linux/fs/ext4/xattr.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/fs/ext4/xattr.c
   4 *
   5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   6 *
   7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   9 * Extended attributes for symlinks and special files added per
  10 *  suggestion of Luka Renko <luka.renko@hermes.si>.
  11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  12 *  Red Hat Inc.
  13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  14 *  and Andreas Gruenbacher <agruen@suse.de>.
  15 */
  16
  17/*
  18 * Extended attributes are stored directly in inodes (on file systems with
  19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  20 * field contains the block number if an inode uses an additional block. All
  21 * attributes must fit in the inode and one additional block. Blocks that
  22 * contain the identical set of attributes may be shared among several inodes.
  23 * Identical blocks are detected by keeping a cache of blocks that have
  24 * recently been accessed.
  25 *
  26 * The attributes in inodes and on blocks have a different header; the entries
  27 * are stored in the same format:
  28 *
  29 *   +------------------+
  30 *   | header           |
  31 *   | entry 1          | |
  32 *   | entry 2          | | growing downwards
  33 *   | entry 3          | v
  34 *   | four null bytes  |
  35 *   | . . .            |
  36 *   | value 1          | ^
  37 *   | value 3          | | growing upwards
  38 *   | value 2          | |
  39 *   +------------------+
  40 *
  41 * The header is followed by multiple entry descriptors. In disk blocks, the
  42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
  43 * attribute values are aligned to the end of the block in no specific order.
  44 *
  45 * Locking strategy
  46 * ----------------
  47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  48 * EA blocks are only changed if they are exclusive to an inode, so
  49 * holding xattr_sem also means that nothing but the EA block's reference
  50 * count can change. Multiple writers to the same block are synchronized
  51 * by the buffer lock.
  52 */
  53
  54#include <linux/init.h>
  55#include <linux/fs.h>
  56#include <linux/slab.h>
  57#include <linux/mbcache.h>
  58#include <linux/quotaops.h>
  59#include <linux/iversion.h>
  60#include "ext4_jbd2.h"
  61#include "ext4.h"
  62#include "xattr.h"
  63#include "acl.h"
  64
  65#ifdef EXT4_XATTR_DEBUG
  66# define ea_idebug(inode, fmt, ...)                                     \
  67        printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
  68               inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  69# define ea_bdebug(bh, fmt, ...)                                        \
  70        printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
  71               bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  72#else
  73# define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  74# define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  75#endif
  76
  77static void ext4_xattr_block_cache_insert(struct mb_cache *,
  78                                          struct buffer_head *);
  79static struct buffer_head *
  80ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  81                            struct mb_cache_entry **);
  82static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  83                                    size_t value_count);
  84static void ext4_xattr_rehash(struct ext4_xattr_header *);
  85
  86static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  87        [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  88#ifdef CONFIG_EXT4_FS_POSIX_ACL
  89        [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  90        [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  91#endif
  92        [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  93#ifdef CONFIG_EXT4_FS_SECURITY
  94        [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  95#endif
  96};
  97
  98const struct xattr_handler *ext4_xattr_handlers[] = {
  99        &ext4_xattr_user_handler,
 100        &ext4_xattr_trusted_handler,
 101#ifdef CONFIG_EXT4_FS_POSIX_ACL
 102        &posix_acl_access_xattr_handler,
 103        &posix_acl_default_xattr_handler,
 104#endif
 105#ifdef CONFIG_EXT4_FS_SECURITY
 106        &ext4_xattr_security_handler,
 107#endif
 108        NULL
 109};
 110
 111#define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 112                                inode->i_sb->s_fs_info)->s_ea_block_cache)
 113
 114#define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 115                                inode->i_sb->s_fs_info)->s_ea_inode_cache)
 116
 117static int
 118ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 119                        struct inode *inode);
 120
 121#ifdef CONFIG_LOCKDEP
 122void ext4_xattr_inode_set_class(struct inode *ea_inode)
 123{
 124        lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 125}
 126#endif
 127
 128static __le32 ext4_xattr_block_csum(struct inode *inode,
 129                                    sector_t block_nr,
 130                                    struct ext4_xattr_header *hdr)
 131{
 132        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 133        __u32 csum;
 134        __le64 dsk_block_nr = cpu_to_le64(block_nr);
 135        __u32 dummy_csum = 0;
 136        int offset = offsetof(struct ext4_xattr_header, h_checksum);
 137
 138        csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 139                           sizeof(dsk_block_nr));
 140        csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 141        csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 142        offset += sizeof(dummy_csum);
 143        csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 144                           EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 145
 146        return cpu_to_le32(csum);
 147}
 148
 149static int ext4_xattr_block_csum_verify(struct inode *inode,
 150                                        struct buffer_head *bh)
 151{
 152        struct ext4_xattr_header *hdr = BHDR(bh);
 153        int ret = 1;
 154
 155        if (ext4_has_metadata_csum(inode->i_sb)) {
 156                lock_buffer(bh);
 157                ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 158                                                        bh->b_blocknr, hdr));
 159                unlock_buffer(bh);
 160        }
 161        return ret;
 162}
 163
 164static void ext4_xattr_block_csum_set(struct inode *inode,
 165                                      struct buffer_head *bh)
 166{
 167        if (ext4_has_metadata_csum(inode->i_sb))
 168                BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 169                                                bh->b_blocknr, BHDR(bh));
 170}
 171
 172static inline const struct xattr_handler *
 173ext4_xattr_handler(int name_index)
 174{
 175        const struct xattr_handler *handler = NULL;
 176
 177        if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 178                handler = ext4_xattr_handler_map[name_index];
 179        return handler;
 180}
 181
 182static int
 183ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 184                         void *value_start)
 185{
 186        struct ext4_xattr_entry *e = entry;
 187
 188        /* Find the end of the names list */
 189        while (!IS_LAST_ENTRY(e)) {
 190                struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 191                if ((void *)next >= end)
 192                        return -EFSCORRUPTED;
 193                e = next;
 194        }
 195
 196        /* Check the values */
 197        while (!IS_LAST_ENTRY(entry)) {
 198                if (entry->e_value_size != 0 &&
 199                    entry->e_value_inum == 0) {
 200                        u16 offs = le16_to_cpu(entry->e_value_offs);
 201                        u32 size = le32_to_cpu(entry->e_value_size);
 202                        void *value;
 203
 204                        /*
 205                         * The value cannot overlap the names, and the value
 206                         * with padding cannot extend beyond 'end'.  Check both
 207                         * the padded and unpadded sizes, since the size may
 208                         * overflow to 0 when adding padding.
 209                         */
 210                        if (offs > end - value_start)
 211                                return -EFSCORRUPTED;
 212                        value = value_start + offs;
 213                        if (value < (void *)e + sizeof(u32) ||
 214                            size > end - value ||
 215                            EXT4_XATTR_SIZE(size) > end - value)
 216                                return -EFSCORRUPTED;
 217                }
 218                entry = EXT4_XATTR_NEXT(entry);
 219        }
 220
 221        return 0;
 222}
 223
 224static inline int
 225ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh)
 226{
 227        int error;
 228
 229        if (buffer_verified(bh))
 230                return 0;
 231
 232        if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 233            BHDR(bh)->h_blocks != cpu_to_le32(1))
 234                return -EFSCORRUPTED;
 235        if (!ext4_xattr_block_csum_verify(inode, bh))
 236                return -EFSBADCRC;
 237        error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 238                                         bh->b_data);
 239        if (!error)
 240                set_buffer_verified(bh);
 241        return error;
 242}
 243
 244static int
 245__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 246                         void *end, const char *function, unsigned int line)
 247{
 248        int error = -EFSCORRUPTED;
 249
 250        if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 251            (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 252                goto errout;
 253        error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 254errout:
 255        if (error)
 256                __ext4_error_inode(inode, function, line, 0,
 257                                   "corrupted in-inode xattr");
 258        return error;
 259}
 260
 261#define xattr_check_inode(inode, header, end) \
 262        __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 263
 264static int
 265ext4_xattr_find_entry(struct ext4_xattr_entry **pentry, int name_index,
 266                      const char *name, int sorted)
 267{
 268        struct ext4_xattr_entry *entry;
 269        size_t name_len;
 270        int cmp = 1;
 271
 272        if (name == NULL)
 273                return -EINVAL;
 274        name_len = strlen(name);
 275        entry = *pentry;
 276        for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 277                cmp = name_index - entry->e_name_index;
 278                if (!cmp)
 279                        cmp = name_len - entry->e_name_len;
 280                if (!cmp)
 281                        cmp = memcmp(name, entry->e_name, name_len);
 282                if (cmp <= 0 && (sorted || cmp == 0))
 283                        break;
 284        }
 285        *pentry = entry;
 286        return cmp ? -ENODATA : 0;
 287}
 288
 289static u32
 290ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 291{
 292        return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 293}
 294
 295static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 296{
 297        return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 298                (u32) inode_peek_iversion_raw(ea_inode);
 299}
 300
 301static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 302{
 303        ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 304        inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 305}
 306
 307static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 308{
 309        return (u32)ea_inode->i_atime.tv_sec;
 310}
 311
 312static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 313{
 314        ea_inode->i_atime.tv_sec = hash;
 315}
 316
 317/*
 318 * Read the EA value from an inode.
 319 */
 320static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 321{
 322        int blocksize = 1 << ea_inode->i_blkbits;
 323        int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 324        int tail_size = (size % blocksize) ?: blocksize;
 325        struct buffer_head *bhs_inline[8];
 326        struct buffer_head **bhs = bhs_inline;
 327        int i, ret;
 328
 329        if (bh_count > ARRAY_SIZE(bhs_inline)) {
 330                bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 331                if (!bhs)
 332                        return -ENOMEM;
 333        }
 334
 335        ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 336                               true /* wait */, bhs);
 337        if (ret)
 338                goto free_bhs;
 339
 340        for (i = 0; i < bh_count; i++) {
 341                /* There shouldn't be any holes in ea_inode. */
 342                if (!bhs[i]) {
 343                        ret = -EFSCORRUPTED;
 344                        goto put_bhs;
 345                }
 346                memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 347                       i < bh_count - 1 ? blocksize : tail_size);
 348        }
 349        ret = 0;
 350put_bhs:
 351        for (i = 0; i < bh_count; i++)
 352                brelse(bhs[i]);
 353free_bhs:
 354        if (bhs != bhs_inline)
 355                kfree(bhs);
 356        return ret;
 357}
 358
 359#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 360
 361static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 362                                 u32 ea_inode_hash, struct inode **ea_inode)
 363{
 364        struct inode *inode;
 365        int err;
 366
 367        inode = ext4_iget(parent->i_sb, ea_ino);
 368        if (IS_ERR(inode)) {
 369                err = PTR_ERR(inode);
 370                ext4_error(parent->i_sb,
 371                           "error while reading EA inode %lu err=%d", ea_ino,
 372                           err);
 373                return err;
 374        }
 375
 376        if (is_bad_inode(inode)) {
 377                ext4_error(parent->i_sb,
 378                           "error while reading EA inode %lu is_bad_inode",
 379                           ea_ino);
 380                err = -EIO;
 381                goto error;
 382        }
 383
 384        if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 385                ext4_error(parent->i_sb,
 386                           "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 387                            ea_ino);
 388                err = -EINVAL;
 389                goto error;
 390        }
 391
 392        ext4_xattr_inode_set_class(inode);
 393
 394        /*
 395         * Check whether this is an old Lustre-style xattr inode. Lustre
 396         * implementation does not have hash validation, rather it has a
 397         * backpointer from ea_inode to the parent inode.
 398         */
 399        if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 400            EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 401            inode->i_generation == parent->i_generation) {
 402                ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 403                ext4_xattr_inode_set_ref(inode, 1);
 404        } else {
 405                inode_lock(inode);
 406                inode->i_flags |= S_NOQUOTA;
 407                inode_unlock(inode);
 408        }
 409
 410        *ea_inode = inode;
 411        return 0;
 412error:
 413        iput(inode);
 414        return err;
 415}
 416
 417static int
 418ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 419                               struct ext4_xattr_entry *entry, void *buffer,
 420                               size_t size)
 421{
 422        u32 hash;
 423
 424        /* Verify stored hash matches calculated hash. */
 425        hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 426        if (hash != ext4_xattr_inode_get_hash(ea_inode))
 427                return -EFSCORRUPTED;
 428
 429        if (entry) {
 430                __le32 e_hash, tmp_data;
 431
 432                /* Verify entry hash. */
 433                tmp_data = cpu_to_le32(hash);
 434                e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 435                                               &tmp_data, 1);
 436                if (e_hash != entry->e_hash)
 437                        return -EFSCORRUPTED;
 438        }
 439        return 0;
 440}
 441
 442/*
 443 * Read xattr value from the EA inode.
