fs/ext4/ialloc.c

   1 /*
   2  *  linux/fs/ext4/ialloc.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  BSD ufs-inspired inode and directory allocation by
  10  *  Stephen Tweedie (sct@redhat.com), 1993
  11  *  Big-endian to little-endian byte-swapping/bitmaps by
  12  *        David S. Miller (davem@caip.rutgers.edu), 1995
  13  */
  14
  15 #include <linux/time.h>
  16 #include <linux/fs.h>
  17 #include <linux/stat.h>
  18 #include <linux/string.h>
  19 #include <linux/quotaops.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/random.h>
  22 #include <linux/bitops.h>
  23 #include <linux/blkdev.h>
  24 #include <linux/cred.h>
  25
  26 #include <asm/byteorder.h>
  27
  28 #include "ext4.h"
  29 #include "ext4_jbd2.h"
  30 #include "xattr.h"
  31 #include "acl.h"
  32
  33 #include <trace/events/ext4.h>
  34
  35 /*
  36  * ialloc.c contains the inodes allocation and deallocation routines
  37  */
  38
  39 /*
  40  * The free inodes are managed by bitmaps.  A file system contains several
  41  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  42  * block for inodes, N blocks for the inode table and data blocks.
  43  *
  44  * The file system contains group descriptors which are located after the
  45  * super block.  Each descriptor contains the number of the bitmap block and
  46  * the free blocks count in the block.
  47  */
  48
  49 /*
  50  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  51  * need to use it within a single byte (to ensure we get endianness right).
  52  * We can use memset for the rest of the bitmap as there are no other users.
  53  */
  54 void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  55 {
  56         int i;
  57
  58         if (start_bit >= end_bit)
  59                 return;
  60
  61         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  62         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  63                 ext4_set_bit(i, bitmap);
  64         if (i < end_bit)
  65                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  66 }
  67
  68 /* Initializes an uninitialized inode bitmap */
  69 static int ext4_init_inode_bitmap(struct super_block *sb,
  70                                        struct buffer_head *bh,
  71                                        ext4_group_t block_group,
  72                                        struct ext4_group_desc *gdp)
  73 {
  74         struct ext4_group_info *grp;
  75         struct ext4_sb_info *sbi = EXT4_SB(sb);
  76         J_ASSERT_BH(bh, buffer_locked(bh));
  77
  78         /* If checksum is bad mark all blocks and inodes use to prevent
  79          * allocation, essentially implementing a per-group read-only flag. */
  80         if (!ext4_group_desc_csum_verify(sb, block_group, gdp)) {
  81                 grp = ext4_get_group_info(sb, block_group);
  82                 if (!EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
  83                         percpu_counter_sub(&sbi->s_freeclusters_counter,
  84                                            grp->bb_free);
  85                 set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT, &grp->bb_state);
  86                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
  87                         int count;
  88                         count = ext4_free_inodes_count(sb, gdp);
  89                         percpu_counter_sub(&sbi->s_freeinodes_counter,
  90                                            count);
  91                 }
  92                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
  93                 return -EFSBADCRC;
  94         }
  95
  96         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  97         ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  98                         bh->b_data);
  99         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bh,
 100                                    EXT4_INODES_PER_GROUP(sb) / 8);
 101         ext4_group_desc_csum_set(sb, block_group, gdp);
 102
 103         return 0;
 104 }
 105
 106 void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate)
 107 {
 108         if (uptodate) {
 109                 set_buffer_uptodate(bh);
 110                 set_bitmap_uptodate(bh);
 111         }
 112         unlock_buffer(bh);
 113         put_bh(bh);
 114 }
 115
 116 static int ext4_validate_inode_bitmap(struct super_block *sb,
 117                                       struct ext4_group_desc *desc,
 118                                       ext4_group_t block_group,
 119                                       struct buffer_head *bh)
 120 {
 121         ext4_fsblk_t    blk;
 122         struct ext4_group_info *grp = ext4_get_group_info(sb, block_group);
 123         struct ext4_sb_info *sbi = EXT4_SB(sb);
 124
 125         if (buffer_verified(bh))
 126                 return 0;
 127         if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 128                 return -EFSCORRUPTED;
 129
 130         ext4_lock_group(sb, block_group);
 131         blk = ext4_inode_bitmap(sb, desc);
 132         if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh,
 133                                            EXT4_INODES_PER_GROUP(sb) / 8)) {
 134                 ext4_unlock_group(sb, block_group);
 135                 ext4_error(sb, "Corrupt inode bitmap - block_group = %u, "
 136                            "inode_bitmap = %llu", block_group, blk);
 137                 grp = ext4_get_group_info(sb, block_group);
 138                 if (!EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 139                         int count;
 140                         count = ext4_free_inodes_count(sb, desc);
 141                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 142                                            count);
 143                 }
 144                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 145                 return -EFSBADCRC;
 146         }
 147         set_buffer_verified(bh);
 148         ext4_unlock_group(sb, block_group);
 149         return 0;
 150 }
 151
 152 /*
 153  * Read the inode allocation bitmap for a given block_group, reading
 154  * into the specified slot in the superblock's bitmap cache.
 155  *
 156  * Return buffer_head of bitmap on success or NULL.
