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/jbd2.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 <asm/byteorder.h>
  26 #include "ext4.h"
  27 #include "ext4_jbd2.h"
  28 #include "xattr.h"
  29 #include "acl.h"
  30 #include "group.h"
  31
  32 /*
  33  * ialloc.c contains the inodes allocation and deallocation routines
  34  */
  35
  36 /*
  37  * The free inodes are managed by bitmaps.  A file system contains several
  38  * blocks groups.  Each group contains 1 bitmap block for blocks, 1 bitmap
  39  * block for inodes, N blocks for the inode table and data blocks.
  40  *
  41  * The file system contains group descriptors which are located after the
  42  * super block.  Each descriptor contains the number of the bitmap block and
  43  * the free blocks count in the block.
  44  */
  45
  46 /*
  47  * To avoid calling the atomic setbit hundreds or thousands of times, we only
  48  * need to use it within a single byte (to ensure we get endianness right).
  49  * We can use memset for the rest of the bitmap as there are no other users.
  50  */
  51 void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  52 {
  53         int i;
  54
  55         if (start_bit >= end_bit)
  56                 return;
  57
  58         ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
  59         for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
  60                 ext4_set_bit(i, bitmap);
  61         if (i < end_bit)
  62                 memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
  63 }
  64
  65 /* Initializes an uninitialized inode bitmap */
  66 unsigned ext4_init_inode_bitmap(struct super_block *sb, struct buffer_head *bh,
  67                                 ext4_group_t block_group,
  68                                 struct ext4_group_desc *gdp)
  69 {
  70         struct ext4_sb_info *sbi = EXT4_SB(sb);
  71
  72         J_ASSERT_BH(bh, buffer_locked(bh));
  73
  74         /* If checksum is bad mark all blocks and inodes use to prevent
  75          * allocation, essentially implementing a per-group read-only flag. */
  76         if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
  77                 ext4_error(sb, __func__, "Checksum bad for group %lu\n",
  78                            block_group);
  79                 gdp->bg_free_blocks_count = 0;
  80                 gdp->bg_free_inodes_count = 0;
  81                 gdp->bg_itable_unused = 0;
  82                 memset(bh->b_data, 0xff, sb->s_blocksize);
  83                 return 0;
  84         }
  85
  86         memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
  87         mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
  88                         bh->b_data);
  89
  90         return EXT4_INODES_PER_GROUP(sb);
  91 }
  92
  93 /*
  94  * Read the inode allocation bitmap for a given block_group, reading
  95  * into the specified slot in the superblock's bitmap cache.
  96  *
  97  * Return buffer_head of bitmap on success or NULL.
  98  */
  99 static struct buffer_head *
 100 ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group)
 101 {
 102         struct ext4_group_desc *desc;
 103         struct buffer_head *bh = NULL;
 104         ext4_fsblk_t bitmap_blk;
 105
 106         desc = ext4_get_group_desc(sb, block_group, NULL);
 107         if (!desc)
 108                 return NULL;
 109         bitmap_blk = ext4_inode_bitmap(sb, desc);
 110         bh = sb_getblk(sb, bitmap_blk);
 111         if (unlikely(!bh)) {
 112                 ext4_error(sb, __func__,
 113                             "Cannot read inode bitmap - "
 114                             "block_group = %lu, inode_bitmap = %llu",
 115                             block_group, bitmap_blk);
 116                 return NULL;
 117         }
 118         if (bitmap_uptodate(bh))
 119                 return bh;
 120
 121         lock_buffer(bh);
 122         if (bitmap_uptodate(bh)) {
 123                 unlock_buffer(bh);
 124                 return bh;
 125         }
 126         spin_lock(sb_bgl_lock(EXT4_SB(sb), block_group));
 127         if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 128                 ext4_init_inode_bitmap(sb, bh, block_group, desc);
 129                 set_bitmap_uptodate(bh);
 130                 set_buffer_uptodate(bh);
 131                 unlock_buffer(bh);
 132                 spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 133                 return bh;
 134         }
 135         spin_unlock(sb_bgl_lock(EXT4_SB(sb), block_group));
 136         if (buffer_uptodate(bh)) {
 137                 /*
 138                  * if not uninit if bh is uptodate,
 139                  * bitmap is also uptodate
 140                  */
 141                 set_bitmap_uptodate(bh);
 142                 unlock_buffer(bh);
 143                 return bh;
 144         }
 145         /*
 146          * submit the buffer_head for read. We can
 147          * safely mark the bitmap as uptodate now.
 148          * We do it here so the bitmap uptodate bit
 149          * get set with buffer lock held.
 150          */
 151         set_bitmap_uptodate(bh);
 152         if (bh_submit_read(bh) < 0) {
 153                 put_bh(bh);
 154                 ext4_error(sb, __func__,
 155                             "Cannot read inode bitmap - "
 156                             "block_group = %lu, inode_bitmap = %llu",
 157                             block_group, bitmap_blk);
 158                 return NULL;
 159         }
 160         return bh;
 161 }
 162
 163 /*
 164  * NOTE! When we get the inode, we're the only people
 165  * that have access to it, and as such there are no
 166  * race conditions we have to worry about. The inode
 167  * is not on the hash-lists, and it cannot be reached
 168  * through the filesystem because the directory entry
 169  * has been deleted earlier.