 444 */
 445static int
 446ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 447                     void *buffer, size_t size)
 448{
 449        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 450        struct inode *ea_inode;
 451        int err;
 452
 453        err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 454                                    le32_to_cpu(entry->e_hash), &ea_inode);
 455        if (err) {
 456                ea_inode = NULL;
 457                goto out;
 458        }
 459
 460        if (i_size_read(ea_inode) != size) {
 461                ext4_warning_inode(ea_inode,
 462                                   "ea_inode file size=%llu entry size=%zu",
 463                                   i_size_read(ea_inode), size);
 464                err = -EFSCORRUPTED;
 465                goto out;
 466        }
 467
 468        err = ext4_xattr_inode_read(ea_inode, buffer, size);
 469        if (err)
 470                goto out;
 471
 472        if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 473                err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 474                                                     size);
 475                if (err) {
 476                        ext4_warning_inode(ea_inode,
 477                                           "EA inode hash validation failed");
 478                        goto out;
 479                }
 480
 481                if (ea_inode_cache)
 482                        mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 483                                        ext4_xattr_inode_get_hash(ea_inode),
 484                                        ea_inode->i_ino, true /* reusable */);
 485        }
 486out:
 487        iput(ea_inode);
 488        return err;
 489}
 490
 491static int
 492ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 493                     void *buffer, size_t buffer_size)
 494{
 495        struct buffer_head *bh = NULL;
 496        struct ext4_xattr_entry *entry;
 497        size_t size;
 498        int error;
 499        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 500
 501        ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 502                  name_index, name, buffer, (long)buffer_size);
 503
 504        error = -ENODATA;
 505        if (!EXT4_I(inode)->i_file_acl)
 506                goto cleanup;
 507        ea_idebug(inode, "reading block %llu",
 508                  (unsigned long long)EXT4_I(inode)->i_file_acl);
 509        bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 510        if (!bh)
 511                goto cleanup;
 512        ea_bdebug(bh, "b_count=%d, refcount=%d",
 513                atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 514        if (ext4_xattr_check_block(inode, bh)) {
 515                EXT4_ERROR_INODE(inode, "bad block %llu",
 516                                 EXT4_I(inode)->i_file_acl);
 517                error = -EFSCORRUPTED;
 518                goto cleanup;
 519        }
 520        ext4_xattr_block_cache_insert(ea_block_cache, bh);
 521        entry = BFIRST(bh);
 522        error = ext4_xattr_find_entry(&entry, name_index, name, 1);
 523        if (error)
 524                goto cleanup;
 525        size = le32_to_cpu(entry->e_value_size);
 526        if (buffer) {
 527                error = -ERANGE;
 528                if (size > buffer_size)
 529                        goto cleanup;
 530                if (entry->e_value_inum) {
 531                        error = ext4_xattr_inode_get(inode, entry, buffer,
 532                                                     size);
 533                        if (error)
 534                                goto cleanup;
 535                } else {
 536                        memcpy(buffer, bh->b_data +
 537                               le16_to_cpu(entry->e_value_offs), size);
 538                }
 539        }
 540        error = size;
 541
 542cleanup:
 543        brelse(bh);
 544        return error;
 545}
 546
 547int
 548ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 549                     void *buffer, size_t buffer_size)
 550{
 551        struct ext4_xattr_ibody_header *header;
 552        struct ext4_xattr_entry *entry;
 553        struct ext4_inode *raw_inode;
 554        struct ext4_iloc iloc;
 555        size_t size;
 556        void *end;
 557        int error;
 558
 559        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 560                return -ENODATA;
 561        error = ext4_get_inode_loc(inode, &iloc);
 562        if (error)
 563                return error;
 564        raw_inode = ext4_raw_inode(&iloc);
 565        header = IHDR(inode, raw_inode);
 566        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 567        error = xattr_check_inode(inode, header, end);
 568        if (error)
 569                goto cleanup;
 570        entry = IFIRST(header);
 571        error = ext4_xattr_find_entry(&entry, name_index, name, 0);
 572        if (error)
 573                goto cleanup;
 574        size = le32_to_cpu(entry->e_value_size);
 575        if (buffer) {
 576                error = -ERANGE;
 577                if (size > buffer_size)
 578                        goto cleanup;
 579                if (entry->e_value_inum) {
 580                        error = ext4_xattr_inode_get(inode, entry, buffer,
 581                                                     size);
 582                        if (error)
 583                                goto cleanup;
 584                } else {
 585                        memcpy(buffer, (void *)IFIRST(header) +
 586                               le16_to_cpu(entry->e_value_offs), size);
 587                }
 588        }
 589        error = size;
 590
 591cleanup:
 592        brelse(iloc.bh);
 593        return error;
 594}
 595
 596/*
 597 * ext4_xattr_get()
 598 *
 599 * Copy an extended attribute into the buffer
 600 * provided, or compute the buffer size required.
 601 * Buffer is NULL to compute the size of the buffer required.
 602 *
 603 * Returns a negative error number on failure, or the number of bytes
 604 * used / required on success.
 605 */
 606int
 607ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 608               void *buffer, size_t buffer_size)
 609{
 610        int error;
 611
 612        if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 613                return -EIO;
 614
 615        if (strlen(name) > 255)
 616                return -ERANGE;
 617
 618        down_read(&EXT4_I(inode)->xattr_sem);
 619        error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 620                                     buffer_size);
 621        if (error == -ENODATA)
 622                error = ext4_xattr_block_get(inode, name_index, name, buffer,
 623                                             buffer_size);
 624        up_read(&EXT4_I(inode)->xattr_sem);
 625        return error;
 626}
 627
 628static int
 629ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 630                        char *buffer, size_t buffer_size)
 631{
 632        size_t rest = buffer_size;
 633
 634        for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 635                const struct xattr_handler *handler =
 636                        ext4_xattr_handler(entry->e_name_index);
 637
 638                if (handler && (!handler->list || handler->list(dentry))) {
 639                        const char *prefix = handler->prefix ?: handler->name;
 640                        size_t prefix_len = strlen(prefix);
 641                        size_t size = prefix_len + entry->e_name_len + 1;
 642
 643                        if (buffer) {
 644                                if (size > rest)
 645                                        return -ERANGE;
 646                                memcpy(buffer, prefix, prefix_len);
 647                                buffer += prefix_len;
 648                                memcpy(buffer, entry->e_name, entry->e_name_len);
 649                                buffer += entry->e_name_len;
 650                                *buffer++ = 0;
 651                        }
 652                        rest -= size;
 653                }
 654        }
 655        return buffer_size - rest;  /* total size */
 656}
 657
 658static int
 659ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 660{
 661        struct inode *inode = d_inode(dentry);
 662        struct buffer_head *bh = NULL;
 663        int error;
 664
 665        ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 666                  buffer, (long)buffer_size);
 667
 668        error = 0;
 669        if (!EXT4_I(inode)->i_file_acl)
 670                goto cleanup;
 671        ea_idebug(inode, "reading block %llu",
 672                  (unsigned long long)EXT4_I(inode)->i_file_acl);
 673        bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 674        error = -EIO;
 675        if (!bh)
 676                goto cleanup;
 677        ea_bdebug(bh, "b_count=%d, refcount=%d",
 678                atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 679        if (ext4_xattr_check_block(inode, bh)) {
 680                EXT4_ERROR_INODE(inode, "bad block %llu",
 681                                 EXT4_I(inode)->i_file_acl);
 682                error = -EFSCORRUPTED;
 683                goto cleanup;
 684        }
 685        ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 686        error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 687
 688cleanup:
 689        brelse(bh);
 690
 691        return error;
 692}
 693
 694static int
 695ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 696{
 697        struct inode *inode = d_inode(dentry);
 698        struct ext4_xattr_ibody_header *header;
 699        struct ext4_inode *raw_inode;
 700        struct ext4_iloc iloc;
 701        void *end;
 702        int error;
 703
 704        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 705                return 0;
 706        error = ext4_get_inode_loc(inode, &iloc);
 707        if (error)
 708                return error;
 709        raw_inode = ext4_raw_inode(&iloc);
 710        header = IHDR(inode, raw_inode);
 711        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 712        error = xattr_check_inode(inode, header, end);
 713        if (error)
 714                goto cleanup;
 715        error = ext4_xattr_list_entries(dentry, IFIRST(header),
 716                                        buffer, buffer_size);
 717
 718cleanup:
 719        brelse(iloc.bh);
 720        return error;
 721}
 722
 723/*
 724 * Inode operation listxattr()
 725 *
 726 * d_inode(dentry)->i_rwsem: don't care
 727 *
 728 * Copy a list of attribute names into the buffer
 729 * provided, or compute the buffer size required.
 730 * Buffer is NULL to compute the size of the buffer required.
 731 *
 732 * Returns a negative error number on failure, or the number of bytes
 733 * used / required on success.
 734 */
 735ssize_t
 736ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 737{
 738        int ret, ret2;
 739
 740        down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 741        ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 742        if (ret < 0)
 743                goto errout;
 744        if (buffer) {
 745                buffer += ret;
 746                buffer_size -= ret;
 747        }
 748        ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 749        if (ret < 0)
 750                goto errout;
 751        ret += ret2;
 752errout:
 753        up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 754        return ret;
 755}
 756
 757/*
 758 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 759 * not set, set it.
 760 */
 761static void ext4_xattr_update_super_block(handle_t *handle,
 762                                          struct super_block *sb)
 763{
 764        if (ext4_has_feature_xattr(sb))
 765                return;
 766
 767        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 768        if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 769                ext4_set_feature_xattr(sb);
 770                ext4_handle_dirty_super(handle, sb);
 771        }
 772}
 773
 774int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 775{
 776        struct ext4_iloc iloc = { .bh = NULL };
 777        struct buffer_head *bh = NULL;
 778        struct ext4_inode *raw_inode;
 779        struct ext4_xattr_ibody_header *header;
 780        struct ext4_xattr_entry *entry;
 781        qsize_t ea_inode_refs = 0;
 782        void *end;
 783        int ret;
 784
 785        lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 786
 787        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 788                ret = ext4_get_inode_loc(inode, &iloc);
 789                if (ret)
 790                        goto out;
 791                raw_inode = ext4_raw_inode(&iloc);
 792                header = IHDR(inode, raw_inode);
 793                end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 794                ret = xattr_check_inode(inode, header, end);
 795                if (ret)
 796                        goto out;
 797
 798                for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 799                     entry = EXT4_XATTR_NEXT(entry))
 800                        if (entry->e_value_inum)
 801                                ea_inode_refs++;
 802        }
 803
 804        if (EXT4_I(inode)->i_file_acl) {
 805                bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 806                if (!bh) {
 807                        ret = -EIO;
 808                        goto out;
 809                }
 810
 811                if (ext4_xattr_check_block(inode, bh)) {
 812                        ret = -EFSCORRUPTED;
 813                        goto out;
 814                }
 815
 816                for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 817                     entry = EXT4_XATTR_NEXT(entry))
 818                        if (entry->e_value_inum)
 819                                ea_inode_refs++;
 820        }
 821        *usage = ea_inode_refs + 1;
 822        ret = 0;
 823out:
 824        brelse(iloc.bh);
 825        brelse(bh);
 826        return ret;
 827}
 828
 829static inline size_t round_up_cluster(struct inode *inode, size_t length)
 830{
 831        struct super_block *sb = inode->i_sb;
 832        size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 833                                    inode->i_blkbits);
 834        size_t mask = ~(cluster_size - 1);
 835
 836        return (length + cluster_size - 1) & mask;
 837}
 838
 839static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 840{
 841        int err;
 842
 843        err = dquot_alloc_inode(inode);
 844        if (err)
 845                return err;
 846        err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 847        if (err)
 848                dquot_free_inode(inode);
 849        return err;
 850}
 851
 852static void ext4_xattr_inode_free_quota(struct inode *parent,
 853                                        struct inode *ea_inode,
 854                                        size_t len)
 855{
 856        if (ea_inode &&
 857            ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 858                return;
 859        dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 860        dquot_free_inode(parent);
 861}
 862
 863int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 864                             struct buffer_head *block_bh, size_t value_len,
 865                             bool is_create)
 866{
 867        int credits;
 868        int blocks;
 869
 870        /*
 871         * 1) Owner inode update
 872         * 2) Ref count update on old xattr block
 873         * 3) new xattr block
 874         * 4) block bitmap update for new xattr block
 875         * 5) group descriptor for new xattr block
 876         * 6) block bitmap update for old xattr block
 877         * 7) group descriptor for old block
 878         *
 879         * 6 & 7 can happen if we have two racing threads T_a and T_b
 880         * which are each trying to set an xattr on inodes I_a and I_b
 881         * which were both initially sharing an xattr block.
