linux/fs/ext4/namei.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/ext4/namei.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 *  from
  10 *
  11 *  linux/fs/minix/namei.c
  12 *
  13 *  Copyright (C) 1991, 1992  Linus Torvalds
  14 *
  15 *  Big-endian to little-endian byte-swapping/bitmaps by
  16 *        David S. Miller (davem@caip.rutgers.edu), 1995
  17 *  Directory entry file type support and forward compatibility hooks
  18 *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19 *  Hash Tree Directory indexing (c)
  20 *      Daniel Phillips, 2001
  21 *  Hash Tree Directory indexing porting
  22 *      Christopher Li, 2002
  23 *  Hash Tree Directory indexing cleanup
  24 *      Theodore Ts'o, 2002
  25 */
  26
  27#include <linux/fs.h>
  28#include <linux/pagemap.h>
  29#include <linux/jbd2.h>
  30#include <linux/time.h>
  31#include <linux/fcntl.h>
  32#include <linux/stat.h>
  33#include <linux/string.h>
  34#include <linux/quotaops.h>
  35#include <linux/buffer_head.h>
  36#include <linux/bio.h>
  37#include "ext4.h"
  38#include "ext4_jbd2.h"
  39
  40#include "xattr.h"
  41#include "acl.h"
  42
  43#include <trace/events/ext4.h>
  44/*
  45 * define how far ahead to read directories while searching them.
  46 */
  47#define NAMEI_RA_CHUNKS  2
  48#define NAMEI_RA_BLOCKS  4
  49#define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  50#define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  51
  52static struct buffer_head *ext4_append(handle_t *handle,
  53                                        struct inode *inode,
  54                                        ext4_lblk_t *block, int *err)
  55{
  56        struct buffer_head *bh;
  57
  58        *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
  59
  60        bh = ext4_bread(handle, inode, *block, 1, err);
  61        if (bh) {
  62                inode->i_size += inode->i_sb->s_blocksize;
  63                EXT4_I(inode)->i_disksize = inode->i_size;
  64                *err = ext4_journal_get_write_access(handle, bh);
  65                if (*err) {
  66                        brelse(bh);
  67                        bh = NULL;
  68                }
  69        }
  70        return bh;
  71}
  72
  73#ifndef assert
  74#define assert(test) J_ASSERT(test)
  75#endif
  76
  77#ifdef DX_DEBUG
  78#define dxtrace(command) command
  79#else
  80#define dxtrace(command)
  81#endif
  82
  83struct fake_dirent
  84{
  85        __le32 inode;
  86        __le16 rec_len;
  87        u8 name_len;
  88        u8 file_type;
  89};
  90
  91struct dx_countlimit
  92{
  93        __le16 limit;
  94        __le16 count;
  95};
  96
  97struct dx_entry
  98{
  99        __le32 hash;
 100        __le32 block;
 101};
 102
 103/*
 104 * dx_root_info is laid out so that if it should somehow get overlaid by a
 105 * dirent the two low bits of the hash version will be zero.  Therefore, the
 106 * hash version mod 4 should never be 0.  Sincerely, the paranoia department.
 107 */
 108
 109struct dx_root
 110{
 111        struct fake_dirent dot;
 112        char dot_name[4];
 113        struct fake_dirent dotdot;
 114        char dotdot_name[4];
 115        struct dx_root_info
 116        {
 117                __le32 reserved_zero;
 118                u8 hash_version;
 119                u8 info_length; /* 8 */
 120                u8 indirect_levels;
 121                u8 unused_flags;
 122        }
 123        info;
 124        struct dx_entry entries[0];
 125};
 126
 127struct dx_node
 128{
 129        struct fake_dirent fake;
 130        struct dx_entry entries[0];
 131};
 132
 133
 134struct dx_frame
 135{
 136        struct buffer_head *bh;
 137        struct dx_entry *entries;
 138        struct dx_entry *at;
 139};
 140
 141struct dx_map_entry
 142{
 143        u32 hash;
 144        u16 offs;
 145        u16 size;
 146};
 147
 148/*
 149 * This goes at the end of each htree block.
 150 */
 151struct dx_tail {
 152        u32 dt_reserved;
 153        __le32 dt_checksum;     /* crc32c(uuid+inum+dirblock) */
 154};
 155
 156static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
 157static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
 158static inline unsigned dx_get_hash(struct dx_entry *entry);
 159static void dx_set_hash(struct dx_entry *entry, unsigned value);
 160static unsigned dx_get_count(struct dx_entry *entries);
 161static unsigned dx_get_limit(struct dx_entry *entries);
 162static void dx_set_count(struct dx_entry *entries, unsigned value);
 163static void dx_set_limit(struct dx_entry *entries, unsigned value);
 164static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
 165static unsigned dx_node_limit(struct inode *dir);
 166static struct dx_frame *dx_probe(const struct qstr *d_name,
 167                                 struct inode *dir,
 168                                 struct dx_hash_info *hinfo,
 169                                 struct dx_frame *frame,
 170                                 int *err);
 171static void dx_release(struct dx_frame *frames);
 172static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 173                       struct dx_hash_info *hinfo, struct dx_map_entry map[]);
 174static void dx_sort_map(struct dx_map_entry *map, unsigned count);
 175static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
 176                struct dx_map_entry *offsets, int count, unsigned blocksize);
 177static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
 178static void dx_insert_block(struct dx_frame *frame,
 179                                        u32 hash, ext4_lblk_t block);
 180static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 181                                 struct dx_frame *frame,
 182                                 struct dx_frame *frames,
 183                                 __u32 *start_hash);
 184static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
 185                const struct qstr *d_name,
 186                struct ext4_dir_entry_2 **res_dir,
 187                int *err);
 188static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 189                             struct inode *inode);
 190
 191/* checksumming functions */
 192#define EXT4_DIRENT_TAIL(block, blocksize) \
 193        ((struct ext4_dir_entry_tail *)(((void *)(block)) + \
 194                                        ((blocksize) - \
 195                                         sizeof(struct ext4_dir_entry_tail))))
 196
 197static void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
 198                                   unsigned int blocksize)
 199{
 200        memset(t, 0, sizeof(struct ext4_dir_entry_tail));
 201        t->det_rec_len = ext4_rec_len_to_disk(
 202                        sizeof(struct ext4_dir_entry_tail), blocksize);
 203        t->det_reserved_ft = EXT4_FT_DIR_CSUM;
 204}
 205
 206/* Walk through a dirent block to find a checksum "dirent" at the tail */
 207static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
 208                                                   struct ext4_dir_entry *de)
 209{
 210        struct ext4_dir_entry_tail *t;
 211
 212#ifdef PARANOID
 213        struct ext4_dir_entry *d, *top;
 214
 215        d = de;
 216        top = (struct ext4_dir_entry *)(((void *)de) +
 217                (EXT4_BLOCK_SIZE(inode->i_sb) -
 218                sizeof(struct ext4_dir_entry_tail)));
 219        while (d < top && d->rec_len)
 220                d = (struct ext4_dir_entry *)(((void *)d) +
 221                    le16_to_cpu(d->rec_len));
 222
 223        if (d != top)
 224                return NULL;
 225
 226        t = (struct ext4_dir_entry_tail *)d;
 227#else
 228        t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
 229#endif
 230
 231        if (t->det_reserved_zero1 ||
 232            le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
 233            t->det_reserved_zero2 ||
 234            t->det_reserved_ft != EXT4_FT_DIR_CSUM)
 235                return NULL;
 236
 237        return t;
 238}
 239
 240static __le32 ext4_dirent_csum(struct inode *inode,
 241                               struct ext4_dir_entry *dirent, int size)
 242{
 243        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 244        struct ext4_inode_info *ei = EXT4_I(inode);
 245        __u32 csum;
 246
 247        csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 248        return cpu_to_le32(csum);
 249}
 250
 251int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
 252{
 253        struct ext4_dir_entry_tail *t;
 254
 255        if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 256                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 257                return 1;
 258
 259        t = get_dirent_tail(inode, dirent);
 260        if (!t) {
 261                EXT4_ERROR_INODE(inode, "metadata_csum set but no space in dir "
 262                                 "leaf for checksum.  Please run e2fsck -D.");
 263                return 0;
 264        }
 265
 266        if (t->det_checksum != ext4_dirent_csum(inode, dirent,
 267                                                (void *)t - (void *)dirent))
 268                return 0;
 269
 270        return 1;
 271}
 272
 273static void ext4_dirent_csum_set(struct inode *inode,
 274                                 struct ext4_dir_entry *dirent)
 275{
 276        struct ext4_dir_entry_tail *t;
 277
 278        if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 279                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 280                return;
 281
 282        t = get_dirent_tail(inode, dirent);
 283        if (!t) {
 284                EXT4_ERROR_INODE(inode, "metadata_csum set but no space in dir "
 285                                 "leaf for checksum.  Please run e2fsck -D.");
 286                return;
 287        }
 288
 289        t->det_checksum = ext4_dirent_csum(inode, dirent,
 290                                           (void *)t - (void *)dirent);
 291}
 292
 293static inline int ext4_handle_dirty_dirent_node(handle_t *handle,
 294                                                struct inode *inode,
 295                                                struct buffer_head *bh)
 296{
 297        ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 298        return ext4_handle_dirty_metadata(handle, inode, bh);
 299}
 300
 301static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
 302                                               struct ext4_dir_entry *dirent,
 303                                               int *offset)
 304{
 305        struct ext4_dir_entry *dp;
 306        struct dx_root_info *root;
 307        int count_offset;
 308
 309        if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
 310                count_offset = 8;
 311        else if (le16_to_cpu(dirent->rec_len) == 12) {
 312                dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
 313                if (le16_to_cpu(dp->rec_len) !=
 314                    EXT4_BLOCK_SIZE(inode->i_sb) - 12)
 315                        return NULL;
 316                root = (struct dx_root_info *)(((void *)dp + 12));
 317                if (root->reserved_zero ||
 318                    root->info_length != sizeof(struct dx_root_info))
 319                        return NULL;
 320                count_offset = 32;
 321        } else
 322                return NULL;
 323
 324        if (offset)
 325                *offset = count_offset;
 326        return (struct dx_countlimit *)(((void *)dirent) + count_offset);
 327}
 328
 329static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
 330                           int count_offset, int count, struct dx_tail *t)
 331{
 332        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 333        struct ext4_inode_info *ei = EXT4_I(inode);
 334        __u32 csum, old_csum;
 335        int size;
 336
 337        size = count_offset + (count * sizeof(struct dx_entry));
 338        old_csum = t->dt_checksum;
 339        t->dt_checksum = 0;
 340        csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
 341        csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
 342        t->dt_checksum = old_csum;
 343
 344        return cpu_to_le32(csum);
 345}
 346
 347static int ext4_dx_csum_verify(struct inode *inode,
 348                               struct ext4_dir_entry *dirent)
 349{
 350        struct dx_countlimit *c;
 351        struct dx_tail *t;
 352        int count_offset, limit, count;
 353
 354        if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 355                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 356                return 1;
 357
 358        c = get_dx_countlimit(inode, dirent, &count_offset);
 359        if (!c) {
 360                EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 361                return 1;
 362        }
 363        limit = le16_to_cpu(c->limit);
 364        count = le16_to_cpu(c->count);
 365        if (count_offset + (limit * sizeof(struct dx_entry)) >
 366            EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 367                EXT4_ERROR_INODE(inode, "metadata_csum set but no space for "
 368                                 "tree checksum found.  Run e2fsck -D.");
 369                return 1;
 370        }
 371        t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 372
 373        if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
 374                                            count, t))
 375                return 0;
 376        return 1;
 377}
 378
 379static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
 380{
 381        struct dx_countlimit *c;
 382        struct dx_tail *t;
 383        int count_offset, limit, count;
 384
 385        if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
 386                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 387                return;
 388
 389        c = get_dx_countlimit(inode, dirent, &count_offset);
 390        if (!c) {
 391                EXT4_ERROR_INODE(inode, "dir seems corrupt?  Run e2fsck -D.");
 392                return;
 393        }
 394        limit = le16_to_cpu(c->limit);
 395        count = le16_to_cpu(c->count);
 396        if (count_offset + (limit * sizeof(struct dx_entry)) >
 397            EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
 398                EXT4_ERROR_INODE(inode, "metadata_csum set but no space for "
 399                                 "tree checksum.  Run e2fsck -D.");
 400                return;
 401        }
 402        t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
 403
 404        t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
 405}
 406
 407static inline int ext4_handle_dirty_dx_node(handle_t *handle,
 408                                            struct inode *inode,
 409                                            struct buffer_head *bh)
 410{
 411        ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
 412        return ext4_handle_dirty_metadata(handle, inode, bh);
 413}
 414
 415/*
 416 * p is at least 6 bytes before the end of page
 417 */
 418static inline struct ext4_dir_entry_2 *
 419ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
 420{
 421        return (struct ext4_dir_entry_2 *)((char *)p +
 422                ext4_rec_len_from_disk(p->rec_len, blocksize));
 423}
 424
 425/*
 426 * Future: use high four bits of block for coalesce-on-delete flags
 427 * Mask them off for now.
