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        [EXT4_XATTR_INDEX_HURD]              = &ext4_xattr_hurd_handler,
  97};
  98
  99const struct xattr_handler *ext4_xattr_handlers[] = {
 100        &ext4_xattr_user_handler,
 101        &ext4_xattr_trusted_handler,
 102#ifdef CONFIG_EXT4_FS_POSIX_ACL
 103        &posix_acl_access_xattr_handler,
 104        &posix_acl_default_xattr_handler,
 105#endif
 106#ifdef CONFIG_EXT4_FS_SECURITY
 107        &ext4_xattr_security_handler,
 108#endif
 109        &ext4_xattr_hurd_handler,
 110        NULL
 111};
 112
 113#define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 114                                inode->i_sb->s_fs_info)->s_ea_block_cache)
 115
 116#define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 117                                inode->i_sb->s_fs_info)->s_ea_inode_cache)
 118
 119static int
 120ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 121                        struct inode *inode);
 122
 123#ifdef CONFIG_LOCKDEP
 124void ext4_xattr_inode_set_class(struct inode *ea_inode)
 125{
 126        lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 127}
 128#endif
 129
 130static __le32 ext4_xattr_block_csum(struct inode *inode,
 131                                    sector_t block_nr,
 132                                    struct ext4_xattr_header *hdr)
 133{
 134        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 135        __u32 csum;
 136        __le64 dsk_block_nr = cpu_to_le64(block_nr);
 137        __u32 dummy_csum = 0;
 138        int offset = offsetof(struct ext4_xattr_header, h_checksum);
 139
 140        csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 141                           sizeof(dsk_block_nr));
 142        csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 143        csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 144        offset += sizeof(dummy_csum);
 145        csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 146                           EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 147
 148        return cpu_to_le32(csum);
 149}
 150
 151static int ext4_xattr_block_csum_verify(struct inode *inode,
 152                                        struct buffer_head *bh)
 153{
 154        struct ext4_xattr_header *hdr = BHDR(bh);
 155        int ret = 1;
 156
 157        if (ext4_has_metadata_csum(inode->i_sb)) {
 158                lock_buffer(bh);
 159                ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 160                                                        bh->b_blocknr, hdr));
 161                unlock_buffer(bh);
 162        }
 163        return ret;
 164}
 165
 166static void ext4_xattr_block_csum_set(struct inode *inode,
 167                                      struct buffer_head *bh)
 168{
 169        if (ext4_has_metadata_csum(inode->i_sb))
 170                BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 171                                                bh->b_blocknr, BHDR(bh));
 172}
 173
 174static inline const struct xattr_handler *
 175ext4_xattr_handler(int name_index)
 176{
 177        const struct xattr_handler *handler = NULL;
 178
 179        if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 180                handler = ext4_xattr_handler_map[name_index];
 181        return handler;
 182}
 183
 184static int
 185ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 186                         void *value_start)
 187{
 188        struct ext4_xattr_entry *e = entry;
 189
 190        /* Find the end of the names list */
 191        while (!IS_LAST_ENTRY(e)) {
 192                struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 193                if ((void *)next >= end)
 194                        return -EFSCORRUPTED;
 195                if (strnlen(e->e_name, e->e_name_len) != e->e_name_len)
 196                        return -EFSCORRUPTED;
 197                e = next;
 198        }
 199
 200        /* Check the values */
 201        while (!IS_LAST_ENTRY(entry)) {
 202                u32 size = le32_to_cpu(entry->e_value_size);
 203
 204                if (size > EXT4_XATTR_SIZE_MAX)
 205                        return -EFSCORRUPTED;
 206
 207                if (size != 0 && entry->e_value_inum == 0) {
 208                        u16 offs = le16_to_cpu(entry->e_value_offs);
 209                        void *value;
 210
 211                        /*
 212                         * The value cannot overlap the names, and the value
 213                         * with padding cannot extend beyond 'end'.  Check both
 214                         * the padded and unpadded sizes, since the size may
 215                         * overflow to 0 when adding padding.