 882         */
 883        credits = 7;
 884
 885        /* Quota updates. */
 886        credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 887
 888        /*
 889         * In case of inline data, we may push out the data to a block,
 890         * so we need to reserve credits for this eventuality
 891         */
 892        if (inode && ext4_has_inline_data(inode))
 893                credits += ext4_writepage_trans_blocks(inode) + 1;
 894
 895        /* We are done if ea_inode feature is not enabled. */
 896        if (!ext4_has_feature_ea_inode(sb))
 897                return credits;
 898
 899        /* New ea_inode, inode map, block bitmap, group descriptor. */
 900        credits += 4;
 901
 902        /* Data blocks. */
 903        blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 904
 905        /* Indirection block or one level of extent tree. */
 906        blocks += 1;
 907
 908        /* Block bitmap and group descriptor updates for each block. */
 909        credits += blocks * 2;
 910
 911        /* Blocks themselves. */
 912        credits += blocks;
 913
 914        if (!is_create) {
 915                /* Dereference ea_inode holding old xattr value.
 916                 * Old ea_inode, inode map, block bitmap, group descriptor.
 917                 */
 918                credits += 4;
 919
 920                /* Data blocks for old ea_inode. */
 921                blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 922
 923                /* Indirection block or one level of extent tree for old
 924                 * ea_inode.
 925                 */
 926                blocks += 1;
 927
 928                /* Block bitmap and group descriptor updates for each block. */
 929                credits += blocks * 2;
 930        }
 931
 932        /* We may need to clone the existing xattr block in which case we need
 933         * to increment ref counts for existing ea_inodes referenced by it.
 934         */
 935        if (block_bh) {
 936                struct ext4_xattr_entry *entry = BFIRST(block_bh);
 937
 938                for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 939                        if (entry->e_value_inum)
 940                                /* Ref count update on ea_inode. */
 941                                credits += 1;
 942        }
 943        return credits;
 944}
 945
 946static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 947                                     int credits, struct buffer_head *bh,
 948                                     bool dirty, bool block_csum)
 949{
 950        int error;
 951
 952        if (!ext4_handle_valid(handle))
 953                return 0;
 954
 955        if (handle->h_buffer_credits >= credits)
 956                return 0;
 957
 958        error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 959        if (!error)
 960                return 0;
 961        if (error < 0) {
 962                ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 963                return error;
 964        }
 965
 966        if (bh && dirty) {
 967                if (block_csum)
 968                        ext4_xattr_block_csum_set(inode, bh);
 969                error = ext4_handle_dirty_metadata(handle, NULL, bh);
 970                if (error) {
 971                        ext4_warning(inode->i_sb, "Handle metadata (error %d)",
 972                                     error);
 973                        return error;
 974                }
 975        }
 976
 977        error = ext4_journal_restart(handle, credits);
 978        if (error) {
 979                ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
 980                return error;
 981        }
 982
 983        if (bh) {
 984                error = ext4_journal_get_write_access(handle, bh);
 985                if (error) {
 986                        ext4_warning(inode->i_sb,
 987                                     "Get write access failed (error %d)",
 988                                     error);
 989                        return error;
 990                }
 991        }
 992        return 0;
 993}
 994
 995static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
 996                                       int ref_change)
 997{
 998        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
 999        struct ext4_iloc iloc;
1000        s64 ref_count;
1001        u32 hash;
1002        int ret;
1003
1004        inode_lock(ea_inode);
1005
1006        ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1007        if (ret) {
1008                iloc.bh = NULL;
1009                goto out;
1010        }
1011
1012        ref_count = ext4_xattr_inode_get_ref(ea_inode);
1013        ref_count += ref_change;
1014        ext4_xattr_inode_set_ref(ea_inode, ref_count);
1015
1016        if (ref_change > 0) {
1017                WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1018                          ea_inode->i_ino, ref_count);
1019
1020                if (ref_count == 1) {
1021                        WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1022                                  ea_inode->i_ino, ea_inode->i_nlink);
1023
1024                        set_nlink(ea_inode, 1);
1025                        ext4_orphan_del(handle, ea_inode);
1026
1027                        if (ea_inode_cache) {
1028                                hash = ext4_xattr_inode_get_hash(ea_inode);
1029                                mb_cache_entry_create(ea_inode_cache,
1030                                                      GFP_NOFS, hash,
1031                                                      ea_inode->i_ino,
1032                                                      true /* reusable */);
1033                        }
1034                }
1035        } else {
1036                WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1037                          ea_inode->i_ino, ref_count);
1038
1039                if (ref_count == 0) {
1040                        WARN_ONCE(ea_inode->i_nlink != 1,
1041                                  "EA inode %lu i_nlink=%u",
1042                                  ea_inode->i_ino, ea_inode->i_nlink);
1043
1044                        clear_nlink(ea_inode);
1045                        ext4_orphan_add(handle, ea_inode);
1046
1047                        if (ea_inode_cache) {
1048                                hash = ext4_xattr_inode_get_hash(ea_inode);
1049                                mb_cache_entry_delete(ea_inode_cache, hash,
1050                                                      ea_inode->i_ino);
1051                        }
1052                }
1053        }
1054
1055        ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1056        iloc.bh = NULL;
1057        if (ret)
1058                ext4_warning_inode(ea_inode,
1059                                   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1060out:
1061        brelse(iloc.bh);
1062        inode_unlock(ea_inode);
1063        return ret;
1064}
1065
1066static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1067{
1068        return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1069}
1070
1071static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1072{
1073        return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1074}
1075
1076static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1077                                        struct ext4_xattr_entry *first)
1078{
1079        struct inode *ea_inode;
1080        struct ext4_xattr_entry *entry;
1081        struct ext4_xattr_entry *failed_entry;
1082        unsigned int ea_ino;
1083        int err, saved_err;
1084
1085        for (entry = first; !IS_LAST_ENTRY(entry);
1086             entry = EXT4_XATTR_NEXT(entry)) {
1087                if (!entry->e_value_inum)
1088                        continue;
1089                ea_ino = le32_to_cpu(entry->e_value_inum);
1090                err = ext4_xattr_inode_iget(parent, ea_ino,
1091                                            le32_to_cpu(entry->e_hash),
1092                                            &ea_inode);
1093                if (err)
1094                        goto cleanup;
1095                err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1096                if (err) {
1097                        ext4_warning_inode(ea_inode, "inc ref error %d", err);
1098                        iput(ea_inode);
1099                        goto cleanup;
1100                }
1101                iput(ea_inode);
1102        }
1103        return 0;
1104
1105cleanup:
1106        saved_err = err;
1107        failed_entry = entry;
1108
1109        for (entry = first; entry != failed_entry;
1110             entry = EXT4_XATTR_NEXT(entry)) {
1111                if (!entry->e_value_inum)
1112                        continue;
1113                ea_ino = le32_to_cpu(entry->e_value_inum);
1114                err = ext4_xattr_inode_iget(parent, ea_ino,
1115                                            le32_to_cpu(entry->e_hash),
1116                                            &ea_inode);
1117                if (err) {
1118                        ext4_warning(parent->i_sb,
1119                                     "cleanup ea_ino %u iget error %d", ea_ino,
1120                                     err);
1121                        continue;
1122                }
1123                err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1124                if (err)
1125                        ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1126                                           err);
1127                iput(ea_inode);
1128        }
1129        return saved_err;
1130}
1131
1132static void
1133ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1134                             struct buffer_head *bh,
1135                             struct ext4_xattr_entry *first, bool block_csum,
1136                             struct ext4_xattr_inode_array **ea_inode_array,
1137                             int extra_credits, bool skip_quota)
1138{
1139        struct inode *ea_inode;
1140        struct ext4_xattr_entry *entry;
1141        bool dirty = false;
1142        unsigned int ea_ino;
1143        int err;
1144        int credits;
1145
1146        /* One credit for dec ref on ea_inode, one for orphan list addition, */
1147        credits = 2 + extra_credits;
1148
1149        for (entry = first; !IS_LAST_ENTRY(entry);
1150             entry = EXT4_XATTR_NEXT(entry)) {
1151                if (!entry->e_value_inum)
1152                        continue;
1153                ea_ino = le32_to_cpu(entry->e_value_inum);
1154                err = ext4_xattr_inode_iget(parent, ea_ino,
1155                                            le32_to_cpu(entry->e_hash),
1156                                            &ea_inode);
1157                if (err)
1158                        continue;
1159
1160                err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1161                if (err) {
1162                        ext4_warning_inode(ea_inode,
1163                                           "Expand inode array err=%d", err);
1164                        iput(ea_inode);
1165                        continue;
1166                }
1167
1168                err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1169                                                dirty, block_csum);
1170                if (err) {
1171                        ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1172                                           err);
1173                        continue;
1174                }
1175
1176                err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1177                if (err) {
1178                        ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1179                                           err);
1180                        continue;
1181                }
1182
1183                if (!skip_quota)
1184                        ext4_xattr_inode_free_quota(parent, ea_inode,
1185                                              le32_to_cpu(entry->e_value_size));
1186
1187                /*
1188                 * Forget about ea_inode within the same transaction that
1189                 * decrements the ref count. This avoids duplicate decrements in
1190                 * case the rest of the work spills over to subsequent
1191                 * transactions.
1192                 */
1193                entry->e_value_inum = 0;
1194                entry->e_value_size = 0;
1195
1196                dirty = true;
1197        }
1198
1199        if (dirty) {
1200                /*
1201                 * Note that we are deliberately skipping csum calculation for
1202                 * the final update because we do not expect any journal
1203                 * restarts until xattr block is freed.
1204                 */
1205
1206                err = ext4_handle_dirty_metadata(handle, NULL, bh);
1207                if (err)
1208                        ext4_warning_inode(parent,
1209                                           "handle dirty metadata err=%d", err);
1210        }
1211}
1212
1213/*
1214 * Release the xattr block BH: If the reference count is > 1, decrement it;
1215 * otherwise free the block.