 428 */
 429
 430static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
 431{
 432        return le32_to_cpu(entry->block) & 0x00ffffff;
 433}
 434
 435static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
 436{
 437        entry->block = cpu_to_le32(value);
 438}
 439
 440static inline unsigned dx_get_hash(struct dx_entry *entry)
 441{
 442        return le32_to_cpu(entry->hash);
 443}
 444
 445static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
 446{
 447        entry->hash = cpu_to_le32(value);
 448}
 449
 450static inline unsigned dx_get_count(struct dx_entry *entries)
 451{
 452        return le16_to_cpu(((struct dx_countlimit *) entries)->count);
 453}
 454
 455static inline unsigned dx_get_limit(struct dx_entry *entries)
 456{
 457        return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
 458}
 459
 460static inline void dx_set_count(struct dx_entry *entries, unsigned value)
 461{
 462        ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
 463}
 464
 465static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
 466{
 467        ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
 468}
 469
 470static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
 471{
 472        unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
 473                EXT4_DIR_REC_LEN(2) - infosize;
 474
 475        if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 476                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 477                entry_space -= sizeof(struct dx_tail);
 478        return entry_space / sizeof(struct dx_entry);
 479}
 480
 481static inline unsigned dx_node_limit(struct inode *dir)
 482{
 483        unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
 484
 485        if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
 486                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 487                entry_space -= sizeof(struct dx_tail);
 488        return entry_space / sizeof(struct dx_entry);
 489}
 490
 491/*
 492 * Debug
 493 */
 494#ifdef DX_DEBUG
 495static void dx_show_index(char * label, struct dx_entry *entries)
 496{
 497        int i, n = dx_get_count (entries);
 498        printk(KERN_DEBUG "%s index ", label);
 499        for (i = 0; i < n; i++) {
 500                printk("%x->%lu ", i ? dx_get_hash(entries + i) :
 501                                0, (unsigned long)dx_get_block(entries + i));
 502        }
 503        printk("\n");
 504}
 505
 506struct stats
 507{
 508        unsigned names;
 509        unsigned space;
 510        unsigned bcount;
 511};
 512
 513static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
 514                                 int size, int show_names)
 515{
 516        unsigned names = 0, space = 0;
 517        char *base = (char *) de;
 518        struct dx_hash_info h = *hinfo;
 519
 520        printk("names: ");
 521        while ((char *) de < base + size)
 522        {
 523                if (de->inode)
 524                {
 525                        if (show_names)
 526                        {
 527                                int len = de->name_len;
 528                                char *name = de->name;
 529                                while (len--) printk("%c", *name++);
 530                                ext4fs_dirhash(de->name, de->name_len, &h);
 531                                printk(":%x.%u ", h.hash,
 532                                       (unsigned) ((char *) de - base));
 533                        }
 534                        space += EXT4_DIR_REC_LEN(de->name_len);
 535                        names++;
 536                }
 537                de = ext4_next_entry(de, size);
 538        }
 539        printk("(%i)\n", names);
 540        return (struct stats) { names, space, 1 };
 541}
 542
 543struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
 544                             struct dx_entry *entries, int levels)
 545{
 546        unsigned blocksize = dir->i_sb->s_blocksize;
 547        unsigned count = dx_get_count(entries), names = 0, space = 0, i;
 548        unsigned bcount = 0;
 549        struct buffer_head *bh;
 550        int err;
 551        printk("%i indexed blocks...\n", count);
 552        for (i = 0; i < count; i++, entries++)
 553        {
 554                ext4_lblk_t block = dx_get_block(entries);
 555                ext4_lblk_t hash  = i ? dx_get_hash(entries): 0;
 556                u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
 557                struct stats stats;
 558                printk("%s%3u:%03u hash %8x/%8x ",levels?"":"   ", i, block, hash, range);
 559                if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
 560                stats = levels?
 561                   dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
 562                   dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
 563                names += stats.names;
 564                space += stats.space;
 565                bcount += stats.bcount;
 566                brelse(bh);
 567        }
 568        if (bcount)
 569                printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
 570                       levels ? "" : "   ", names, space/bcount,
 571                       (space/bcount)*100/blocksize);
 572        return (struct stats) { names, space, bcount};
 573}
 574#endif /* DX_DEBUG */
 575
 576/*
 577 * Probe for a directory leaf block to search.
 578 *
 579 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
 580 * error in the directory index, and the caller should fall back to
 581 * searching the directory normally.  The callers of dx_probe **MUST**
 582 * check for this error code, and make sure it never gets reflected
 583 * back to userspace.
 584 */
 585static struct dx_frame *
 586dx_probe(const struct qstr *d_name, struct inode *dir,
 587         struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
 588{
 589        unsigned count, indirect;
 590        struct dx_entry *at, *entries, *p, *q, *m;
 591        struct dx_root *root;
 592        struct buffer_head *bh;
 593        struct dx_frame *frame = frame_in;
 594        u32 hash;
 595
 596        frame->bh = NULL;
 597        if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
 598                goto fail;
 599        root = (struct dx_root *) bh->b_data;
 600        if (root->info.hash_version != DX_HASH_TEA &&
 601            root->info.hash_version != DX_HASH_HALF_MD4 &&
 602            root->info.hash_version != DX_HASH_LEGACY) {
 603                ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
 604                             root->info.hash_version);
 605                brelse(bh);
 606                *err = ERR_BAD_DX_DIR;
 607                goto fail;
 608        }
 609        hinfo->hash_version = root->info.hash_version;
 610        if (hinfo->hash_version <= DX_HASH_TEA)
 611                hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
 612        hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 613        if (d_name)
 614                ext4fs_dirhash(d_name->name, d_name->len, hinfo);
 615        hash = hinfo->hash;
 616
 617        if (root->info.unused_flags & 1) {
 618                ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
 619                             root->info.unused_flags);
 620                brelse(bh);
 621                *err = ERR_BAD_DX_DIR;
 622                goto fail;
 623        }
 624
 625        if ((indirect = root->info.indirect_levels) > 1) {
 626                ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
 627                             root->info.indirect_levels);
 628                brelse(bh);
 629                *err = ERR_BAD_DX_DIR;
 630                goto fail;
 631        }
 632
 633        if (!buffer_verified(bh) &&
 634            !ext4_dx_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data)) {
 635                ext4_warning(dir->i_sb, "Root failed checksum");
 636                brelse(bh);
 637                *err = ERR_BAD_DX_DIR;
 638                goto fail;
 639        }
 640        set_buffer_verified(bh);
 641
 642        entries = (struct dx_entry *) (((char *)&root->info) +
 643                                       root->info.info_length);
 644
 645        if (dx_get_limit(entries) != dx_root_limit(dir,
 646                                                   root->info.info_length)) {
 647                ext4_warning(dir->i_sb, "dx entry: limit != root limit");
 648                brelse(bh);
 649                *err = ERR_BAD_DX_DIR;
 650                goto fail;
 651        }
 652
 653        dxtrace(printk("Look up %x", hash));
 654        while (1)
 655        {
 656                count = dx_get_count(entries);
 657                if (!count || count > dx_get_limit(entries)) {
 658                        ext4_warning(dir->i_sb,
 659                                     "dx entry: no count or count > limit");
 660                        brelse(bh);
 661                        *err = ERR_BAD_DX_DIR;
 662                        goto fail2;
 663                }
 664
 665                p = entries + 1;
 666                q = entries + count - 1;
 667                while (p <= q)
 668                {
 669                        m = p + (q - p)/2;
 670                        dxtrace(printk("."));
 671                        if (dx_get_hash(m) > hash)
 672                                q = m - 1;
 673                        else
 674                                p = m + 1;
 675                }
 676
 677                if (0) // linear search cross check
 678                {
 679                        unsigned n = count - 1;
 680                        at = entries;
 681                        while (n--)
 682                        {
 683                                dxtrace(printk(","));
 684                                if (dx_get_hash(++at) > hash)
 685                                {
 686                                        at--;
 687                                        break;
 688                                }
 689                        }
 690                        assert (at == p - 1);
 691                }
 692
 693                at = p - 1;
 694                dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
 695                frame->bh = bh;
 696                frame->entries = entries;
 697                frame->at = at;
 698                if (!indirect--) return frame;
 699                if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
 700                        goto fail2;
 701                at = entries = ((struct dx_node *) bh->b_data)->entries;
 702
 703                if (!buffer_verified(bh) &&
 704                    !ext4_dx_csum_verify(dir,
 705                                         (struct ext4_dir_entry *)bh->b_data)) {
 706                        ext4_warning(dir->i_sb, "Node failed checksum");
 707                        brelse(bh);
 708                        *err = ERR_BAD_DX_DIR;
 709                        goto fail;
 710                }
 711                set_buffer_verified(bh);
 712
 713                if (dx_get_limit(entries) != dx_node_limit (dir)) {
 714                        ext4_warning(dir->i_sb,
 715                                     "dx entry: limit != node limit");
 716                        brelse(bh);
 717                        *err = ERR_BAD_DX_DIR;
 718                        goto fail2;
 719                }
 720                frame++;
 721                frame->bh = NULL;
 722        }
 723fail2:
 724        while (frame >= frame_in) {
 725                brelse(frame->bh);
 726                frame--;
 727        }
 728fail:
 729        if (*err == ERR_BAD_DX_DIR)
 730                ext4_warning(dir->i_sb,
 731                             "Corrupt dir inode %lu, running e2fsck is "
 732                             "recommended.", dir->i_ino);
 733        return NULL;
 734}
 735
 736static void dx_release (struct dx_frame *frames)
 737{
 738        if (frames[0].bh == NULL)
 739                return;
 740
 741        if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
 742                brelse(frames[1].bh);
 743        brelse(frames[0].bh);
 744}
 745
 746/*
 747 * This function increments the frame pointer to search the next leaf
 748 * block, and reads in the necessary intervening nodes if the search
 749 * should be necessary.  Whether or not the search is necessary is
 750 * controlled by the hash parameter.  If the hash value is even, then
 751 * the search is only continued if the next block starts with that
 752 * hash value.  This is used if we are searching for a specific file.
 753 *
 754 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
 755 *
 756 * This function returns 1 if the caller should continue to search,
 757 * or 0 if it should not.  If there is an error reading one of the
 758 * index blocks, it will a negative error code.
 759 *
 760 * If start_hash is non-null, it will be filled in with the starting
 761 * hash of the next page.
 762 */
 763static int ext4_htree_next_block(struct inode *dir, __u32 hash,
 764                                 struct dx_frame *frame,
 765                                 struct dx_frame *frames,
 766                                 __u32 *start_hash)
 767{
 768        struct dx_frame *p;
 769        struct buffer_head *bh;
 770        int err, num_frames = 0;
 771        __u32 bhash;
 772
 773        p = frame;
 774        /*
 775         * Find the next leaf page by incrementing the frame pointer.
 776         * If we run out of entries in the interior node, loop around and
 777         * increment pointer in the parent node.  When we break out of
 778         * this loop, num_frames indicates the number of interior
 779         * nodes need to be read.
 780         */
 781        while (1) {
 782                if (++(p->at) < p->entries + dx_get_count(p->entries))
 783                        break;
 784                if (p == frames)
 785                        return 0;
 786                num_frames++;
 787                p--;
 788        }
 789
 790        /*
 791         * If the hash is 1, then continue only if the next page has a
 792         * continuation hash of any value.  This is used for readdir
 793         * handling.  Otherwise, check to see if the hash matches the
 794         * desired contiuation hash.  If it doesn't, return since
 795         * there's no point to read in the successive index pages.