 216                         */
 217                        if (offs > end - value_start)
 218                                return -EFSCORRUPTED;
 219                        value = value_start + offs;
 220                        if (value < (void *)e + sizeof(u32) ||
 221                            size > end - value ||
 222                            EXT4_XATTR_SIZE(size) > end - value)
 223                                return -EFSCORRUPTED;
 224                }
 225                entry = EXT4_XATTR_NEXT(entry);
 226        }
 227
 228        return 0;
 229}
 230
 231static inline int
 232__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 233                         const char *function, unsigned int line)
 234{
 235        int error = -EFSCORRUPTED;
 236
 237        if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 238            BHDR(bh)->h_blocks != cpu_to_le32(1))
 239                goto errout;
 240        if (buffer_verified(bh))
 241                return 0;
 242
 243        error = -EFSBADCRC;
 244        if (!ext4_xattr_block_csum_verify(inode, bh))
 245                goto errout;
 246        error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 247                                         bh->b_data);
 248errout:
 249        if (error)
 250                __ext4_error_inode(inode, function, line, 0, -error,
 251                                   "corrupted xattr block %llu",
 252                                   (unsigned long long) bh->b_blocknr);
 253        else
 254                set_buffer_verified(bh);
 255        return error;
 256}
 257
 258#define ext4_xattr_check_block(inode, bh) \
 259        __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 260
 261
 262static int
 263__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 264                         void *end, const char *function, unsigned int line)
 265{
 266        int error = -EFSCORRUPTED;
 267
 268        if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 269            (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 270                goto errout;
 271        error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 272errout:
 273        if (error)
 274                __ext4_error_inode(inode, function, line, 0, -error,
 275                                   "corrupted in-inode xattr");
 276        return error;
 277}
 278
 279#define xattr_check_inode(inode, header, end) \
 280        __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 281
 282static int
 283xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 284                 void *end, int name_index, const char *name, int sorted)
 285{
 286        struct ext4_xattr_entry *entry, *next;
 287        size_t name_len;
 288        int cmp = 1;
 289
 290        if (name == NULL)
 291                return -EINVAL;
 292        name_len = strlen(name);
 293        for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 294                next = EXT4_XATTR_NEXT(entry);
 295                if ((void *) next >= end) {
 296                        EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 297                        return -EFSCORRUPTED;
 298                }
 299                cmp = name_index - entry->e_name_index;
 300                if (!cmp)
 301                        cmp = name_len - entry->e_name_len;
 302                if (!cmp)
 303                        cmp = memcmp(name, entry->e_name, name_len);
 304                if (cmp <= 0 && (sorted || cmp == 0))
 305                        break;
 306        }
 307        *pentry = entry;
 308        return cmp ? -ENODATA : 0;
 309}
 310
 311static u32
 312ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 313{
 314        return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 315}
 316
 317static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 318{
 319        return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 320                (u32) inode_peek_iversion_raw(ea_inode);
 321}
 322
 323static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 324{
 325        ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 326        inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 327}
 328
 329static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 330{
 331        return (u32)ea_inode->i_atime.tv_sec;
 332}
 333
 334static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 335{
 336        ea_inode->i_atime.tv_sec = hash;
 337}
 338
 339/*
 340 * Read the EA value from an inode.
 341 */
 342static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 343{
 344        int blocksize = 1 << ea_inode->i_blkbits;
 345        int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 346        int tail_size = (size % blocksize) ?: blocksize;
 347        struct buffer_head *bhs_inline[8];
 348        struct buffer_head **bhs = bhs_inline;
 349        int i, ret;
 350
 351        if (bh_count > ARRAY_SIZE(bhs_inline)) {
 352                bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 353                if (!bhs)
 354                        return -ENOMEM;
 355        }
 356
 357        ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 358                               true /* wait */, bhs);
 359        if (ret)
 360                goto free_bhs;
 361
 362        for (i = 0; i < bh_count; i++) {
 363                /* There shouldn't be any holes in ea_inode. */
 364                if (!bhs[i]) {
 365                        ret = -EFSCORRUPTED;
 366                        goto put_bhs;
 367                }
 368                memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 369                       i < bh_count - 1 ? blocksize : tail_size);
 370        }
 371        ret = 0;
 372put_bhs:
 373        for (i = 0; i < bh_count; i++)
 374                brelse(bhs[i]);
 375free_bhs:
 376        if (bhs != bhs_inline)
 377                kfree(bhs);
 378        return ret;
 379}
 380
 381#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 382
 383static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 384                                 u32 ea_inode_hash, struct inode **ea_inode)
 385{
 386        struct inode *inode;
 387        int err;
 388
 389        inode = ext4_iget(parent->i_sb, ea_ino, EXT4_IGET_NORMAL);
 390        if (IS_ERR(inode)) {
 391                err = PTR_ERR(inode);
 392                ext4_error(parent->i_sb,
 393                           "error while reading EA inode %lu err=%d", ea_ino,
 394                           err);
 395                return err;
 396        }
 397
 398        if (is_bad_inode(inode)) {
 399                ext4_error(parent->i_sb,
 400                           "error while reading EA inode %lu is_bad_inode",
 401                           ea_ino);
 402                err = -EIO;
 403                goto error;
 404        }
 405
 406        if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 407                ext4_error(parent->i_sb,
 408                           "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 409                            ea_ino);
 410                err = -EINVAL;
 411                goto error;
 412        }
 413
 414        ext4_xattr_inode_set_class(inode);
 415
 416        /*
 417         * Check whether this is an old Lustre-style xattr inode. Lustre
 418         * implementation does not have hash validation, rather it has a
 419         * backpointer from ea_inode to the parent inode.
 420         */
 421        if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 422            EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 423            inode->i_generation == parent->i_generation) {
 424                ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 425                ext4_xattr_inode_set_ref(inode, 1);
 426        } else {
 427                inode_lock(inode);
 428                inode->i_flags |= S_NOQUOTA;
 429                inode_unlock(inode);
 430        }
 431
 432        *ea_inode = inode;
 433        return 0;
 434error:
 435        iput(inode);
 436        return err;
 437}
 438
 439static int
 440ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 441                               struct ext4_xattr_entry *entry, void *buffer,
 442                               size_t size)
 443{
 444        u32 hash;
 445
 446        /* Verify stored hash matches calculated hash. */
 447        hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 448        if (hash != ext4_xattr_inode_get_hash(ea_inode))
 449                return -EFSCORRUPTED;
 450
 451        if (entry) {
 452                __le32 e_hash, tmp_data;
 453
 454                /* Verify entry hash. */
 455                tmp_data = cpu_to_le32(hash);
 456                e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 457                                               &tmp_data, 1);
 458                if (e_hash != entry->e_hash)
 459                        return -EFSCORRUPTED;
 460        }
 461        return 0;
 462}
 463
 464/*
 465 * Read xattr value from the EA inode.
