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