, avefreei, grp_free;
 420         ext4_fsblk_t freeb, avefreec;
 421         unsigned int ndirs;
 422         int max_dirs, min_inodes;
 423         ext4_grpblk_t min_clusters;
 424         ext4_group_t i, grp, g, ngroups;
 425         struct ext4_group_desc *desc;
 426         struct orlov_stats stats;
 427         int flex_size = ext4_flex_bg_size(sbi);
 428         struct dx_hash_info hinfo;
 429
 430         ngroups = real_ngroups;
 431         if (flex_size > 1) {
 432                 ngroups = (real_ngroups + flex_size - 1) >>
 433                         sbi->s_log_groups_per_flex;
 434                 parent_group >>= sbi->s_log_groups_per_flex;
 435         }
 436
 437         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 438         avefreei = freei / ngroups;
 439         freeb = EXT4_C2B(sbi,
 440                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 441         avefreec = freeb;
 442         do_div(avefreec, ngroups);
 443         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 444
 445         if (S_ISDIR(mode) &&
 446             ((parent == d_inode(sb->s_root)) ||
 447              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 448                 int best_ndir = inodes_per_group;
 449                 int ret = -1;
 450
 451                 if (qstr) {
 452                         hinfo.hash_version = DX_HASH_HALF_MD4;
 453                         hinfo.seed = sbi->s_hash_seed;
 454                         ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo);
 455                         grp = hinfo.hash;
 456                 } else
 457                         grp = prandom_u32();
 458                 parent_group = (unsigned)grp % ngroups;
 459                 for (i = 0; i < ngroups; i++) {
 460                         g = (parent_group + i) % ngroups;
 461                         get_orlov_stats(sb, g, flex_size, &stats);
 462                         if (!stats.free_inodes)
 463                                 continue;
 464                         if (stats.used_dirs >= best_ndir)
 465                                 continue;
 466                         if (stats.free_inodes < avefreei)
 467                                 continue;
 468                         if (stats.free_clusters < avefreec)
 469                                 continue;
 470                         grp = g;
 471                         ret = 0;
 472                         best_ndir = stats.used_dirs;
 473                 }
 474                 if (ret)
 475                         goto fallback;
 476         found_flex_bg:
 477                 if (flex_size == 1) {
 478                         *group = grp;
 479                         return 0;
 480                 }
 481
 482                 /*
 483                  * We pack inodes at the beginning of the flexgroup's
 484                  * inode tables.  Block allocation decisions will do
 485                  * something similar, although regular files will
 486                  * start at 2nd block group of the flexgroup.  See
 487                  * ext4_ext_find_goal() and ext4_find_near().
 488                  */
 489                 grp *= flex_size;
 490                 for (i = 0; i < flex_size; i++) {
 491                         if (grp+i >= real_ngroups)
 492                                 break;
 493                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 494                         if (desc && ext4_free_inodes_count(sb, desc)) {
 495                                 *group = grp+i;
 496                                 return 0;
 497                         }
 498                 }
 499                 goto fallback;
 500         }
 501
 502         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 503         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 504         if (min_inodes < 1)
 505                 min_inodes = 1;
 506         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 507
 508         /*
 509          * Start looking in the flex group where we last allocated an
 510          * inode for this parent directory
 511          */
 512         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 513                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 514                 if (flex_size > 1)
 515                         parent_group >>= sbi->s_log_groups_per_flex;
 516         }
 517
 518         for (i = 0; i < ngroups; i++) {
 519                 grp = (parent_group + i) % ngroups;
 520                 get_orlov_stats(sb, grp, flex_size, &stats);
 521                 if (stats.used_dirs >= max_dirs)
 522                         continue;
 523                 if (stats.free_inodes < min_inodes)
 524                         continue;
 525                 if (stats.free_clusters < min_clusters)
 526                         continue;
 527                 goto found_flex_bg;
 528         }
 529
 530 fallback:
 531         ngroups = real_ngroups;
 532         avefreei = freei / ngroups;
 533 fallback_retry:
 534         parent_group = EXT4_I(parent)->i_block_group;
 535         for (i = 0; i < ngroups; i++) {
 536                 grp = (parent_group + i) % ngroups;
 537                 desc = ext4_get_group_desc(sb, grp, NULL);
 538                 if (desc) {
 539                         grp_free = ext4_free_inodes_count(sb, desc);
 540                         if (grp_free && grp_free >= avefreei) {
 541                                 *group = grp;
 542                                 return 0;
 543                         }
 544                 }
 545         }
 546
 547         if (avefreei) {
 548                 /*
 549                  * The free-inodes counter is approximate, and for really small
 550                  * filesystems the above test can fail to find any blockgroups
 551                  */
 552                 avefreei = 0;
 553                 goto fallback_retry;
 554         }
 555
 556         return -1;
 557 }
 558
 559 static int find_group_other(struct super_block *sb, struct inode *parent,
 560                             ext4_group_t *group, umode_t mode)
 561 {
 562         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 563         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 564         struct ext4_group_desc *desc;
 565         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 566
 567         /*
 568          * Try to place the inode is the same flex group as its
 569          * parent.  If we can't find space, use the Orlov algorithm to
 570          * find another flex group, and store that information in the
 571          * parent directory's inode information so that use that flex
 572          * group for future allocations.