 170  *
 171  * HOWEVER: we must make sure that we get no aliases,
 172  * which means that we have to call "clear_inode()"
 173  * _before_ we mark the inode not in use in the inode
 174  * bitmaps. Otherwise a newly created file might use
 175  * the same inode number (not actually the same pointer
 176  * though), and then we'd have two inodes sharing the
 177  * same inode number and space on the harddisk.
 178  */
 179 void ext4_free_inode (handle_t *handle, struct inode * inode)
 180 {
 181         struct super_block * sb = inode->i_sb;
 182         int is_directory;
 183         unsigned long ino;
 184         struct buffer_head *bitmap_bh = NULL;
 185         struct buffer_head *bh2;
 186         ext4_group_t block_group;
 187         unsigned long bit;
 188         struct ext4_group_desc * gdp;
 189         struct ext4_super_block * es;
 190         struct ext4_sb_info *sbi;
 191         int fatal = 0, err, cleared;
 192         ext4_group_t flex_group;
 193
 194         if (atomic_read(&inode->i_count) > 1) {
 195                 printk ("ext4_free_inode: inode has count=%d\n",
 196                                         atomic_read(&inode->i_count));
 197                 return;
 198         }
 199         if (inode->i_nlink) {
 200                 printk ("ext4_free_inode: inode has nlink=%d\n",
 201                         inode->i_nlink);
 202                 return;
 203         }
 204         if (!sb) {
 205                 printk("ext4_free_inode: inode on nonexistent device\n");
 206                 return;
 207         }
 208         sbi = EXT4_SB(sb);
 209
 210         ino = inode->i_ino;
 211         ext4_debug ("freeing inode %lu\n", ino);
 212
 213         /*
 214          * Note: we must free any quota before locking the superblock,
 215          * as writing the quota to disk may need the lock as well.
 216          */
 217         DQUOT_INIT(inode);
 218         ext4_xattr_delete_inode(handle, inode);
 219         DQUOT_FREE_INODE(inode);
 220         DQUOT_DROP(inode);
 221
 222         is_directory = S_ISDIR(inode->i_mode);
 223
 224         /* Do this BEFORE marking the inode not in use or returning an error */
 225         clear_inode (inode);
 226
 227         es = EXT4_SB(sb)->s_es;
 228         if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
 229                 ext4_error (sb, "ext4_free_inode",
 230                             "reserved or nonexistent inode %lu", ino);
 231                 goto error_return;
 232         }
 233         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 234         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 235         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 236         if (!bitmap_bh)
 237                 goto error_return;
 238
 239         BUFFER_TRACE(bitmap_bh, "get_write_access");
 240         fatal = ext4_journal_get_write_access(handle, bitmap_bh);
 241         if (fatal)
 242                 goto error_return;
 243
 244         /* Ok, now we can actually update the inode bitmaps.. */
 245         spin_lock(sb_bgl_lock(sbi, block_group));
 246         cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
 247         spin_unlock(sb_bgl_lock(sbi, block_group));
 248         if (!cleared)
 249                 ext4_error(sb, "ext4_free_inode",
 250                            "bit already cleared for inode %lu", ino);
 251         else {
 252                 gdp = ext4_get_group_desc (sb, block_group, &bh2);
 253
 254                 BUFFER_TRACE(bh2, "get_write_access");
 255                 fatal = ext4_journal_get_write_access(handle, bh2);
 256                 if (fatal) goto error_return;
 257
 258                 if (gdp) {
 259                         spin_lock(sb_bgl_lock(sbi, block_group));
 260                         le16_add_cpu(&gdp->bg_free_inodes_count, 1);
 261                         if (is_directory)
 262                                 le16_add_cpu(&gdp->bg_used_dirs_count, -1);
 263                         gdp->bg_checksum = ext4_group_desc_csum(sbi,
 264                                                         block_group, gdp);
 265                         spin_unlock(sb_bgl_lock(sbi, block_group));
 266                         percpu_counter_inc(&sbi->s_freeinodes_counter);
 267                         if (is_directory)
 268                                 percpu_counter_dec(&sbi->s_dirs_counter);
 269
 270                         if (sbi->s_log_groups_per_flex) {
 271                                 flex_group = ext4_flex_group(sbi, block_group);
 272                                 spin_lock(sb_bgl_lock(sbi, flex_group));
 273                                 sbi->s_flex_groups[flex_group].free_inodes++;
 274                                 spin_unlock(sb_bgl_lock(sbi, flex_group));
 275                         }
 276                 }
 277                 BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
 278                 err = ext4_journal_dirty_metadata(handle, bh2);
 279                 if (!fatal) fatal = err;
 280         }
 281         BUFFER_TRACE(bitmap_bh, "call ext4_journal_dirty_metadata");
 282         err = ext4_journal_dirty_metadata(handle, bitmap_bh);
 283         if (!fatal)
 284                 fatal = err;
 285         sb->s_dirt = 1;
 286 error_return:
 287         brelse(bitmap_bh);
 288         ext4_std_error(sb, fatal);
 289 }
 290
 291 /*
 292  * There are two policies for allocating an inode.  If the new inode is
 293  * a directory, then a forward search is made for a block group with both
 294  * free space and a low directory-to-inode ratio; if that fails, then of
 295  * the groups with above-average free space, that group with the fewest
 296  * directories already is chosen.