 466 */
 467static int
 468ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 469                     void *buffer, size_t size)
 470{
 471        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 472        struct inode *ea_inode;
 473        int err;
 474
 475        err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 476                                    le32_to_cpu(entry->e_hash), &ea_inode);
 477        if (err) {
 478                ea_inode = NULL;
 479                goto out;
 480        }
 481
 482        if (i_size_read(ea_inode) != size) {
 483                ext4_warning_inode(ea_inode,
 484                                   "ea_inode file size=%llu entry size=%zu",
 485                                   i_size_read(ea_inode), size);
 486                err = -EFSCORRUPTED;
 487                goto out;
 488        }
 489
 490        err = ext4_xattr_inode_read(ea_inode, buffer, size);
 491        if (err)
 492                goto out;
 493
 494        if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 495                err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 496                                                     size);
 497                if (err) {
 498                        ext4_warning_inode(ea_inode,
 499                                           "EA inode hash validation failed");
 500                        goto out;
 501                }
 502
 503                if (ea_inode_cache)
 504                        mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 505                                        ext4_xattr_inode_get_hash(ea_inode),
 506                                        ea_inode->i_ino, true /* reusable */);
 507        }
 508out:
 509        iput(ea_inode);
 510        return err;
 511}
 512
 513static int
 514ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 515                     void *buffer, size_t buffer_size)
 516{
 517        struct buffer_head *bh = NULL;
 518        struct ext4_xattr_entry *entry;
 519        size_t size;
 520        void *end;
 521        int error;
 522        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 523
 524        ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 525                  name_index, name, buffer, (long)buffer_size);
 526
 527        if (!EXT4_I(inode)->i_file_acl)
 528                return -ENODATA;
 529        ea_idebug(inode, "reading block %llu",
 530                  (unsigned long long)EXT4_I(inode)->i_file_acl);
 531        bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 532        if (IS_ERR(bh))
 533                return PTR_ERR(bh);
 534        ea_bdebug(bh, "b_count=%d, refcount=%d",
 535                atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 536        error = ext4_xattr_check_block(inode, bh);
 537        if (error)
 538                goto cleanup;
 539        ext4_xattr_block_cache_insert(ea_block_cache, bh);
 540        entry = BFIRST(bh);
 541        end = bh->b_data + bh->b_size;
 542        error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 543        if (error)
 544                goto cleanup;
 545        size = le32_to_cpu(entry->e_value_size);
 546        error = -ERANGE;
 547        if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 548                goto cleanup;
 549        if (buffer) {
 550                if (size > buffer_size)
 551                        goto cleanup;
 552                if (entry->e_value_inum) {
 553                        error = ext4_xattr_inode_get(inode, entry, buffer,
 554                                                     size);
 555                        if (error)
 556                                goto cleanup;
 557                } else {
 558                        u16 offset = le16_to_cpu(entry->e_value_offs);
 559                        void *p = bh->b_data + offset;
 560
 561                        if (unlikely(p + size > end))
 562                                goto cleanup;
 563                        memcpy(buffer, p, size);
 564                }
 565        }
 566        error = size;
 567
 568cleanup:
 569        brelse(bh);
 570        return error;
 571}
 572
 573int
 574ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 575                     void *buffer, size_t buffer_size)
 576{
 577        struct ext4_xattr_ibody_header *header;
 578        struct ext4_xattr_entry *entry;
 579        struct ext4_inode *raw_inode;
 580        struct ext4_iloc iloc;
 581        size_t size;
 582        void *end;
 583        int error;
 584
 585        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 586                return -ENODATA;
 587        error = ext4_get_inode_loc(inode, &iloc);
 588        if (error)
 589                return error;
 590        raw_inode = ext4_raw_inode(&iloc);
 591        header = IHDR(inode, raw_inode);
 592        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 593        error = xattr_check_inode(inode, header, end);
 594        if (error)
 595                goto cleanup;
 596        entry = IFIRST(header);
 597        error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 598        if (error)
 599                goto cleanup;
 600        size = le32_to_cpu(entry->e_value_size);
 601        error = -ERANGE;
 602        if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 603                goto cleanup;
 604        if (buffer) {
 605                if (size > buffer_size)
 606                        goto cleanup;
 607                if (entry->e_value_inum) {
 608                        error = ext4_xattr_inode_get(inode, entry, buffer,
 609                                                     size);
 610                        if (error)
 611                                goto cleanup;
 612                } else {
 613                        u16 offset = le16_to_cpu(entry->e_value_offs);
 614                        void *p = (void *)IFIRST(header) + offset;
 615
 616                        if (unlikely(p + size > end))
 617                                goto cleanup;
 618                        memcpy(buffer, p, size);
 619                }
 620        }
 621        error = size;
 622
 623cleanup:
 624        brelse(iloc.bh);
 625        return error;
 626}
 627
 628/*
 629 * ext4_xattr_get()
 630 *
 631 * Copy an extended attribute into the buffer
 632 * provided, or compute the buffer size required.
 633 * Buffer is NULL to compute the size of the buffer required.
 634 *
 635 * Returns a negative error number on failure, or the number of bytes
 636 * used / required on success.