 573          */
 574         if (flex_size > 1) {
 575                 int retry = 0;
 576
 577         try_again:
 578                 parent_group &= ~(flex_size-1);
 579                 last = parent_group + flex_size;
 580                 if (last > ngroups)
 581                         last = ngroups;
 582                 for  (i = parent_group; i < last; i++) {
 583                         desc = ext4_get_group_desc(sb, i, NULL);
 584                         if (desc && ext4_free_inodes_count(sb, desc)) {
 585                                 *group = i;
 586                                 return 0;
 587                         }
 588                 }
 589                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 590                         retry = 1;
 591                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 592                         goto try_again;
 593                 }
 594                 /*
 595                  * If this didn't work, use the Orlov search algorithm
 596                  * to find a new flex group; we pass in the mode to
 597                  * avoid the topdir algorithms.
 598                  */
 599                 *group = parent_group + flex_size;
 600                 if (*group > ngroups)
 601                         *group = 0;
 602                 return find_group_orlov(sb, parent, group, mode, NULL);
 603         }
 604
 605         /*
 606          * Try to place the inode in its parent directory
 607          */
 608         *group = parent_group;
 609         desc = ext4_get_group_desc(sb, *group, NULL);
 610         if (desc && ext4_free_inodes_count(sb, desc) &&
 611             ext4_free_group_clusters(sb, desc))
 612                 return 0;
 613
 614         /*
 615          * We're going to place this inode in a different blockgroup from its
 616          * parent.  We want to cause files in a common directory to all land in
 617          * the same blockgroup.  But we want files which are in a different
 618          * directory which shares a blockgroup with our parent to land in a
 619          * different blockgroup.
 620          *
 621          * So add our directory's i_ino into the starting point for the hash.
 622          */
 623         *group = (*group + parent->i_ino) % ngroups;
 624
 625         /*
 626          * Use a quadratic hash to find a group with a free inode and some free
 627          * blocks.
 628          */
 629         for (i = 1; i < ngroups; i <<= 1) {
 630                 *group += i;
 631                 if (*group >= ngroups)
 632                         *group -= ngroups;
 633                 desc = ext4_get_group_desc(sb, *group, NULL);
 634                 if (desc && ext4_free_inodes_count(sb, desc) &&
 635                     ext4_free_group_clusters(sb, desc))
 636                         return 0;
 637         }
 638
 639         /*
 640          * That failed: try linear search for a free inode, even if that group
 641          * has no free blocks.
 642          */
 643         *group = parent_group;
 644         for (i = 0; i < ngroups; i++) {
 645                 if (++*group >= ngroups)
 646                         *group = 0;
 647                 desc = ext4_get_group_desc(sb, *group, NULL);
 648                 if (desc && ext4_free_inodes_count(sb, desc))
 649                         return 0;
 650         }
 651
 652         return -1;
 653 }
 654
 655 /*
 656  * In no journal mode, if an inode has recently been deleted, we want
 657  * to avoid reusing it until we're reasonably sure the inode table
 658  * block has been written back to disk.  (Yes, these values are
 659  * somewhat arbitrary...)