 157  */
 158 static struct buffer_head *
 159 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 160 {
 161         struct ext4_group_desc *desc;
 162         struct buffer_head *bh = NULL;
 163         ext4_fsblk_t bitmap_blk;
 164         int err;
 165
 166         desc = ext4_get_group_desc(sb, block_group, NULL);
 167         if (!desc)
 168                 return ERR_PTR(-EFSCORRUPTED);
 169
 170         bitmap_blk = ext4_inode_bitmap(sb, desc);
 171         bh = sb_getblk(sb, bitmap_blk);
 172         if (unlikely(!bh)) {
 173                 ext4_error(sb, "Cannot read inode bitmap - "
 174                             "block_group = %u, inode_bitmap = %llu",
 175                             block_group, bitmap_blk);
 176                 return ERR_PTR(-EIO);
 177         }
 178         if (bitmap_uptodate(bh))
 179                 goto verify;
 180
 181         lock_buffer(bh);
 182         if (bitmap_uptodate(bh)) {
 183                 unlock_buffer(bh);
 184                 goto verify;
 185         }
 186
 187         ext4_lock_group(sb, block_group);
 188         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 189                 err = ext4_init_inode_bitmap(sb, bh, block_group, desc);
 190                 set_bitmap_uptodate(bh);
 191                 set_buffer_uptodate(bh);
 192                 set_buffer_verified(bh);
 193                 ext4_unlock_group(sb, block_group);
 194                 unlock_buffer(bh);
 195                 if (err) {
 196                         ext4_error(sb, "Failed to init inode bitmap for group "
 197                                    "%u: %d", block_group, err);
 198                         goto out;
 199                 }
 200                 return bh;
 201         }
 202         ext4_unlock_group(sb, block_group);
 203
 204         if (buffer_uptodate(bh)) {
 205                 /*
 206                  * if not uninit if bh is uptodate,
 207                  * bitmap is also uptodate
 208                  */
 209                 set_bitmap_uptodate(bh);
 210                 unlock_buffer(bh);
 211                 goto verify;
 212         }
 213         /*
 214          * submit the buffer_head for reading
 215          */
 216         trace_ext4_load_inode_bitmap(sb, block_group);
 217         bh->b_end_io = ext4_end_bitmap_read;
 218         get_bh(bh);
 219         submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
 220         wait_on_buffer(bh);
 221         if (!buffer_uptodate(bh)) {
 222                 put_bh(bh);
 223                 ext4_error(sb, "Cannot read inode bitmap - "
 224                            "block_group = %u, inode_bitmap = %llu",
 225                            block_group, bitmap_blk);
 226                 return ERR_PTR(-EIO);
 227         }
 228
 229 verify:
 230         err = ext4_validate_inode_bitmap(sb, desc, block_group, bh);
 231         if (err)
 232                 goto out;
 233         return bh;
 234 out:
 235         put_bh(bh);
 236         return ERR_PTR(err);
 237 }
 238
 239 /*
 240  * NOTE! When we get the inode, we're the only people
 241  * that have access to it, and as such there are no
 242  * race conditions we have to worry about. The inode
 243  * is not on the hash-lists, and it cannot be reached
 244  * through the filesystem because the directory entry
 245  * has been deleted earlier.
 246  *
 247  * HOWEVER: we must make sure that we get no aliases,
 248  * which means that we have to call "clear_inode()"
 249  * _before_ we mark the inode not in use in the inode
 250  * bitmaps. Otherwise a newly created file might use
 251  * the same inode number (not actually the same pointer
 252  * though), and then we'd have two inodes sharing the
 253  * same inode number and space on the harddisk.
 254  */
 255 void ext4_free_inode(handle_t *handle, struct inode *inode)
 256 {
 257         struct super_block *sb = inode->i_sb;
 258         int is_directory;
 259         unsigned long ino;
 260         struct buffer_head *bitmap_bh = NULL;
 261         struct buffer_head *bh2;
 262         ext4_group_t block_group;
 263         unsigned long bit;
 264         struct ext4_group_desc *gdp;
 265         struct ext4_super_block *es;
 266         struct ext4_sb_info *sbi;
 267         int fatal = 0, err, count, cleared;
 268         struct ext4_group_info *grp;
 269
 270         if (!sb) {
 271                 printk(KERN_ERR "EXT4-fs: %s:%d: inode on "
 272                        "nonexistent device\n", __func__, __LINE__);
 273                 return;
 274         }
 275         if (atomic_read(&inode->i_count) > 1) {
 276                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d",
 277                          __func__, __LINE__, inode->i_ino,
 278                          atomic_read(&inode->i_count));
 279                 return;
 280         }
 281         if (inode->i_nlink) {
 282                 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n",
 283                          __func__, __LINE__, inode->i_ino, inode->i_nlink);
 284                 return;
 285         }
 286         sbi = EXT4_SB(sb);
 287
 288         ino = inode->i_ino;
 289         ext4_debug("freeing inode %lu\n", ino);
 290         trace_ext4_free_inode(inode);
 291
 292         /*
 293          * Note: we must free any quota before locking the superblock,
 294          * as writing the quota to disk may need the lock as well.
 295          */
 296         dquot_initialize(inode);
 297         dquot_free_inode(inode);
 298         dquot_drop(inode);
 299
 300         is_directory = S_ISDIR(inode->i_mode);
 301
 302         /* Do this BEFORE marking the inode not in use or returning an error */
 303         ext4_clear_inode(inode);
 304
 305         es = EXT4_SB(sb)->s_es;
 306         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 307                 ext4_error(sb, "reserved or nonexistent inode %lu", ino);
 308                 goto error_return;
 309         }
 310         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 311         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 312         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 313         /* Don't bother if the inode bitmap is corrupt. */
 314         grp = ext4_get_group_info(sb, block_group);
 315         if (IS_ERR(bitmap_bh)) {
 316                 fatal = PTR_ERR(bitmap_bh);
 317                 bitmap_bh = NULL;
 318                 goto error_return;
 319         }
 320         if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
 321                 fatal = -EFSCORRUPTED;
 322                 goto error_return;
 323         }
 324
 325         BUFFER_TRACE(bitmap_bh, "get_write_access");
 326         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 327         if (fatal)
 328                 goto error_return;
 329
 330         fatal = -ESRCH;
 331         gdp = ext4_get_group_desc(sb, block_group, &bh2);
 332         if (gdp) {
 333                 BUFFER_TRACE(bh2, "get_write_access");
 334                 fatal = ext4_journal_get_write_access(handle, bh2);
 335         }
 336         ext4_lock_group(sb, block_group);
 337         cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data);
 338         if (fatal || !cleared) {
 339                 ext4_unlock_group(sb, block_group);
 340                 goto out;
 341         }
 342
 343         count = ext4_free_inodes_count(sb, gdp) + 1;
 344         ext4_free_inodes_set(sb, gdp, count);
 345         if (is_directory) {
 346                 count = ext4_used_dirs_count(sb, gdp) - 1;
 347                 ext4_used_dirs_set(sb, gdp, count);
 348                 percpu_counter_dec(&sbi->s_dirs_counter);
 349         }
 350         ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh,
 351                                    EXT4_INODES_PER_GROUP(sb) / 8);
 352         ext4_group_desc_csum_set(sb, block_group, gdp);
 353         ext4_unlock_group(sb, block_group);
 354
 355         percpu_counter_inc(&sbi->s_freeinodes_counter);
 356         if (sbi->s_log_groups_per_flex) {
 357                 ext4_group_t f = ext4_flex_group(sbi, block_group);
 358
 359                 atomic_inc(&sbi->s_flex_groups[f].free_inodes);
 360                 if (is_directory)
 361                         atomic_dec(&sbi->s_flex_groups[f].used_dirs);
 362         }
 363         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 364         fatal = ext4_handle_dirty_metadata(handle, NULL, bh2);
 365 out:
 366         if (cleared) {
 367                 BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata");
 368                 err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh);
 369                 if (!fatal)
 370                         fatal = err;
 371         } else {
 372                 ext4_error(sb, "bit already cleared for inode %lu", ino);
 373                 if (gdp && !EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) {
 374                         int count;
 375                         count = ext4_free_inodes_count(sb, gdp);
 376                         percpu_counter_sub(&sbi->s_freeinodes_counter,
 377                                            count);
 378                 }
 379                 set_bit(EXT4_GROUP_INFO_IBITMAP_CORRUPT_BIT, &grp->bb_state);
 380         }
 381
 382 error_return:
 383         brelse(bitmap_bh);
 384         ext4_std_error(sb, fatal);
 385 }
 386
 387 struct orlov_stats {
 388         __u64 free_clusters;
 389         __u32 free_inodes;
 390         __u32 used_dirs;
 391 };
 392
 393 /*
 394  * Helper function for Orlov's allocator; returns critical information
 395  * for a particular block group or flex_bg.  If flex_size is 1, then g
 396  * is a block group number; otherwise it is flex_bg number.