 637 */
 638int
 639ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 640               void *buffer, size_t buffer_size)
 641{
 642        int error;
 643
 644        if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 645                return -EIO;
 646
 647        if (strlen(name) > 255)
 648                return -ERANGE;
 649
 650        down_read(&EXT4_I(inode)->xattr_sem);
 651        error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 652                                     buffer_size);
 653        if (error == -ENODATA)
 654                error = ext4_xattr_block_get(inode, name_index, name, buffer,
 655                                             buffer_size);
 656        up_read(&EXT4_I(inode)->xattr_sem);
 657        return error;
 658}
 659
 660static int
 661ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 662                        char *buffer, size_t buffer_size)
 663{
 664        size_t rest = buffer_size;
 665
 666        for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 667                const struct xattr_handler *handler =
 668                        ext4_xattr_handler(entry->e_name_index);
 669
 670                if (handler && (!handler->list || handler->list(dentry))) {
 671                        const char *prefix = handler->prefix ?: handler->name;
 672                        size_t prefix_len = strlen(prefix);
 673                        size_t size = prefix_len + entry->e_name_len + 1;
 674
 675                        if (buffer) {
 676                                if (size > rest)
 677                                        return -ERANGE;
 678                                memcpy(buffer, prefix, prefix_len);
 679                                buffer += prefix_len;
 680                                memcpy(buffer, entry->e_name, entry->e_name_len);
 681                                buffer += entry->e_name_len;
 682                                *buffer++ = 0;
 683                        }
 684                        rest -= size;
 685                }
 686        }
 687        return buffer_size - rest;  /* total size */
 688}
 689
 690static int
 691ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 692{
 693        struct inode *inode = d_inode(dentry);
 694        struct buffer_head *bh = NULL;
 695        int error;
 696
 697        ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 698                  buffer, (long)buffer_size);
 699
 700        if (!EXT4_I(inode)->i_file_acl)
 701                return 0;
 702        ea_idebug(inode, "reading block %llu",
 703                  (unsigned long long)EXT4_I(inode)->i_file_acl);
 704        bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 705        if (IS_ERR(bh))
 706                return PTR_ERR(bh);
 707        ea_bdebug(bh, "b_count=%d, refcount=%d",
 708                atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 709        error = ext4_xattr_check_block(inode, bh);
 710        if (error)
 711                goto cleanup;
 712        ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 713        error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer,
 714                                        buffer_size);
 715cleanup:
 716        brelse(bh);
 717        return error;
 718}
 719
 720static int
 721ext4_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 722{
 723        struct inode *inode = d_inode(dentry);
 724        struct ext4_xattr_ibody_header *header;
 725        struct ext4_inode *raw_inode;
 726        struct ext4_iloc iloc;
 727        void *end;
 728        int error;
 729
 730        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 731                return 0;
 732        error = ext4_get_inode_loc(inode, &iloc);
 733        if (error)
 734                return error;
 735        raw_inode = ext4_raw_inode(&iloc);
 736        header = IHDR(inode, raw_inode);
 737        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 738        error = xattr_check_inode(inode, header, end);
 739        if (error)
 740                goto cleanup;
 741        error = ext4_xattr_list_entries(dentry, IFIRST(header),
 742                                        buffer, buffer_size);
 743
 744cleanup:
 745        brelse(iloc.bh);
 746        return error;
 747}
 748
 749/*
 750 * Inode operation listxattr()
 751 *
 752 * d_inode(dentry)->i_rwsem: don't care
 753 *
 754 * Copy a list of attribute names into the buffer
 755 * provided, or compute the buffer size required.
 756 * Buffer is NULL to compute the size of the buffer required.
 757 *
 758 * Returns a negative error number on failure, or the number of bytes
 759 * used / required on success.
 760 */
 761ssize_t
 762ext4_listxattr(struct dentry *dentry, char *buffer, size_t buffer_size)
 763{
 764        int ret, ret2;
 765
 766        down_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 767        ret = ret2 = ext4_xattr_ibody_list(dentry, buffer, buffer_size);
 768        if (ret < 0)
 769                goto errout;
 770        if (buffer) {
 771                buffer += ret;
 772                buffer_size -= ret;
 773        }
 774        ret = ext4_xattr_block_list(dentry, buffer, buffer_size);
 775        if (ret < 0)
 776                goto errout;
 777        ret += ret2;
 778errout:
 779        up_read(&EXT4_I(d_inode(dentry))->xattr_sem);
 780        return ret;
 781}
 782
 783/*
 784 * If the EXT4_FEATURE_COMPAT_EXT_ATTR feature of this file system is
 785 * not set, set it.
 786 */
 787static void ext4_xattr_update_super_block(handle_t *handle,
 788                                          struct super_block *sb)
 789{
 790        if (ext4_has_feature_xattr(sb))
 791                return;
 792
 793        BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
 794        if (ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh) == 0) {
 795                lock_buffer(EXT4_SB(sb)->s_sbh);
 796                ext4_set_feature_xattr(sb);
 797                ext4_superblock_csum_set(sb);
 798                unlock_buffer(EXT4_SB(sb)->s_sbh);
 799                ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 800        }
 801}
 802
 803int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 804{
 805        struct ext4_iloc iloc = { .bh = NULL };
 806        struct buffer_head *bh = NULL;
 807        struct ext4_inode *raw_inode;
 808        struct ext4_xattr_ibody_header *header;
 809        struct ext4_xattr_entry *entry;
 810        qsize_t ea_inode_refs = 0;
 811        void *end;
 812        int ret;
 813
 814        lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 815
 816        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 817                ret = ext4_get_inode_loc(inode, &iloc);
 818                if (ret)
 819                        goto out;
 820                raw_inode = ext4_raw_inode(&iloc);
 821                header = IHDR(inode, raw_inode);
 822                end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 823                ret = xattr_check_inode(inode, header, end);
 824                if (ret)
 825                        goto out;
 826
 827                for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 828                     entry = EXT4_XATTR_NEXT(entry))
 829                        if (entry->e_value_inum)
 830                                ea_inode_refs++;
 831        }
 832
 833        if (EXT4_I(inode)->i_file_acl) {
 834                bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