1216 */
1217static void
1218ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1219                         struct buffer_head *bh,
1220                         struct ext4_xattr_inode_array **ea_inode_array,
1221                         int extra_credits)
1222{
1223        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1224        u32 hash, ref;
1225        int error = 0;
1226
1227        BUFFER_TRACE(bh, "get_write_access");
1228        error = ext4_journal_get_write_access(handle, bh);
1229        if (error)
1230                goto out;
1231
1232        lock_buffer(bh);
1233        hash = le32_to_cpu(BHDR(bh)->h_hash);
1234        ref = le32_to_cpu(BHDR(bh)->h_refcount);
1235        if (ref == 1) {
1236                ea_bdebug(bh, "refcount now=0; freeing");
1237                /*
1238                 * This must happen under buffer lock for
1239                 * ext4_xattr_block_set() to reliably detect freed block
1240                 */
1241                if (ea_block_cache)
1242                        mb_cache_entry_delete(ea_block_cache, hash,
1243                                              bh->b_blocknr);
1244                get_bh(bh);
1245                unlock_buffer(bh);
1246
1247                if (ext4_has_feature_ea_inode(inode->i_sb))
1248                        ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1249                                                     BFIRST(bh),
1250                                                     true /* block_csum */,
1251                                                     ea_inode_array,
1252                                                     extra_credits,
1253                                                     true /* skip_quota */);
1254                ext4_free_blocks(handle, inode, bh, 0, 1,
1255                                 EXT4_FREE_BLOCKS_METADATA |
1256                                 EXT4_FREE_BLOCKS_FORGET);
1257        } else {
1258                ref--;
1259                BHDR(bh)->h_refcount = cpu_to_le32(ref);
1260                if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1261                        struct mb_cache_entry *ce;
1262
1263                        if (ea_block_cache) {
1264                                ce = mb_cache_entry_get(ea_block_cache, hash,
1265                                                        bh->b_blocknr);
1266                                if (ce) {
1267                                        ce->e_reusable = 1;
1268                                        mb_cache_entry_put(ea_block_cache, ce);
1269                                }
1270                        }
1271                }
1272
1273                ext4_xattr_block_csum_set(inode, bh);
1274                /*
1275                 * Beware of this ugliness: Releasing of xattr block references
1276                 * from different inodes can race and so we have to protect
1277                 * from a race where someone else frees the block (and releases
1278                 * its journal_head) before we are done dirtying the buffer. In
1279                 * nojournal mode this race is harmless and we actually cannot
1280                 * call ext4_handle_dirty_metadata() with locked buffer as
1281                 * that function can call sync_dirty_buffer() so for that case
1282                 * we handle the dirtying after unlocking the buffer.
1283                 */
1284                if (ext4_handle_valid(handle))
1285                        error = ext4_handle_dirty_metadata(handle, inode, bh);
1286                unlock_buffer(bh);
1287                if (!ext4_handle_valid(handle))
1288                        error = ext4_handle_dirty_metadata(handle, inode, bh);
1289                if (IS_SYNC(inode))
1290                        ext4_handle_sync(handle);
1291                dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1292                ea_bdebug(bh, "refcount now=%d; releasing",
1293                          le32_to_cpu(BHDR(bh)->h_refcount));
1294        }
1295out:
1296        ext4_std_error(inode->i_sb, error);
1297        return;
1298}
1299
1300/*
1301 * Find the available free space for EAs. This also returns the total number of
1302 * bytes used by EA entries.
1303 */
1304static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1305                                    size_t *min_offs, void *base, int *total)
1306{
1307        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1308                if (!last->e_value_inum && last->e_value_size) {
1309                        size_t offs = le16_to_cpu(last->e_value_offs);
1310                        if (offs < *min_offs)
1311                                *min_offs = offs;
1312                }
1313                if (total)
1314                        *total += EXT4_XATTR_LEN(last->e_name_len);
1315        }
1316        return (*min_offs - ((void *)last - base) - sizeof(__u32));
1317}
1318
1319/*
1320 * Write the value of the EA in an inode.
1321 */
1322static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1323                                  const void *buf, int bufsize)
1324{
1325        struct buffer_head *bh = NULL;
1326        unsigned long block = 0;
1327        int blocksize = ea_inode->i_sb->s_blocksize;
1328        int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1329        int csize, wsize = 0;
1330        int ret = 0;
1331        int retries = 0;
1332
1333retry:
1334        while (ret >= 0 && ret < max_blocks) {
1335                struct ext4_map_blocks map;
1336                map.m_lblk = block += ret;
1337                map.m_len = max_blocks -= ret;
1338
1339                ret = ext4_map_blocks(handle, ea_inode, &map,
1340                                      EXT4_GET_BLOCKS_CREATE);
1341                if (ret <= 0) {
1342                        ext4_mark_inode_dirty(handle, ea_inode);
1343                        if (ret == -ENOSPC &&
1344                            ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1345                                ret = 0;
1346                                goto retry;
1347                        }
1348                        break;
1349                }
1350        }
1351
1352        if (ret < 0)
1353                return ret;
1354
1355        block = 0;
1356        while (wsize < bufsize) {
1357                if (bh != NULL)
1358                        brelse(bh);
1359                csize = (bufsize - wsize) > blocksize ? blocksize :
1360                                                                bufsize - wsize;
1361                bh = ext4_getblk(handle, ea_inode, block, 0);
1362                if (IS_ERR(bh))
1363                        return PTR_ERR(bh);
1364                ret = ext4_journal_get_write_access(handle, bh);
1365                if (ret)
1366                        goto out;
1367
1368                memcpy(bh->b_data, buf, csize);
1369                set_buffer_uptodate(bh);
1370                ext4_handle_dirty_metadata(handle, ea_inode, bh);
1371
1372                buf += csize;
1373                wsize += csize;
1374                block += 1;
1375        }
1376
1377        inode_lock(ea_inode);
1378        i_size_write(ea_inode, wsize);
1379        ext4_update_i_disksize(ea_inode, wsize);
1380        inode_unlock(ea_inode);
1381
1382        ext4_mark_inode_dirty(handle, ea_inode);
1383
1384out:
1385        brelse(bh);
1386
1387        return ret;
1388}
1389
1390/*
1391 * Create an inode to store the value of a large EA.
1392 */
1393static struct inode *ext4_xattr_inode_create(handle_t *handle,
1394                                             struct inode *inode, u32 hash)
1395{
1396        struct inode *ea_inode = NULL;
1397        uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1398        int err;
1399
1400        /*
1401         * Let the next inode be the goal, so we try and allocate the EA inode
1402         * in the same group, or nearby one.
1403         */
1404        ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1405                                  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1406                                  EXT4_EA_INODE_FL);
1407        if (!IS_ERR(ea_inode)) {
1408                ea_inode->i_op = &ext4_file_inode_operations;
1409                ea_inode->i_fop = &ext4_file_operations;
1410                ext4_set_aops(ea_inode);
1411                ext4_xattr_inode_set_class(ea_inode);
1412                unlock_new_inode(ea_inode);
1413                ext4_xattr_inode_set_ref(ea_inode, 1);
1414                ext4_xattr_inode_set_hash(ea_inode, hash);
1415                err = ext4_mark_inode_dirty(handle, ea_inode);
1416                if (!err)
1417                        err = ext4_inode_attach_jinode(ea_inode);
1418                if (err) {
1419                        iput(ea_inode);
1420                        return ERR_PTR(err);
1421                }
1422
1423                /*
1424                 * Xattr inodes are shared therefore quota charging is performed
1425                 * at a higher level.
1426                 */
1427                dquot_free_inode(ea_inode);
1428                dquot_drop(ea_inode);
1429                inode_lock(ea_inode);
1430                ea_inode->i_flags |= S_NOQUOTA;
1431                inode_unlock(ea_inode);
1432        }
1433
1434        return ea_inode;
1435}
1436
1437static struct inode *
1438ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1439                            size_t value_len, u32 hash)
1440{
1441        struct inode *ea_inode;
1442        struct mb_cache_entry *ce;
1443        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1444        void *ea_data;
1445
1446        if (!ea_inode_cache)
1447                return NULL;
1448
1449        ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1450        if (!ce)
1451                return NULL;
1452
1453        ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1454        if (!ea_data) {
1455                mb_cache_entry_put(ea_inode_cache, ce);
1456                return NULL;
1457        }
1458
1459        while (ce) {
1460                ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1461                if (!IS_ERR(ea_inode) &&
1462                    !is_bad_inode(ea_inode) &&
1463                    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1464                    i_size_read(ea_inode) == value_len &&
1465                    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1466                    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1467                                                    value_len) &&
1468                    !memcmp(value, ea_data, value_len)) {
1469                        mb_cache_entry_touch(ea_inode_cache, ce);
1470                        mb_cache_entry_put(ea_inode_cache, ce);
1471                        kvfree(ea_data);
1472                        return ea_inode;
1473                }
1474
1475                if (!IS_ERR(ea_inode))
1476                        iput(ea_inode);
1477                ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1478        }
1479        kvfree(ea_data);
1480        return NULL;
1481}
1482
1483/*
1484 * Add value of the EA in an inode.
1485 */
1486static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1487                                          const void *value, size_t value_len,
1488                                          struct inode **ret_inode)
1489{
1490        struct inode *ea_inode;
1491        u32 hash;
1492        int err;
1493
1494        hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1495        ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1496        if (ea_inode) {
1497                err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1498                if (err) {
1499                        iput(ea_inode);
1500                        return err;
1501                }
1502
1503                *ret_inode = ea_inode;
1504                return 0;
1505        }
1506
1507        /* Create an inode for the EA value */
1508        ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1509        if (IS_ERR(ea_inode))
1510                return PTR_ERR(ea_inode);
1511
1512        err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1513        if (err) {
1514                ext4_xattr_inode_dec_ref(handle, ea_inode);
1515                iput(ea_inode);
1516                return err;
1517        }
1518
1519        if (EA_INODE_CACHE(inode))
1520                mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1521                                      ea_inode->i_ino, true /* reusable */);
1522
1523        *ret_inode = ea_inode;
1524        return 0;
1525}
1526
1527/*
1528 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1529 * feature is enabled.
1530 */
1531#define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1532
1533static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1534                                struct ext4_xattr_search *s,
1535                                handle_t *handle, struct inode *inode,
1536                                bool is_block)
1537{
1538        struct ext4_xattr_entry *last;
1539        struct ext4_xattr_entry *here = s->here;
1540        size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1541        int in_inode = i->in_inode;
1542        struct inode *old_ea_inode = NULL;
1543        struct inode *new_ea_inode = NULL;
1544        size_t old_size, new_size;
1545        int ret;
1546
1547        /* Space used by old and new values. */
1548        old_size = (!s->not_found && !here->e_value_inum) ?
1549                        EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1550        new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1551
1552        /*
1553         * Optimization for the simple case when old and new values have the
1554         * same padded sizes. Not applicable if external inodes are involved.
1555         */
1556        if (new_size && new_size == old_size) {
1557                size_t offs = le16_to_cpu(here->e_value_offs);
1558                void *val = s->base + offs;
1559
1560                here->e_value_size = cpu_to_le32(i->value_len);
1561                if (i->value == EXT4_ZERO_XATTR_VALUE) {
1562                        memset(val, 0, new_size);
1563                } else {
1564                        memcpy(val, i->value, i->value_len);
1565                        /* Clear padding bytes. */
1566                        memset(val + i->value_len, 0, new_size - i->value_len);
1567                }
1568                goto update_hash;
1569        }
1570
1571        /* Compute min_offs and last. */
1572        last = s->first;
1573        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1574                if (!last->e_value_inum && last->e_value_size) {
1575                        size_t offs = le16_to_cpu(last->e_value_offs);
1576                        if (offs < min_offs)
1577                                min_offs = offs;
1578                }
1579        }
1580
1581        /* Check whether we have enough space. */
1582        if (i->value) {
1583                size_t free;
1584
1585                free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1586                if (!s->not_found)
1587                        free += EXT4_XATTR_LEN(name_len) + old_size;
1588
1589                if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1590                        ret = -ENOSPC;
1591                        goto out;
1592                }
1593
1594                /*
1595                 * If storing the value in an external inode is an option,
1596                 * reserve space for xattr entries/names in the external
1597                 * attribute block so that a long value does not occupy the
1598                 * whole space and prevent futher entries being added.
1599                 */
1600                if (ext4_has_feature_ea_inode(inode->i_sb) &&
1601                    new_size && is_block &&
1602                    (min_offs + old_size - new_size) <
1603                                        EXT4_XATTR_BLOCK_RESERVE(inode)) {
1604                        ret = -ENOSPC;
1605                        goto out;
1606                }
1607        }
1608
1609        /*
1610         * Getting access to old and new ea inodes is subject to failures.
1611         * Finish that work before doing any modifications to the xattr data.