 397  */
 398 static void get_orlov_stats(struct super_block *sb, ext4_group_t g,
 399                             int flex_size, struct orlov_stats *stats)
 400 {
 401         struct ext4_group_desc *desc;
 402         struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups;
 403
 404         if (flex_size > 1) {
 405                 stats->free_inodes = atomic_read(&flex_group[g].free_inodes);
 406                 stats->free_clusters = atomic64_read(&flex_group[g].free_clusters);
 407                 stats->used_dirs = atomic_read(&flex_group[g].used_dirs);
 408                 return;
 409         }
 410
 411         desc = ext4_get_group_desc(sb, g, NULL);
 412         if (desc) {
 413                 stats->free_inodes = ext4_free_inodes_count(sb, desc);
 414                 stats->free_clusters = ext4_free_group_clusters(sb, desc);
 415                 stats->used_dirs = ext4_used_dirs_count(sb, desc);
 416         } else {
 417                 stats->free_inodes = 0;
 418                 stats->free_clusters = 0;
 419                 stats->used_dirs = 0;
 420         }
 421 }
 422
 423 /*
 424  * Orlov's allocator for directories.
 425  *
 426  * We always try to spread first-level directories.
 427  *
 428  * If there are blockgroups with both free inodes and free blocks counts
 429  * not worse than average we return one with smallest directory count.
 430  * Otherwise we simply return a random group.
 431  *
 432  * For the rest rules look so:
 433  *
 434  * It's OK to put directory into a group unless
 435  * it has too many directories already (max_dirs) or
 436  * it has too few free inodes left (min_inodes) or
 437  * it has too few free blocks left (min_blocks) or
 438  * Parent's group is preferred, if it doesn't satisfy these
 439  * conditions we search cyclically through the rest. If none
 440  * of the groups look good we just look for a group with more
 441  * free inodes than average (starting at parent's group).
 442  */
 443
 444 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 445                             ext4_group_t *group, umode_t mode,
 446                             const struct qstr *qstr)
 447 {
 448         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 449         struct ext4_sb_info *sbi = EXT4_SB(sb);
 450         ext4_group_t real_ngroups = ext4_get_groups_count(sb);
 451         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 452         unsigned int freei, avefreei, grp_free;
 453         ext4_fsblk_t freeb, avefreec;
 454         unsigned int ndirs;
 455         int max_dirs, min_inodes;
 456         ext4_grpblk_t min_clusters;
 457         ext4_group_t i, grp, g, ngroups;
 458         struct ext4_group_desc *desc;
 459         struct orlov_stats stats;
 460         int flex_size = ext4_flex_bg_size(sbi);
 461         struct dx_hash_info hinfo;
 462
 463         ngroups = real_ngroups;
 464         if (flex_size > 1) {
 465                 ngroups = (real_ngroups + flex_size - 1) >>
 466                         sbi->s_log_groups_per_flex;
 467                 parent_group >>= sbi->s_log_groups_per_flex;
 468         }
 469
 470         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 471         avefreei = freei / ngroups;
 472         freeb = EXT4_C2B(sbi,
 473                 percpu_counter_read_positive(&sbi->s_freeclusters_counter));
 474         avefreec = freeb;
 475         do_div(avefreec, ngroups);
 476         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 477
 478         if (S_ISDIR(mode) &&
 479             ((parent == d_inode(sb->s_root)) ||
 480              (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) {
 481                 int best_ndir = inodes_per_group;
 482                 int ret = -1;
 483
 484                 if (qstr) {
 485                         hinfo.hash_version = DX_HASH_HALF_MD4;
 486                         hinfo.seed = sbi->s_hash_seed;
 487                         ext4fs_dirhash(qstr->name, qstr->len, &hinfo);
 488                         grp = hinfo.hash;
 489                 } else
 490                         grp = prandom_u32();
 491                 parent_group = (unsigned)grp % ngroups;
 492                 for (i = 0; i < ngroups; i++) {
 493                         g = (parent_group + i) % ngroups;
 494                         get_orlov_stats(sb, g, flex_size, &stats);
 495                         if (!stats.free_inodes)
 496                                 continue;
 497                         if (stats.used_dirs >= best_ndir)
 498                                 continue;
 499                         if (stats.free_inodes < avefreei)
 500                                 continue;
 501                         if (stats.free_clusters < avefreec)
 502                                 continue;
 503                         grp = g;
 504                         ret = 0;
 505                         best_ndir = stats.used_dirs;
 506                 }
 507                 if (ret)
 508                         goto fallback;
 509         found_flex_bg:
 510                 if (flex_size == 1) {
 511                         *group = grp;
 512                         return 0;
 513                 }
 514
 515                 /*
 516                  * We pack inodes at the beginning of the flexgroup's
 517                  * inode tables.  Block allocation decisions will do
 518                  * something similar, although regular files will
 519                  * start at 2nd block group of the flexgroup.  See
 520                  * ext4_ext_find_goal() and ext4_find_near().