 835                if (IS_ERR(bh)) {
 836                        ret = PTR_ERR(bh);
 837                        bh = NULL;
 838                        goto out;
 839                }
 840
 841                ret = ext4_xattr_check_block(inode, bh);
 842                if (ret)
 843                        goto out;
 844
 845                for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 846                     entry = EXT4_XATTR_NEXT(entry))
 847                        if (entry->e_value_inum)
 848                                ea_inode_refs++;
 849        }
 850        *usage = ea_inode_refs + 1;
 851        ret = 0;
 852out:
 853        brelse(iloc.bh);
 854        brelse(bh);
 855        return ret;
 856}
 857
 858static inline size_t round_up_cluster(struct inode *inode, size_t length)
 859{
 860        struct super_block *sb = inode->i_sb;
 861        size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 862                                    inode->i_blkbits);
 863        size_t mask = ~(cluster_size - 1);
 864
 865        return (length + cluster_size - 1) & mask;
 866}
 867
 868static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 869{
 870        int err;
 871
 872        err = dquot_alloc_inode(inode);
 873        if (err)
 874                return err;
 875        err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 876        if (err)
 877                dquot_free_inode(inode);
 878        return err;
 879}
 880
 881static void ext4_xattr_inode_free_quota(struct inode *parent,
 882                                        struct inode *ea_inode,
 883                                        size_t len)
 884{
 885        if (ea_inode &&
 886            ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 887                return;
 888        dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 889        dquot_free_inode(parent);
 890}
 891
 892int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 893                             struct buffer_head *block_bh, size_t value_len,
 894                             bool is_create)
 895{
 896        int credits;
 897        int blocks;
 898
 899        /*
 900         * 1) Owner inode update
 901         * 2) Ref count update on old xattr block
 902         * 3) new xattr block
 903         * 4) block bitmap update for new xattr block
 904         * 5) group descriptor for new xattr block
 905         * 6) block bitmap update for old xattr block
 906         * 7) group descriptor for old block
 907         *
 908         * 6 & 7 can happen if we have two racing threads T_a and T_b
 909         * which are each trying to set an xattr on inodes I_a and I_b
 910         * which were both initially sharing an xattr block.
 911         */
 912        credits = 7;
 913
 914        /* Quota updates. */
 915        credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 916
 917        /*
 918         * In case of inline data, we may push out the data to a block,
 919         * so we need to reserve credits for this eventuality
 920         */
 921        if (inode && ext4_has_inline_data(inode))
 922                credits += ext4_writepage_trans_blocks(inode) + 1;
 923
 924        /* We are done if ea_inode feature is not enabled. */
 925        if (!ext4_has_feature_ea_inode(sb))
 926                return credits;
 927
 928        /* New ea_inode, inode map, block bitmap, group descriptor. */
 929        credits += 4;
 930
 931        /* Data blocks. */
 932        blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 933
 934        /* Indirection block or one level of extent tree. */
 935        blocks += 1;
 936
 937        /* Block bitmap and group descriptor updates for each block. */
 938        credits += blocks * 2;
 939
 940        /* Blocks themselves. */
 941        credits += blocks;
 942
 943        if (!is_create) {
 944                /* Dereference ea_inode holding old xattr value.
 945                 * Old ea_inode, inode map, block bitmap, group descriptor.
 946                 */
 947                credits += 4;
 948
 949                /* Data blocks for old ea_inode. */
 950                blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 951
 952                /* Indirection block or one level of extent tree for old
 953                 * ea_inode.
 954                 */
 955                blocks += 1;
 956
 957                /* Block bitmap and group descriptor updates for each block. */
 958                credits += blocks * 2;
 959        }
 960
 961        /* We may need to clone the existing xattr block in which case we need
 962         * to increment ref counts for existing ea_inodes referenced by it.
 963         */
 964        if (block_bh) {
 965                struct ext4_xattr_entry *entry = BFIRST(block_bh);
 966
 967                for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 968                        if (entry->e_value_inum)
 969                                /* Ref count update on ea_inode. */
 970                                credits += 1;
 971        }
 972        return credits;
 973}
 974
 975static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
 976                                       int ref_change)
 977{
 978        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
 979        struct ext4_iloc iloc;
 980        s64 ref_count;
 981        u32 hash;
 982        int ret;
 983
 984        inode_lock(ea_inode);
 985
 986        ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
 987        if (ret)
 988                goto out;
 989
 990        ref_count = ext4_xattr_inode_get_ref(ea_inode);
 991        ref_count += ref_change;
 992        ext4_xattr_inode_set_ref(ea_inode, ref_count);
 993
 994        if (ref_change > 0) {
 995                WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
 996                          ea_inode->i_ino, ref_count);
 997
 998                if (ref_count == 1) {
 999                        WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1000                                  ea_inode->i_ino, ea_inode->i_nlink);
1001
1002                        set_nlink(ea_inode, 1);
1003                        ext4_orphan_del(handle, ea_inode);
1004
1005                        if (ea_inode_cache) {
1006                                hash = ext4_xattr_inode_get_hash(ea_inode);
1007                                mb_cache_entry_create(ea_inode_cache,
1008                                                      GFP_NOFS, hash,
1009                                                      ea_inode->i_ino,
1010                                                      true /* reusable */);
1011                        }
1012                }
1013        } else {
1014                WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1015                          ea_inode->i_ino, ref_count);
1016
1017                if (ref_count == 0) {
1018                        WARN_ONCE(ea_inode->i_nlink != 1,
1019                                  "EA inode %lu i_nlink=%u",
1020                                  ea_inode->i_ino, ea_inode->i_nlink);
1021
1022                        clear_nlink(ea_inode);
1023                        ext4_orphan_add(handle, ea_inode);
1024
1025                        if (ea_inode_cache) {
1026                                hash = ext4_xattr_inode_get_hash(ea_inode);
1027                                mb_cache_entry_delete(ea_inode_cache, hash,
1028                                                      ea_inode->i_ino);
1029                        }