 796         */
 797        bhash = dx_get_hash(p->at);
 798        if (start_hash)
 799                *start_hash = bhash;
 800        if ((hash & 1) == 0) {
 801                if ((bhash & ~1) != hash)
 802                        return 0;
 803        }
 804        /*
 805         * If the hash is HASH_NB_ALWAYS, we always go to the next
 806         * block so no check is necessary
 807         */
 808        while (num_frames--) {
 809                if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
 810                                      0, &err)))
 811                        return err; /* Failure */
 812
 813                if (!buffer_verified(bh) &&
 814                    !ext4_dx_csum_verify(dir,
 815                                         (struct ext4_dir_entry *)bh->b_data)) {
 816                        ext4_warning(dir->i_sb, "Node failed checksum");
 817                        return -EIO;
 818                }
 819                set_buffer_verified(bh);
 820
 821                p++;
 822                brelse(p->bh);
 823                p->bh = bh;
 824                p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
 825        }
 826        return 1;
 827}
 828
 829
 830/*
 831 * This function fills a red-black tree with information from a
 832 * directory block.  It returns the number directory entries loaded
 833 * into the tree.  If there is an error it is returned in err.
 834 */
 835static int htree_dirblock_to_tree(struct file *dir_file,
 836                                  struct inode *dir, ext4_lblk_t block,
 837                                  struct dx_hash_info *hinfo,
 838                                  __u32 start_hash, __u32 start_minor_hash)
 839{
 840        struct buffer_head *bh;
 841        struct ext4_dir_entry_2 *de, *top;
 842        int err, count = 0;
 843
 844        dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
 845                                                        (unsigned long)block));
 846        if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
 847                return err;
 848
 849        if (!buffer_verified(bh) &&
 850            !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
 851                return -EIO;
 852        set_buffer_verified(bh);
 853
 854        de = (struct ext4_dir_entry_2 *) bh->b_data;
 855        top = (struct ext4_dir_entry_2 *) ((char *) de +
 856                                           dir->i_sb->s_blocksize -
 857                                           EXT4_DIR_REC_LEN(0));
 858        for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
 859                if (ext4_check_dir_entry(dir, NULL, de, bh,
 860                                (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
 861                                         + ((char *)de - bh->b_data))) {
 862                        /* On error, skip the f_pos to the next block. */
 863                        dir_file->f_pos = (dir_file->f_pos |
 864                                        (dir->i_sb->s_blocksize - 1)) + 1;
 865                        brelse(bh);
 866                        return count;
 867                }
 868                ext4fs_dirhash(de->name, de->name_len, hinfo);
 869                if ((hinfo->hash < start_hash) ||
 870                    ((hinfo->hash == start_hash) &&
 871                     (hinfo->minor_hash < start_minor_hash)))
 872                        continue;
 873                if (de->inode == 0)
 874                        continue;
 875                if ((err = ext4_htree_store_dirent(dir_file,
 876                                   hinfo->hash, hinfo->minor_hash, de)) != 0) {
 877                        brelse(bh);
 878                        return err;
 879                }
 880                count++;
 881        }
 882        brelse(bh);
 883        return count;
 884}
 885
 886
 887/*
 888 * This function fills a red-black tree with information from a
 889 * directory.  We start scanning the directory in hash order, starting
 890 * at start_hash and start_minor_hash.
 891 *
 892 * This function returns the number of entries inserted into the tree,
 893 * or a negative error code.
 894 */
 895int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 896                         __u32 start_minor_hash, __u32 *next_hash)
 897{
 898        struct dx_hash_info hinfo;
 899        struct ext4_dir_entry_2 *de;
 900        struct dx_frame frames[2], *frame;
 901        struct inode *dir;
 902        ext4_lblk_t block;
 903        int count = 0;
 904        int ret, err;
 905        __u32 hashval;
 906
 907        dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
 908                       start_hash, start_minor_hash));
 909        dir = dir_file->f_path.dentry->d_inode;
 910        if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
 911                hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
 912                if (hinfo.hash_version <= DX_HASH_TEA)
 913                        hinfo.hash_version +=
 914                                EXT4_SB(dir->i_sb)->s_hash_unsigned;
 915                hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
 916                count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
 917                                               start_hash, start_minor_hash);
 918                *next_hash = ~0;
 919                return count;
 920        }
 921        hinfo.hash = start_hash;
 922        hinfo.minor_hash = 0;
 923        frame = dx_probe(NULL, dir, &hinfo, frames, &err);
 924        if (!frame)
 925                return err;
 926
 927        /* Add '.' and '..' from the htree header */
 928        if (!start_hash && !start_minor_hash) {
 929                de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 930                if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
 931                        goto errout;
 932                count++;
 933        }
 934        if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
 935                de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
 936                de = ext4_next_entry(de, dir->i_sb->s_blocksize);
 937                if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
 938                        goto errout;
 939                count++;
 940        }
 941
 942        while (1) {
 943                block = dx_get_block(frame->at);
 944                ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
 945                                             start_hash, start_minor_hash);
 946                if (ret < 0) {
 947                        err = ret;
 948                        goto errout;
 949                }
 950                count += ret;
 951                hashval = ~0;
 952                ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
 953                                            frame, frames, &hashval);
 954                *next_hash = hashval;
 955                if (ret < 0) {
 956                        err = ret;
 957                        goto errout;
 958                }
 959                /*
 960                 * Stop if:  (a) there are no more entries, or
 961                 * (b) we have inserted at least one entry and the
 962                 * next hash value is not a continuation
 963                 */
 964                if ((ret == 0) ||
 965                    (count && ((hashval & 1) == 0)))
 966                        break;
 967        }
 968        dx_release(frames);
 969        dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
 970                       "next hash: %x\n", count, *next_hash));
 971        return count;
 972errout:
 973        dx_release(frames);
 974        return (err);
 975}
 976
 977
 978/*
 979 * Directory block splitting, compacting
 980 */
 981
 982/*
 983 * Create map of hash values, offsets, and sizes, stored at end of block.
 984 * Returns number of entries mapped.
 985 */
 986static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
 987                       struct dx_hash_info *hinfo,
 988                       struct dx_map_entry *map_tail)
 989{
 990        int count = 0;
 991        char *base = (char *) de;
 992        struct dx_hash_info h = *hinfo;
 993
 994        while ((char *) de < base + blocksize) {
 995                if (de->name_len && de->inode) {
 996                        ext4fs_dirhash(de->name, de->name_len, &h);
 997                        map_tail--;
 998                        map_tail->hash = h.hash;
 999                        map_tail->offs = ((char *) de - base)>>2;
1000                        map_tail->size = le16_to_cpu(de->rec_len);
1001                        count++;
1002                        cond_resched();
1003                }
1004                /* XXX: do we need to check rec_len == 0 case? -Chris */
1005                de = ext4_next_entry(de, blocksize);
1006        }
1007        return count;
1008}
1009
1010/* Sort map by hash value */
1011static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1012{
1013        struct dx_map_entry *p, *q, *top = map + count - 1;
1014        int more;
1015        /* Combsort until bubble sort doesn't suck */
1016        while (count > 2) {
1017                count = count*10/13;
1018                if (count - 9 < 2) /* 9, 10 -> 11 */
1019                        count = 11;
1020                for (p = top, q = p - count; q >= map; p--, q--)
1021                        if (p->hash < q->hash)
1022                                swap(*p, *q);
1023        }
1024        /* Garden variety bubble sort */
1025        do {
1026                more = 0;
1027                q = top;
1028                while (q-- > map) {
1029                        if (q[1].hash >= q[0].hash)
1030                                continue;
1031                        swap(*(q+1), *q);
1032                        more = 1;
1033                }
1034        } while(more);
1035}
1036
1037static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
1038{
1039        struct dx_entry *entries = frame->entries;
1040        struct dx_entry *old = frame->at, *new = old + 1;
1041        int count = dx_get_count(entries);
1042
1043        assert(count < dx_get_limit(entries));
1044        assert(old < entries + count);
1045        memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1046        dx_set_hash(new, hash);
1047        dx_set_block(new, block);
1048        dx_set_count(entries, count + 1);
1049}
1050
1051static void ext4_update_dx_flag(struct inode *inode)
1052{
1053        if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
1054                                     EXT4_FEATURE_COMPAT_DIR_INDEX))
1055                ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
1056}
1057
1058/*
1059 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
1060 *
1061 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
1062 * `de != NULL' is guaranteed by caller.
1063 */
1064static inline int ext4_match (int len, const char * const name,
1065                              struct ext4_dir_entry_2 * de)
1066{
1067        if (len != de->name_len)
1068                return 0;
1069        if (!de->inode)
1070                return 0;
1071        return !memcmp(name, de->name, len);
1072}
1073
1074/*
1075 * Returns 0 if not found, -1 on failure, and 1 on success
1076 */
1077static inline int search_dirblock(struct buffer_head *bh,
1078                                  struct inode *dir,
1079                                  const struct qstr *d_name,
1080                                  unsigned int offset,
1081                                  struct ext4_dir_entry_2 ** res_dir)
1082{
1083        struct ext4_dir_entry_2 * de;
1084        char * dlimit;
1085        int de_len;
1086        const char *name = d_name->name;
1087        int namelen = d_name->len;
1088
1089        de = (struct ext4_dir_entry_2 *) bh->b_data;
1090        dlimit = bh->b_data + dir->i_sb->s_blocksize;
1091        while ((char *) de < dlimit) {
1092                /* this code is executed quadratically often */
1093                /* do minimal checking `by hand' */
1094
1095                if ((char *) de + namelen <= dlimit &&
1096                    ext4_match (namelen, name, de)) {
1097                        /* found a match - just to be sure, do a full check */
1098                        if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1099                                return -1;
1100                        *res_dir = de;
1101                        return 1;
1102                }
1103                /* prevent looping on a bad block */
1104                de_len = ext4_rec_len_from_disk(de->rec_len,
1105                                                dir->i_sb->s_blocksize);
1106                if (de_len <= 0)
1107                        return -1;
1108                offset += de_len;
1109                de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
1110        }
1111        return 0;
1112}
1113
1114
1115/*
1116 *      ext4_find_entry()
1117 *
1118 * finds an entry in the specified directory with the wanted name. It
1119 * returns the cache buffer in which the entry was found, and the entry
1120 * itself (as a parameter - res_dir). It does NOT read the inode of the
1121 * entry - you'll have to do that yourself if you want to.
1122 *
1123 * The returned buffer_head has ->b_count elevated.  The caller is expected
1124 * to brelse() it when appropriate.
1125 */
1126static struct buffer_head * ext4_find_entry (struct inode *dir,
1127                                        const struct qstr *d_name,
1128                                        struct ext4_dir_entry_2 ** res_dir)
1129{
1130        struct super_block *sb;
1131        struct buffer_head *bh_use[NAMEI_RA_SIZE];
1132        struct buffer_head *bh, *ret = NULL;
1133        ext4_lblk_t start, block, b;
1134        const u8 *name = d_name->name;
1135        int ra_max = 0;         /* Number of bh's in the readahead
1136                                   buffer, bh_use[] */
1137        int ra_ptr = 0;         /* Current index into readahead
1138                                   buffer */
1139        int num = 0;
1140        ext4_lblk_t  nblocks;
1141        int i, err;
1142        int namelen;
1143
1144        *res_dir = NULL;
1145        sb = dir->i_sb;
1146        namelen = d_name->len;
1147        if (namelen > EXT4_NAME_LEN)
1148                return NULL;
1149        if ((namelen <= 2) && (name[0] == '.') &&
1150            (name[1] == '.' || name[1] == '\0')) {
1151                /*
1152                 * "." or ".." will only be in the first block
1153                 * NFS may look up ".."; "." should be handled by the VFS
1154                 */
1155                block = start = 0;
1156                nblocks = 1;
1157                goto restart;
1158        }
1159        if (is_dx(dir)) {
1160                bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
1161                /*
1162                 * On success, or if the error was file not found,
1163                 * return.  Otherwise, fall back to doing a search the
1164                 * old fashioned way.
1165                 */
1166                if (bh || (err != ERR_BAD_DX_DIR))
1167                        return bh;
1168                dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1169                               "falling back\n"));
1170        }
1171        nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1172        start = EXT4_I(dir)->i_dir_start_lookup;
1173        if (start >= nblocks)
1174                start = 0;
1175        block = start;
1176restart:
1177        do {
1178                /*
1179                 * We deal with the read-ahead logic here.