 521                  */
 522                 grp *= flex_size;
 523                 for (i = 0; i < flex_size; i++) {
 524                         if (grp+i >= real_ngroups)
 525                                 break;
 526                         desc = ext4_get_group_desc(sb, grp+i, NULL);
 527                         if (desc && ext4_free_inodes_count(sb, desc)) {
 528                                 *group = grp+i;
 529                                 return 0;
 530                         }
 531                 }
 532                 goto fallback;
 533         }
 534
 535         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 536         min_inodes = avefreei - inodes_per_group*flex_size / 4;
 537         if (min_inodes < 1)
 538                 min_inodes = 1;
 539         min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4;
 540
 541         /*
 542          * Start looking in the flex group where we last allocated an
 543          * inode for this parent directory
 544          */
 545         if (EXT4_I(parent)->i_last_alloc_group != ~0) {
 546                 parent_group = EXT4_I(parent)->i_last_alloc_group;
 547                 if (flex_size > 1)
 548                         parent_group >>= sbi->s_log_groups_per_flex;
 549         }
 550
 551         for (i = 0; i < ngroups; i++) {
 552                 grp = (parent_group + i) % ngroups;
 553                 get_orlov_stats(sb, grp, flex_size, &stats);
 554                 if (stats.used_dirs >= max_dirs)
 555                         continue;
 556                 if (stats.free_inodes < min_inodes)
 557                         continue;
 558                 if (stats.free_clusters < min_clusters)
 559                         continue;
 560                 goto found_flex_bg;
 561         }
 562
 563 fallback:
 564         ngroups = real_ngroups;
 565         avefreei = freei / ngroups;
 566 fallback_retry:
 567         parent_group = EXT4_I(parent)->i_block_group;
 568         for (i = 0; i < ngroups; i++) {
 569                 grp = (parent_group + i) % ngroups;
 570                 desc = ext4_get_group_desc(sb, grp, NULL);
 571                 if (desc) {
 572                         grp_free = ext4_free_inodes_count(sb, desc);
 573                         if (grp_free && grp_free >= avefreei) {
 574                                 *group = grp;
 575                                 return 0;
 576                         }
 577                 }
 578         }
 579
 580         if (avefreei) {
 581                 /*
 582                  * The free-inodes counter is approximate, and for really small
 583                  * filesystems the above test can fail to find any blockgroups
 584                  */
 585                 avefreei = 0;
 586                 goto fallback_retry;
 587         }
 588
 589         return -1;
 590 }
 591
 592 static int find_group_other(struct super_block *sb, struct inode *parent,
 593                             ext4_group_t *group, umode_t mode)
 594 {
 595         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 596         ext4_group_t i, last, ngroups = ext4_get_groups_count(sb);
 597         struct ext4_group_desc *desc;
 598         int flex_size = ext4_flex_bg_size(EXT4_SB(sb));
 599
 600         /*
 601          * Try to place the inode is the same flex group as its
 602          * parent.  If we can't find space, use the Orlov algorithm to
 603          * find another flex group, and store that information in the
 604          * parent directory's inode information so that use that flex
 605          * group for future allocations.
 606          */
 607         if (flex_size > 1) {
 608                 int retry = 0;
 609
 610         try_again:
 611                 parent_group &= ~(flex_size-1);
 612                 last = parent_group + flex_size;
 613                 if (last > ngroups)
 614                         last = ngroups;
 615                 for  (i = parent_group; i < last; i++) {
 616                         desc = ext4_get_group_desc(sb, i, NULL);
 617                         if (desc && ext4_free_inodes_count(sb, desc)) {
 618                                 *group = i;
 619                                 return 0;
 620                         }
 621                 }
 622                 if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) {
 623                         retry = 1;
 624                         parent_group = EXT4_I(parent)->i_last_alloc_group;
 625                         goto try_again;
 626                 }
 627                 /*
 628                  * If this didn't work, use the Orlov search algorithm
 629                  * to find a new flex group; we pass in the mode to
 630                  * avoid the topdir algorithms.
 631                  */
 632                 *group = parent_group + flex_size;
 633                 if (*group > ngroups)
 634                         *group = 0;
 635                 return find_group_orlov(sb, parent, group, mode, NULL);
 636         }
 637
 638         /*
 639          * Try to place the inode in its parent directory
 640          */
 641         *group = parent_group;
 642         desc = ext4_get_group_desc(sb, *group, NULL);
 643         if (desc && ext4_free_inodes_count(sb, desc) &&
 644             ext4_free_group_clusters(sb, desc))
 645                 return 0;
 646
 647         /*
 648          * We're going to place this inode in a different blockgroup from its
 649          * parent.  We want to cause files in a common directory to all land in
 650          * the same blockgroup.  But we want files which are in a different
 651          * directory which shares a blockgroup with our parent to land in a
 652          * different blockgroup.
 653          *
 654          * So add our directory's i_ino into the starting point for the hash.
 655          */
 656         *group = (*group + parent->i_ino) % ngroups;
 657
 658         /*
 659          * Use a quadratic hash to find a group with a free inode and some free
 660          * blocks.
 661          */
 662         for (i = 1; i < ngroups; i <<= 1) {
 663                 *group += i;
 664                 if (*group >= ngroups)
 665                         *group -= ngroups;
 666                 desc = ext4_get_group_desc(sb, *group, NULL);
 667                 if (desc && ext4_free_inodes_count(sb, desc) &&
 668                     ext4_free_group_clusters(sb, desc))
 669                         return 0;
 670         }
 671
 672         /*
 673          * That failed: try linear search for a free inode, even if that group
 674          * has no free blocks.
 675          */
 676         *group = parent_group;
 677         for (i = 0; i < ngroups; i++) {
 678                 if (++*group >= ngroups)
 679                         *group = 0;
 680                 desc = ext4_get_group_desc(sb, *group, NULL);
 681                 if (desc && ext4_free_inodes_count(sb, desc))
 682                         return 0;
 683         }
 684
 685         return -1;
 686 }
 687
 688 /*
 689  * In no journal mode, if an inode has recently been deleted, we want
 690  * to avoid reusing it until we're reasonably sure the inode table
 691  * block has been written back to disk.  (Yes, these values are
 692  * somewhat arbitrary...)
 693  */
 694 #define RECENTCY_MIN    5
 695 #define RECENTCY_DIRTY  300
 696
 697 static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino)
 698 {
 699         struct ext4_group_desc  *gdp;
 700         struct ext4_inode       *raw_inode;
 701         struct buffer_head      *bh;
 702         int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
 703         int offset, ret = 0;
 704         int recentcy = RECENTCY_MIN;
 705         u32 dtime, now;
 706
 707         gdp = ext4_get_group_desc(sb, group, NULL);
 708         if (unlikely(!gdp))
 709                 return 0;
 710
 711         bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) +
 712                        (ino / inodes_per_block));
 713         if (!bh || !buffer_uptodate(bh))
 714                 /*
 715                  * If the block is not in the buffer cache, then it
 716                  * must have been written out.