1030                }
1031        }
1032
1033        ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1034        if (ret)
1035                ext4_warning_inode(ea_inode,
1036                                   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1037out:
1038        inode_unlock(ea_inode);
1039        return ret;
1040}
1041
1042static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1043{
1044        return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1045}
1046
1047static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1048{
1049        return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1050}
1051
1052static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1053                                        struct ext4_xattr_entry *first)
1054{
1055        struct inode *ea_inode;
1056        struct ext4_xattr_entry *entry;
1057        struct ext4_xattr_entry *failed_entry;
1058        unsigned int ea_ino;
1059        int err, saved_err;
1060
1061        for (entry = first; !IS_LAST_ENTRY(entry);
1062             entry = EXT4_XATTR_NEXT(entry)) {
1063                if (!entry->e_value_inum)
1064                        continue;
1065                ea_ino = le32_to_cpu(entry->e_value_inum);
1066                err = ext4_xattr_inode_iget(parent, ea_ino,
1067                                            le32_to_cpu(entry->e_hash),
1068                                            &ea_inode);
1069                if (err)
1070                        goto cleanup;
1071                err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1072                if (err) {
1073                        ext4_warning_inode(ea_inode, "inc ref error %d", err);
1074                        iput(ea_inode);
1075                        goto cleanup;
1076                }
1077                iput(ea_inode);
1078        }
1079        return 0;
1080
1081cleanup:
1082        saved_err = err;
1083        failed_entry = entry;
1084
1085        for (entry = first; entry != failed_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                        ext4_warning(parent->i_sb,
1095                                     "cleanup ea_ino %u iget error %d", ea_ino,
1096                                     err);
1097                        continue;
1098                }
1099                err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1100                if (err)
1101                        ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1102                                           err);
1103                iput(ea_inode);
1104        }
1105        return saved_err;
1106}
1107
1108static int ext4_xattr_restart_fn(handle_t *handle, struct inode *inode,
1109                        struct buffer_head *bh, bool block_csum, bool dirty)
1110{
1111        int error;
1112
1113        if (bh && dirty) {
1114                if (block_csum)
1115                        ext4_xattr_block_csum_set(inode, bh);
1116                error = ext4_handle_dirty_metadata(handle, NULL, bh);
1117                if (error) {
1118                        ext4_warning(inode->i_sb, "Handle metadata (error %d)",
1119                                     error);
1120                        return error;
1121                }
1122        }
1123        return 0;
1124}
1125
1126static void
1127ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1128                             struct buffer_head *bh,
1129                             struct ext4_xattr_entry *first, bool block_csum,
1130                             struct ext4_xattr_inode_array **ea_inode_array,
1131                             int extra_credits, bool skip_quota)
1132{
1133        struct inode *ea_inode;
1134        struct ext4_xattr_entry *entry;
1135        bool dirty = false;
1136        unsigned int ea_ino;
1137        int err;
1138        int credits;
1139
1140        /* One credit for dec ref on ea_inode, one for orphan list addition, */
1141        credits = 2 + extra_credits;
1142
1143        for (entry = first; !IS_LAST_ENTRY(entry);
1144             entry = EXT4_XATTR_NEXT(entry)) {
1145                if (!entry->e_value_inum)
1146                        continue;
1147                ea_ino = le32_to_cpu(entry->e_value_inum);
1148                err = ext4_xattr_inode_iget(parent, ea_ino,
1149                                            le32_to_cpu(entry->e_hash),
1150                                            &ea_inode);
1151                if (err)
1152                        continue;
1153
1154                err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1155                if (err) {
1156                        ext4_warning_inode(ea_inode,
1157                                           "Expand inode array err=%d", err);
1158                        iput(ea_inode);
1159                        continue;
1160                }
1161
1162                err = ext4_journal_ensure_credits_fn(handle, credits, credits,
1163                        ext4_free_metadata_revoke_credits(parent->i_sb, 1),
1164                        ext4_xattr_restart_fn(handle, parent, bh, block_csum,
1165                                              dirty));
1166                if (err < 0) {
1167                        ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1168                                           err);
1169                        continue;
1170                }
1171                if (err > 0) {
1172                        err = ext4_journal_get_write_access(handle, bh);
1173                        if (err) {
1174                                ext4_warning_inode(ea_inode,
1175                                                "Re-get write access err=%d",
1176                                                err);
1177                                continue;
1178                        }
1179                }
1180
1181                err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1182                if (err) {
1183                        ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1184                                           err);
1185                        continue;
1186                }
1187
1188                if (!skip_quota)
1189                        ext4_xattr_inode_free_quota(parent, ea_inode,
1190                                              le32_to_cpu(entry->e_value_size));
1191
1192                /*
1193                 * Forget about ea_inode within the same transaction that
1194                 * decrements the ref count. This avoids duplicate decrements in
1195                 * case the rest of the work spills over to subsequent
1196                 * transactions.
1197                 */
1198                entry->e_value_inum = 0;
1199                entry->e_value_size = 0;
1200
1201                dirty = true;
1202        }
1203
1204        if (dirty) {
1205                /*
1206                 * Note that we are deliberately skipping csum calculation for
1207                 * the final update because we do not expect any journal
1208                 * restarts until xattr block is freed.
1209                 */
1210
1211                err = ext4_handle_dirty_metadata(handle, NULL, bh);
1212                if (err)
1213                        ext4_warning_inode(parent,
1214                                           "handle dirty metadata err=%d", err);
1215        }
1216}
1217
1218/*
1219 * Release the xattr block BH: If the reference count is > 1, decrement it;
1220 * otherwise free the block.