1180                 */
1181                if (ra_ptr >= ra_max) {
1182                        /* Refill the readahead buffer */
1183                        ra_ptr = 0;
1184                        b = block;
1185                        for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1186                                /*
1187                                 * Terminate if we reach the end of the
1188                                 * directory and must wrap, or if our
1189                                 * search has finished at this block.
1190                                 */
1191                                if (b >= nblocks || (num && block == start)) {
1192                                        bh_use[ra_max] = NULL;
1193                                        break;
1194                                }
1195                                num++;
1196                                bh = ext4_getblk(NULL, dir, b++, 0, &err);
1197                                bh_use[ra_max] = bh;
1198                                if (bh)
1199                                        ll_rw_block(READ | REQ_META | REQ_PRIO,
1200                                                    1, &bh);
1201                        }
1202                }
1203                if ((bh = bh_use[ra_ptr++]) == NULL)
1204                        goto next;
1205                wait_on_buffer(bh);
1206                if (!buffer_uptodate(bh)) {
1207                        /* read error, skip block & hope for the best */
1208                        EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1209                                         (unsigned long) block);
1210                        brelse(bh);
1211                        goto next;
1212                }
1213                if (!buffer_verified(bh) &&
1214                    !ext4_dirent_csum_verify(dir,
1215                                (struct ext4_dir_entry *)bh->b_data)) {
1216                        EXT4_ERROR_INODE(dir, "checksumming directory "
1217                                         "block %lu", (unsigned long)block);
1218                        brelse(bh);
1219                        goto next;
1220                }
1221                set_buffer_verified(bh);
1222                i = search_dirblock(bh, dir, d_name,
1223                            block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
1224                if (i == 1) {
1225                        EXT4_I(dir)->i_dir_start_lookup = block;
1226                        ret = bh;
1227                        goto cleanup_and_exit;
1228                } else {
1229                        brelse(bh);
1230                        if (i < 0)
1231                                goto cleanup_and_exit;
1232                }
1233        next:
1234                if (++block >= nblocks)
1235                        block = 0;
1236        } while (block != start);
1237
1238        /*
1239         * If the directory has grown while we were searching, then
1240         * search the last part of the directory before giving up.
1241         */
1242        block = nblocks;
1243        nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1244        if (block < nblocks) {
1245                start = 0;
1246                goto restart;
1247        }
1248
1249cleanup_and_exit:
1250        /* Clean up the read-ahead blocks */
1251        for (; ra_ptr < ra_max; ra_ptr++)
1252                brelse(bh_use[ra_ptr]);
1253        return ret;
1254}
1255
1256static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
1257                       struct ext4_dir_entry_2 **res_dir, int *err)
1258{
1259        struct super_block * sb = dir->i_sb;
1260        struct dx_hash_info     hinfo;
1261        struct dx_frame frames[2], *frame;
1262        struct buffer_head *bh;
1263        ext4_lblk_t block;
1264        int retval;
1265
1266        if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1267                return NULL;
1268        do {
1269                block = dx_get_block(frame->at);
1270                if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
1271                        goto errout;
1272
1273                if (!buffer_verified(bh) &&
1274                    !ext4_dirent_csum_verify(dir,
1275                                (struct ext4_dir_entry *)bh->b_data)) {
1276                        EXT4_ERROR_INODE(dir, "checksumming directory "
1277                                         "block %lu", (unsigned long)block);
1278                        brelse(bh);
1279                        *err = -EIO;
1280                        goto errout;
1281                }
1282                set_buffer_verified(bh);
1283                retval = search_dirblock(bh, dir, d_name,
1284                                         block << EXT4_BLOCK_SIZE_BITS(sb),
1285                                         res_dir);
1286                if (retval == 1) {      /* Success! */
1287                        dx_release(frames);
1288                        return bh;
1289                }
1290                brelse(bh);
1291                if (retval == -1) {
1292                        *err = ERR_BAD_DX_DIR;
1293                        goto errout;
1294                }
1295
1296                /* Check to see if we should continue to search */
1297                retval = ext4_htree_next_block(dir, hinfo.hash, frame,
1298                                               frames, NULL);
1299                if (retval < 0) {
1300                        ext4_warning(sb,
1301                             "error reading index page in directory #%lu",
1302                             dir->i_ino);
1303                        *err = retval;
1304                        goto errout;
1305                }
1306        } while (retval == 1);
1307
1308        *err = -ENOENT;
1309errout:
1310        dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
1311        dx_release (frames);
1312        return NULL;
1313}
1314
1315static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
1316{
1317        struct inode *inode;
1318        struct ext4_dir_entry_2 *de;
1319        struct buffer_head *bh;
1320
1321        if (dentry->d_name.len > EXT4_NAME_LEN)
1322                return ERR_PTR(-ENAMETOOLONG);
1323
1324        bh = ext4_find_entry(dir, &dentry->d_name, &de);
1325        inode = NULL;
1326        if (bh) {
1327                __u32 ino = le32_to_cpu(de->inode);
1328                brelse(bh);
1329                if (!ext4_valid_inum(dir->i_sb, ino)) {
1330                        EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
1331                        return ERR_PTR(-EIO);
1332                }
1333                if (unlikely(ino == dir->i_ino)) {
1334                        EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1335                                         dentry->d_name.len,
1336                                         dentry->d_name.name);
1337                        return ERR_PTR(-EIO);
1338                }
1339                inode = ext4_iget(dir->i_sb, ino);
1340                if (inode == ERR_PTR(-ESTALE)) {
1341                        EXT4_ERROR_INODE(dir,
1342                                         "deleted inode referenced: %u",
1343                                         ino);
1344                        return ERR_PTR(-EIO);
1345                }
1346        }
1347        return d_splice_alias(inode, dentry);
1348}
1349
1350
1351struct dentry *ext4_get_parent(struct dentry *child)
1352{
1353        __u32 ino;
1354        static const struct qstr dotdot = QSTR_INIT("..", 2);
1355        struct ext4_dir_entry_2 * de;
1356        struct buffer_head *bh;
1357
1358        bh = ext4_find_entry(child->d_inode, &dotdot, &de);
1359        if (!bh)
1360                return ERR_PTR(-ENOENT);
1361        ino = le32_to_cpu(de->inode);
1362        brelse(bh);
1363
1364        if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1365                EXT4_ERROR_INODE(child->d_inode,
1366                                 "bad parent inode number: %u", ino);
1367                return ERR_PTR(-EIO);
1368        }
1369
1370        return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
1371}
1372
1373#define S_SHIFT 12
1374static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1375        [S_IFREG >> S_SHIFT]    = EXT4_FT_REG_FILE,
1376        [S_IFDIR >> S_SHIFT]    = EXT4_FT_DIR,
1377        [S_IFCHR >> S_SHIFT]    = EXT4_FT_CHRDEV,
1378        [S_IFBLK >> S_SHIFT]    = EXT4_FT_BLKDEV,
1379        [S_IFIFO >> S_SHIFT]    = EXT4_FT_FIFO,
1380        [S_IFSOCK >> S_SHIFT]   = EXT4_FT_SOCK,
1381        [S_IFLNK >> S_SHIFT]    = EXT4_FT_SYMLINK,
1382};
1383
1384static inline void ext4_set_de_type(struct super_block *sb,
1385                                struct ext4_dir_entry_2 *de,
1386                                umode_t mode) {
1387        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1388                de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1389}
1390
1391/*
1392 * Move count entries from end of map between two memory locations.
1393 * Returns pointer to last entry moved.
1394 */
1395static struct ext4_dir_entry_2 *
1396dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1397                unsigned blocksize)
1398{
1399        unsigned rec_len = 0;
1400
1401        while (count--) {
1402                struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1403                                                (from + (map->offs<<2));
1404                rec_len = EXT4_DIR_REC_LEN(de->name_len);
1405                memcpy (to, de, rec_len);
1406                ((struct ext4_dir_entry_2 *) to)->rec_len =
1407                                ext4_rec_len_to_disk(rec_len, blocksize);
1408                de->inode = 0;
1409                map++;
1410                to += rec_len;
1411        }
1412        return (struct ext4_dir_entry_2 *) (to - rec_len);
1413}
1414
1415/*
1416 * Compact each dir entry in the range to the minimal rec_len.
1417 * Returns pointer to last entry in range.
1418 */
1419static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
1420{
1421        struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
1422        unsigned rec_len = 0;
1423
1424        prev = to = de;
1425        while ((char*)de < base + blocksize) {
1426                next = ext4_next_entry(de, blocksize);
1427                if (de->inode && de->name_len) {
1428                        rec_len = EXT4_DIR_REC_LEN(de->name_len);
1429                        if (de > to)
1430                                memmove(to, de, rec_len);
1431                        to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
1432                        prev = to;
1433                        to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
1434                }
1435                de = next;
1436        }
1437        return prev;
1438}
1439
1440/*
1441 * Split a full leaf block to make room for a new dir entry.
1442 * Allocate a new block, and move entries so that they are approx. equally full.
1443 * Returns pointer to de in block into which the new entry will be inserted.
1444 */
1445static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1446                        struct buffer_head **bh,struct dx_frame *frame,
1447                        struct dx_hash_info *hinfo, int *error)
1448{
1449        unsigned blocksize = dir->i_sb->s_blocksize;
1450        unsigned count, continued;
1451        struct buffer_head *bh2;
1452        ext4_lblk_t newblock;
1453        u32 hash2;
1454        struct dx_map_entry *map;
1455        char *data1 = (*bh)->b_data, *data2;
1456        unsigned split, move, size;
1457        struct ext4_dir_entry_2 *de = NULL, *de2;
1458        struct ext4_dir_entry_tail *t;
1459        int     csum_size = 0;
1460        int     err = 0, i;
1461
1462        if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1463                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1464                csum_size = sizeof(struct ext4_dir_entry_tail);
1465
1466        bh2 = ext4_append (handle, dir, &newblock, &err);
1467        if (!(bh2)) {
1468                brelse(*bh);
1469                *bh = NULL;
1470                goto errout;
1471        }
1472
1473        BUFFER_TRACE(*bh, "get_write_access");
1474        err = ext4_journal_get_write_access(handle, *bh);
1475        if (err)
1476                goto journal_error;
1477
1478        BUFFER_TRACE(frame->bh, "get_write_access");
1479        err = ext4_journal_get_write_access(handle, frame->bh);
1480        if (err)
1481                goto journal_error;
1482
1483        data2 = bh2->b_data;
1484
1485        /* create map in the end of data2 block */
1486        map = (struct dx_map_entry *) (data2 + blocksize);
1487        count = dx_make_map((struct ext4_dir_entry_2 *) data1,
1488                             blocksize, hinfo, map);
1489        map -= count;
1490        dx_sort_map(map, count);
1491        /* Split the existing block in the middle, size-wise */
1492        size = 0;
1493        move = 0;
1494        for (i = count-1; i >= 0; i--) {
1495                /* is more than half of this entry in 2nd half of the block? */
1496                if (size + map[i].size/2 > blocksize/2)
1497                        break;
1498                size += map[i].size;
1499                move++;
1500        }
1501        /* map index at which we will split */
1502        split = count - move;
1503        hash2 = map[split].hash;
1504        continued = hash2 == map[split - 1].hash;
1505        dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1506                        (unsigned long)dx_get_block(frame->at),
1507                                        hash2, split, count-split));
1508
1509        /* Fancy dance to stay within two buffers */
1510        de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
1511        de = dx_pack_dirents(data1, blocksize);
1512        de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1513                                           (char *) de,
1514                                           blocksize);
1515        de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1516                                            (char *) de2,
1517                                            blocksize);
1518        if (csum_size) {
1519                t = EXT4_DIRENT_TAIL(data2, blocksize);
1520                initialize_dirent_tail(t, blocksize);
1521
1522                t = EXT4_DIRENT_TAIL(data1, blocksize);
1523                initialize_dirent_tail(t, blocksize);
1524        }
1525
1526        dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1527        dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
1528
1529        /* Which block gets the new entry? */
1530        if (hinfo->hash >= hash2)
1531        {
1532                swap(*bh, bh2);
1533                de = de2;
1534        }
1535        dx_insert_block(frame, hash2 + continued, newblock);
1536        err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
1537        if (err)
1538                goto journal_error;
1539        err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1540        if (err)
1541                goto journal_error;
1542        brelse(bh2);
1543        dxtrace(dx_show_index("frame", frame->entries));
1544        return de;
1545
1546journal_error:
1547        brelse(*bh);
1548        brelse(bh2);
1549        *bh = NULL;
1550        ext4_std_error(dir->i_sb, err);
1551errout:
1552        *error = err;
1553        return NULL;
1554}
1555
1556/*
1557 * Add a new entry into a directory (leaf) block.  If de is non-NULL,
1558 * it points to a directory entry which is guaranteed to be large
1559 * enough for new directory entry.  If de is NULL, then
1560 * add_dirent_to_buf will attempt search the directory block for
1561 * space.  It will return -ENOSPC if no space is available, and -EIO
1562 * and -EEXIST if directory entry already exists.