 717                  */
 718                 goto out;
 719
 720         offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb);
 721         raw_inode = (struct ext4_inode *) (bh->b_data + offset);
 722
 723         /* i_dtime is only 32 bits on disk, but we only care about relative
 724          * times in the range of a few minutes (i.e. long enough to sync a
 725          * recently-deleted inode to disk), so using the low 32 bits of the
 726          * clock (a 68 year range) is enough, see time_before32() */
 727         dtime = le32_to_cpu(raw_inode->i_dtime);
 728         now = ktime_get_real_seconds();
 729         if (buffer_dirty(bh))
 730                 recentcy += RECENTCY_DIRTY;
 731
 732         if (dtime && time_before32(dtime, now) &&
 733             time_before32(now, dtime + recentcy))
 734                 ret = 1;
 735 out:
 736         brelse(bh);
 737         return ret;
 738 }
 739
 740 static int find_inode_bit(struct super_block *sb, ext4_group_t group,
 741                           struct buffer_head *bitmap, unsigned long *ino)
 742 {
 743 next:
 744         *ino = ext4_find_next_zero_bit((unsigned long *)
 745                                        bitmap->b_data,
 746                                        EXT4_INODES_PER_GROUP(sb), *ino);
 747         if (*ino >= EXT4_INODES_PER_GROUP(sb))
 748                 return 0;
 749
 750         if ((EXT4_SB(sb)->s_journal == NULL) &&
 751             recently_deleted(sb, group, *ino)) {
 752                 *ino = *ino + 1;
 753                 if (*ino < EXT4_INODES_PER_GROUP(sb))
 754                         goto next;
 755                 return 0;
 756         }
 757
 758         return 1;
 759 }
 760
 761 /*
 762  * There are two policies for allocating an inode.  If the new inode is
 763  * a directory, then a forward search is made for a block group with both
 764  * free space and a low directory-to-inode ratio; if that fails, then of
 765  * the groups with above-average free space, that group with the fewest
 766  * directories already is chosen.
 767  *
 768  * For other inodes, search forward from the parent directory's block
 769  * group to find a free inode.
 770  */
 771 struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir,
 772                                umode_t mode, const struct qstr *qstr,
 773                                __u32 goal, uid_t *owner, __u32 i_flags,
 774                                int handle_type, unsigned int line_no,
 775                                int nblocks)
 776 {
 777         struct super_block *sb;
 778         struct buffer_head *inode_bitmap_bh = NULL;
 779         struct buffer_head *group_desc_bh;
 780         ext4_group_t ngroups, group = 0;
 781         unsigned long ino = 0;
 782         struct inode *inode;
 783         struct ext4_group_desc *gdp = NULL;
 784         struct ext4_inode_info *ei;
 785         struct ext4_sb_info *sbi;
 786         int ret2, err;
 787         struct inode *ret;
 788         ext4_group_t i;
 789         ext4_group_t flex_group;
 790         struct ext4_group_info *grp;
 791         int encrypt = 0;
 792
 793         /* Cannot create files in a deleted directory */
 794         if (!dir || !dir->i_nlink)
 795                 return ERR_PTR(-EPERM);
 796
 797         sb = dir->i_sb;
 798         sbi = EXT4_SB(sb);
 799
 800         if (unlikely(ext4_forced_shutdown(sbi)))
 801                 return ERR_PTR(-EIO);
 802
 803         if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) &&
 804             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) &&
 805             !(i_flags & EXT4_EA_INODE_FL)) {
 806                 err = fscrypt_get_encryption_info(dir);
 807                 if (err)
 808                         return ERR_PTR(err);
 809                 if (!fscrypt_has_encryption_key(dir))
 810                         return ERR_PTR(-ENOKEY);
 811                 encrypt = 1;
 812         }
 813
 814         if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) {
 815 #ifdef CONFIG_EXT4_FS_POSIX_ACL
 816                 struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT);
 817
 818                 if (p) {
 819                         int acl_size = p->a_count * sizeof(ext4_acl_entry);
 820
 821                         nblocks += (S_ISDIR(mode) ? 2 : 1) *
 822                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 823                                         NULL /* block_bh */, acl_size,
 824                                         true /* is_create */);
 825                         posix_acl_release(p);
 826                 }
 827 #endif
 828
 829 #ifdef CONFIG_SECURITY
 830                 {
 831                         int num_security_xattrs = 1;
 832
 833 #ifdef CONFIG_INTEGRITY
 834                         num_security_xattrs++;
 835 #endif
 836                         /*
 837                          * We assume that security xattrs are never
 838                          * more than 1k.  In practice they are under
 839                          * 128 bytes.
 840                          */
 841                         nblocks += num_security_xattrs *
 842                                 __ext4_xattr_set_credits(sb, NULL /* inode */,
 843                                         NULL /* block_bh */, 1024,
 844                                         true /* is_create */);
 845                 }
 846 #endif
 847                 if (encrypt)
 848                         nblocks += __ext4_xattr_set_credits(sb,
 849                                         NULL /* inode */, NULL /* block_bh */,
 850                                         FSCRYPT_SET_CONTEXT_MAX_SIZE,
 851                                         true /* is_create */);
 852         }
 853
 854         ngroups = ext4_get_groups_count(sb);
 855         trace_ext4_request_inode(dir, mode);
 856         inode = new_inode(sb);
 857         if (!inode)
 858                 return ERR_PTR(-ENOMEM);
 859         ei = EXT4_I(inode);
 860
 861         /*
 862          * Initialize owners and quota early so that we don't have to account
 863          * for quota initialization worst case in standard inode creating
 864          * transaction
 865          */
 866         if (owner) {
 867                 inode->i_mode = mode;
 868                 i_uid_write(inode, owner[0]);
 869                 i_gid_write(inode, owner[1]);
 870         } else if (test_opt(sb, GRPID)) {
 871                 inode->i_mode = mode;
 872                 inode->i_uid = current_fsuid();
 873                 inode->i_gid = dir->i_gid;
 874         } else
 875                 inode_init_owner(inode, dir, mode);
 876
 877         if (ext4_has_feature_project(sb) &&
 878             ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT))
 879                 ei->i_projid = EXT4_I(dir)->i_projid;
 880         else
 881                 ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID);
 882
 883         err = dquot_initialize(inode);
 884         if (err)
 885                 goto out;
 886
 887         if (!goal)
 888                 goal = sbi->s_inode_goal;
 889
 890         if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) {
 891                 group = (goal - 1) / EXT4_INODES_PER_GROUP(sb);
 892                 ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb);
 893                 ret2 = 0;
 894                 goto got_group;
 895         }
 896
 897         if (S_ISDIR(mode))
 898                 ret2 = find_group_orlov(sb, dir, &group, mode, qstr);
 899         else
 900                 ret2 = find_group_other(sb, dir, &group, mode);
 901
 902 got_group:
 903         EXT4_I(dir)->i_last_alloc_group = group;
 904         err = -ENOSPC;
 905         if (ret2 == -1)
 906                 goto out;
 907
 908         /*
 909          * Normally we will only go through one pass of this loop,
 910          * unless we get unlucky and it turns out the group we selected
 911          * had its last inode grabbed by someone else.