 297  *
 298  * For other inodes, search forward from the parent directory\'s block
 299  * group to find a free inode.
 300  */
 301 static int find_group_dir(struct super_block *sb, struct inode *parent,
 302                                 ext4_group_t *best_group)
 303 {
 304         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
 305         unsigned int freei, avefreei;
 306         struct ext4_group_desc *desc, *best_desc = NULL;
 307         ext4_group_t group;
 308         int ret = -1;
 309
 310         freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
 311         avefreei = freei / ngroups;
 312
 313         for (group = 0; group < ngroups; group++) {
 314                 desc = ext4_get_group_desc (sb, group, NULL);
 315                 if (!desc || !desc->bg_free_inodes_count)
 316                         continue;
 317                 if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
 318                         continue;
 319                 if (!best_desc ||
 320                     (le16_to_cpu(desc->bg_free_blocks_count) >
 321                      le16_to_cpu(best_desc->bg_free_blocks_count))) {
 322                         *best_group = group;
 323                         best_desc = desc;
 324                         ret = 0;
 325                 }
 326         }
 327         return ret;
 328 }
 329
 330 #define free_block_ratio 10
 331
 332 static int find_group_flex(struct super_block *sb, struct inode *parent,
 333                            ext4_group_t *best_group)
 334 {
 335         struct ext4_sb_info *sbi = EXT4_SB(sb);
 336         struct ext4_group_desc *desc;
 337         struct buffer_head *bh;
 338         struct flex_groups *flex_group = sbi->s_flex_groups;
 339         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 340         ext4_group_t parent_fbg_group = ext4_flex_group(sbi, parent_group);
 341         ext4_group_t ngroups = sbi->s_groups_count;
 342         int flex_size = ext4_flex_bg_size(sbi);
 343         ext4_group_t best_flex = parent_fbg_group;
 344         int blocks_per_flex = sbi->s_blocks_per_group * flex_size;
 345         int flexbg_free_blocks;
 346         int flex_freeb_ratio;
 347         ext4_group_t n_fbg_groups;
 348         ext4_group_t i;
 349
 350         n_fbg_groups = (sbi->s_groups_count + flex_size - 1) >>
 351                 sbi->s_log_groups_per_flex;
 352
 353 find_close_to_parent:
 354         flexbg_free_blocks = flex_group[best_flex].free_blocks;
 355         flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 356         if (flex_group[best_flex].free_inodes &&
 357             flex_freeb_ratio > free_block_ratio)
 358                 goto found_flexbg;
 359
 360         if (best_flex && best_flex == parent_fbg_group) {
 361                 best_flex--;
 362                 goto find_close_to_parent;
 363         }
 364
 365         for (i = 0; i < n_fbg_groups; i++) {
 366                 if (i == parent_fbg_group || i == parent_fbg_group - 1)
 367                         continue;
 368
 369                 flexbg_free_blocks = flex_group[i].free_blocks;
 370                 flex_freeb_ratio = flexbg_free_blocks * 100 / blocks_per_flex;
 371
 372                 if (flex_freeb_ratio > free_block_ratio &&
 373                     flex_group[i].free_inodes) {
 374                         best_flex = i;
 375                         goto found_flexbg;
 376                 }
 377
 378                 if (flex_group[best_flex].free_inodes == 0 ||
 379                     (flex_group[i].free_blocks >
 380                      flex_group[best_flex].free_blocks &&
 381                      flex_group[i].free_inodes))
 382                         best_flex = i;
 383         }
 384
 385         if (!flex_group[best_flex].free_inodes ||
 386             !flex_group[best_flex].free_blocks)
 387                 return -1;
 388
 389 found_flexbg:
 390         for (i = best_flex * flex_size; i < ngroups &&
 391                      i < (best_flex + 1) * flex_size; i++) {
 392                 desc = ext4_get_group_desc(sb, i, &bh);
 393                 if (le16_to_cpu(desc->bg_free_inodes_count)) {
 394                         *best_group = i;
 395                         goto out;
 396                 }
 397         }
 398
 399         return -1;
 400 out:
 401         return 0;
 402 }
 403
 404 /*
 405  * Orlov's allocator for directories.
 406  *
 407  * We always try to spread first-level directories.
 408  *
 409  * If there are blockgroups with both free inodes and free blocks counts
 410  * not worse than average we return one with smallest directory count.
 411  * Otherwise we simply return a random group.
 412  *
 413  * For the rest rules look so:
 414  *
 415  * It's OK to put directory into a group unless
 416  * it has too many directories already (max_dirs) or
 417  * it has too few free inodes left (min_inodes) or
 418  * it has too few free blocks left (min_blocks) or
 419  * it's already running too large debt (max_debt).