1563 */
1564static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1565                             struct inode *inode, struct ext4_dir_entry_2 *de,
1566                             struct buffer_head *bh)
1567{
1568        struct inode    *dir = dentry->d_parent->d_inode;
1569        const char      *name = dentry->d_name.name;
1570        int             namelen = dentry->d_name.len;
1571        unsigned int    offset = 0;
1572        unsigned int    blocksize = dir->i_sb->s_blocksize;
1573        unsigned short  reclen;
1574        int             nlen, rlen, err;
1575        char            *top;
1576        int             csum_size = 0;
1577
1578        if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1579                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1580                csum_size = sizeof(struct ext4_dir_entry_tail);
1581
1582        reclen = EXT4_DIR_REC_LEN(namelen);
1583        if (!de) {
1584                de = (struct ext4_dir_entry_2 *)bh->b_data;
1585                top = bh->b_data + (blocksize - csum_size) - reclen;
1586                while ((char *) de <= top) {
1587                        if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
1588                                return -EIO;
1589                        if (ext4_match(namelen, name, de))
1590                                return -EEXIST;
1591                        nlen = EXT4_DIR_REC_LEN(de->name_len);
1592                        rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1593                        if ((de->inode? rlen - nlen: rlen) >= reclen)
1594                                break;
1595                        de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
1596                        offset += rlen;
1597                }
1598                if ((char *) de > top)
1599                        return -ENOSPC;
1600        }
1601        BUFFER_TRACE(bh, "get_write_access");
1602        err = ext4_journal_get_write_access(handle, bh);
1603        if (err) {
1604                ext4_std_error(dir->i_sb, err);
1605                return err;
1606        }
1607
1608        /* By now the buffer is marked for journaling */
1609        nlen = EXT4_DIR_REC_LEN(de->name_len);
1610        rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
1611        if (de->inode) {
1612                struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
1613                de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1614                de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
1615                de = de1;
1616        }
1617        de->file_type = EXT4_FT_UNKNOWN;
1618        de->inode = cpu_to_le32(inode->i_ino);
1619        ext4_set_de_type(dir->i_sb, de, inode->i_mode);
1620        de->name_len = namelen;
1621        memcpy(de->name, name, namelen);
1622        /*
1623         * XXX shouldn't update any times until successful
1624         * completion of syscall, but too many callers depend
1625         * on this.
1626         *
1627         * XXX similarly, too many callers depend on
1628         * ext4_new_inode() setting the times, but error
1629         * recovery deletes the inode, so the worst that can
1630         * happen is that the times are slightly out of date
1631         * and/or different from the directory change time.
1632         */
1633        dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
1634        ext4_update_dx_flag(dir);
1635        dir->i_version++;
1636        ext4_mark_inode_dirty(handle, dir);
1637        BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1638        err = ext4_handle_dirty_dirent_node(handle, dir, bh);
1639        if (err)
1640                ext4_std_error(dir->i_sb, err);
1641        return 0;
1642}
1643
1644/*
1645 * This converts a one block unindexed directory to a 3 block indexed
1646 * directory, and adds the dentry to the indexed directory.
1647 */
1648static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1649                            struct inode *inode, struct buffer_head *bh)
1650{
1651        struct inode    *dir = dentry->d_parent->d_inode;
1652        const char      *name = dentry->d_name.name;
1653        int             namelen = dentry->d_name.len;
1654        struct buffer_head *bh2;
1655        struct dx_root  *root;
1656        struct dx_frame frames[2], *frame;
1657        struct dx_entry *entries;
1658        struct ext4_dir_entry_2 *de, *de2;
1659        struct ext4_dir_entry_tail *t;
1660        char            *data1, *top;
1661        unsigned        len;
1662        int             retval;
1663        unsigned        blocksize;
1664        struct dx_hash_info hinfo;
1665        ext4_lblk_t  block;
1666        struct fake_dirent *fde;
1667        int             csum_size = 0;
1668
1669        if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1670                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1671                csum_size = sizeof(struct ext4_dir_entry_tail);
1672
1673        blocksize =  dir->i_sb->s_blocksize;
1674        dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
1675        retval = ext4_journal_get_write_access(handle, bh);
1676        if (retval) {
1677                ext4_std_error(dir->i_sb, retval);
1678                brelse(bh);
1679                return retval;
1680        }
1681        root = (struct dx_root *) bh->b_data;
1682
1683        /* The 0th block becomes the root, move the dirents out */
1684        fde = &root->dotdot;
1685        de = (struct ext4_dir_entry_2 *)((char *)fde +
1686                ext4_rec_len_from_disk(fde->rec_len, blocksize));
1687        if ((char *) de >= (((char *) root) + blocksize)) {
1688                EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
1689                brelse(bh);
1690                return -EIO;
1691        }
1692        len = ((char *) root) + (blocksize - csum_size) - (char *) de;
1693
1694        /* Allocate new block for the 0th block's dirents */
1695        bh2 = ext4_append(handle, dir, &block, &retval);
1696        if (!(bh2)) {
1697                brelse(bh);
1698                return retval;
1699        }
1700        ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
1701        data1 = bh2->b_data;
1702
1703        memcpy (data1, de, len);
1704        de = (struct ext4_dir_entry_2 *) data1;
1705        top = data1 + len;
1706        while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
1707                de = de2;
1708        de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1709                                           (char *) de,
1710                                           blocksize);
1711
1712        if (csum_size) {
1713                t = EXT4_DIRENT_TAIL(data1, blocksize);
1714                initialize_dirent_tail(t, blocksize);
1715        }
1716
1717        /* Initialize the root; the dot dirents already exist */
1718        de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1719        de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1720                                           blocksize);
1721        memset (&root->info, 0, sizeof(root->info));
1722        root->info.info_length = sizeof(root->info);
1723        root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
1724        entries = root->entries;
1725        dx_set_block(entries, 1);
1726        dx_set_count(entries, 1);
1727        dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
1728
1729        /* Initialize as for dx_probe */
1730        hinfo.hash_version = root->info.hash_version;
1731        if (hinfo.hash_version <= DX_HASH_TEA)
1732                hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
1733        hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1734        ext4fs_dirhash(name, namelen, &hinfo);
1735        frame = frames;
1736        frame->entries = entries;
1737        frame->at = entries;
1738        frame->bh = bh;
1739        bh = bh2;
1740
1741        ext4_handle_dirty_dx_node(handle, dir, frame->bh);
1742        ext4_handle_dirty_dirent_node(handle, dir, bh);
1743
1744        de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1745        if (!de) {
1746                /*
1747                 * Even if the block split failed, we have to properly write
1748                 * out all the changes we did so far. Otherwise we can end up
1749                 * with corrupted filesystem.
1750                 */
1751                ext4_mark_inode_dirty(handle, dir);
1752                dx_release(frames);
1753                return retval;
1754        }
1755        dx_release(frames);
1756
1757        retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1758        brelse(bh);
1759        return retval;
1760}
1761
1762/*
1763 *      ext4_add_entry()
1764 *
1765 * adds a file entry to the specified directory, using the same
1766 * semantics as ext4_find_entry(). It returns NULL if it failed.
1767 *
1768 * NOTE!! The inode part of 'de' is left at 0 - which means you
1769 * may not sleep between calling this and putting something into
1770 * the entry, as someone else might have used it while you slept.
1771 */
1772static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1773                          struct inode *inode)
1774{
1775        struct inode *dir = dentry->d_parent->d_inode;
1776        struct buffer_head *bh;
1777        struct ext4_dir_entry_2 *de;
1778        struct ext4_dir_entry_tail *t;
1779        struct super_block *sb;
1780        int     retval;
1781        int     dx_fallback=0;
1782        unsigned blocksize;
1783        ext4_lblk_t block, blocks;
1784        int     csum_size = 0;
1785
1786        if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1787                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1788                csum_size = sizeof(struct ext4_dir_entry_tail);
1789
1790        sb = dir->i_sb;
1791        blocksize = sb->s_blocksize;
1792        if (!dentry->d_name.len)
1793                return -EINVAL;
1794        if (is_dx(dir)) {
1795                retval = ext4_dx_add_entry(handle, dentry, inode);
1796                if (!retval || (retval != ERR_BAD_DX_DIR))
1797                        return retval;
1798                ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
1799                dx_fallback++;
1800                ext4_mark_inode_dirty(handle, dir);
1801        }
1802        blocks = dir->i_size >> sb->s_blocksize_bits;
1803        for (block = 0; block < blocks; block++) {
1804                bh = ext4_bread(handle, dir, block, 0, &retval);
1805                if(!bh)
1806                        return retval;
1807                if (!buffer_verified(bh) &&
1808                    !ext4_dirent_csum_verify(dir,
1809                                (struct ext4_dir_entry *)bh->b_data))
1810                        return -EIO;
1811                set_buffer_verified(bh);
1812                retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1813                if (retval != -ENOSPC) {
1814                        brelse(bh);
1815                        return retval;
1816                }
1817
1818                if (blocks == 1 && !dx_fallback &&
1819                    EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1820                        return make_indexed_dir(handle, dentry, inode, bh);
1821                brelse(bh);
1822        }
1823        bh = ext4_append(handle, dir, &block, &retval);
1824        if (!bh)
1825                return retval;
1826        de = (struct ext4_dir_entry_2 *) bh->b_data;
1827        de->inode = 0;
1828        de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1829
1830        if (csum_size) {
1831                t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1832                initialize_dirent_tail(t, blocksize);
1833        }
1834
1835        retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1836        brelse(bh);
1837        if (retval == 0)
1838                ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
1839        return retval;
1840}
1841
1842/*
1843 * Returns 0 for success, or a negative error value
1844 */
1845static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
1846                             struct inode *inode)
1847{
1848        struct dx_frame frames[2], *frame;
1849        struct dx_entry *entries, *at;
1850        struct dx_hash_info hinfo;
1851        struct buffer_head *bh;
1852        struct inode *dir = dentry->d_parent->d_inode;
1853        struct super_block *sb = dir->i_sb;
1854        struct ext4_dir_entry_2 *de;
1855        int err;
1856
1857        frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
1858        if (!frame)
1859                return err;
1860        entries = frame->entries;
1861        at = frame->at;
1862
1863        if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1864                goto cleanup;
1865
1866        if (!buffer_verified(bh) &&
1867            !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
1868                goto journal_error;
1869        set_buffer_verified(bh);
1870
1871        BUFFER_TRACE(bh, "get_write_access");
1872        err = ext4_journal_get_write_access(handle, bh);
1873        if (err)
1874                goto journal_error;
1875
1876        err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1877        if (err != -ENOSPC)
1878                goto cleanup;
1879
1880        /* Block full, should compress but for now just split */
1881        dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
1882                       dx_get_count(entries), dx_get_limit(entries)));
1883        /* Need to split index? */
1884        if (dx_get_count(entries) == dx_get_limit(entries)) {
1885                ext4_lblk_t newblock;
1886                unsigned icount = dx_get_count(entries);
1887                int levels = frame - frames;
1888                struct dx_entry *entries2;
1889                struct dx_node *node2;
1890                struct buffer_head *bh2;
1891
1892                if (levels && (dx_get_count(frames->entries) ==
1893                               dx_get_limit(frames->entries))) {
1894                        ext4_warning(sb, "Directory index full!");
1895                        err = -ENOSPC;
1896                        goto cleanup;
1897                }
1898                bh2 = ext4_append (handle, dir, &newblock, &err);
1899                if (!(bh2))
1900                        goto cleanup;
1901                node2 = (struct dx_node *)(bh2->b_data);
1902                entries2 = node2->entries;
1903                memset(&node2->fake, 0, sizeof(struct fake_dirent));
1904                node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1905                                                           sb->s_blocksize);
1906                BUFFER_TRACE(frame->bh, "get_write_access");
1907                err = ext4_journal_get_write_access(handle, frame->bh);
1908                if (err)
1909                        goto journal_error;
1910                if (levels) {
1911                        unsigned icount1 = icount/2, icount2 = icount - icount1;
1912                        unsigned hash2 = dx_get_hash(entries + icount1);
1913                        dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1914                                       icount1, icount2));
1915
1916                        BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1917                        err = ext4_journal_get_write_access(handle,
1918                                                             frames[0].bh);
1919                        if (err)
1920                                goto journal_error;
1921
1922                        memcpy((char *) entries2, (char *) (entries + icount1),
1923                               icount2 * sizeof(struct dx_entry));
1924                        dx_set_count(entries, icount1);
1925                        dx_set_count(entries2, icount2);
1926                        dx_set_limit(entries2, dx_node_limit(dir));
1927
1928                        /* Which index block gets the new entry? */
1929                        if (at - entries >= icount1) {
1930                                frame->at = at = at - entries - icount1 + entries2;
1931                                frame->entries = entries = entries2;
1932                                swap(frame->bh, bh2);
1933                        }
1934                        dx_insert_block(frames + 0, hash2, newblock);
1935                        dxtrace(dx_show_index("node", frames[1].entries));
1936                        dxtrace(dx_show_index("node",
1937                               ((struct dx_node *) bh2->b_data)->entries));
1938                        err = ext4_handle_dirty_dx_node(handle, dir, bh2);
1939                        if (err)
1940                                goto journal_error;
1941                        brelse (bh2);
1942                } else {
1943                        dxtrace(printk(KERN_DEBUG
1944                                       "Creating second level index...\n"));
1945                        memcpy((char *) entries2, (char *) entries,
1946                               icount * sizeof(struct dx_entry));