 912          */
 913         for (i = 0; i < ngroups; i++, ino = 0) {
 914                 err = -EIO;
 915
 916                 gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
 917                 if (!gdp)
 918                         goto out;
 919
 920                 /*
 921                  * Check free inodes count before loading bitmap.
 922                  */
 923                 if (ext4_free_inodes_count(sb, gdp) == 0)
 924                         goto next_group;
 925
 926                 grp = ext4_get_group_info(sb, group);
 927                 /* Skip groups with already-known suspicious inode tables */
 928                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
 929                         goto next_group;
 930
 931                 brelse(inode_bitmap_bh);
 932                 inode_bitmap_bh = ext4_read_inode_bitmap(sb, group);
 933                 /* Skip groups with suspicious inode tables */
 934                 if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) ||
 935                     IS_ERR(inode_bitmap_bh)) {
 936                         inode_bitmap_bh = NULL;
 937                         goto next_group;
 938                 }
 939
 940 repeat_in_this_group:
 941                 ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 942                 if (!ret2)
 943                         goto next_group;
 944
 945                 if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) {
 946                         ext4_error(sb, "reserved inode found cleared - "
 947                                    "inode=%lu", ino + 1);
 948                         goto next_group;
 949                 }
 950
 951                 if (!handle) {
 952                         BUG_ON(nblocks <= 0);
 953                         handle = __ext4_journal_start_sb(dir->i_sb, line_no,
 954                                                          handle_type, nblocks,
 955                                                          0);
 956                         if (IS_ERR(handle)) {
 957                                 err = PTR_ERR(handle);
 958                                 ext4_std_error(sb, err);
 959                                 goto out;
 960                         }
 961                 }
 962                 BUFFER_TRACE(inode_bitmap_bh, "get_write_access");
 963                 err = ext4_journal_get_write_access(handle, inode_bitmap_bh);
 964                 if (err) {
 965                         ext4_std_error(sb, err);
 966                         goto out;
 967                 }
 968                 ext4_lock_group(sb, group);
 969                 ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data);
 970                 if (ret2) {
 971                         /* Someone already took the bit. Repeat the search
 972                          * with lock held.
 973                          */
 974                         ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino);
 975                         if (ret2) {
 976                                 ext4_set_bit(ino, inode_bitmap_bh->b_data);
 977                                 ret2 = 0;
 978                         } else {
 979                                 ret2 = 1; /* we didn't grab the inode */
 980                         }
 981                 }
 982                 ext4_unlock_group(sb, group);
 983                 ino++;          /* the inode bitmap is zero-based */
 984                 if (!ret2)
 985                         goto got; /* we grabbed the inode! */
 986
 987                 if (ino < EXT4_INODES_PER_GROUP(sb))
 988                         goto repeat_in_this_group;
 989 next_group:
 990                 if (++group == ngroups)
 991                         group = 0;
 992         }
 993         err = -ENOSPC;
 994         goto out;
 995
 996 got:
 997         BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata");
 998         err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh);
 999         if (err) {
1000                 ext4_std_error(sb, err);
1001                 goto out;
1002         }
1003
1004         BUFFER_TRACE(group_desc_bh, "get_write_access");
1005         err = ext4_journal_get_write_access(handle, group_desc_bh);
1006         if (err) {
1007                 ext4_std_error(sb, err);
1008                 goto out;
1009         }
1010
1011         /* We may have to initialize the block bitmap if it isn't already */
1012         if (ext4_has_group_desc_csum(sb) &&
1013             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1014                 struct buffer_head *block_bitmap_bh;
1015
1016                 block_bitmap_bh = ext4_read_block_bitmap(sb, group);
1017                 if (IS_ERR(block_bitmap_bh)) {
1018                         err = PTR_ERR(block_bitmap_bh);
1019                         goto out;
1020                 }
1021                 BUFFER_TRACE(block_bitmap_bh, "get block bitmap access");
1022                 err = ext4_journal_get_write_access(handle, block_bitmap_bh);
1023                 if (err) {
1024                         brelse(block_bitmap_bh);
1025                         ext4_std_error(sb, err);
1026                         goto out;
1027                 }
1028
1029                 BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap");
1030                 err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh);
1031
1032                 /* recheck and clear flag under lock if we still need to */
1033                 ext4_lock_group(sb, group);
1034                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
1035                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
1036                         ext4_free_group_clusters_set(sb, gdp,
1037                                 ext4_free_clusters_after_init(sb, group, gdp));
1038                         ext4_block_bitmap_csum_set(sb, group, gdp,
1039                                                    block_bitmap_bh);
1040                         ext4_group_desc_csum_set(sb, group, gdp);
1041                 }
1042                 ext4_unlock_group(sb, group);
1043                 brelse(block_bitmap_bh);
1044
1045                 if (err) {
1046                         ext4_std_error(sb, err);
1047                         goto out;
1048                 }
1049         }
1050
1051         /* Update the relevant bg descriptor fields */
1052         if (ext4_has_group_desc_csum(sb)) {
1053                 int free;
1054                 struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1055
1056                 down_read(&grp->alloc_sem); /* protect vs itable lazyinit */
1057                 ext4_lock_group(sb, group); /* while we modify the bg desc */
1058                 free = EXT4_INODES_PER_GROUP(sb) -
1059                         ext4_itable_unused_count(sb, gdp);
1060                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
1061                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
1062                         free = 0;
1063                 }
1064                 /*
1065                  * Check the relative inode number against the last used
1066                  * relative inode number in this group. if it is greater
1067                  * we need to update the bg_itable_unused count
1068                  */
1069                 if (ino > free)
1070                         ext4_itable_unused_set(sb, gdp,
1071                                         (EXT4_INODES_PER_GROUP(sb) - ino));
1072                 up_read(&grp->alloc_sem);
1073         } else {
1074                 ext4_lock_group(sb, group);
1075         }
1076
1077         ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1);
1078         if (S_ISDIR(mode)) {
1079                 ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1);
1080                 if (sbi->s_log_groups_per_flex) {
1081                         ext4_group_t f = ext4_flex_group(sbi, group);
1082
1083                         atomic_inc(&sbi->s_flex_groups[f].used_dirs);
1084                 }
1085         }
1086         if (ext4_has_group_desc_csum(sb)) {
1087                 ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh,
1088                                            EXT4_INODES_PER_GROUP(sb) / 8);
1089                 ext4_group_desc_csum_set(sb, group, gdp);
1090         }
1091         ext4_unlock_group(sb, group);
1092
1093         BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata");
1094         err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh);
1095         if (err) {
1096                 ext4_std_error(sb, err);
1097                 goto out;
1098         }
1099
1100         percpu_counter_dec(&sbi->s_freeinodes_counter);
1101         if (S_ISDIR(mode))
1102                 percpu_counter_inc(&sbi->s_dirs_counter);
1103
1104         if (sbi->s_log_groups_per_flex) {
1105                 flex_group = ext4_flex_group(sbi, group);
1106                 atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes);
1107         }
1108
1109         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
1110         /* This is the optimal IO size (for stat), not the fs block size */
1111         inode->i_blocks = 0;
1112         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
1113                                                        current_time(inode);
1114
1115         memset(ei->i_data, 0, sizeof(ei->i_data));
1116         ei->i_dir_start_lookup = 0;
1117         ei->i_disksize = 0;
1118
1119         /* Don't inherit extent flag from directory, amongst others. */
1120         ei->i_flags =
1121                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
1122         ei->i_flags |= i_flags;
1123         ei->i_file_acl = 0;
1124         ei->i_dtime = 0;
1125         ei->i_block_group = group;
1126         ei->i_last_alloc_group = ~0;
1127
1128         ext4_set_inode_flags(inode);
1129         if (IS_DIRSYNC(inode))
1130                 ext4_handle_sync(handle);
1131         if (insert_inode_locked(inode) < 0) {
1132                 /*
1133                  * Likely a bitmap corruption causing inode to be allocated
1134                  * twice.