 660  */
 661 #define RECENTCY_MIN    5
 662 #define RECENTCY_DIRTY  300
 663
 664 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 665 {
 666         struct ext4_group_desc  *gdp;
 667         struct ext4_inode       *raw_inode;
 668         struct buffer_head      *bh;
 669         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 670         int offset, ret = 0;
 671         int recentcy = RECENTCY_MIN;
 672         u32 dtime, now;
 673
 674         gdp = ext4_get_group_desc(sb, group, NULL);
 675         if (unlikely(!gdp))
 676                 return 0;
 677
 678         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 679                        (ino / inodes_per_block));
 680         if (!bh || !buffer_uptodate(bh))
 681                 /*
 682                  * If the block is not in the buffer cache, then it
 683                  * must have been written out.
 684                  */
 685                 goto out;
 686
 687         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 688         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 689
 690         /* i_dtime is only 32 bits on disk, but we only care about relative
 691          * times in the range of a few minutes (i.e. long enough to sync a
 692          * recently-deleted inode to disk), so using the low 32 bits of the
 693          * clock (a 68 year range) is enough, see time_before32() */
 694         dtime = le32_to_cpu(raw_inode->i_dtime);
 695         now = ktime_get_real_seconds();
 696         if (buffer_dirty(bh))
 697                 recentcy += RECENTCY_DIRTY;
 698
 699         if (dtime && time_before32(dtime, now) &&
 700             time_before32(now, dtime + recentcy))
 701                 ret = 1;
 702 out:
 703         brelse(bh);
 704         return ret;
 705 }
 706
 707 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 708                           struct buffer_head *bitmap, unsigned long *ino)
 709 {
 710 next:
 711         *ino = ext4_find_next_zero_bit((unsigned long *)
 712                                        bitmap->b_data,
 713                                        EXT4_INODES_PER_GROUP(sb), *ino);
 714         if (*ino >= EXT4_INODES_PER_GROUP(sb))
 715                 return 0;
 716
 717         if ((EXT4_SB(sb)->s_journal == NULL) &&
 718             recently_deleted(sb, group, *ino)) {
 719                 *ino = *ino + 1;
 720                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 721                         goto next;
 722                 return 0;
 723         }
 724
 725         return 1;
 726 }
 727
 728 /*
 729  * There are two policies for allocating an inode.  If the new inode is
 730  * a directory, then a forward search is made for a block group with both
 731  * free space and a low directory-to-inode ratio; if that fails, then of
 732  * the groups with above-average free space, that group with the fewest
 733  * directories already is chosen.
 734  *
 735  * For other inodes, search forward from the parent directory's block
 736  * group to find a free inode.
 737  */
 738 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 739                                umode_t mode, const struct qstr *qstr,
 740                                __u32 goal, uid_t *owner, __u32 i_flags,
 741                                int handle_type, unsigned int line_no,
 742                                int nblocks)
 743 {
 744         struct super_block *sb;
 745         struct buffer_head *inode_bitmap_bh = NULL;
 746         struct buffer_head *group_desc_bh;
 747         ext4_group_t ngroups, group = 0;
 748         unsigned long ino = 0;
 749         struct inode *inode;
 750         struct ext4_group_desc *gdp = NULL;
 751         struct ext4_inode_info *ei;
 752         struct ext4_sb_info *sbi;
 753         int ret2, err;
 754         struct inode *ret;
 755         ext4_group_t i;
 756         ext4_group_t flex_group;
 757         struct ext4_group_info *grp;
 758         int encrypt = 0;
 759
 760         /* Cannot create files in a deleted directory */
 761         if (!dir || !dir->i_nlink)
 762                 return ERR_PTR(-EPERM);
 763
 764         sb = dir->i_sb;
 765         sbi = EXT4_SB(sb);
 766
 767         if (unlikely(ext4_forced_shutdown(sbi)))
 768                 return ERR_PTR(-EIO);
 769
 770         if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
 771             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
 772             !(i_flags & EXT4_EA_INODE_FL)) {
 773                 err = fscrypt_get_encryption_info(dir);
 774                 if (err)
 775                         return ERR_PTR(err);
 776                 if (!fscrypt_has_encryption_key(dir))
 777                         return ERR_PTR(-ENOKEY);
 778                 encrypt = 1;
 779         }
 780
 781         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 782 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 783                 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 784
 785                 if (IS_ERR(p))
 786                         return ERR_CAST(p);
 787                 if (p) {
 788                         int acl_size = p->a_count * sizeof(ext4_acl_entry);
 789
 790                         nblocks += (S_ISDIR(mode) ? 2 : 1) *
 791                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 792                                         NULL /* block_bh */, acl_size,
 793                                         true /* is_create */);
 794                         posix_acl_release(p);
 795                 }
 796 #endif
 797
 798 #ifdef CONFIG_SECURITY
 799                 {
 800                         int num_security_xattrs = 1;
 801
 802 #ifdef CONFIG_INTEGRITY
 803                         num_security_xattrs++;
 804 #endif
 805                         /*
 806                          * We assume that security xattrs are never
 807                          * more than 1k.  In practice they are under
 808                          * 128 bytes.