1947                        dx_set_limit(entries2, dx_node_limit(dir));
1948
1949                        /* Set up root */
1950                        dx_set_count(entries, 1);
1951                        dx_set_block(entries + 0, newblock);
1952                        ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1953
1954                        /* Add new access path frame */
1955                        frame = frames + 1;
1956                        frame->at = at = at - entries + entries2;
1957                        frame->entries = entries = entries2;
1958                        frame->bh = bh2;
1959                        err = ext4_journal_get_write_access(handle,
1960                                                             frame->bh);
1961                        if (err)
1962                                goto journal_error;
1963                }
1964                err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
1965                if (err) {
1966                        ext4_std_error(inode->i_sb, err);
1967                        goto cleanup;
1968                }
1969        }
1970        de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1971        if (!de)
1972                goto cleanup;
1973        err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1974        goto cleanup;
1975
1976journal_error:
1977        ext4_std_error(dir->i_sb, err);
1978cleanup:
1979        if (bh)
1980                brelse(bh);
1981        dx_release(frames);
1982        return err;
1983}
1984
1985/*
1986 * ext4_delete_entry deletes a directory entry by merging it with the
1987 * previous entry
1988 */
1989static int ext4_delete_entry(handle_t *handle,
1990                             struct inode *dir,
1991                             struct ext4_dir_entry_2 *de_del,
1992                             struct buffer_head *bh)
1993{
1994        struct ext4_dir_entry_2 *de, *pde;
1995        unsigned int blocksize = dir->i_sb->s_blocksize;
1996        int csum_size = 0;
1997        int i, err;
1998
1999        if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2000                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2001                csum_size = sizeof(struct ext4_dir_entry_tail);
2002
2003        i = 0;
2004        pde = NULL;
2005        de = (struct ext4_dir_entry_2 *) bh->b_data;
2006        while (i < bh->b_size - csum_size) {
2007                if (ext4_check_dir_entry(dir, NULL, de, bh, i))
2008                        return -EIO;
2009                if (de == de_del)  {
2010                        BUFFER_TRACE(bh, "get_write_access");
2011                        err = ext4_journal_get_write_access(handle, bh);
2012                        if (unlikely(err)) {
2013                                ext4_std_error(dir->i_sb, err);
2014                                return err;
2015                        }
2016                        if (pde)
2017                                pde->rec_len = ext4_rec_len_to_disk(
2018                                        ext4_rec_len_from_disk(pde->rec_len,
2019                                                               blocksize) +
2020                                        ext4_rec_len_from_disk(de->rec_len,
2021                                                               blocksize),
2022                                        blocksize);
2023                        else
2024                                de->inode = 0;
2025                        dir->i_version++;
2026                        BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
2027                        err = ext4_handle_dirty_dirent_node(handle, dir, bh);
2028                        if (unlikely(err)) {
2029                                ext4_std_error(dir->i_sb, err);
2030                                return err;
2031                        }
2032                        return 0;
2033                }
2034                i += ext4_rec_len_from_disk(de->rec_len, blocksize);
2035                pde = de;
2036                de = ext4_next_entry(de, blocksize);
2037        }
2038        return -ENOENT;
2039}
2040
2041/*
2042 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2043 * since this indicates that nlinks count was previously 1.
2044 */
2045static void ext4_inc_count(handle_t *handle, struct inode *inode)
2046{
2047        inc_nlink(inode);
2048        if (is_dx(inode) && inode->i_nlink > 1) {
2049                /* limit is 16-bit i_links_count */
2050                if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
2051                        set_nlink(inode, 1);
2052                        EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2053                                              EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2054                }
2055        }
2056}
2057
2058/*
2059 * If a directory had nlink == 1, then we should let it be 1. This indicates
2060 * directory has >EXT4_LINK_MAX subdirs.
2061 */
2062static void ext4_dec_count(handle_t *handle, struct inode *inode)
2063{
2064        if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2065                drop_nlink(inode);
2066}
2067
2068
2069static int ext4_add_nondir(handle_t *handle,
2070                struct dentry *dentry, struct inode *inode)
2071{
2072        int err = ext4_add_entry(handle, dentry, inode);
2073        if (!err) {
2074                ext4_mark_inode_dirty(handle, inode);
2075                unlock_new_inode(inode);
2076                d_instantiate(dentry, inode);
2077                return 0;
2078        }
2079        drop_nlink(inode);
2080        unlock_new_inode(inode);
2081        iput(inode);
2082        return err;
2083}
2084
2085/*
2086 * By the time this is called, we already have created
2087 * the directory cache entry for the new file, but it
2088 * is so far negative - it has no inode.
2089 *
2090 * If the create succeeds, we fill in the inode information
2091 * with d_instantiate().
2092 */
2093static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2094                       bool excl)
2095{
2096        handle_t *handle;
2097        struct inode *inode;
2098        int err, retries = 0;
2099
2100        dquot_initialize(dir);
2101
2102retry:
2103        handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2104                                        EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2105                                        EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2106        if (IS_ERR(handle))
2107                return PTR_ERR(handle);
2108
2109        if (IS_DIRSYNC(dir))
2110                ext4_handle_sync(handle);
2111
2112        inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2113        err = PTR_ERR(inode);
2114        if (!IS_ERR(inode)) {
2115                inode->i_op = &ext4_file_inode_operations;
2116                inode->i_fop = &ext4_file_operations;
2117                ext4_set_aops(inode);
2118                err = ext4_add_nondir(handle, dentry, inode);
2119        }
2120        ext4_journal_stop(handle);
2121        if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2122                goto retry;
2123        return err;
2124}
2125
2126static int ext4_mknod(struct inode *dir, struct dentry *dentry,
2127                      umode_t mode, dev_t rdev)
2128{
2129        handle_t *handle;
2130        struct inode *inode;
2131        int err, retries = 0;
2132
2133        if (!new_valid_dev(rdev))
2134                return -EINVAL;
2135
2136        dquot_initialize(dir);
2137
2138retry:
2139        handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2140                                        EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2141                                        EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2142        if (IS_ERR(handle))
2143                return PTR_ERR(handle);
2144
2145        if (IS_DIRSYNC(dir))
2146                ext4_handle_sync(handle);
2147
2148        inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
2149        err = PTR_ERR(inode);
2150        if (!IS_ERR(inode)) {
2151                init_special_inode(inode, inode->i_mode, rdev);
2152#ifdef CONFIG_EXT4_FS_XATTR
2153                inode->i_op = &ext4_special_inode_operations;
2154#endif
2155                err = ext4_add_nondir(handle, dentry, inode);
2156        }
2157        ext4_journal_stop(handle);
2158        if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2159                goto retry;
2160        return err;
2161}
2162
2163static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2164{
2165        handle_t *handle;
2166        struct inode *inode;
2167        struct buffer_head *dir_block = NULL;
2168        struct ext4_dir_entry_2 *de;
2169        struct ext4_dir_entry_tail *t;
2170        unsigned int blocksize = dir->i_sb->s_blocksize;
2171        int csum_size = 0;
2172        int err, retries = 0;
2173
2174        if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2175                                       EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2176                csum_size = sizeof(struct ext4_dir_entry_tail);
2177
2178        if (EXT4_DIR_LINK_MAX(dir))
2179                return -EMLINK;
2180
2181        dquot_initialize(dir);
2182
2183retry:
2184        handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2185                                        EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2186                                        EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
2187        if (IS_ERR(handle))
2188                return PTR_ERR(handle);
2189
2190        if (IS_DIRSYNC(dir))
2191                ext4_handle_sync(handle);
2192
2193        inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
2194                               &dentry->d_name, 0, NULL);
2195        err = PTR_ERR(inode);
2196        if (IS_ERR(inode))
2197                goto out_stop;
2198
2199        inode->i_op = &ext4_dir_inode_operations;
2200        inode->i_fop = &ext4_dir_operations;
2201        inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
2202        dir_block = ext4_bread(handle, inode, 0, 1, &err);
2203        if (!dir_block)
2204                goto out_clear_inode;
2205        BUFFER_TRACE(dir_block, "get_write_access");
2206        err = ext4_journal_get_write_access(handle, dir_block);
2207        if (err)
2208                goto out_clear_inode;
2209        de = (struct ext4_dir_entry_2 *) dir_block->b_data;
2210        de->inode = cpu_to_le32(inode->i_ino);
2211        de->name_len = 1;
2212        de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2213                                           blocksize);
2214        strcpy(de->name, ".");
2215        ext4_set_de_type(dir->i_sb, de, S_IFDIR);
2216        de = ext4_next_entry(de, blocksize);
2217        de->inode = cpu_to_le32(dir->i_ino);
2218        de->rec_len = ext4_rec_len_to_disk(blocksize -
2219                                           (csum_size + EXT4_DIR_REC_LEN(1)),
2220                                           blocksize);
2221        de->name_len = 2;
2222        strcpy(de->name, "..");
2223        ext4_set_de_type(dir->i_sb, de, S_IFDIR);
2224        set_nlink(inode, 2);
2225
2226        if (csum_size) {
2227                t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2228                initialize_dirent_tail(t, blocksize);
2229        }
2230
2231        BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
2232        err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
2233        if (err)
2234                goto out_clear_inode;
2235        set_buffer_verified(dir_block);
2236        err = ext4_mark_inode_dirty(handle, inode);
2237        if (!err)
2238                err = ext4_add_entry(handle, dentry, inode);
2239        if (err) {
2240out_clear_inode:
2241                clear_nlink(inode);
2242                unlock_new_inode(inode);
2243                ext4_mark_inode_dirty(handle, inode);
2244                iput(inode);
2245                goto out_stop;
2246        }
2247        ext4_inc_count(handle, dir);
2248        ext4_update_dx_flag(dir);
2249        err = ext4_mark_inode_dirty(handle, dir);
2250        if (err)
2251                goto out_clear_inode;
2252        unlock_new_inode(inode);
2253        d_instantiate(dentry, inode);
2254out_stop:
2255        brelse(dir_block);
2256        ext4_journal_stop(handle);
2257        if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2258                goto retry;
2259        return err;
2260}
2261
2262/*
2263 * routine to check that the specified directory is empty (for rmdir)
2264 */
2265static int empty_dir(struct inode *inode)
2266{
2267        unsigned int offset;
2268        struct buffer_head *bh;
2269        struct ext4_dir_entry_2 *de, *de1;
2270        struct super_block *sb;
2271        int err = 0;
2272
2273        sb = inode->i_sb;
2274        if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
2275            !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
2276                if (err)
2277                        EXT4_ERROR_INODE(inode,
2278                                "error %d reading directory lblock 0", err);
2279                else
2280                        ext4_warning(inode->i_sb,
2281                                     "bad directory (dir #%lu) - no data block",
2282                                     inode->i_ino);
2283                return 1;
2284        }
2285        if (!buffer_verified(bh) &&
2286            !ext4_dirent_csum_verify(inode,
2287                        (struct ext4_dir_entry *)bh->b_data)) {
2288                EXT4_ERROR_INODE(inode, "checksum error reading directory "
2289                                 "lblock 0");
2290                return -EIO;
2291        }
2292        set_buffer_verified(bh);
2293        de = (struct ext4_dir_entry_2 *) bh->b_data;
2294        de1 = ext4_next_entry(de, sb->s_blocksize);
2295        if (le32_to_cpu(de->inode) != inode->i_ino ||
2296                        !le32_to_cpu(de1->inode) ||
2297                        strcmp(".", de->name) ||
2298                        strcmp("..", de1->name)) {
2299                ext4_warning(inode->i_sb,
2300                             "bad directory (dir #%lu) - no `.' or `..'",
2301                             inode->i_ino);
2302                brelse(bh);
2303                return 1;
2304        }
2305        offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2306                 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2307        de = ext4_next_entry(de1, sb->s_blocksize);
2308        while (offset < inode->i_size) {
2309                if (!bh ||
2310                    (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2311                        unsigned int lblock;
2312                        err = 0;
2313                        brelse(bh);
2314                        lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2315                        bh = ext4_bread(NULL, inode, lblock, 0, &err);
2316                        if (!bh) {
2317                                if (err)
2318                                        EXT4_ERROR_INODE(inode,
2319                                                "error %d reading directory "
2320                                                "lblock %u", err, lblock);
2321                                offset += sb->s_blocksize;
2322                                continue;
2323                        }
2324                        if (!buffer_verified(bh) &&
2325                            !ext4_dirent_csum_verify(inode,
2326                                        (struct ext4_dir_entry *)bh->b_data)) {
2327                                EXT4_ERROR_INODE(inode, "checksum error "
2328                                                 "reading directory lblock 0");
2329                                return -EIO;
2330                        }
2331                        set_buffer_verified(bh);
2332                        de = (struct ext4_dir_entry_2 *) bh->b_data;
2333                }
2334                if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
2335                        de = (struct ext4_dir_entry_2 *)(bh->b_data +
2336                                                         sb->s_blocksize);
2337                        offset = (offset | (sb->s_blocksize - 1)) + 1;
2338                        continue;
2339                }
2340                if (le32_to_cpu(de->inode)) {
2341                        brelse(bh);
2342                        return 0;
2343                }
2344                offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2345                de = ext4_next_entry(de, sb->s_blocksize);
2346        }
2347        brelse(bh);
2348        return 1;
2349}
2350
2351/* ext4_orphan_add() links an unlinked or truncated inode into a list of
2352 * such inodes, starting at the superblock, in case we crash before the
2353 * file is closed/deleted, or in case the inode truncate spans multiple
2354 * transactions and the last transaction is not recovered after a crash.