1612         */
1613        if (!s->not_found && here->e_value_inum) {
1614                ret = ext4_xattr_inode_iget(inode,
1615                                            le32_to_cpu(here->e_value_inum),
1616                                            le32_to_cpu(here->e_hash),
1617                                            &old_ea_inode);
1618                if (ret) {
1619                        old_ea_inode = NULL;
1620                        goto out;
1621                }
1622        }
1623        if (i->value && in_inode) {
1624                WARN_ON_ONCE(!i->value_len);
1625
1626                ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1627                if (ret)
1628                        goto out;
1629
1630                ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1631                                                     i->value_len,
1632                                                     &new_ea_inode);
1633                if (ret) {
1634                        new_ea_inode = NULL;
1635                        ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1636                        goto out;
1637                }
1638        }
1639
1640        if (old_ea_inode) {
1641                /* We are ready to release ref count on the old_ea_inode. */
1642                ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1643                if (ret) {
1644                        /* Release newly required ref count on new_ea_inode. */
1645                        if (new_ea_inode) {
1646                                int err;
1647
1648                                err = ext4_xattr_inode_dec_ref(handle,
1649                                                               new_ea_inode);
1650                                if (err)
1651                                        ext4_warning_inode(new_ea_inode,
1652                                                  "dec ref new_ea_inode err=%d",
1653                                                  err);
1654                                ext4_xattr_inode_free_quota(inode, new_ea_inode,
1655                                                            i->value_len);
1656                        }
1657                        goto out;
1658                }
1659
1660                ext4_xattr_inode_free_quota(inode, old_ea_inode,
1661                                            le32_to_cpu(here->e_value_size));
1662        }
1663
1664        /* No failures allowed past this point. */
1665
1666        if (!s->not_found && here->e_value_offs) {
1667                /* Remove the old value. */
1668                void *first_val = s->base + min_offs;
1669                size_t offs = le16_to_cpu(here->e_value_offs);
1670                void *val = s->base + offs;
1671
1672                memmove(first_val + old_size, first_val, val - first_val);
1673                memset(first_val, 0, old_size);
1674                min_offs += old_size;
1675
1676                /* Adjust all value offsets. */
1677                last = s->first;
1678                while (!IS_LAST_ENTRY(last)) {
1679                        size_t o = le16_to_cpu(last->e_value_offs);
1680
1681                        if (!last->e_value_inum &&
1682                            last->e_value_size && o < offs)
1683                                last->e_value_offs = cpu_to_le16(o + old_size);
1684                        last = EXT4_XATTR_NEXT(last);
1685                }
1686        }
1687
1688        if (!i->value) {
1689                /* Remove old name. */
1690                size_t size = EXT4_XATTR_LEN(name_len);
1691
1692                last = ENTRY((void *)last - size);
1693                memmove(here, (void *)here + size,
1694                        (void *)last - (void *)here + sizeof(__u32));
1695                memset(last, 0, size);
1696        } else if (s->not_found) {
1697                /* Insert new name. */
1698                size_t size = EXT4_XATTR_LEN(name_len);
1699                size_t rest = (void *)last - (void *)here + sizeof(__u32);
1700
1701                memmove((void *)here + size, here, rest);
1702                memset(here, 0, size);
1703                here->e_name_index = i->name_index;
1704                here->e_name_len = name_len;
1705                memcpy(here->e_name, i->name, name_len);
1706        } else {
1707                /* This is an update, reset value info. */
1708                here->e_value_inum = 0;
1709                here->e_value_offs = 0;
1710                here->e_value_size = 0;
1711        }
1712
1713        if (i->value) {
1714                /* Insert new value. */
1715                if (in_inode) {
1716                        here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1717                } else if (i->value_len) {
1718                        void *val = s->base + min_offs - new_size;
1719
1720                        here->e_value_offs = cpu_to_le16(min_offs - new_size);
1721                        if (i->value == EXT4_ZERO_XATTR_VALUE) {
1722                                memset(val, 0, new_size);
1723                        } else {
1724                                memcpy(val, i->value, i->value_len);
1725                                /* Clear padding bytes. */
1726                                memset(val + i->value_len, 0,
1727                                       new_size - i->value_len);
1728                        }
1729                }
1730                here->e_value_size = cpu_to_le32(i->value_len);
1731        }
1732
1733update_hash:
1734        if (i->value) {
1735                __le32 hash = 0;
1736
1737                /* Entry hash calculation. */
1738                if (in_inode) {
1739                        __le32 crc32c_hash;
1740
1741                        /*
1742                         * Feed crc32c hash instead of the raw value for entry
1743                         * hash calculation. This is to avoid walking
1744                         * potentially long value buffer again.
1745                         */
1746                        crc32c_hash = cpu_to_le32(
1747                                       ext4_xattr_inode_get_hash(new_ea_inode));
1748                        hash = ext4_xattr_hash_entry(here->e_name,
1749                                                     here->e_name_len,
1750                                                     &crc32c_hash, 1);
1751                } else if (is_block) {
1752                        __le32 *value = s->base + le16_to_cpu(
1753                                                        here->e_value_offs);
1754
1755                        hash = ext4_xattr_hash_entry(here->e_name,
1756                                                     here->e_name_len, value,
1757                                                     new_size >> 2);
1758                }
1759                here->e_hash = hash;
1760        }
1761
1762        if (is_block)
1763                ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1764
1765        ret = 0;
1766out:
1767        iput(old_ea_inode);
1768        iput(new_ea_inode);
1769        return ret;
1770}
1771
1772struct ext4_xattr_block_find {
1773        struct ext4_xattr_search s;
1774        struct buffer_head *bh;
1775};
1776
1777static int
1778ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1779                      struct ext4_xattr_block_find *bs)
1780{
1781        struct super_block *sb = inode->i_sb;
1782        int error;
1783
1784        ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1785                  i->name_index, i->name, i->value, (long)i->value_len);
1786
1787        if (EXT4_I(inode)->i_file_acl) {
1788                /* The inode already has an extended attribute block. */
1789                bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1790                error = -EIO;
1791                if (!bs->bh)
1792                        goto cleanup;
1793                ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1794                        atomic_read(&(bs->bh->b_count)),
1795                        le32_to_cpu(BHDR(bs->bh)->h_refcount));
1796                if (ext4_xattr_check_block(inode, bs->bh)) {
1797                        EXT4_ERROR_INODE(inode, "bad block %llu",
1798                                         EXT4_I(inode)->i_file_acl);
1799                        error = -EFSCORRUPTED;
1800                        goto cleanup;
1801                }
1802                /* Find the named attribute. */
1803                bs->s.base = BHDR(bs->bh);
1804                bs->s.first = BFIRST(bs->bh);
1805                bs->s.end = bs->bh->b_data + bs->bh->b_size;
1806                bs->s.here = bs->s.first;
1807                error = ext4_xattr_find_entry(&bs->s.here, i->name_index,
1808                                              i->name, 1);
1809                if (error && error != -ENODATA)
1810                        goto cleanup;
1811                bs->s.not_found = error;
1812        }
1813        error = 0;
1814
1815cleanup:
1816        return error;
1817}
1818
1819static int
1820ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1821                     struct ext4_xattr_info *i,
1822                     struct ext4_xattr_block_find *bs)
1823{
1824        struct super_block *sb = inode->i_sb;
1825        struct buffer_head *new_bh = NULL;
1826        struct ext4_xattr_search s_copy = bs->s;
1827        struct ext4_xattr_search *s = &s_copy;
1828        struct mb_cache_entry *ce = NULL;
1829        int error = 0;
1830        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1831        struct inode *ea_inode = NULL, *tmp_inode;
1832        size_t old_ea_inode_quota = 0;
1833        unsigned int ea_ino;
1834
1835
1836#define header(x) ((struct ext4_xattr_header *)(x))
1837
1838        if (s->base) {
1839                BUFFER_TRACE(bs->bh, "get_write_access");
1840                error = ext4_journal_get_write_access(handle, bs->bh);
1841                if (error)
1842                        goto cleanup;
1843                lock_buffer(bs->bh);
1844
1845                if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1846                        __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1847
1848                        /*
1849                         * This must happen under buffer lock for
1850                         * ext4_xattr_block_set() to reliably detect modified
1851                         * block
1852                         */
1853                        if (ea_block_cache)
1854                                mb_cache_entry_delete(ea_block_cache, hash,
1855                                                      bs->bh->b_blocknr);
1856                        ea_bdebug(bs->bh, "modifying in-place");
1857                        error = ext4_xattr_set_entry(i, s, handle, inode,
1858                                                     true /* is_block */);
1859                        ext4_xattr_block_csum_set(inode, bs->bh);
1860                        unlock_buffer(bs->bh);
1861                        if (error == -EFSCORRUPTED)
1862                                goto bad_block;
1863                        if (!error)
1864                                error = ext4_handle_dirty_metadata(handle,
1865                                                                   inode,
1866                                                                   bs->bh);
1867                        if (error)
1868                                goto cleanup;
1869                        goto inserted;
1870                } else {
1871                        int offset = (char *)s->here - bs->bh->b_data;
1872
1873                        unlock_buffer(bs->bh);
1874                        ea_bdebug(bs->bh, "cloning");
1875                        s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1876                        error = -ENOMEM;
1877                        if (s->base == NULL)
1878                                goto cleanup;
1879                        memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1880                        s->first = ENTRY(header(s->base)+1);
1881                        header(s->base)->h_refcount = cpu_to_le32(1);
1882                        s->here = ENTRY(s->base + offset);
1883                        s->end = s->base + bs->bh->b_size;
1884
1885                        /*
1886                         * If existing entry points to an xattr inode, we need
1887                         * to prevent ext4_xattr_set_entry() from decrementing
1888                         * ref count on it because the reference belongs to the
1889                         * original block. In this case, make the entry look
1890                         * like it has an empty value.
1891                         */
1892                        if (!s->not_found && s->here->e_value_inum) {
1893                                ea_ino = le32_to_cpu(s->here->e_value_inum);
1894                                error = ext4_xattr_inode_iget(inode, ea_ino,
1895                                              le32_to_cpu(s->here->e_hash),
1896                                              &tmp_inode);
1897                                if (error)
1898                                        goto cleanup;
1899
1900                                if (!ext4_test_inode_state(tmp_inode,
1901                                                EXT4_STATE_LUSTRE_EA_INODE)) {
1902                                        /*
1903                                         * Defer quota free call for previous
1904                                         * inode until success is guaranteed.
1905                                         */
1906                                        old_ea_inode_quota = le32_to_cpu(
1907                                                        s->here->e_value_size);
1908                                }
1909                                iput(tmp_inode);
1910
1911                                s->here->e_value_inum = 0;
1912                                s->here->e_value_size = 0;
1913                        }
1914                }
1915        } else {
1916                /* Allocate a buffer where we construct the new block. */
1917                s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1918                /* assert(header == s->base) */
1919                error = -ENOMEM;
1920                if (s->base == NULL)
1921                        goto cleanup;
1922                header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1923                header(s->base)->h_blocks = cpu_to_le32(1);
1924                header(s->base)->h_refcount = cpu_to_le32(1);
1925                s->first = ENTRY(header(s->base)+1);
1926                s->here = ENTRY(header(s->base)+1);
1927                s->end = s->base + sb->s_blocksize;
1928        }
1929
1930        error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1931        if (error == -EFSCORRUPTED)
1932                goto bad_block;
1933        if (error)
1934                goto cleanup;
1935
1936        if (i->value && s->here->e_value_inum) {
1937                /*
1938                 * A ref count on ea_inode has been taken as part of the call to
1939                 * ext4_xattr_set_entry() above. We would like to drop this
1940                 * extra ref but we have to wait until the xattr block is
1941                 * initialized and has its own ref count on the ea_inode.
1942                 */
1943                ea_ino = le32_to_cpu(s->here->e_value_inum);
1944                error = ext4_xattr_inode_iget(inode, ea_ino,
1945                                              le32_to_cpu(s->here->e_hash),
1946                                              &ea_inode);
1947                if (error) {
1948                        ea_inode = NULL;
1949                        goto cleanup;
1950                }
1951        }
1952
1953inserted:
1954        if (!IS_LAST_ENTRY(s->first)) {
1955                new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1956                                                     &ce);
1957                if (new_bh) {
1958                        /* We found an identical block in the cache. */
1959                        if (new_bh == bs->bh)
1960                                ea_bdebug(new_bh, "keeping");
1961                        else {
1962                                u32 ref;
1963
1964                                WARN_ON_ONCE(dquot_initialize_needed(inode));
1965
1966                                /* The old block is released after updating
1967                                   the inode. */
1968                                error = dquot_alloc_block(inode,
1969                                                EXT4_C2B(EXT4_SB(sb), 1));
1970                                if (error)
1971                                        goto cleanup;
1972                                BUFFER_TRACE(new_bh, "get_write_access");
1973                                error = ext4_journal_get_write_access(handle,
1974                                                                      new_bh);
1975                                if (error)
1976                                        goto cleanup_dquot;
1977                                lock_buffer(new_bh);
1978                                /*
1979                                 * We have to be careful about races with
1980                                 * freeing, rehashing or adding references to
1981                                 * xattr block. Once we hold buffer lock xattr
1982                                 * block's state is stable so we can check
1983                                 * whether the block got freed / rehashed or
1984                                 * not.  Since we unhash mbcache entry under
1985                                 * buffer lock when freeing / rehashing xattr
1986                                 * block, checking whether entry is still
1987                                 * hashed is reliable. Same rules hold for
1988                                 * e_reusable handling.