1135                  */
1136                 err = -EIO;
1137                 ext4_error(sb, "failed to insert inode %lu: doubly allocated?",
1138                            inode->i_ino);
1139                 goto out;
1140         }
1141         spin_lock(&sbi->s_next_gen_lock);
1142         inode->i_generation = sbi->s_next_generation++;
1143         spin_unlock(&sbi->s_next_gen_lock);
1144
1145         /* Precompute checksum seed for inode metadata */
1146         if (ext4_has_metadata_csum(sb)) {
1147                 __u32 csum;
1148                 __le32 inum = cpu_to_le32(inode->i_ino);
1149                 __le32 gen = cpu_to_le32(inode->i_generation);
1150                 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
1151                                    sizeof(inum));
1152                 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
1153                                               sizeof(gen));
1154         }
1155
1156         ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
1157         ext4_set_inode_state(inode, EXT4_STATE_NEW);
1158
1159         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
1160         ei->i_inline_off = 0;
1161         if (ext4_has_feature_inline_data(sb))
1162                 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
1163         ret = inode;
1164         err = dquot_alloc_inode(inode);
1165         if (err)
1166                 goto fail_drop;
1167
1168         /*
1169          * Since the encryption xattr will always be unique, create it first so
1170          * that it's less likely to end up in an external xattr block and
1171          * prevent its deduplication.
1172          */
1173         if (encrypt) {
1174                 err = fscrypt_inherit_context(dir, inode, handle, true);
1175                 if (err)
1176                         goto fail_free_drop;
1177         }
1178
1179         if (!(ei->i_flags & EXT4_EA_INODE_FL)) {
1180                 err = ext4_init_acl(handle, inode, dir);
1181                 if (err)
1182                         goto fail_free_drop;
1183
1184                 err = ext4_init_security(handle, inode, dir, qstr);
1185                 if (err)
1186                         goto fail_free_drop;
1187         }
1188
1189         if (ext4_has_feature_extents(sb)) {
1190                 /* set extent flag only for directory, file and normal symlink*/
1191                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
1192                         ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS);
1193                         ext4_ext_tree_init(handle, inode);
1194                 }
1195         }
1196
1197         if (ext4_handle_valid(handle)) {
1198                 ei->i_sync_tid = handle->h_transaction->t_tid;
1199                 ei->i_datasync_tid = handle->h_transaction->t_tid;
1200         }
1201
1202         err = ext4_mark_inode_dirty(handle, inode);
1203         if (err) {
1204                 ext4_std_error(sb, err);
1205                 goto fail_free_drop;
1206         }
1207
1208         ext4_debug("allocating inode %lu\n", inode->i_ino);
1209         trace_ext4_allocate_inode(inode, dir, mode);
1210         brelse(inode_bitmap_bh);
1211         return ret;
1212
1213 fail_free_drop:
1214         dquot_free_inode(inode);
1215 fail_drop:
1216         clear_nlink(inode);
1217         unlock_new_inode(inode);
1218 out:
1219         dquot_drop(inode);
1220         inode->i_flags |= S_NOQUOTA;
1221         iput(inode);
1222         brelse(inode_bitmap_bh);
1223         return ERR_PTR(err);
1224 }
1225
1226 /* Verify that we are loading a valid orphan from disk */
1227 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
1228 {
1229         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
1230         ext4_group_t block_group;
1231         int bit;
1232         struct buffer_head *bitmap_bh = NULL;
1233         struct inode *inode = NULL;
1234         int err = -EFSCORRUPTED;
1235
1236         if (ino < EXT4_FIRST_INO(sb) || ino > max_ino)
1237                 goto bad_orphan;
1238
1239         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
1240         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
1241         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
1242         if (IS_ERR(bitmap_bh)) {
1243                 ext4_error(sb, "inode bitmap error %ld for orphan %lu",
1244                            ino, PTR_ERR(bitmap_bh));
1245                 return (struct inode *) bitmap_bh;
1246         }
1247
1248         /* Having the inode bit set should be a 100% indicator that this
1249          * is a valid orphan (no e2fsck run on fs).  Orphans also include
1250          * inodes that were being truncated, so we can't check i_nlink==0.
1251          */
1252         if (!ext4_test_bit(bit, bitmap_bh->b_data))
1253                 goto bad_orphan;
1254
1255         inode = ext4_iget(sb, ino);
1256         if (IS_ERR(inode)) {
1257                 err = PTR_ERR(inode);
1258                 ext4_error(sb, "couldn't read orphan inode %lu (err %d)",
1259                            ino, err);
1260                 return inode;
1261         }
1262
1263         /*
1264          * If the orphans has i_nlinks > 0 then it should be able to
1265          * be truncated, otherwise it won't be removed from the orphan
1266          * list during processing and an infinite loop will result.
1267          * Similarly, it must not be a bad inode.