 809                          */
 810                         nblocks += num_security_xattrs *
 811                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 812                                         NULL /* block_bh */, 1024,
 813                                         true /* is_create */);
 814                 }
 815 #endif
 816                 if (encrypt)
 817                         nblocks += __ext4_xattr_set_credits(sb,
 818                                         NULL /* inode */, NULL /* block_bh */,
 819                                         FSCRYPT_SET_CONTEXT_MAX_SIZE,
 820                                         true /* is_create */);
 821         }
 822
 823         ngroups = ext4_get_groups_count(sb);
 824         trace_ext4_request_inode(dir, mode);
 825         inode = new_inode(sb);
 826         if (!inode)
 827                 return ERR_PTR(-ENOMEM);
 828         ei = EXT4_I(inode);
 829
 830         /*
 831          * Initialize owners and quota early so that we don't have to account
 832          * for quota initialization worst case in standard inode creating
 833          * transaction
 834          */
 835         if (owner) {
 836                 inode->i_mode = mode;
 837                 i_uid_write(inode, owner[0]);
 838                 i_gid_write(inode, owner[1]);
 839         } else if (test_opt(sb, GRPID)) {
 840                 inode->i_mode = mode;
 841                 inode->i_uid = current_fsuid();
 842                 inode->i_gid = dir->i_gid;
 843         } else
 844                 inode_init_owner(inode, dir, mode);
 845
 846         if (ext4_has_feature_project(sb) &&
 847             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 848                 ei->i_projid = EXT4_I(dir)->i_projid;
 849         else
 850                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 851
 852         err = dquot_initialize(inode);
 853         if (err)
 854                 goto out;
 855
 856         if (!goal)
 857                 goal = sbi->s_inode_goal;
 858
 859         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 860                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 861                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 862                 ret2 = 0;
 863                 goto got_group;
 864         }
 865
 866         if (S_ISDIR(mode))
 867                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 868         else
 869                 ret2 = find_group_other(sb, dir, &group, mode);
 870
 871 got_group:
 872         EXT4_I(dir)->i_last_alloc_group = group;
 873         err = -ENOSPC;
 874         if (ret2 == -1)
 875                 goto out;
 876
 877         /*
 878          * Normally we will only go through one pass of this loop,
 879          * unless we get unlucky and it turns out the group we selected
 880          * had its last inode grabbed by someone else.
 881          */
 882         for (i = 0; i < ngroups; i++, ino = 0) {
 883                 err = -EIO;
 884
 885                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 886                 if (!gdp)
 887                         goto out;
 888
 889                 /*
 890                  * Check free inodes count before loading bitmap.