1221 */
1222static void
1223ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1224                         struct buffer_head *bh,
1225                         struct ext4_xattr_inode_array **ea_inode_array,
1226                         int extra_credits)
1227{
1228        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1229        u32 hash, ref;
1230        int error = 0;
1231
1232        BUFFER_TRACE(bh, "get_write_access");
1233        error = ext4_journal_get_write_access(handle, bh);
1234        if (error)
1235                goto out;
1236
1237        lock_buffer(bh);
1238        hash = le32_to_cpu(BHDR(bh)->h_hash);
1239        ref = le32_to_cpu(BHDR(bh)->h_refcount);
1240        if (ref == 1) {
1241                ea_bdebug(bh, "refcount now=0; freeing");
1242                /*
1243                 * This must happen under buffer lock for
1244                 * ext4_xattr_block_set() to reliably detect freed block
1245                 */
1246                if (ea_block_cache)
1247                        mb_cache_entry_delete(ea_block_cache, hash,
1248                                              bh->b_blocknr);
1249                get_bh(bh);
1250                unlock_buffer(bh);
1251
1252                if (ext4_has_feature_ea_inode(inode->i_sb))
1253                        ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1254                                                     BFIRST(bh),
1255                                                     true /* block_csum */,
1256                                                     ea_inode_array,
1257                                                     extra_credits,
1258                                                     true /* skip_quota */);
1259                ext4_free_blocks(handle, inode, bh, 0, 1,
1260                                 EXT4_FREE_BLOCKS_METADATA |
1261                                 EXT4_FREE_BLOCKS_FORGET);
1262        } else {
1263                ref--;
1264                BHDR(bh)->h_refcount = cpu_to_le32(ref);
1265                if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1266                        struct mb_cache_entry *ce;
1267
1268                        if (ea_block_cache) {
1269                                ce = mb_cache_entry_get(ea_block_cache, hash,
1270                                                        bh->b_blocknr);
1271                                if (ce) {
1272                                        ce->e_reusable = 1;
1273                                        mb_cache_entry_put(ea_block_cache, ce);
1274                                }
1275                        }
1276                }
1277
1278                ext4_xattr_block_csum_set(inode, bh);
1279                /*
1280                 * Beware of this ugliness: Releasing of xattr block references
1281                 * from different inodes can race and so we have to protect
1282                 * from a race where someone else frees the block (and releases
1283                 * its journal_head) before we are done dirtying the buffer. In
1284                 * nojournal mode this race is harmless and we actually cannot
1285                 * call ext4_handle_dirty_metadata() with locked buffer as
1286                 * that function can call sync_dirty_buffer() so for that case
1287                 * we handle the dirtying after unlocking the buffer.
1288                 */
1289                if (ext4_handle_valid(handle))
1290                        error = ext4_handle_dirty_metadata(handle, inode, bh);
1291                unlock_buffer(bh);
1292                if (!ext4_handle_valid(handle))
1293                        error = ext4_handle_dirty_metadata(handle, inode, bh);
1294                if (IS_SYNC(inode))
1295                        ext4_handle_sync(handle);
1296                dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1297                ea_bdebug(bh, "refcount now=%d; releasing",
1298                          le32_to_cpu(BHDR(bh)->h_refcount));
1299        }
1300out:
1301        ext4_std_error(inode->i_sb, error);
1302        return;
1303}
1304
1305/*
1306 * Find the available free space for EAs. This also returns the total number of
1307 * bytes used by EA entries.
1308 */
1309static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1310                                    size_t *min_offs, void *base, int *total)
1311{
1312        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1313                if (!last->e_value_inum && last->e_value_size) {
1314                        size_t offs = le16_to_cpu(last->e_value_offs);
1315                        if (offs < *min_offs)
1316                                *min_offs = offs;
1317                }
1318                if (total)
1319                        *total += EXT4_XATTR_LEN(last->e_name_len);
1320        }
1321        return (*min_offs - ((void *)last - base) - sizeof(__u32));
1322}
1323
1324/*
1325 * Write the value of the EA in an inode.
1326 */
1327static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1328                                  const void *buf, int bufsize)
1329{
1330        struct buffer_head *bh = NULL;
1331        unsigned long block = 0;
1332        int blocksize = ea_inode->i_sb->s_blocksize;
1333        int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1334        int csize, wsize = 0;
1335        int ret = 0, ret2 = 0;
1336        int retries = 0;
1337
1338retry:
1339        while (ret >= 0 && ret < max_blocks) {
1340                struct ext4_map_blocks map;
1341                map.m_lblk = block += ret;
1342                map.m_len = max_blocks -= ret;
1343
1344                ret = ext4_map_blocks(handle, ea_inode, &map,
1345                                      EXT4_GET_BLOCKS_CREATE);
1346                if (ret <= 0) {
1347                        ext4_mark_inode_dirty(handle, ea_inode);
1348                        if (ret == -ENOSPC &&
1349                            ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1350                                ret = 0;
1351                                goto retry;
1352                        }
1353                        break;
1354                }
1355        }
1356
1357        if (ret < 0)
1358                return ret;
1359
1360        block = 0;
1361        while (wsize < bufsize) {
1362                brelse(bh);
1363                csize = (bufsize - wsize) > blocksize ? blocksize :
1364                                                                bufsize - wsize;
1365                bh = ext4_getblk(handle, ea_inode, block, 0);
1366                if (IS_ERR(bh))
1367                        return PTR_ERR(bh);
1368                if (!bh) {
1369                        WARN_ON_ONCE(1);
1370                        EXT4_ERROR_INODE(ea_inode,
1371                                         "ext4_getblk() return bh = NULL");
1372                        return -EFSCORRUPTED;
1373                }
1374                ret = ext4_journal_get_write_access(handle, bh);
1375                if (ret)
1376                        goto out;
1377
1378                memcpy(bh->b_data, buf, csize);
1379                set_buffer_uptodate(bh);
1380                ext4_handle_dirty_metadata(handle, ea_inode, bh);
1381
1382                buf += csize;
1383                wsize += csize;
1384                block += 1;
1385        }
1386
1387        inode_lock(ea_inode);
1388        i_size_write(ea_inode, wsize);
1389        ext4_update_i_disksize(ea_inode, wsize);
1390        inode_unlock(ea_inode);
1391
1392        ret2 = ext4_mark_inode_dirty(handle, ea_inode);
1393        if (unlikely(ret2 && !ret))
1394                ret = ret2;
1395
1396out:
1397        brelse(bh);
1398
1399        return ret;
1400}
1401
1402/*
1403 * Create an inode to store the value of a large EA.