 420  * Parent's group is preferred, if it doesn't satisfy these
 421  * conditions we search cyclically through the rest. If none
 422  * of the groups look good we just look for a group with more
 423  * free inodes than average (starting at parent's group).
 424  *
 425  * Debt is incremented each time we allocate a directory and decremented
 426  * when we allocate an inode, within 0--255.
 427  */
 428
 429 #define INODE_COST 64
 430 #define BLOCK_COST 256
 431
 432 static int find_group_orlov(struct super_block *sb, struct inode *parent,
 433                                 ext4_group_t *group)
 434 {
 435         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 436         struct ext4_sb_info *sbi = EXT4_SB(sb);
 437         struct ext4_super_block *es = sbi->s_es;
 438         ext4_group_t ngroups = sbi->s_groups_count;
 439         int inodes_per_group = EXT4_INODES_PER_GROUP(sb);
 440         unsigned int freei, avefreei;
 441         ext4_fsblk_t freeb, avefreeb;
 442         ext4_fsblk_t blocks_per_dir;
 443         unsigned int ndirs;
 444         int max_debt, max_dirs, min_inodes;
 445         ext4_grpblk_t min_blocks;
 446         ext4_group_t i;
 447         struct ext4_group_desc *desc;
 448
 449         freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
 450         avefreei = freei / ngroups;
 451         freeb = percpu_counter_read_positive(&sbi->s_freeblocks_counter);
 452         avefreeb = freeb;
 453         do_div(avefreeb, ngroups);
 454         ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter);
 455
 456         if ((parent == sb->s_root->d_inode) ||
 457             (EXT4_I(parent)->i_flags & EXT4_TOPDIR_FL)) {
 458                 int best_ndir = inodes_per_group;
 459                 ext4_group_t grp;
 460                 int ret = -1;
 461
 462                 get_random_bytes(&grp, sizeof(grp));
 463                 parent_group = (unsigned)grp % ngroups;
 464                 for (i = 0; i < ngroups; i++) {
 465                         grp = (parent_group + i) % ngroups;
 466                         desc = ext4_get_group_desc(sb, grp, NULL);
 467                         if (!desc || !desc->bg_free_inodes_count)
 468                                 continue;
 469                         if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
 470                                 continue;
 471                         if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
 472                                 continue;
 473                         if (le16_to_cpu(desc->bg_free_blocks_count) < avefreeb)
 474                                 continue;
 475                         *group = grp;
 476                         ret = 0;
 477                         best_ndir = le16_to_cpu(desc->bg_used_dirs_count);
 478                 }
 479                 if (ret == 0)
 480                         return ret;
 481                 goto fallback;
 482         }
 483
 484         blocks_per_dir = ext4_blocks_count(es) - freeb;
 485         do_div(blocks_per_dir, ndirs);
 486
 487         max_dirs = ndirs / ngroups + inodes_per_group / 16;
 488         min_inodes = avefreei - inodes_per_group / 4;
 489         min_blocks = avefreeb - EXT4_BLOCKS_PER_GROUP(sb) / 4;
 490
 491         max_debt = EXT4_BLOCKS_PER_GROUP(sb);
 492         max_debt /= max_t(int, blocks_per_dir, BLOCK_COST);
 493         if (max_debt * INODE_COST > inodes_per_group)
 494                 max_debt = inodes_per_group / INODE_COST;
 495         if (max_debt > 255)
 496                 max_debt = 255;
 497         if (max_debt == 0)
 498                 max_debt = 1;
 499
 500         for (i = 0; i < ngroups; i++) {
 501                 *group = (parent_group + i) % ngroups;
 502                 desc = ext4_get_group_desc(sb, *group, NULL);
 503                 if (!desc || !desc->bg_free_inodes_count)
 504                         continue;
 505                 if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
 506                         continue;
 507                 if (le16_to_cpu(desc->bg_free_inodes_count) < min_inodes)
 508                         continue;
 509                 if (le16_to_cpu(desc->bg_free_blocks_count) < min_blocks)
 510                         continue;
 511                 return 0;
 512         }
 513
 514 fallback:
 515         for (i = 0; i < ngroups; i++) {
 516                 *group = (parent_group + i) % ngroups;
 517                 desc = ext4_get_group_desc(sb, *group, NULL);
 518                 if (desc && desc->bg_free_inodes_count &&
 519                         le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
 520                         return 0;
 521         }
 522
 523         if (avefreei) {
 524                 /*
 525                  * The free-inodes counter is approximate, and for really small
 526                  * filesystems the above test can fail to find any blockgroups
 527                  */
 528                 avefreei = 0;
 529                 goto fallback;
 530         }
 531
 532         return -1;
 533 }
 534
 535 static int find_group_other(struct super_block *sb, struct inode *parent,
 536                                 ext4_group_t *group)
 537 {
 538         ext4_group_t parent_group = EXT4_I(parent)->i_block_group;
 539         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
 540         struct ext4_group_desc *desc;
 541         ext4_group_t i;
 542
 543         /*
 544          * Try to place the inode in its parent directory
 545          */
 546         *group = parent_group;
 547         desc = ext4_get_group_desc(sb, *group, NULL);
 548         if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
 549                         le16_to_cpu(desc->bg_free_blocks_count))
 550                 return 0;
 551
 552         /*
 553          * We're going to place this inode in a different blockgroup from its
 554          * parent.  We want to cause files in a common directory to all land in
 555          * the same blockgroup.  But we want files which are in a different
 556          * directory which shares a blockgroup with our parent to land in a
 557          * different blockgroup.