 891                  */
 892                 if (ext4_free_inodes_count(sb, gdp) == 0)
 893                         goto next_group;
 894
 895                 grp = ext4_get_group_info(sb, group);
 896                 /* Skip groups with already-known suspicious inode tables */
 897                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 898                         goto next_group;
 899
 900                 brelse(inode_bitmap_bh);
 901                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 902                 /* Skip groups with suspicious inode tables */
 903                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 904                     IS_ERR(inode_bitmap_bh)) {
 905                         inode_bitmap_bh = NULL;
 906                         goto next_group;
 907                 }
 908
 909 repeat_in_this_group:
 910                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 911                 if (!ret2)
 912                         goto next_group;
 913
 914                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 915                         ext4_error(sb, "reserved inode found cleared - "
 916                                    "inode=%lu", ino + 1);
 917                         ext4_mark_group_bitmap_corrupted(sb, group,
 918                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
 919                         goto next_group;
 920                 }
 921
 922                 if (!handle) {
 923                         BUG_ON(nblocks <= 0);
 924                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 925                                  handle_type, nblocks, 0,
 926                                  ext4_trans_default_revoke_credits(sb));
 927                         if (IS_ERR(handle)) {
 928                                 err = PTR_ERR(handle);
 929                                 ext4_std_error(sb, err);
 930                                 goto out;
 931                         }
 932                 }
 933                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 934                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 935                 if (err) {
 936                         ext4_std_error(sb, err);
 937                         goto out;
 938                 }
 939                 ext4_lock_group(sb, group);
 940                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 941                 if (ret2) {
 942                         /* Someone already took the bit. Repeat the search
 943                          * with lock held.
 944                          */
 945                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 946                         if (ret2) {
 947                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
 948                                 ret2 = 0;
 949                         } else {
 950                                 ret2 = 1; /* we didn't grab the inode */
 951                         }
 952                 }
 953                 ext4_unlock_group(sb, group);
 954                 ino++;          /* the inode bitmap is zero-based */
 955                 if (!ret2)
 956                         goto got; /* we grabbed the inode! */
 957
 958                 if (ino < EXT4_INODES_PER_GROUP(sb))
 959                         goto repeat_in_this_group;
 960 next_group:
 961                 if (++group == ngroups)
 962                         group = 0;
 963         }
 964         err = -ENOSPC;
 965         goto out;
 966
 967 got:
 968         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 969         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 970         if (err) {
 971                 ext4_std_error(sb, err);
 972                 goto out;
 973         }
 974
 975         BUFFER_TRACE(group_desc_bh, "get_write_access");
 976         err = ext4_journal_get_write_access(handle, group_desc_bh);
 977         if (err) {
 978                 ext4_std_error(sb, err);
 979                 goto out;
 980         }
 981
 982         /* We may have to initialize the block bitmap if it isn't already */
 983         if (ext4_has_group_desc_csum(sb) &&
 984             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 985                 struct buffer_head *block_bitmap_bh;
 986
 987                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
 988                 if (IS_ERR(block_bitmap_bh)) {
 989                         err = PTR_ERR(block_bitmap_bh);
 990                         goto out;
 991                 }
 992                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
 993                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
 994                 if (err) {
 995                         brelse(block_bitmap_bh);
 996                         ext4_std_error(sb, err);
 997                         goto out;
 998                 }
 999
1000                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1001                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1002
1003                 /* recheck and clear flag under lock if we still need to */
1004                 ext4_lock_group(sb, group);
1005                 if (ext4_has_group_desc_csum(sb) &&
1006                     (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) {
1007                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1008                         ext4_free_group_clusters_set(sb, gdp,
1009                                 ext4_free_clusters_after_init(sb, group, gdp));
1010                         ext4_block_bitmap_csum_set(sb, group, gdp,
1011                                                    block_bitmap_bh);