1268          */
1269         if ((inode->i_nlink && !ext4_can_truncate(inode)) ||
1270             is_bad_inode(inode))
1271                 goto bad_orphan;
1272
1273         if (NEXT_ORPHAN(inode) > max_ino)
1274                 goto bad_orphan;
1275         brelse(bitmap_bh);
1276         return inode;
1277
1278 bad_orphan:
1279         ext4_error(sb, "bad orphan inode %lu", ino);
1280         if (bitmap_bh)
1281                 printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1282                        bit, (unsigned long long)bitmap_bh->b_blocknr,
1283                        ext4_test_bit(bit, bitmap_bh->b_data));
1284         if (inode) {
1285                 printk(KERN_ERR "is_bad_inode(inode)=%d\n",
1286                        is_bad_inode(inode));
1287                 printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n",
1288                        NEXT_ORPHAN(inode));
1289                 printk(KERN_ERR "max_ino=%lu\n", max_ino);
1290                 printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink);
1291                 /* Avoid freeing blocks if we got a bad deleted inode */
1292                 if (inode->i_nlink == 0)
1293                         inode->i_blocks = 0;
1294                 iput(inode);
1295         }
1296         brelse(bitmap_bh);
1297         return ERR_PTR(err);
1298 }
1299
1300 unsigned long ext4_count_free_inodes(struct super_block *sb)
1301 {
1302         unsigned long desc_count;
1303         struct ext4_group_desc *gdp;
1304         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1305 #ifdef EXT4FS_DEBUG
1306         struct ext4_super_block *es;
1307         unsigned long bitmap_count, x;
1308         struct buffer_head *bitmap_bh = NULL;
1309
1310         es = EXT4_SB(sb)->s_es;
1311         desc_count = 0;
1312         bitmap_count = 0;
1313         gdp = NULL;
1314         for (i = 0; i < ngroups; i++) {
1315                 gdp = ext4_get_group_desc(sb, i, NULL);
1316                 if (!gdp)
1317                         continue;
1318                 desc_count += ext4_free_inodes_count(sb, gdp);
1319                 brelse(bitmap_bh);
1320                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1321                 if (IS_ERR(bitmap_bh)) {
1322                         bitmap_bh = NULL;
1323                         continue;
1324                 }
1325
1326                 x = ext4_count_free(bitmap_bh->b_data,
1327                                     EXT4_INODES_PER_GROUP(sb) / 8);
1328                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1329                         (unsigned long) i, ext4_free_inodes_count(sb, gdp), x);
1330                 bitmap_count += x;
1331         }
1332         brelse(bitmap_bh);
1333         printk(KERN_DEBUG "ext4_count_free_inodes: "
1334                "stored = %u, computed = %lu, %lu\n",
1335                le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1336         return desc_count;
1337 #else
1338         desc_count = 0;
1339         for (i = 0; i < ngroups; i++) {
1340                 gdp = ext4_get_group_desc(sb, i, NULL);
1341                 if (!gdp)
1342                         continue;
1343                 desc_count += ext4_free_inodes_count(sb, gdp);
1344                 cond_resched();
1345         }
1346         return desc_count;
1347 #endif
1348 }
1349
1350 /* Called at mount-time, super-block is locked */
1351 unsigned long ext4_count_dirs(struct super_block * sb)
1352 {
1353         unsigned long count = 0;
1354         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
1355
1356         for (i = 0; i < ngroups; i++) {
1357                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1358                 if (!gdp)
1359                         continue;
1360                 count += ext4_used_dirs_count(sb, gdp);
1361         }
1362         return count;
1363 }
1364
1365 /*
1366  * Zeroes not yet zeroed inode table - just write zeroes through the whole
1367  * inode table. Must be called without any spinlock held. The only place
1368  * where it is called from on active part of filesystem is ext4lazyinit
1369  * thread, so we do not need any special locks, however we have to prevent
1370  * inode allocation from the current group, so we take alloc_sem lock, to
1371  * block ext4_new_inode() until we are finished.
1372  */
1373 int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
1374                                  int barrier)
1375 {
1376         struct ext4_group_info *grp = ext4_get_group_info(sb, group);
1377         struct ext4_sb_info *sbi = EXT4_SB(sb);
1378         struct ext4_group_desc *gdp = NULL;
1379         struct buffer_head *group_desc_bh;
1380         handle_t *handle;
1381         ext4_fsblk_t blk;
1382         int num, ret = 0, used_blks = 0;
1383
1384         /* This should not happen, but just to be sure check this */
1385         if (sb_rdonly(sb)) {
1386                 ret = 1;
1387                 goto out;
1388         }
1389
1390         gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
1391         if (!gdp)
1392                 goto out;
1393
1394         /*
1395          * We do not need to lock this, because we are the only one
1396          * handling this flag.
1397          */
1398         if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED))
1399                 goto out;
1400
1401         handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1);
1402         if (IS_ERR(handle)) {
1403                 ret = PTR_ERR(handle);
1404                 goto out;
1405         }
1406
1407         down_write(&grp->alloc_sem);
1408         /*
1409          * If inode bitmap was already initialized there may be some
1410          * used inodes so we need to skip blocks with used inodes in
1411          * inode table.
1412          */
1413         if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)))
1414                 used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) -
1415                             ext4_itable_unused_count(sb, gdp)),
1416                             sbi->s_inodes_per_block);
1417
1418         if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) {
1419                 ext4_error(sb, "Something is wrong with group %u: "
1420                            "used itable blocks: %d; "
1421                            "itable unused count: %u",
1422                            group, used_blks,
1423                            ext4_itable_unused_count(sb, gdp));
1424                 ret = 1;
1425                 goto err_out;
1426         }
1427
1428         blk = ext4_inode_table(sb, gdp) + used_blks;
1429         num = sbi->s_itb_per_group - used_blks;
1430
1431         BUFFER_TRACE(group_desc_bh, "get_write_access");
1432         ret = ext4_journal_get_write_access(handle,
1433                                             group_desc_bh);
1434         if (ret)
1435                 goto err_out;
1436
1437         /*
1438          * Skip zeroout if the inode table is full. But we set the ZEROED
1439          * flag anyway, because obviously, when it is full it does not need
1440          * further zeroing.
1441          */
1442         if (unlikely(num == 0))
1443                 goto skip_zeroout;
1444
1445         ext4_debug("going to zero out inode table in group %d\n",
1446                    group);
1447         ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS);
1448         if (ret < 0)
1449                 goto err_out;
1450         if (barrier)
1451                 blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL);
1452
1453 skip_zeroout:
1454         ext4_lock_group(sb, group);
1455         gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED);
1456         ext4_group_desc_csum_set(sb, group, gdp);
1457         ext4_unlock_group(sb, group);
1458
1459         BUFFER_TRACE(group_desc_bh,
1460                      "call ext4_handle_dirty_metadata");
1461         ret = ext4_handle_dirty_metadata(handle, NULL,
1462                                          group_desc_bh);
1463
1464 err_out:
1465         up_write(&grp->alloc_sem);
1466         ext4_journal_stop(handle);
1467 out:
1468         return ret;
1469 }