 558          *
 559          * So add our directory's i_ino into the starting point for the hash.
 560          */
 561         *group = (*group + parent->i_ino) % ngroups;
 562
 563         /*
 564          * Use a quadratic hash to find a group with a free inode and some free
 565          * blocks.
 566          */
 567         for (i = 1; i < ngroups; i <<= 1) {
 568                 *group += i;
 569                 if (*group >= ngroups)
 570                         *group -= ngroups;
 571                 desc = ext4_get_group_desc(sb, *group, NULL);
 572                 if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
 573                                 le16_to_cpu(desc->bg_free_blocks_count))
 574                         return 0;
 575         }
 576
 577         /*
 578          * That failed: try linear search for a free inode, even if that group
 579          * has no free blocks.
 580          */
 581         *group = parent_group;
 582         for (i = 0; i < ngroups; i++) {
 583                 if (++*group >= ngroups)
 584                         *group = 0;
 585                 desc = ext4_get_group_desc(sb, *group, NULL);
 586                 if (desc && le16_to_cpu(desc->bg_free_inodes_count))
 587                         return 0;
 588         }
 589
 590         return -1;
 591 }
 592
 593 /*
 594  * claim the inode from the inode bitmap. If the group
 595  * is uninit we need to take the groups's sb_bgl_lock
 596  * and clear the uninit flag. The inode bitmap update
 597  * and group desc uninit flag clear should be done
 598  * after holding sb_bgl_lock so that ext4_read_inode_bitmap
 599  * doesn't race with the ext4_claim_inode
 600  */
 601 static int ext4_claim_inode(struct super_block *sb,
 602                         struct buffer_head *inode_bitmap_bh,
 603                         unsigned long ino, ext4_group_t group, int mode)
 604 {
 605         int free = 0, retval = 0;
 606         struct ext4_sb_info *sbi = EXT4_SB(sb);
 607         struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
 608
 609         spin_lock(sb_bgl_lock(sbi, group));
 610         if (ext4_set_bit(ino, inode_bitmap_bh->b_data)) {
 611                 /* not a free inode */
 612                 retval = 1;
 613                 goto err_ret;
 614         }
 615         ino++;
 616         if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
 617                         ino > EXT4_INODES_PER_GROUP(sb)) {
 618                 spin_unlock(sb_bgl_lock(sbi, group));
 619                 ext4_error(sb, __func__,
 620                            "reserved inode or inode > inodes count - "
 621                            "block_group = %lu, inode=%lu", group,
 622                            ino + group * EXT4_INODES_PER_GROUP(sb));
 623                 return 1;
 624         }
 625         /* If we didn't allocate from within the initialized part of the inode
 626          * table then we need to initialize up to this inode. */
 627         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
 628
 629                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
 630                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
 631                         /* When marking the block group with
 632                          * ~EXT4_BG_INODE_UNINIT we don't want to depend
 633                          * on the value of bg_itable_unused even though
 634                          * mke2fs could have initialized the same for us.
 635                          * Instead we calculated the value below
 636                          */
 637
 638                         free = 0;
 639                 } else {
 640                         free = EXT4_INODES_PER_GROUP(sb) -
 641                                 le16_to_cpu(gdp->bg_itable_unused);
 642                 }
 643
 644                 /*
 645                  * Check the relative inode number against the last used
 646                  * relative inode number in this group. if it is greater
 647                  * we need to  update the bg_itable_unused count
 648                  *
 649                  */
 650                 if (ino > free)
 651                         gdp->bg_itable_unused =
 652                                 cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino);
 653         }
 654         le16_add_cpu(&gdp->bg_free_inodes_count, -1);
 655         if (S_ISDIR(mode)) {
 656                 le16_add_cpu(&gdp->bg_used_dirs_count, 1);
 657         }
 658         gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
 659 err_ret:
 660         spin_unlock(sb_bgl_lock(sbi, group));
 661         return retval;
 662 }
 663
 664 /*
 665  * There are two policies for allocating an inode.  If the new inode is
 666  * a directory, then a forward search is made for a block group with both
 667  * free space and a low directory-to-inode ratio; if that fails, then of
 668  * the groups with above-average free space, that group with the fewest
 669  * directories already is chosen.
 670  *
 671  * For other inodes, search forward from the parent directory's block
 672  * group to find a free inode.