1012                         ext4_group_desc_csum_set(sb, group, gdp);
1013                 }
1014                 ext4_unlock_group(sb, group);
1015                 brelse(block_bitmap_bh);
1016
1017                 if (err) {
1018                         ext4_std_error(sb, err);
1019                         goto out;
1020                 }
1021         }
1022
1023         /* Update the relevant bg descriptor fields */
1024         if (ext4_has_group_desc_csum(sb)) {
1025                 int free;
1026                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1027
1028                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1029                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1030                 free = EXT4_INODES_PER_GROUP(sb) -
1031                         ext4_itable_unused_count(sb, gdp);
1032                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1033                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1034                         free = 0;
1035                 }
1036                 /*
1037                  * Check the relative inode number against the last used
1038                  * relative inode number in this group. if it is greater
1039                  * we need to update the bg_itable_unused count
1040                  */
1041                 if (ino > free)
1042                         ext4_itable_unused_set(sb, gdp,
1043                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1044                 up_read(&grp->alloc_sem);
1045         } else {
1046                 ext4_lock_group(sb, group);
1047         }
1048
1049         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1050         if (S_ISDIR(mode)) {
1051                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1052                 if (sbi->s_log_groups_per_flex) {
1053                         ext4_group_t f = ext4_flex_group(sbi, group);
1054
1055                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1056                 }
1057         }
1058         if (ext4_has_group_desc_csum(sb)) {
1059                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1060                                            EXT4_INODES_PER_GROUP(sb) / 8);
1061                 ext4_group_desc_csum_set(sb, group, gdp);
1062         }
1063         ext4_unlock_group(sb, group);
1064
1065         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1066         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1067         if (err) {
1068                 ext4_std_error(sb, err);
1069                 goto out;
1070         }
1071
1072         percpu_counter_dec(&sbi->s_freeinodes_counter);
1073         if (S_ISDIR(mode))
1074                 percpu_counter_inc(&sbi->s_dirs_counter);
1075
1076         if (sbi->s_log_groups_per_flex) {
1077                 flex_group = ext4_flex_group(sbi, group);
1078                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1079         }
1080
1081         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1082         /* This is the optimal IO size (for stat), not the fs block size */
1083         inode->i_blocks = 0;
1084         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1085         ei->i_crtime = inode->i_mtime;
1086
1087         memset(ei->i_data, 0, sizeof(ei->i_data));
1088         ei->i_dir_start_lookup = 0;
1089         ei->i_disksize = 0;
1090
1091         /* Don't inherit extent flag from directory, amongst others. */
1092         ei->i_flags =
1093                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1094         ei->i_flags |= i_flags;
1095         ei->i_file_acl = 0;
1096         ei->i_dtime = 0;
1097         ei->i_block_group = group;
1098         ei->i_last_alloc_group = ~0;
1099
1100         ext4_set_inode_flags(inode);
1101         if (IS_DIRSYNC(inode))
1102                 ext4_handle_sync(handle);
1103         if (insert_inode_locked(inode) < 0) {
1104                 /*
1105                  * Likely a bitmap corruption causing inode to be allocated
1106                  * twice.
1107                  */
1108                 err = -EIO;
1109                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1110                            inode->i_ino);
1111                 ext4_mark_group_bitmap_corrupted(sb, group,
1112                                         EXT4_GROUP_INFO_IBITMAP_CORRUPT);
1113                 goto out;
1114         }
1115         inode->i_generation = prandom_u32();
1116
1117         /* Precompute checksum seed for inode metadata */
1118         if (ext4_has_metadata_csum(sb)) {
1119                 __u32 csum;
1120                 __le32 inum = cpu_to_le32(inode->i_ino);
1121                 __le32 gen = cpu_to_le32(inode->i_generation);
1122                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1123                                    sizeof(inum));
1124                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1125                                               sizeof(gen));
1126         }
1127
1128         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1129         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1130
1131         ei->i_extra_isize = sbi->s_want_extra_isize;
1132         ei->i_inline_off = 0;
1133         if (ext4_has_feature_inline_data(sb))
1134                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1135         ret = inode;
1136         err = dquot_alloc_inode(inode);
1137         if (err)
1138                 goto fail_drop;
1139
1140         /*
1141          * Since the encryption xattr will always be unique, create it first so
1142          * that it's less likely to end up in an external xattr block and
1143          * prevent its deduplication.