2355 *
2356 * At filesystem recovery time, we walk this list deleting unlinked
2357 * inodes and truncating linked inodes in ext4_orphan_cleanup().
2358 */
2359int ext4_orphan_add(handle_t *handle, struct inode *inode)
2360{
2361        struct super_block *sb = inode->i_sb;
2362        struct ext4_iloc iloc;
2363        int err = 0, rc;
2364
2365        if (!ext4_handle_valid(handle))
2366                return 0;
2367
2368        mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
2369        if (!list_empty(&EXT4_I(inode)->i_orphan))
2370                goto out_unlock;
2371
2372        /*
2373         * Orphan handling is only valid for files with data blocks
2374         * being truncated, or files being unlinked. Note that we either
2375         * hold i_mutex, or the inode can not be referenced from outside,
2376         * so i_nlink should not be bumped due to race
2377         */
2378        J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2379                  S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
2380
2381        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2382        err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
2383        if (err)
2384                goto out_unlock;
2385
2386        err = ext4_reserve_inode_write(handle, inode, &iloc);
2387        if (err)
2388                goto out_unlock;
2389        /*
2390         * Due to previous errors inode may be already a part of on-disk
2391         * orphan list. If so skip on-disk list modification.
2392         */
2393        if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2394                (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2395                        goto mem_insert;
2396
2397        /* Insert this inode at the head of the on-disk orphan list... */
2398        NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2399        EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
2400        err = ext4_handle_dirty_super(handle, sb);
2401        rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
2402        if (!err)
2403                err = rc;
2404
2405        /* Only add to the head of the in-memory list if all the
2406         * previous operations succeeded.  If the orphan_add is going to
2407         * fail (possibly taking the journal offline), we can't risk
2408         * leaving the inode on the orphan list: stray orphan-list
2409         * entries can cause panics at unmount time.
2410         *
2411         * This is safe: on error we're going to ignore the orphan list
2412         * anyway on the next recovery. */
2413mem_insert:
2414        if (!err)
2415                list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2416
2417        jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2418        jbd_debug(4, "orphan inode %lu will point to %d\n",
2419                        inode->i_ino, NEXT_ORPHAN(inode));
2420out_unlock:
2421        mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
2422        ext4_std_error(inode->i_sb, err);
2423        return err;
2424}
2425
2426/*
2427 * ext4_orphan_del() removes an unlinked or truncated inode from the list
2428 * of such inodes stored on disk, because it is finally being cleaned up.
2429 */
2430int ext4_orphan_del(handle_t *handle, struct inode *inode)
2431{
2432        struct list_head *prev;
2433        struct ext4_inode_info *ei = EXT4_I(inode);
2434        struct ext4_sb_info *sbi;
2435        __u32 ino_next;
2436        struct ext4_iloc iloc;
2437        int err = 0;
2438
2439        /* ext4_handle_valid() assumes a valid handle_t pointer */
2440        if (handle && !ext4_handle_valid(handle))
2441                return 0;
2442
2443        mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2444        if (list_empty(&ei->i_orphan))
2445                goto out;
2446
2447        ino_next = NEXT_ORPHAN(inode);
2448        prev = ei->i_orphan.prev;
2449        sbi = EXT4_SB(inode->i_sb);
2450
2451        jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2452
2453        list_del_init(&ei->i_orphan);
2454
2455        /* If we're on an error path, we may not have a valid
2456         * transaction handle with which to update the orphan list on
2457         * disk, but we still need to remove the inode from the linked
2458         * list in memory. */
2459        if (sbi->s_journal && !handle)
2460                goto out;
2461
2462        err = ext4_reserve_inode_write(handle, inode, &iloc);
2463        if (err)
2464                goto out_err;
2465
2466        if (prev == &sbi->s_orphan) {
2467                jbd_debug(4, "superblock will point to %u\n", ino_next);
2468                BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2469                err = ext4_journal_get_write_access(handle, sbi->s_sbh);
2470                if (err)
2471                        goto out_brelse;
2472                sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2473                err = ext4_handle_dirty_super(handle, inode->i_sb);
2474        } else {
2475                struct ext4_iloc iloc2;
2476                struct inode *i_prev =
2477                        &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
2478
2479                jbd_debug(4, "orphan inode %lu will point to %u\n",
2480                          i_prev->i_ino, ino_next);
2481                err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
2482                if (err)
2483                        goto out_brelse;
2484                NEXT_ORPHAN(i_prev) = ino_next;
2485                err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
2486        }
2487        if (err)
2488                goto out_brelse;
2489        NEXT_ORPHAN(inode) = 0;
2490        err = ext4_mark_iloc_dirty(handle, inode, &iloc);
2491
2492out_err:
2493        ext4_std_error(inode->i_sb, err);
2494out:
2495        mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2496        return err;
2497
2498out_brelse:
2499        brelse(iloc.bh);
2500        goto out_err;
2501}
2502
2503static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
2504{
2505        int retval;
2506        struct inode *inode;
2507        struct buffer_head *bh;
2508        struct ext4_dir_entry_2 *de;
2509        handle_t *handle;
2510
2511        /* Initialize quotas before so that eventual writes go in
2512         * separate transaction */
2513        dquot_initialize(dir);
2514        dquot_initialize(dentry->d_inode);
2515
2516        handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2517        if (IS_ERR(handle))
2518                return PTR_ERR(handle);
2519
2520        retval = -ENOENT;
2521        bh = ext4_find_entry(dir, &dentry->d_name, &de);
2522        if (!bh)
2523                goto end_rmdir;
2524
2525        if (IS_DIRSYNC(dir))
2526                ext4_handle_sync(handle);
2527
2528        inode = dentry->d_inode;
2529
2530        retval = -EIO;
2531        if (le32_to_cpu(de->inode) != inode->i_ino)
2532                goto end_rmdir;
2533
2534        retval = -ENOTEMPTY;
2535        if (!empty_dir(inode))
2536                goto end_rmdir;
2537
2538        retval = ext4_delete_entry(handle, dir, de, bh);
2539        if (retval)
2540                goto end_rmdir;
2541        if (!EXT4_DIR_LINK_EMPTY(inode))
2542                ext4_warning(inode->i_sb,
2543                             "empty directory has too many links (%d)",
2544                             inode->i_nlink);
2545        inode->i_version++;
2546        clear_nlink(inode);
2547        /* There's no need to set i_disksize: the fact that i_nlink is
2548         * zero will ensure that the right thing happens during any
2549         * recovery. */
2550        inode->i_size = 0;
2551        ext4_orphan_add(handle, inode);
2552        inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
2553        ext4_mark_inode_dirty(handle, inode);
2554        ext4_dec_count(handle, dir);
2555        ext4_update_dx_flag(dir);
2556        ext4_mark_inode_dirty(handle, dir);
2557
2558end_rmdir:
2559        ext4_journal_stop(handle);
2560        brelse(bh);
2561        return retval;
2562}
2563
2564static int ext4_unlink(struct inode *dir, struct dentry *dentry)
2565{
2566        int retval;
2567        struct inode *inode;
2568        struct buffer_head *bh;
2569        struct ext4_dir_entry_2 *de;
2570        handle_t *handle;
2571
2572        trace_ext4_unlink_enter(dir, dentry);
2573        /* Initialize quotas before so that eventual writes go
2574         * in separate transaction */
2575        dquot_initialize(dir);
2576        dquot_initialize(dentry->d_inode);
2577
2578        handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
2579        if (IS_ERR(handle))
2580                return PTR_ERR(handle);
2581
2582        if (IS_DIRSYNC(dir))
2583                ext4_handle_sync(handle);
2584
2585        retval = -ENOENT;
2586        bh = ext4_find_entry(dir, &dentry->d_name, &de);
2587        if (!bh)
2588                goto end_unlink;
2589
2590        inode = dentry->d_inode;
2591
2592        retval = -EIO;
2593        if (le32_to_cpu(de->inode) != inode->i_ino)
2594                goto end_unlink;
2595
2596        if (!inode->i_nlink) {
2597                ext4_warning(inode->i_sb,
2598                             "Deleting nonexistent file (%lu), %d",
2599                             inode->i_ino, inode->i_nlink);
2600                set_nlink(inode, 1);
2601        }
2602        retval = ext4_delete_entry(handle, dir, de, bh);
2603        if (retval)
2604                goto end_unlink;
2605        dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
2606        ext4_update_dx_flag(dir);
2607        ext4_mark_inode_dirty(handle, dir);
2608        drop_nlink(inode);
2609        if (!inode->i_nlink)
2610                ext4_orphan_add(handle, inode);
2611        inode->i_ctime = ext4_current_time(inode);
2612        ext4_mark_inode_dirty(handle, inode);
2613        retval = 0;
2614
2615end_unlink:
2616        ext4_journal_stop(handle);
2617        brelse(bh);
2618        trace_ext4_unlink_exit(dentry, retval);
2619        return retval;
2620}
2621
2622static int ext4_symlink(struct inode *dir,
2623                        struct dentry *dentry, const char *symname)
2624{
2625        handle_t *handle;
2626        struct inode *inode;
2627        int l, err, retries = 0;
2628        int credits;
2629
2630        l = strlen(symname)+1;
2631        if (l > dir->i_sb->s_blocksize)
2632                return -ENAMETOOLONG;
2633
2634        dquot_initialize(dir);
2635
2636        if (l > EXT4_N_BLOCKS * 4) {
2637                /*
2638                 * For non-fast symlinks, we just allocate inode and put it on
2639                 * orphan list in the first transaction => we need bitmap,
2640                 * group descriptor, sb, inode block, quota blocks, and
2641                 * possibly selinux xattr blocks.