 673  */
 674 struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
 675 {
 676         struct super_block *sb;
 677         struct buffer_head *bitmap_bh = NULL;
 678         struct buffer_head *bh2;
 679         ext4_group_t group = 0;
 680         unsigned long ino = 0;
 681         struct inode * inode;
 682         struct ext4_group_desc * gdp = NULL;
 683         struct ext4_super_block * es;
 684         struct ext4_inode_info *ei;
 685         struct ext4_sb_info *sbi;
 686         int ret2, err = 0;
 687         struct inode *ret;
 688         ext4_group_t i;
 689         int free = 0;
 690         static int once = 1;
 691         ext4_group_t flex_group;
 692
 693         /* Cannot create files in a deleted directory */
 694         if (!dir || !dir->i_nlink)
 695                 return ERR_PTR(-EPERM);
 696
 697         sb = dir->i_sb;
 698         inode = new_inode(sb);
 699         if (!inode)
 700                 return ERR_PTR(-ENOMEM);
 701         ei = EXT4_I(inode);
 702
 703         sbi = EXT4_SB(sb);
 704         es = sbi->s_es;
 705
 706         if (sbi->s_log_groups_per_flex) {
 707                 ret2 = find_group_flex(sb, dir, &group);
 708                 if (ret2 == -1) {
 709                         ret2 = find_group_other(sb, dir, &group);
 710                         if (ret2 == 0 && once) {
 711                                 once = 0;
 712                                 printk(KERN_NOTICE "ext4: find_group_flex "
 713                                        "failed, fallback succeeded dir %lu\n",
 714                                        dir->i_ino);
 715                         }
 716                 }
 717                 goto got_group;
 718         }
 719
 720         if (S_ISDIR(mode)) {
 721                 if (test_opt (sb, OLDALLOC))
 722                         ret2 = find_group_dir(sb, dir, &group);
 723                 else
 724                         ret2 = find_group_orlov(sb, dir, &group);
 725         } else
 726                 ret2 = find_group_other(sb, dir, &group);
 727
 728 got_group:
 729         err = -ENOSPC;
 730         if (ret2 == -1)
 731                 goto out;
 732
 733         for (i = 0; i < sbi->s_groups_count; i++) {
 734                 err = -EIO;
 735
 736                 gdp = ext4_get_group_desc(sb, group, &bh2);
 737                 if (!gdp)
 738                         goto fail;
 739
 740                 brelse(bitmap_bh);
 741                 bitmap_bh = ext4_read_inode_bitmap(sb, group);
 742                 if (!bitmap_bh)
 743                         goto fail;
 744
 745                 ino = 0;
 746
 747 repeat_in_this_group:
 748                 ino = ext4_find_next_zero_bit((unsigned long *)
 749                                 bitmap_bh->b_data, EXT4_INODES_PER_GROUP(sb), ino);
 750                 if (ino < EXT4_INODES_PER_GROUP(sb)) {
 751
 752                         BUFFER_TRACE(bitmap_bh, "get_write_access");
 753                         err = ext4_journal_get_write_access(handle, bitmap_bh);
 754                         if (err)
 755                                 goto fail;
 756
 757                         BUFFER_TRACE(bh2, "get_write_access");
 758                         err = ext4_journal_get_write_access(handle, bh2);
 759                         if (err)
 760                                 goto fail;
 761                         if (!ext4_claim_inode(sb, bitmap_bh,
 762                                                 ino, group, mode)) {
 763                                 /* we won it */
 764                                 BUFFER_TRACE(bitmap_bh,
 765                                         "call ext4_journal_dirty_metadata");
 766                                 err = ext4_journal_dirty_metadata(handle,
 767                                                                 bitmap_bh);
 768                                 if (err)
 769                                         goto fail;
 770                                 /* zero bit is inode number 1*/
 771                                 ino++;
 772                                 goto got;
 773                         }
 774                         /* we lost it */
 775                         jbd2_journal_release_buffer(handle, bitmap_bh);
 776                         jbd2_journal_release_buffer(handle, bh2);
 777
 778                         if (++ino < EXT4_INODES_PER_GROUP(sb))
 779                                 goto repeat_in_this_group;
 780                 }
 781
 782                 /*
 783                  * This case is possible in concurrent environment.  It is very
 784                  * rare.  We cannot repeat the find_group_xxx() call because
 785                  * that will simply return the same blockgroup, because the
 786                  * group descriptor metadata has not yet been updated.
 787                  * So we just go onto the next blockgroup.