1144          */
1145         if (encrypt) {
1146                 err = fscrypt_inherit_context(dir, inode, handle, true);
1147                 if (err)
1148                         goto fail_free_drop;
1149         }
1150
1151         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1152                 err = ext4_init_acl(handle, inode, dir);
1153                 if (err)
1154                         goto fail_free_drop;
1155
1156                 err = ext4_init_security(handle, inode, dir, qstr);
1157                 if (err)
1158                         goto fail_free_drop;
1159         }
1160
1161         if (ext4_has_feature_extents(sb)) {
1162                 /* set extent flag only for directory, file and normal symlink*/
1163                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1164                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1165                         ext4_ext_tree_init(handle, inode);
1166                 }
1167         }
1168
1169         if (ext4_handle_valid(handle)) {
1170                 ei->i_sync_tid = handle->h_transaction->t_tid;
1171                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1172         }
1173
1174         err = ext4_mark_inode_dirty(handle, inode);
1175         if (err) {
1176                 ext4_std_error(sb, err);
1177                 goto fail_free_drop;
1178         }
1179
1180         ext4_debug("allocating inode %lu\n", inode->i_ino);
1181         trace_ext4_allocate_inode(inode, dir, mode);
1182         brelse(inode_bitmap_bh);
1183         return ret;
1184
1185 fail_free_drop:
1186         dquot_free_inode(inode);
1187 fail_drop:
1188         clear_nlink(inode);
1189         unlock_new_inode(inode);
1190 out:
1191         dquot_drop(inode);
1192         inode->i_flags |= S_NOQUOTA;
1193         iput(inode);
1194         brelse(inode_bitmap_bh);
1195         return ERR_PTR(err);
1196 }
1197
1198 /* Verify that we are loading a valid orphan from disk */
1199 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1200 {
1201         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1202         ext4_group_t block_group;
1203         int bit;
1204         struct buffer_head *bitmap_bh = NULL;
1205         struct inode *inode = NULL;
1206         int err = -EFSCORRUPTED;
1207
1208         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1209                 goto bad_orphan;
1210
1211         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1212         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1213         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1214         if (IS_ERR(bitmap_bh))
1215                 return ERR_CAST(bitmap_bh);
1216
1217         /* Having the inode bit set should be a 100% indicator that this
1218          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1219          * inodes that were being truncated, so we can't check i_nlink==0.
1220          */
1221         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1222                 goto bad_orphan;
1223
1224         inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL);
1225         if (IS_ERR(inode)) {
1226                 err = PTR_ERR(inode);
1227                 ext4_set_errno(sb, -err);
1228                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1229                            ino, err);
1230                 return inode;
1231         }
1232
1233         /*
1234          * If the orphans has i_nlinks > 0 then it should be able to
1235          * be truncated, otherwise it won't be removed from the orphan
1236          * list during processing and an infinite loop will result.
1237          * Similarly, it must not be a bad inode.
1238          */
1239         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1240             is_bad_inode(inode))
1241                 goto bad_orphan;
1242
1243         if (NEXT_ORPHAN(inode) > max_ino)
1244                 goto bad_orphan;
1245         brelse(bitmap_bh);
1246         return inode;
1247
1248 bad_orphan:
1249         ext4_error(sb, "bad orphan inode %lu", ino);
1250         if (bitmap_bh)
1251                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1252                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1253                        ext4_test_bit(bit, bitmap_bh->b_data));
1254         if (inode) {
1255                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1256                        is_bad_inode(inode));
1257                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1258                        NEXT_ORPHAN(inode));
1259                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1260                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1261                 /* Avoid freeing blocks if we got a bad deleted inode */
1262                 if (inode->i_nlink == 0)
1263                         inode->i_blocks = 0;
1264                 iput(inode);
1265         }
1266         brelse(bitmap_bh);
1267         return ERR_PTR(err);
1268 }
1269
1270 unsigned long ext4_count_free_inodes(struct super_block *sb)
1271 {
1272         unsigned long desc_count;
1273         struct ext4_group_desc *gdp;
1274         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1275 #ifdef EXT4FS_DEBUG
1276         struct ext4_super_block *es;
1277         unsigned long bitmap_count, x;
1278         struct buffer_head *bitmap_bh = NULL;
1279
1280         es = EXT4_SB(sb)->s_es;
1281         desc_count = 0;
1282         bitmap_count = 0;
1283         gdp = NULL;
1284         for (i = 0; i < ngroups; i++) {
1285                 gdp = ext4_get_group_desc(sb, i, NULL);
1286                 if (!gdp)
1287                         continue;
1288                 desc_count += ext4_free_inodes_count(sb, gdp);
1289                 brelse(bitmap_bh);
1290                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1291                 if (IS_ERR(bitmap_bh)) {
1292                         bitmap_bh = NULL;
1293                         continue;
1294                 }
1295
1296                 x = ext4_count_free(bitmap_bh->b_data,
1297                                     EXT4_INODES_PER_GROUP(sb) / 8);
1298                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1299                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1300                 bitmap_count += x;
1301         }
1302         brelse(bitmap_bh);
1303         printk(KERN_DEBUG "ext4_count_free_inodes: "
1304                "stored = %u, computed = %lu, %lu\n",
1305                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1306         return desc_count;
1307 #else
1308         desc_count = 0;
1309         for (i = 0; i < ngroups; i++) {
1310                 gdp = ext4_get_group_desc(sb, i, NULL);
1311                 if (!gdp)
1312                         continue;
1313                 desc_count += ext4_free_inodes_count(sb, gdp);
1314                 cond_resched();
1315         }
1316         return desc_count;
1317 #endif
1318 }
1319
1320 /* Called at mount-time, super-block is locked */
1321 unsigned long ext4_count_dirs(struct super_block * sb)
1322 {
1323         unsigned long count = 0;
1324         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1325
1326         for (i = 0; i < ngroups; i++) {
1327                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1328                 if (!gdp)
1329                         continue;
1330                 count += ext4_used_dirs_count(sb, gdp);
1331         }
1332         return count;
1333 }
1334
1335 /*
1336  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1337  * inode table. Must be called without any spinlock held. The only place
1338  * where it is called from on active part of filesystem is ext4lazyinit
1339  * thread, so we do not need any special locks, however we have to prevent
1340  * inode allocation from the current group, so we take alloc_sem lock, to
1341  * block ext4_new_inode() until we are finished.