2642                 */
2643                credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2644                          EXT4_XATTR_TRANS_BLOCKS;
2645        } else {
2646                /*
2647                 * Fast symlink. We have to add entry to directory
2648                 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2649                 * allocate new inode (bitmap, group descriptor, inode block,
2650                 * quota blocks, sb is already counted in previous macros).
2651                 */
2652                credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2653                          EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2654                          EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2655        }
2656retry:
2657        handle = ext4_journal_start(dir, credits);
2658        if (IS_ERR(handle))
2659                return PTR_ERR(handle);
2660
2661        if (IS_DIRSYNC(dir))
2662                ext4_handle_sync(handle);
2663
2664        inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2665                               &dentry->d_name, 0, NULL);
2666        err = PTR_ERR(inode);
2667        if (IS_ERR(inode))
2668                goto out_stop;
2669
2670        if (l > EXT4_N_BLOCKS * 4) {
2671                inode->i_op = &ext4_symlink_inode_operations;
2672                ext4_set_aops(inode);
2673                /*
2674                 * We cannot call page_symlink() with transaction started
2675                 * because it calls into ext4_write_begin() which can wait
2676                 * for transaction commit if we are running out of space
2677                 * and thus we deadlock. So we have to stop transaction now
2678                 * and restart it when symlink contents is written.
2679                 * 
2680                 * To keep fs consistent in case of crash, we have to put inode
2681                 * to orphan list in the mean time.
2682                 */
2683                drop_nlink(inode);
2684                err = ext4_orphan_add(handle, inode);
2685                ext4_journal_stop(handle);
2686                if (err)
2687                        goto err_drop_inode;
2688                err = __page_symlink(inode, symname, l, 1);
2689                if (err)
2690                        goto err_drop_inode;
2691                /*
2692                 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2693                 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2694                 */
2695                handle = ext4_journal_start(dir,
2696                                EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2697                                EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2698                if (IS_ERR(handle)) {
2699                        err = PTR_ERR(handle);
2700                        goto err_drop_inode;
2701                }
2702                set_nlink(inode, 1);
2703                err = ext4_orphan_del(handle, inode);
2704                if (err) {
2705                        ext4_journal_stop(handle);
2706                        clear_nlink(inode);
2707                        goto err_drop_inode;
2708                }
2709        } else {
2710                /* clear the extent format for fast symlink */
2711                ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
2712                inode->i_op = &ext4_fast_symlink_inode_operations;
2713                memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
2714                inode->i_size = l-1;
2715        }
2716        EXT4_I(inode)->i_disksize = inode->i_size;
2717        err = ext4_add_nondir(handle, dentry, inode);
2718out_stop:
2719        ext4_journal_stop(handle);
2720        if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2721                goto retry;
2722        return err;
2723err_drop_inode:
2724        unlock_new_inode(inode);
2725        iput(inode);
2726        return err;
2727}
2728
2729static int ext4_link(struct dentry *old_dentry,
2730                     struct inode *dir, struct dentry *dentry)
2731{
2732        handle_t *handle;
2733        struct inode *inode = old_dentry->d_inode;
2734        int err, retries = 0;
2735
2736        if (inode->i_nlink >= EXT4_LINK_MAX)
2737                return -EMLINK;
2738
2739        dquot_initialize(dir);
2740
2741retry:
2742        handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2743                                        EXT4_INDEX_EXTRA_TRANS_BLOCKS);
2744        if (IS_ERR(handle))
2745                return PTR_ERR(handle);
2746
2747        if (IS_DIRSYNC(dir))
2748                ext4_handle_sync(handle);
2749
2750        inode->i_ctime = ext4_current_time(inode);
2751        ext4_inc_count(handle, inode);
2752        ihold(inode);
2753
2754        err = ext4_add_entry(handle, dentry, inode);
2755        if (!err) {
2756                ext4_mark_inode_dirty(handle, inode);
2757                d_instantiate(dentry, inode);
2758        } else {
2759                drop_nlink(inode);
2760                iput(inode);
2761        }
2762        ext4_journal_stop(handle);
2763        if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
2764                goto retry;
2765        return err;
2766}
2767
2768#define PARENT_INO(buffer, size) \
2769        (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
2770
2771/*
2772 * Anybody can rename anything with this: the permission checks are left to the
2773 * higher-level routines.
2774 */
2775static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2776                       struct inode *new_dir, struct dentry *new_dentry)
2777{
2778        handle_t *handle;
2779        struct inode *old_inode, *new_inode;
2780        struct buffer_head *old_bh, *new_bh, *dir_bh;
2781        struct ext4_dir_entry_2 *old_de, *new_de;
2782        int retval, force_da_alloc = 0;
2783
2784        dquot_initialize(old_dir);
2785        dquot_initialize(new_dir);
2786
2787        old_bh = new_bh = dir_bh = NULL;
2788
2789        /* Initialize quotas before so that eventual writes go
2790         * in separate transaction */
2791        if (new_dentry->d_inode)
2792                dquot_initialize(new_dentry->d_inode);
2793        handle = ext4_journal_start(old_dir, 2 *
2794                                        EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2795                                        EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
2796        if (IS_ERR(handle))
2797                return PTR_ERR(handle);
2798
2799        if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2800                ext4_handle_sync(handle);
2801
2802        old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
2803        /*
2804         *  Check for inode number is _not_ due to possible IO errors.
2805         *  We might rmdir the source, keep it as pwd of some process
2806         *  and merrily kill the link to whatever was created under the
2807         *  same name. Goodbye sticky bit ;-<
2808         */
2809        old_inode = old_dentry->d_inode;
2810        retval = -ENOENT;
2811        if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2812                goto end_rename;
2813
2814        new_inode = new_dentry->d_inode;
2815        new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
2816        if (new_bh) {
2817                if (!new_inode) {
2818                        brelse(new_bh);
2819                        new_bh = NULL;
2820                }
2821        }
2822        if (S_ISDIR(old_inode->i_mode)) {
2823                if (new_inode) {
2824                        retval = -ENOTEMPTY;
2825                        if (!empty_dir(new_inode))
2826                                goto end_rename;
2827                }
2828                retval = -EIO;
2829                dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
2830                if (!dir_bh)
2831                        goto end_rename;
2832                if (!buffer_verified(dir_bh) &&
2833                    !ext4_dirent_csum_verify(old_inode,
2834                                (struct ext4_dir_entry *)dir_bh->b_data))
2835                        goto end_rename;
2836                set_buffer_verified(dir_bh);
2837                if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2838                                old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
2839                        goto end_rename;
2840                retval = -EMLINK;
2841                if (!new_inode && new_dir != old_dir &&
2842                    EXT4_DIR_LINK_MAX(new_dir))
2843                        goto end_rename;
2844                BUFFER_TRACE(dir_bh, "get_write_access");
2845                retval = ext4_journal_get_write_access(handle, dir_bh);
2846                if (retval)
2847                        goto end_rename;
2848        }
2849        if (!new_bh) {
2850                retval = ext4_add_entry(handle, new_dentry, old_inode);
2851                if (retval)
2852                        goto end_rename;
2853        } else {
2854                BUFFER_TRACE(new_bh, "get write access");
2855                retval = ext4_journal_get_write_access(handle, new_bh);
2856                if (retval)
2857                        goto end_rename;
2858                new_de->inode = cpu_to_le32(old_inode->i_ino);
2859                if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2860                                              EXT4_FEATURE_INCOMPAT_FILETYPE))
2861                        new_de->file_type = old_de->file_type;
2862                new_dir->i_version++;
2863                new_dir->i_ctime = new_dir->i_mtime =
2864                                        ext4_current_time(new_dir);
2865                ext4_mark_inode_dirty(handle, new_dir);
2866                BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2867                retval = ext4_handle_dirty_dirent_node(handle, new_dir, new_bh);
2868                if (unlikely(retval)) {
2869                        ext4_std_error(new_dir->i_sb, retval);
2870                        goto end_rename;
2871                }
2872                brelse(new_bh);
2873                new_bh = NULL;
2874        }
2875
2876        /*
2877         * Like most other Unix systems, set the ctime for inodes on a
2878         * rename.
2879         */
2880        old_inode->i_ctime = ext4_current_time(old_inode);
2881        ext4_mark_inode_dirty(handle, old_inode);
2882
2883        /*
2884         * ok, that's it
2885         */
2886        if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2887            old_de->name_len != old_dentry->d_name.len ||
2888            strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2889            (retval = ext4_delete_entry(handle, old_dir,
2890                                        old_de, old_bh)) == -ENOENT) {
2891                /* old_de could have moved from under us during htree split, so
2892                 * make sure that we are deleting the right entry.  We might
2893                 * also be pointing to a stale entry in the unused part of
2894                 * old_bh so just checking inum and the name isn't enough. */
2895                struct buffer_head *old_bh2;
2896                struct ext4_dir_entry_2 *old_de2;
2897
2898                old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
2899                if (old_bh2) {
2900                        retval = ext4_delete_entry(handle, old_dir,
2901                                                   old_de2, old_bh2);
2902                        brelse(old_bh2);
2903                }
2904        }
2905        if (retval) {
2906                ext4_warning(old_dir->i_sb,
2907                                "Deleting old file (%lu), %d, error=%d",
2908                                old_dir->i_ino, old_dir->i_nlink, retval);
2909        }
2910
2911        if (new_inode) {
2912                ext4_dec_count(handle, new_inode);
2913                new_inode->i_ctime = ext4_current_time(new_inode);
2914        }
2915        old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
2916        ext4_update_dx_flag(old_dir);
2917        if (dir_bh) {
2918                PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2919                                                cpu_to_le32(new_dir->i_ino);
2920                BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2921                if (is_dx(old_inode)) {
2922                        retval = ext4_handle_dirty_dx_node(handle,
2923                                                           old_inode,
2924                                                           dir_bh);
2925                } else {
2926                        retval = ext4_handle_dirty_dirent_node(handle,
2927                                                               old_inode,
2928                                                               dir_bh);
2929                }
2930                if (retval) {
2931                        ext4_std_error(old_dir->i_sb, retval);
2932                        goto end_rename;
2933                }
2934                ext4_dec_count(handle, old_dir);
2935                if (new_inode) {
2936                        /* checked empty_dir above, can't have another parent,
2937                         * ext4_dec_count() won't work for many-linked dirs */
2938                        clear_nlink(new_inode);
2939                } else {
2940                        ext4_inc_count(handle, new_dir);
2941                        ext4_update_dx_flag(new_dir);
2942                        ext4_mark_inode_dirty(handle, new_dir);
2943                }
2944        }
2945        ext4_mark_inode_dirty(handle, old_dir);
2946        if (new_inode) {
2947                ext4_mark_inode_dirty(handle, new_inode);
2948                if (!new_inode->i_nlink)
2949                        ext4_orphan_add(handle, new_inode);
2950                if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2951                        force_da_alloc = 1;
2952        }
2953        retval = 0;
2954
2955end_rename:
2956        brelse(dir_bh);
2957        brelse(old_bh);
2958        brelse(new_bh);
2959        ext4_journal_stop(handle);
2960        if (retval == 0 && force_da_alloc)
2961                ext4_alloc_da_blocks(old_inode);
2962        return retval;
2963}
2964
2965/*
2966 * directories can handle most operations...
2967 */
2968const struct inode_operations ext4_dir_inode_operations = {
2969        .create         = ext4_create,
2970        .lookup         = ext4_lookup,
2971        .link           = ext4_link,
2972        .unlink         = ext4_unlink,
2973        .symlink        = ext4_symlink,
2974        .mkdir          = ext4_mkdir,
2975        .rmdir          = ext4_rmdir,
2976        .mknod          = ext4_mknod,
2977        .rename         = ext4_rename,
2978        .setattr        = ext4_setattr,
2979#ifdef CONFIG_EXT4_FS_XATTR
2980        .setxattr       = generic_setxattr,
2981        .getxattr       = generic_getxattr,
2982        .listxattr      = ext4_listxattr,
2983        .removexattr    = generic_removexattr,
2984#endif
2985        .get_acl        = ext4_get_acl,
2986        .fiemap         = ext4_fiemap,
2987};
2988
2989const struct inode_operations ext4_special_inode_operations = {
2990        .setattr        = ext4_setattr,
2991#ifdef CONFIG_EXT4_FS_XATTR
2992        .setxattr       = generic_setxattr,
2993        .getxattr       = generic_getxattr,
2994        .listxattr      = ext4_listxattr,
2995        .removexattr    = generic_removexattr,
2996#endif
2997        .get_acl        = ext4_get_acl,
2998};
2999
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.