 788                  */
 789                 if (++group == sbi->s_groups_count)
 790                         group = 0;
 791         }
 792         err = -ENOSPC;
 793         goto out;
 794
 795 got:
 796         /* We may have to initialize the block bitmap if it isn't already */
 797         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
 798             gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 799                 struct buffer_head *block_bh = ext4_read_block_bitmap(sb, group);
 800
 801                 BUFFER_TRACE(block_bh, "get block bitmap access");
 802                 err = ext4_journal_get_write_access(handle, block_bh);
 803                 if (err) {
 804                         brelse(block_bh);
 805                         goto fail;
 806                 }
 807
 808                 free = 0;
 809                 spin_lock(sb_bgl_lock(sbi, group));
 810                 /* recheck and clear flag under lock if we still need to */
 811                 if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
 812                         gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
 813                         free = ext4_free_blocks_after_init(sb, group, gdp);
 814                         gdp->bg_free_blocks_count = cpu_to_le16(free);
 815                         gdp->bg_checksum = ext4_group_desc_csum(sbi, group,
 816                                                                 gdp);
 817                 }
 818                 spin_unlock(sb_bgl_lock(sbi, group));
 819
 820                 /* Don't need to dirty bitmap block if we didn't change it */
 821                 if (free) {
 822                         BUFFER_TRACE(block_bh, "dirty block bitmap");
 823                         err = ext4_journal_dirty_metadata(handle, block_bh);
 824                 }
 825
 826                 brelse(block_bh);
 827                 if (err)
 828                         goto fail;
 829         }
 830         BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata");
 831         err = ext4_journal_dirty_metadata(handle, bh2);
 832         if (err)
 833                 goto fail;
 834
 835         percpu_counter_dec(&sbi->s_freeinodes_counter);
 836         if (S_ISDIR(mode))
 837                 percpu_counter_inc(&sbi->s_dirs_counter);
 838         sb->s_dirt = 1;
 839
 840         if (sbi->s_log_groups_per_flex) {
 841                 flex_group = ext4_flex_group(sbi, group);
 842                 spin_lock(sb_bgl_lock(sbi, flex_group));
 843                 sbi->s_flex_groups[flex_group].free_inodes--;
 844                 spin_unlock(sb_bgl_lock(sbi, flex_group));
 845         }
 846
 847         inode->i_uid = current->fsuid;
 848         if (test_opt (sb, GRPID))
 849                 inode->i_gid = dir->i_gid;
 850         else if (dir->i_mode & S_ISGID) {
 851                 inode->i_gid = dir->i_gid;
 852                 if (S_ISDIR(mode))
 853                         mode |= S_ISGID;
 854         } else
 855                 inode->i_gid = current->fsgid;
 856         inode->i_mode = mode;
 857
 858         inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
 859         /* This is the optimal IO size (for stat), not the fs block size */
 860         inode->i_blocks = 0;
 861         inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
 862                                                        ext4_current_time(inode);
 863
 864         memset(ei->i_data, 0, sizeof(ei->i_data));
 865         ei->i_dir_start_lookup = 0;
 866         ei->i_disksize = 0;
 867
 868         /*
 869          * Don't inherit extent flag from directory, amongst others. We set
 870          * extent flag on newly created directory and file only if -o extent
 871          * mount option is specified
 872          */
 873         ei->i_flags =
 874                 ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED);
 875         ei->i_file_acl = 0;
 876         ei->i_dtime = 0;
 877         ei->i_block_alloc_info = NULL;
 878         ei->i_block_group = group;
 879
 880         ext4_set_inode_flags(inode);
 881         if (IS_DIRSYNC(inode))
 882                 handle->h_sync = 1;
 883         insert_inode_hash(inode);
 884         spin_lock(&sbi->s_next_gen_lock);
 885         inode->i_generation = sbi->s_next_generation++;
 886         spin_unlock(&sbi->s_next_gen_lock);
 887
 888         ei->i_state = EXT4_STATE_NEW;
 889
 890         ei->i_extra_isize = EXT4_SB(sb)->s_want_extra_isize;
 891
 892         ret = inode;
 893         if(DQUOT_ALLOC_INODE(inode)) {
 894                 err = -EDQUOT;
 895                 goto fail_drop;
 896         }
 897
 898         err = ext4_init_acl(handle, inode, dir);
 899         if (err)
 900                 goto fail_free_drop;
 901
 902         err = ext4_init_security(handle,inode, dir);
 903         if (err)
 904                 goto fail_free_drop;
 905
 906         if (test_opt(sb, EXTENTS)) {
 907                 /* set extent flag only for directory, file and normal symlink*/
 908                 if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) {
 909                         EXT4_I(inode)->i_flags |= EXT4_EXTENTS_FL;
 910                         ext4_ext_tree_init(handle, inode);
 911                 }
 912         }
 913
 914         err = ext4_mark_inode_dirty(handle, inode);
 915         if (err) {
 916                 ext4_std_error(sb, err);
 917                 goto fail_free_drop;
 918         }
 919
 920         ext4_debug("allocating inode %lu\n", inode->i_ino);
 921         goto really_out;
 922 fail:
 923         ext4_std_error(sb, err);
 924 out:
 925         iput(inode);
 926         ret = ERR_PTR(err);
 927 really_out:
 928         brelse(bitmap_bh);
 929         return ret;
 930
 931 fail_free_drop:
 932         DQUOT_FREE_INODE(inode);
 933
 934 fail_drop:
 935         DQUOT_DROP(inode);
 936         inode->i_flags |= S_NOQUOTA;
 937         inode->i_nlink = 0;
 938         iput(inode);
 939         brelse(bitmap_bh);
 940         return ERR_PTR(err);
 941 }
 942
 943 /* Verify that we are loading a valid orphan from disk */
 944 struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino)
 945 {
 946         unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count);
 947         ext4_group_t block_group;
 948         int bit;
 949         struct buffer_head *bitmap_bh;
 950         struct inode *inode = NULL;
 951         long err = -EIO;
 952
 953         /* Error cases - e2fsck has already cleaned up for us */
 954         if (ino > max_ino) {
 955                 ext4_warning(sb, __func__,
 956                              "bad orphan ino %lu!  e2fsck was run?", ino);
 957                 goto error;
 958         }
 959
 960         block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
 961         bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb);
 962         bitmap_bh = ext4_read_inode_bitmap(sb, block_group);
 963         if (!bitmap_bh) {
 964                 ext4_warning(sb, __func__,
 965                              "inode bitmap error for orphan %lu", ino);
 966                 goto error;
 967         }
 968
 969         /* Having the inode bit set should be a 100% indicator that this
 970          * is a valid orphan (no e2fsck run on fs).  Orphans also include
 971          * inodes that were being truncated, so we can't check i_nlink==0.