1989                                 */
1990                                if (hlist_bl_unhashed(&ce->e_hash_list) ||
1991                                    !ce->e_reusable) {
1992                                        /*
1993                                         * Undo everything and check mbcache
1994                                         * again.
1995                                         */
1996                                        unlock_buffer(new_bh);
1997                                        dquot_free_block(inode,
1998                                                         EXT4_C2B(EXT4_SB(sb),
1999                                                                  1));
2000                                        brelse(new_bh);
2001                                        mb_cache_entry_put(ea_block_cache, ce);
2002                                        ce = NULL;
2003                                        new_bh = NULL;
2004                                        goto inserted;
2005                                }
2006                                ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2007                                BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2008                                if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2009                                        ce->e_reusable = 0;
2010                                ea_bdebug(new_bh, "reusing; refcount now=%d",
2011                                          ref);
2012                                ext4_xattr_block_csum_set(inode, new_bh);
2013                                unlock_buffer(new_bh);
2014                                error = ext4_handle_dirty_metadata(handle,
2015                                                                   inode,
2016                                                                   new_bh);
2017                                if (error)
2018                                        goto cleanup_dquot;
2019                        }
2020                        mb_cache_entry_touch(ea_block_cache, ce);
2021                        mb_cache_entry_put(ea_block_cache, ce);
2022                        ce = NULL;
2023                } else if (bs->bh && s->base == bs->bh->b_data) {
2024                        /* We were modifying this block in-place. */
2025                        ea_bdebug(bs->bh, "keeping this block");
2026                        ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2027                        new_bh = bs->bh;
2028                        get_bh(new_bh);
2029                } else {
2030                        /* We need to allocate a new block */
2031                        ext4_fsblk_t goal, block;
2032
2033                        WARN_ON_ONCE(dquot_initialize_needed(inode));
2034
2035                        goal = ext4_group_first_block_no(sb,
2036                                                EXT4_I(inode)->i_block_group);
2037
2038                        /* non-extent files can't have physical blocks past 2^32 */
2039                        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2040                                goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2041
2042                        block = ext4_new_meta_blocks(handle, inode, goal, 0,
2043                                                     NULL, &error);
2044                        if (error)
2045                                goto cleanup;
2046
2047                        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2048                                BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2049
2050                        ea_idebug(inode, "creating block %llu",
2051                                  (unsigned long long)block);
2052
2053                        new_bh = sb_getblk(sb, block);
2054                        if (unlikely(!new_bh)) {
2055                                error = -ENOMEM;
2056getblk_failed:
2057                                ext4_free_blocks(handle, inode, NULL, block, 1,
2058                                                 EXT4_FREE_BLOCKS_METADATA);
2059                                goto cleanup;
2060                        }
2061                        error = ext4_xattr_inode_inc_ref_all(handle, inode,
2062                                                      ENTRY(header(s->base)+1));
2063                        if (error)
2064                                goto getblk_failed;
2065                        if (ea_inode) {
2066                                /* Drop the extra ref on ea_inode. */
2067                                error = ext4_xattr_inode_dec_ref(handle,
2068                                                                 ea_inode);
2069                                if (error)
2070                                        ext4_warning_inode(ea_inode,
2071                                                           "dec ref error=%d",
2072                                                           error);
2073                                iput(ea_inode);
2074                                ea_inode = NULL;
2075                        }
2076
2077                        lock_buffer(new_bh);
2078                        error = ext4_journal_get_create_access(handle, new_bh);
2079                        if (error) {
2080                                unlock_buffer(new_bh);
2081                                error = -EIO;
2082                                goto getblk_failed;
2083                        }
2084                        memcpy(new_bh->b_data, s->base, new_bh->b_size);
2085                        ext4_xattr_block_csum_set(inode, new_bh);
2086                        set_buffer_uptodate(new_bh);
2087                        unlock_buffer(new_bh);
2088                        ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2089                        error = ext4_handle_dirty_metadata(handle, inode,
2090                                                           new_bh);
2091                        if (error)
2092                                goto cleanup;
2093                }
2094        }
2095
2096        if (old_ea_inode_quota)
2097                ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2098
2099        /* Update the inode. */
2100        EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2101
2102        /* Drop the previous xattr block. */
2103        if (bs->bh && bs->bh != new_bh) {
2104                struct ext4_xattr_inode_array *ea_inode_array = NULL;
2105
2106                ext4_xattr_release_block(handle, inode, bs->bh,
2107                                         &ea_inode_array,
2108                                         0 /* extra_credits */);
2109                ext4_xattr_inode_array_free(ea_inode_array);
2110        }
2111        error = 0;
2112
2113cleanup:
2114        if (ea_inode) {
2115                int error2;
2116
2117                error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2118                if (error2)
2119                        ext4_warning_inode(ea_inode, "dec ref error=%d",
2120                                           error2);
2121
2122                /* If there was an error, revert the quota charge. */
2123                if (error)
2124                        ext4_xattr_inode_free_quota(inode, ea_inode,
2125                                                    i_size_read(ea_inode));
2126                iput(ea_inode);
2127        }
2128        if (ce)
2129                mb_cache_entry_put(ea_block_cache, ce);
2130        brelse(new_bh);
2131        if (!(bs->bh && s->base == bs->bh->b_data))
2132                kfree(s->base);
2133
2134        return error;
2135
2136cleanup_dquot:
2137        dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2138        goto cleanup;
2139
2140bad_block:
2141        EXT4_ERROR_INODE(inode, "bad block %llu",
2142                         EXT4_I(inode)->i_file_acl);
2143        goto cleanup;
2144
2145#undef header
2146}
2147
2148int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2149                          struct ext4_xattr_ibody_find *is)
2150{
2151        struct ext4_xattr_ibody_header *header;
2152        struct ext4_inode *raw_inode;
2153        int error;
2154
2155        if (EXT4_I(inode)->i_extra_isize == 0)
2156                return 0;
2157        raw_inode = ext4_raw_inode(&is->iloc);
2158        header = IHDR(inode, raw_inode);
2159        is->s.base = is->s.first = IFIRST(header);
2160        is->s.here = is->s.first;
2161        is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2162        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2163                error = xattr_check_inode(inode, header, is->s.end);
2164                if (error)
2165                        return error;
2166                /* Find the named attribute. */
2167                error = ext4_xattr_find_entry(&is->s.here, i->name_index,
2168                                              i->name, 0);
2169                if (error && error != -ENODATA)
2170                        return error;
2171                is->s.not_found = error;
2172        }
2173        return 0;
2174}
2175
2176int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2177                                struct ext4_xattr_info *i,
2178                                struct ext4_xattr_ibody_find *is)
2179{
2180        struct ext4_xattr_ibody_header *header;
2181        struct ext4_xattr_search *s = &is->s;
2182        int error;
2183
2184        if (EXT4_I(inode)->i_extra_isize == 0)
2185                return -ENOSPC;
2186        error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2187        if (error) {
2188                if (error == -ENOSPC &&
2189                    ext4_has_inline_data(inode)) {
2190                        error = ext4_try_to_evict_inline_data(handle, inode,
2191                                        EXT4_XATTR_LEN(strlen(i->name) +
2192                                        EXT4_XATTR_SIZE(i->value_len)));
2193                        if (error)
2194                                return error;
2195                        error = ext4_xattr_ibody_find(inode, i, is);
2196                        if (error)
2197                                return error;
2198                        error = ext4_xattr_set_entry(i, s, handle, inode,
2199                                                     false /* is_block */);
2200                }
2201                if (error)
2202                        return error;
2203        }
2204        header = IHDR(inode, ext4_raw_inode(&is->iloc));
2205        if (!IS_LAST_ENTRY(s->first)) {
2206                header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2207                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2208        } else {
2209                header->h_magic = cpu_to_le32(0);
2210                ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2211        }
2212        return 0;
2213}
2214
2215static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2216                                struct ext4_xattr_info *i,
2217                                struct ext4_xattr_ibody_find *is)
2218{
2219        struct ext4_xattr_ibody_header *header;
2220        struct ext4_xattr_search *s = &is->s;
2221        int error;
2222
2223        if (EXT4_I(inode)->i_extra_isize == 0)
2224                return -ENOSPC;
2225        error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2226        if (error)
2227                return error;
2228        header = IHDR(inode, ext4_raw_inode(&is->iloc));
2229        if (!IS_LAST_ENTRY(s->first)) {
2230                header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2231                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2232        } else {
2233                header->h_magic = cpu_to_le32(0);
2234                ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2235        }
2236        return 0;
2237}
2238
2239static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2240                                 struct ext4_xattr_info *i)
2241{
2242        void *value;
2243
2244        /* When e_value_inum is set the value is stored externally. */
2245        if (s->here->e_value_inum)
2246                return 0;
2247        if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2248                return 0;
2249        value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2250        return !memcmp(value, i->value, i->value_len);
2251}
2252
2253static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2254{
2255        struct buffer_head *bh;
2256        int error;
2257
2258        if (!EXT4_I(inode)->i_file_acl)
2259                return NULL;
2260        bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2261        if (!bh)
2262                return ERR_PTR(-EIO);
2263        error = ext4_xattr_check_block(inode, bh);
2264        if (error)
2265                return ERR_PTR(error);
2266        return bh;
2267}
2268
2269/*
2270 * ext4_xattr_set_handle()
2271 *
2272 * Create, replace or remove an extended attribute for this inode.  Value
2273 * is NULL to remove an existing extended attribute, and non-NULL to
2274 * either replace an existing extended attribute, or create a new extended
2275 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2276 * specify that an extended attribute must exist and must not exist
2277 * previous to the call, respectively.
2278 *
2279 * Returns 0, or a negative error number on failure.
2280 */
2281int
2282ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2283                      const char *name, const void *value, size_t value_len,
2284                      int flags)
2285{
2286        struct ext4_xattr_info i = {
2287                .name_index = name_index,
2288                .name = name,
2289                .value = value,
2290                .value_len = value_len,
2291                .in_inode = 0,
2292        };
2293        struct ext4_xattr_ibody_find is = {
2294                .s = { .not_found = -ENODATA, },
2295        };
2296        struct ext4_xattr_block_find bs = {
2297                .s = { .not_found = -ENODATA, },
2298        };
2299        int no_expand;
2300        int error;
2301
2302        if (!name)
2303                return -EINVAL;
2304        if (strlen(name) > 255)
2305                return -ERANGE;
2306
2307        ext4_write_lock_xattr(inode, &no_expand);
2308
2309        /* Check journal credits under write lock. */
2310        if (ext4_handle_valid(handle)) {
2311                struct buffer_head *bh;
2312                int credits;
2313
2314                bh = ext4_xattr_get_block(inode);
2315                if (IS_ERR(bh)) {
2316                        error = PTR_ERR(bh);
2317                        goto cleanup;
2318                }
2319
2320                credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2321                                                   value_len,
2322                                                   flags & XATTR_CREATE);
2323                brelse(bh);
2324
2325                if (!ext4_handle_has_enough_credits(handle, credits)) {
2326                        error = -ENOSPC;
2327                        goto cleanup;
2328                }
2329        }
2330
2331        error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2332        if (error)
2333                goto cleanup;
2334
2335        if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2336                struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2337                memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2338                ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2339        }
2340
2341        error = ext4_xattr_ibody_find(inode, &i, &is);
2342        if (error)
2343                goto cleanup;
2344        if (is.s.not_found)
2345                error = ext4_xattr_block_find(inode, &i, &bs);
2346        if (error)
2347                goto cleanup;
2348        if (is.s.not_found && bs.s.not_found) {
2349                error = -ENODATA;
2350                if (flags & XATTR_REPLACE)
2351                        goto cleanup;
2352                error = 0;
2353                if (!value)
2354                        goto cleanup;
2355        } else {
2356                error = -EEXIST;
2357                if (flags & XATTR_CREATE)
2358                        goto cleanup;
2359        }
2360
2361        if (!value) {
2362                if (!is.s.not_found)
2363                        error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2364                else if (!bs.s.not_found)
2365                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2366        } else {
2367                error = 0;
2368                /* Xattr value did not change? Save us some work and bail out */
2369                if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2370                        goto cleanup;
2371                if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2372                        goto cleanup;
2373
2374                if (ext4_has_feature_ea_inode(inode->i_sb) &&
2375                    (EXT4_XATTR_SIZE(i.value_len) >
2376                        EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2377                        i.in_inode = 1;
2378retry_inode:
2379                error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2380                if (!error && !bs.s.not_found) {
2381                        i.value = NULL;
2382                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2383                } else if (error == -ENOSPC) {
2384                        if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2385                                error = ext4_xattr_block_find(inode, &i, &bs);
2386                                if (error)
2387                                        goto cleanup;
2388                        }
2389                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2390                        if (!error && !is.s.not_found) {
2391                                i.value = NULL;
2392                                error = ext4_xattr_ibody_set(handle, inode, &i,
2393                                                             &is);
2394                        } else if (error == -ENOSPC) {
2395                                /*
2396                                 * Xattr does not fit in the block, store at
2397                                 * external inode if possible.