1342  */
1343 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1344                                  int barrier)
1345 {
1346         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1347         struct ext4_sb_info *sbi = EXT4_SB(sb);
1348         struct ext4_group_desc *gdp = NULL;
1349         struct buffer_head *group_desc_bh;
1350         handle_t *handle;
1351         ext4_fsblk_t blk;
1352         int num, ret = 0, used_blks = 0;
1353
1354         /* This should not happen, but just to be sure check this */
1355         if (sb_rdonly(sb)) {
1356                 ret = 1;
1357                 goto out;
1358         }
1359
1360         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1361         if (!gdp)
1362                 goto out;
1363
1364         /*
1365          * We do not need to lock this, because we are the only one
1366          * handling this flag.
1367          */
1368         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1369                 goto out;
1370
1371         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1372         if (IS_ERR(handle)) {
1373                 ret = PTR_ERR(handle);
1374                 goto out;
1375         }
1376
1377         down_write(&grp->alloc_sem);
1378         /*
1379          * If inode bitmap was already initialized there may be some
1380          * used inodes so we need to skip blocks with used inodes in
1381          * inode table.
1382          */
1383         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1384                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1385                             ext4_itable_unused_count(sb, gdp)),
1386                             sbi->s_inodes_per_block);
1387
1388         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) ||
1389             ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) -
1390                                ext4_itable_unused_count(sb, gdp)) <
1391                               EXT4_FIRST_INO(sb)))) {
1392                 ext4_error(sb, "Something is wrong with group %u: "
1393                            "used itable blocks: %d; "
1394                            "itable unused count: %u",
1395                            group, used_blks,
1396                            ext4_itable_unused_count(sb, gdp));
1397                 ret = 1;
1398                 goto err_out;
1399         }
1400
1401         blk = ext4_inode_table(sb, gdp) + used_blks;
1402         num = sbi->s_itb_per_group - used_blks;
1403
1404         BUFFER_TRACE(group_desc_bh, "get_write_access");
1405         ret = ext4_journal_get_write_access(handle,
1406                                             group_desc_bh);
1407         if (ret)
1408                 goto err_out;
1409
1410         /*
1411          * Skip zeroout if the inode table is full. But we set the ZEROED
1412          * flag anyway, because obviously, when it is full it does not need
1413          * further zeroing.
1414          */
1415         if (unlikely(num == 0))
1416                 goto skip_zeroout;
1417
1418         ext4_debug("going to zero out inode table in group %d\n",
1419                    group);
1420         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1421         if (ret < 0)
1422                 goto err_out;
1423         if (barrier)
1424                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1425
1426 skip_zeroout:
1427         ext4_lock_group(sb, group);
1428         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1429         ext4_group_desc_csum_set(sb, group, gdp);
1430         ext4_unlock_group(sb, group);
1431
1432         BUFFER_TRACE(group_desc_bh,
1433                      "call ext4_handle_dirty_metadata");
1434         ret = ext4_handle_dirty_metadata(handle, NULL,
1435                                          group_desc_bh);
1436
1437 err_out:
1438         up_write(&grp->alloc_sem);
1439         ext4_journal_stop(handle);
1440 out:
1441         return ret;
1442 }