 972          */
 973         if (!ext4_test_bit(bit, bitmap_bh->b_data))
 974                 goto bad_orphan;
 975
 976         inode = ext4_iget(sb, ino);
 977         if (IS_ERR(inode))
 978                 goto iget_failed;
 979
 980         /*
 981          * If the orphans has i_nlinks > 0 then it should be able to be
 982          * truncated, otherwise it won't be removed from the orphan list
 983          * during processing and an infinite loop will result.
 984          */
 985         if (inode->i_nlink && !ext4_can_truncate(inode))
 986                 goto bad_orphan;
 987
 988         if (NEXT_ORPHAN(inode) > max_ino)
 989                 goto bad_orphan;
 990         brelse(bitmap_bh);
 991         return inode;
 992
 993 iget_failed:
 994         err = PTR_ERR(inode);
 995         inode = NULL;
 996 bad_orphan:
 997         ext4_warning(sb, __func__,
 998                      "bad orphan inode %lu!  e2fsck was run?", ino);
 999         printk(KERN_NOTICE "ext4_test_bit(bit=%d, block=%llu) = %d\n",
1000                bit, (unsigned long long)bitmap_bh->b_blocknr,
1001                ext4_test_bit(bit, bitmap_bh->b_data));
1002         printk(KERN_NOTICE "inode=%p\n", inode);
1003         if (inode) {
1004                 printk(KERN_NOTICE "is_bad_inode(inode)=%d\n",
1005                        is_bad_inode(inode));
1006                 printk(KERN_NOTICE "NEXT_ORPHAN(inode)=%u\n",
1007                        NEXT_ORPHAN(inode));
1008                 printk(KERN_NOTICE "max_ino=%lu\n", max_ino);
1009                 printk(KERN_NOTICE "i_nlink=%u\n", inode->i_nlink);
1010                 /* Avoid freeing blocks if we got a bad deleted inode */
1011                 if (inode->i_nlink == 0)
1012                         inode->i_blocks = 0;
1013                 iput(inode);
1014         }
1015         brelse(bitmap_bh);
1016 error:
1017         return ERR_PTR(err);
1018 }
1019
1020 unsigned long ext4_count_free_inodes (struct super_block * sb)
1021 {
1022         unsigned long desc_count;
1023         struct ext4_group_desc *gdp;
1024         ext4_group_t i;
1025 #ifdef EXT4FS_DEBUG
1026         struct ext4_super_block *es;
1027         unsigned long bitmap_count, x;
1028         struct buffer_head *bitmap_bh = NULL;
1029
1030         es = EXT4_SB(sb)->s_es;
1031         desc_count = 0;
1032         bitmap_count = 0;
1033         gdp = NULL;
1034         for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1035                 gdp = ext4_get_group_desc (sb, i, NULL);
1036                 if (!gdp)
1037                         continue;
1038                 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
1039                 brelse(bitmap_bh);
1040                 bitmap_bh = ext4_read_inode_bitmap(sb, i);
1041                 if (!bitmap_bh)
1042                         continue;
1043
1044                 x = ext4_count_free(bitmap_bh, EXT4_INODES_PER_GROUP(sb) / 8);
1045                 printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n",
1046                         i, le16_to_cpu(gdp->bg_free_inodes_count), x);
1047                 bitmap_count += x;
1048         }
1049         brelse(bitmap_bh);
1050         printk("ext4_count_free_inodes: stored = %u, computed = %lu, %lu\n",
1051                 le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count);
1052         return desc_count;
1053 #else
1054         desc_count = 0;
1055         for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1056                 gdp = ext4_get_group_desc (sb, i, NULL);
1057                 if (!gdp)
1058                         continue;
1059                 desc_count += le16_to_cpu(gdp->bg_free_inodes_count);
1060                 cond_resched();
1061         }
1062         return desc_count;
1063 #endif
1064 }
1065
1066 /* Called at mount-time, super-block is locked */
1067 unsigned long ext4_count_dirs (struct super_block * sb)
1068 {
1069         unsigned long count = 0;
1070         ext4_group_t i;
1071
1072         for (i = 0; i < EXT4_SB(sb)->s_groups_count; i++) {
1073                 struct ext4_group_desc *gdp = ext4_get_group_desc (sb, i, NULL);
1074                 if (!gdp)
1075                         continue;
1076                 count += le16_to_cpu(gdp->bg_used_dirs_count);
1077         }
1078         return count;
1079 }
1080