2398                                 */
2399                                if (ext4_has_feature_ea_inode(inode->i_sb) &&
2400                                    !i.in_inode) {
2401                                        i.in_inode = 1;
2402                                        goto retry_inode;
2403                                }
2404                        }
2405                }
2406        }
2407        if (!error) {
2408                ext4_xattr_update_super_block(handle, inode->i_sb);
2409                inode->i_ctime = current_time(inode);
2410                if (!value)
2411                        no_expand = 0;
2412                error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2413                /*
2414                 * The bh is consumed by ext4_mark_iloc_dirty, even with
2415                 * error != 0.
2416                 */
2417                is.iloc.bh = NULL;
2418                if (IS_SYNC(inode))
2419                        ext4_handle_sync(handle);
2420        }
2421
2422cleanup:
2423        brelse(is.iloc.bh);
2424        brelse(bs.bh);
2425        ext4_write_unlock_xattr(inode, &no_expand);
2426        return error;
2427}
2428
2429int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2430                           bool is_create, int *credits)
2431{
2432        struct buffer_head *bh;
2433        int err;
2434
2435        *credits = 0;
2436
2437        if (!EXT4_SB(inode->i_sb)->s_journal)
2438                return 0;
2439
2440        down_read(&EXT4_I(inode)->xattr_sem);
2441
2442        bh = ext4_xattr_get_block(inode);
2443        if (IS_ERR(bh)) {
2444                err = PTR_ERR(bh);
2445        } else {
2446                *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2447                                                    value_len, is_create);
2448                brelse(bh);
2449                err = 0;
2450        }
2451
2452        up_read(&EXT4_I(inode)->xattr_sem);
2453        return err;
2454}
2455
2456/*
2457 * ext4_xattr_set()
2458 *
2459 * Like ext4_xattr_set_handle, but start from an inode. This extended
2460 * attribute modification is a filesystem transaction by itself.
2461 *
2462 * Returns 0, or a negative error number on failure.
2463 */
2464int
2465ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2466               const void *value, size_t value_len, int flags)
2467{
2468        handle_t *handle;
2469        struct super_block *sb = inode->i_sb;
2470        int error, retries = 0;
2471        int credits;
2472
2473        error = dquot_initialize(inode);
2474        if (error)
2475                return error;
2476
2477retry:
2478        error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2479                                       &credits);
2480        if (error)
2481                return error;
2482
2483        handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2484        if (IS_ERR(handle)) {
2485                error = PTR_ERR(handle);
2486        } else {
2487                int error2;
2488
2489                error = ext4_xattr_set_handle(handle, inode, name_index, name,
2490                                              value, value_len, flags);
2491                error2 = ext4_journal_stop(handle);
2492                if (error == -ENOSPC &&
2493                    ext4_should_retry_alloc(sb, &retries))
2494                        goto retry;
2495                if (error == 0)
2496                        error = error2;
2497        }
2498
2499        return error;
2500}
2501
2502/*
2503 * Shift the EA entries in the inode to create space for the increased
2504 * i_extra_isize.
2505 */
2506static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2507                                     int value_offs_shift, void *to,
2508                                     void *from, size_t n)
2509{
2510        struct ext4_xattr_entry *last = entry;
2511        int new_offs;
2512
2513        /* We always shift xattr headers further thus offsets get lower */
2514        BUG_ON(value_offs_shift > 0);
2515
2516        /* Adjust the value offsets of the entries */
2517        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2518                if (!last->e_value_inum && last->e_value_size) {
2519                        new_offs = le16_to_cpu(last->e_value_offs) +
2520                                                        value_offs_shift;
2521                        last->e_value_offs = cpu_to_le16(new_offs);
2522                }
2523        }
2524        /* Shift the entries by n bytes */
2525        memmove(to, from, n);
2526}
2527
2528/*
2529 * Move xattr pointed to by 'entry' from inode into external xattr block
2530 */
2531static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2532                                    struct ext4_inode *raw_inode,
2533                                    struct ext4_xattr_entry *entry)
2534{
2535        struct ext4_xattr_ibody_find *is = NULL;
2536        struct ext4_xattr_block_find *bs = NULL;
2537        char *buffer = NULL, *b_entry_name = NULL;
2538        size_t value_size = le32_to_cpu(entry->e_value_size);
2539        struct ext4_xattr_info i = {
2540                .value = NULL,
2541                .value_len = 0,
2542                .name_index = entry->e_name_index,
2543                .in_inode = !!entry->e_value_inum,
2544        };
2545        struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2546        int error;
2547
2548        is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2549        bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2550        buffer = kmalloc(value_size, GFP_NOFS);
2551        b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2552        if (!is || !bs || !buffer || !b_entry_name) {
2553                error = -ENOMEM;
2554                goto out;
2555        }
2556
2557        is->s.not_found = -ENODATA;
2558        bs->s.not_found = -ENODATA;
2559        is->iloc.bh = NULL;
2560        bs->bh = NULL;
2561
2562        /* Save the entry name and the entry value */
2563        if (entry->e_value_inum) {
2564                error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2565                if (error)
2566                        goto out;
2567        } else {
2568                size_t value_offs = le16_to_cpu(entry->e_value_offs);
2569                memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2570        }
2571
2572        memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2573        b_entry_name[entry->e_name_len] = '\0';
2574        i.name = b_entry_name;
2575
2576        error = ext4_get_inode_loc(inode, &is->iloc);
2577        if (error)
2578                goto out;
2579
2580        error = ext4_xattr_ibody_find(inode, &i, is);
2581        if (error)
2582                goto out;
2583
2584        /* Remove the chosen entry from the inode */
2585        error = ext4_xattr_ibody_set(handle, inode, &i, is);
2586        if (error)
2587                goto out;
2588
2589        i.value = buffer;
2590        i.value_len = value_size;
2591        error = ext4_xattr_block_find(inode, &i, bs);
2592        if (error)
2593                goto out;
2594
2595        /* Add entry which was removed from the inode into the block */
2596        error = ext4_xattr_block_set(handle, inode, &i, bs);
2597        if (error)
2598                goto out;
2599        error = 0;
2600out:
2601        kfree(b_entry_name);
2602        kfree(buffer);
2603        if (is)
2604                brelse(is->iloc.bh);
2605        kfree(is);
2606        kfree(bs);
2607
2608        return error;
2609}
2610
2611static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2612                                       struct ext4_inode *raw_inode,
2613                                       int isize_diff, size_t ifree,
2614                                       size_t bfree, int *total_ino)
2615{
2616        struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2617        struct ext4_xattr_entry *small_entry;
2618        struct ext4_xattr_entry *entry;
2619        struct ext4_xattr_entry *last;
2620        unsigned int entry_size;        /* EA entry size */
2621        unsigned int total_size;        /* EA entry size + value size */
2622        unsigned int min_total_size;
2623        int error;
2624
2625        while (isize_diff > ifree) {
2626                entry = NULL;
2627                small_entry = NULL;
2628                min_total_size = ~0U;
2629                last = IFIRST(header);
2630                /* Find the entry best suited to be pushed into EA block */
2631                for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2632                        total_size = EXT4_XATTR_LEN(last->e_name_len);
2633                        if (!last->e_value_inum)
2634                                total_size += EXT4_XATTR_SIZE(
2635                                               le32_to_cpu(last->e_value_size));
2636                        if (total_size <= bfree &&
2637                            total_size < min_total_size) {
2638                                if (total_size + ifree < isize_diff) {
2639                                        small_entry = last;
2640                                } else {
2641                                        entry = last;
2642                                        min_total_size = total_size;
2643                                }
2644                        }
2645                }
2646
2647                if (entry == NULL) {
2648                        if (small_entry == NULL)
2649                                return -ENOSPC;
2650                        entry = small_entry;
2651                }
2652
2653                entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2654                total_size = entry_size;
2655                if (!entry->e_value_inum)
2656                        total_size += EXT4_XATTR_SIZE(
2657                                              le32_to_cpu(entry->e_value_size));
2658                error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2659                                                 entry);
2660                if (error)
2661                        return error;
2662
2663                *total_ino -= entry_size;
2664                ifree += total_size;
2665                bfree -= total_size;
2666        }
2667
2668        return 0;
2669}
2670
2671/*
2672 * Expand an inode by new_extra_isize bytes when EAs are present.
2673 * Returns 0 on success or negative error number on failure.
2674 */
2675int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2676                               struct ext4_inode *raw_inode, handle_t *handle)
2677{
2678        struct ext4_xattr_ibody_header *header;
2679        struct buffer_head *bh;
2680        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2681        static unsigned int mnt_count;
2682        size_t min_offs;
2683        size_t ifree, bfree;
2684        int total_ino;
2685        void *base, *end;
2686        int error = 0, tried_min_extra_isize = 0;
2687        int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2688        int isize_diff; /* How much do we need to grow i_extra_isize */
2689
2690retry:
2691        isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2692        if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2693                return 0;
2694
2695        header = IHDR(inode, raw_inode);
2696
2697        /*
2698         * Check if enough free space is available in the inode to shift the
2699         * entries ahead by new_extra_isize.
2700         */
2701
2702        base = IFIRST(header);
2703        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2704        min_offs = end - base;
2705        total_ino = sizeof(struct ext4_xattr_ibody_header);
2706
2707        error = xattr_check_inode(inode, header, end);
2708        if (error)
2709                goto cleanup;
2710
2711        ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2712        if (ifree >= isize_diff)
2713                goto shift;
2714
2715        /*
2716         * Enough free space isn't available in the inode, check if
2717         * EA block can hold new_extra_isize bytes.
2718         */
2719        if (EXT4_I(inode)->i_file_acl) {
2720                bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2721                error = -EIO;
2722                if (!bh)
2723                        goto cleanup;
2724                if (ext4_xattr_check_block(inode, bh)) {
2725                        EXT4_ERROR_INODE(inode, "bad block %llu",
2726                                         EXT4_I(inode)->i_file_acl);
2727                        error = -EFSCORRUPTED;
2728                        brelse(bh);
2729                        goto cleanup;
2730                }
2731                base = BHDR(bh);
2732                end = bh->b_data + bh->b_size;
2733                min_offs = end - base;
2734                bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2735                                              NULL);
2736                brelse(bh);
2737                if (bfree + ifree < isize_diff) {
2738                        if (!tried_min_extra_isize && s_min_extra_isize) {
2739                                tried_min_extra_isize++;
2740                                new_extra_isize = s_min_extra_isize;
2741                                goto retry;
2742                        }
2743                        error = -ENOSPC;
2744                        goto cleanup;
2745                }
2746        } else {
2747                bfree = inode->i_sb->s_blocksize;
2748        }
2749
2750        error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2751                                            isize_diff, ifree, bfree,
2752                                            &total_ino);
2753        if (error) {
2754                if (error == -ENOSPC && !tried_min_extra_isize &&
2755                    s_min_extra_isize) {
2756                        tried_min_extra_isize++;
2757                        new_extra_isize = s_min_extra_isize;
2758                        goto retry;
2759                }
2760                goto cleanup;
2761        }
2762shift:
2763        /* Adjust the offsets and shift the remaining entries ahead */
2764        ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2765                        - new_extra_isize, (void *)raw_inode +
2766                        EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2767                        (void *)header, total_ino);
2768        EXT4_I(inode)->i_extra_isize = new_extra_isize;
2769
2770cleanup:
2771        if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2772                ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2773                             inode->i_ino);
2774                mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2775        }
2776        return error;
2777}
2778
2779#define EIA_INCR 16 /* must be 2^n */
2780#define EIA_MASK (EIA_INCR - 1)
2781
2782/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2783 * If @ea_inode_array is new or full it will be grown and the old
2784 * contents copied over.
2785 */
2786static int
2787ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2788                        struct inode *inode)
2789{
2790        if (*ea_inode_array == NULL) {
2791                /*
2792                 * Start with 15 inodes, so it fits into a power-of-two size.
2793                 * If *ea_inode_array is NULL, this is essentially offsetof()
2794                 */
2795                (*ea_inode_array) =
2796                        kmalloc(offsetof(struct ext4_xattr_inode_array,
2797                                         inodes[EIA_MASK]),
2798                                GFP_NOFS);
2799                if (*ea_inode_array == NULL)
2800                        return -ENOMEM;
2801                (*ea_inode_array)->count = 0;
2802        } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2803                /* expand the array once all 15 + n * 16 slots are full */
2804                struct ext4_xattr_inode_array *new_array = NULL;
2805                int count = (*ea_inode_array)->count;
2806
2807                /* if new_array is NULL, this is essentially offsetof() */
2808                new_array = kmalloc(
2809                                offsetof(struct ext4_xattr_inode_array,
2810                                         inodes[count + EIA_INCR]),
2811                                GFP_NOFS);
2812                if (new_array == NULL)
2813                        return -ENOMEM;
2814                memcpy(new_array, *ea_inode_array,
2815                       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2816                kfree(*ea_inode_array);
2817                *ea_inode_array = new_array;
2818        }
2819        (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2820        return 0;
2821}
2822
2823/*
2824 * ext4_xattr_delete_inode()
2825 *
2826 * Free extended attribute resources associated with this inode. Traverse
2827 * all entries and decrement reference on any xattr inodes associated with this
2828 * inode. This is called immediately before an inode is freed. We have exclusive
2829 * access to the inode. If an orphan inode is deleted it will also release its
2830 * references on xattr block and xattr inodes.
2831 */
2832int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2833                            struct ext4_xattr_inode_array **ea_inode_array,
2834                            int extra_credits)
2835{
2836        struct buffer_head *bh = NULL;
2837        struct ext4_xattr_ibody_header *header;
2838        struct ext4_iloc iloc = { .bh = NULL };
2839        struct ext4_xattr_entry *entry;
2840        struct inode *ea_inode;
2841        int error;
2842
2843        error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2844                                          NULL /* bh */,
2845                                          false /* dirty */,
2846                                          false /* block_csum */);
2847        if (error) {
2848                EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2849                goto cleanup;
2850        }
2851
2852        if (ext4_has_feature_ea_inode(inode->i_sb) &&
2853            ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2854
2855                error = ext4_get_inode_loc(inode, &iloc);
2856                if (error) {
2857                        EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2858                        goto cleanup;
2859                }
2860
2861                error = ext4_journal_get_write_access(handle, iloc.bh);
2862                if (error) {
2863                        EXT4_ERROR_INODE(inode, "write access (error %d)",
2864                                         error);
2865                        goto cleanup;
2866                }
2867
2868                header = IHDR(inode, ext4_raw_inode(&iloc));
2869                if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2870                        ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2871                                                     IFIRST(header),
2872                                                     false /* block_csum */,
2873                                                     ea_inode_array,
2874                                                     extra_credits,
2875                                                     false /* skip_quota */);
2876        }
2877
2878        if (EXT4_I(inode)->i_file_acl) {
2879                bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2880                if (!bh) {
2881                        EXT4_ERROR_INODE(inode, "block %llu read error",
2882                                         EXT4_I(inode)->i_file_acl);
2883                        error = -EIO;
2884                        goto cleanup;
2885                }
2886                error = ext4_xattr_check_block(inode, bh);
2887                if (error) {
2888                        EXT4_ERROR_INODE(inode, "bad block %llu (error %d)",
2889                                         EXT4_I(inode)->i_file_acl, error);
2890                        goto cleanup;
2891                }
2892
2893                if (ext4_has_feature_ea_inode(inode->i_sb)) {
2894                        for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2895                             entry = EXT4_XATTR_NEXT(entry)) {
2896                                if (!entry->e_value_inum)
2897                                        continue;
2898                                error = ext4_xattr_inode_iget(inode,
2899                                              le32_to_cpu(entry->e_value_inum),
2900                                              le32_to_cpu(entry->e_hash),
2901                                              &ea_inode);
2902                                if (error)
2903                                        continue;
2904                                ext4_xattr_inode_free_quota(inode, ea_inode,
2905                                              le32_to_cpu(entry->e_value_size));
2906                                iput(ea_inode);
2907                        }
2908
2909                }
2910
2911                ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2912                                         extra_credits);
2913                /*
2914                 * Update i_file_acl value in the same transaction that releases
2915                 * block.
2916                 */
2917                EXT4_I(inode)->i_file_acl = 0;
2918                error = ext4_mark_inode_dirty(handle, inode);
2919                if (error) {
2920                        EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2921                                         error);
2922                        goto cleanup;
2923                }
2924        }
2925        error = 0;
2926cleanup:
2927        brelse(iloc.bh);
2928        brelse(bh);
2929        return error;
2930}
2931
2932void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2933{
2934        int idx;
2935
2936        if (ea_inode_array == NULL)
2937                return;
2938
2939        for (idx = 0; idx < ea_inode_array->count; ++idx)
2940                iput(ea_inode_array->inodes[idx]);
2941        kfree(ea_inode_array);
2942}
2943
2944/*
2945 * ext4_xattr_block_cache_insert()
2946 *
2947 * Create a new entry in the extended attribute block cache, and insert
2948 * it unless such an entry is already in the cache.
2949 *
2950 * Returns 0, or a negative error number on failure.
2951 */
2952static void
2953ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2954                              struct buffer_head *bh)
2955{
2956        struct ext4_xattr_header *header = BHDR(bh);
2957        __u32 hash = le32_to_cpu(header->h_hash);
2958        int reusable = le32_to_cpu(header->h_refcount) <
2959                       EXT4_XATTR_REFCOUNT_MAX;
2960        int error;
2961
2962        if (!ea_block_cache)
2963                return;
2964        error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2965                                      bh->b_blocknr, reusable);
2966        if (error) {
2967                if (error == -EBUSY)
2968                        ea_bdebug(bh, "already in cache");
2969        } else
2970                ea_bdebug(bh, "inserting [%x]", (int)hash);
2971}
2972
2973/*
2974 * ext4_xattr_cmp()
2975 *
2976 * Compare two extended attribute blocks for equality.
2977 *
2978 * Returns 0 if the blocks are equal, 1 if they differ, and
2979 * a negative error number on errors.
2980 */
2981static int
2982ext4_xattr_cmp(struct ext4_xattr_header *header1,
2983               struct ext4_xattr_header *header2)
2984{
2985        struct ext4_xattr_entry *entry1, *entry2;
2986
2987        entry1 = ENTRY(header1+1);
2988        entry2 = ENTRY(header2+1);
2989        while (!IS_LAST_ENTRY(entry1)) {
2990                if (IS_LAST_ENTRY(entry2))
2991                        return 1;
2992                if (entry1->e_hash != entry2->e_hash ||
2993                    entry1->e_name_index != entry2->e_name_index ||
2994                    entry1->e_name_len != entry2->e_name_len ||
2995                    entry1->e_value_size != entry2->e_value_size ||
2996                    entry1->e_value_inum != entry2->e_value_inum ||
2997                    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2998                        return 1;
2999                if (!entry1->e_value_inum &&
3000                    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3001                           (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3002                           le32_to_cpu(entry1->e_value_size)))
3003                        return 1;
3004
3005                entry1 = EXT4_XATTR_NEXT(entry1);
3006                entry2 = EXT4_XATTR_NEXT(entry2);
3007        }
3008        if (!IS_LAST_ENTRY(entry2))
3009                return 1;
3010        return 0;
3011}
3012
3013/*
3014 * ext4_xattr_block_cache_find()
3015 *
3016 * Find an identical extended attribute block.
3017 *
3018 * Returns a pointer to the block found, or NULL if such a block was
3019 * not found or an error occurred.
3020 */
3021static struct buffer_head *
3022ext4_xattr_block_cache_find(struct inode *inode,
3023                            struct ext4_xattr_header *header,
3024                            struct mb_cache_entry **pce)
3025{
3026        __u32 hash = le32_to_cpu(header->h_hash);
3027        struct mb_cache_entry *ce;
3028        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3029
3030        if (!ea_block_cache)
3031                return NULL;
3032        if (!header->h_hash)
3033                return NULL;  /* never share */
3034        ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3035        ce = mb_cache_entry_find_first(ea_block_cache, hash);
3036        while (ce) {
3037                struct buffer_head *bh;
3038
3039                bh = sb_bread(inode->i_sb, ce->e_value);
3040                if (!bh) {
3041                        EXT4_ERROR_INODE(inode, "block %lu read error",
3042                                         (unsigned long)ce->e_value);
3043                } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3044                        *pce = ce;
3045                        return bh;
3046                }
3047                brelse(bh);
3048                ce = mb_cache_entry_find_next(ea_block_cache, ce);
3049        }
3050        return NULL;
3051}
3052
3053#define NAME_HASH_SHIFT 5
3054#define VALUE_HASH_SHIFT 16
3055
3056/*
3057 * ext4_xattr_hash_entry()
3058 *
3059 * Compute the hash of an extended attribute.
3060 */
3061static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3062                                    size_t value_count)
3063{
3064        __u32 hash = 0;
3065
3066        while (name_len--) {
3067                hash = (hash << NAME_HASH_SHIFT) ^
3068                       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3069                       *name++;
3070        }
3071        while (value_count--) {
3072                hash = (hash << VALUE_HASH_SHIFT) ^
3073                       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3074                       le32_to_cpu(*value++);
3075        }
3076        return cpu_to_le32(hash);
3077}
3078
3079#undef NAME_HASH_SHIFT
3080#undef VALUE_HASH_SHIFT
3081
3082#define BLOCK_HASH_SHIFT 16
3083
3084/*
3085 * ext4_xattr_rehash()
3086 *
3087 * Re-compute the extended attribute hash value after an entry has changed.
3088 */
3089static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3090{
3091        struct ext4_xattr_entry *here;
3092        __u32 hash = 0;
3093
3094        here = ENTRY(header+1);
3095        while (!IS_LAST_ENTRY(here)) {
3096                if (!here->e_hash) {
3097                        /* Block is not shared if an entry's hash value == 0 */
3098                        hash = 0;
3099                        break;
3100                }
3101                hash = (hash << BLOCK_HASH_SHIFT) ^
3102                       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3103                       le32_to_cpu(here->e_hash);
3104                here = EXT4_XATTR_NEXT(here);
3105        }
3106        header->h_hash = cpu_to_le32(hash);
3107}
3108
3109#undef BLOCK_HASH_SHIFT
3110
3111#define HASH_BUCKET_BITS        10
3112
3113struct mb_cache *
3114ext4_xattr_create_cache(void)
3115{
3116        return mb_cache_create(HASH_BUCKET_BITS);
3117}
3118
3119void ext4_xattr_destroy_cache(struct mb_cache *cache)
3120{
3121        if (cache)
3122                mb_cache_destroy(cache);
3123}
3124
3125