1404 */
1405static struct inode *ext4_xattr_inode_create(handle_t *handle,
1406                                             struct inode *inode, u32 hash)
1407{
1408        struct inode *ea_inode = NULL;
1409        uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1410        int err;
1411
1412        /*
1413         * Let the next inode be the goal, so we try and allocate the EA inode
1414         * in the same group, or nearby one.
1415         */
1416        ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1417                                  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1418                                  EXT4_EA_INODE_FL);
1419        if (!IS_ERR(ea_inode)) {
1420                ea_inode->i_op = &ext4_file_inode_operations;
1421                ea_inode->i_fop = &ext4_file_operations;
1422                ext4_set_aops(ea_inode);
1423                ext4_xattr_inode_set_class(ea_inode);
1424                unlock_new_inode(ea_inode);
1425                ext4_xattr_inode_set_ref(ea_inode, 1);
1426                ext4_xattr_inode_set_hash(ea_inode, hash);
1427                err = ext4_mark_inode_dirty(handle, ea_inode);
1428                if (!err)
1429                        err = ext4_inode_attach_jinode(ea_inode);
1430                if (err) {
1431                        iput(ea_inode);
1432                        return ERR_PTR(err);
1433                }
1434
1435                /*
1436                 * Xattr inodes are shared therefore quota charging is performed
1437                 * at a higher level.
1438                 */
1439                dquot_free_inode(ea_inode);
1440                dquot_drop(ea_inode);
1441                inode_lock(ea_inode);
1442                ea_inode->i_flags |= S_NOQUOTA;
1443                inode_unlock(ea_inode);
1444        }
1445
1446        return ea_inode;
1447}
1448
1449static struct inode *
1450ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1451                            size_t value_len, u32 hash)
1452{
1453        struct inode *ea_inode;
1454        struct mb_cache_entry *ce;
1455        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1456        void *ea_data;
1457
1458        if (!ea_inode_cache)
1459                return NULL;
1460
1461        ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1462        if (!ce)
1463                return NULL;
1464
1465        WARN_ON_ONCE(ext4_handle_valid(journal_current_handle()) &&
1466                     !(current->flags & PF_MEMALLOC_NOFS));
1467
1468        ea_data = kvmalloc(value_len, GFP_KERNEL);
1469        if (!ea_data) {
1470                mb_cache_entry_put(ea_inode_cache, ce);
1471                return NULL;
1472        }
1473
1474        while (ce) {
1475                ea_inode = ext4_iget(inode->i_sb, ce->e_value,
1476                                     EXT4_IGET_NORMAL);
1477                if (!IS_ERR(ea_inode) &&
1478                    !is_bad_inode(ea_inode) &&
1479                    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1480                    i_size_read(ea_inode) == value_len &&
1481                    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1482                    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1483                                                    value_len) &&
1484                    !memcmp(value, ea_data, value_len)) {
1485                        mb_cache_entry_touch(ea_inode_cache, ce);
1486                        mb_cache_entry_put(ea_inode_cache, ce);
1487                        kvfree(ea_data);
1488                        return ea_inode;
1489                }
1490
1491                if (!IS_ERR(ea_inode))
1492                        iput(ea_inode);
1493                ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1494        }
1495        kvfree(ea_data);
1496        return NULL;
1497}
1498
1499/*
1500 * Add value of the EA in an inode.
1501 */
1502static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1503                                          const void *value, size_t value_len,
1504                                          struct inode **ret_inode)
1505{
1506        struct inode *ea_inode;
1507        u32 hash;
1508        int err;
1509
1510        hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1511        ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1512        if (ea_inode) {
1513                err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1514                if (err) {
1515                        iput(ea_inode);
1516                        return err;
1517                }
1518
1519                *ret_inode = ea_inode;
1520                return 0;
1521        }
1522
1523        /* Create an inode for the EA value */
1524        ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1525        if (IS_ERR(ea_inode))
1526                return PTR_ERR(ea_inode);
1527
1528        err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1529        if (err) {
1530                ext4_xattr_inode_dec_ref(handle, ea_inode);
1531                iput(ea_inode);
1532                return err;
1533        }
1534
1535        if (EA_INODE_CACHE(inode))
1536                mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1537                                      ea_inode->i_ino, true /* reusable */);
1538
1539        *ret_inode = ea_inode;
1540        return 0;
1541}
1542
1543/*
1544 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1545 * feature is enabled.
1546 */
1547#define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1548
1549static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1550                                struct ext4_xattr_search *s,
1551                                handle_t *handle, struct inode *inode,
1552                                bool is_block)
1553{
1554        struct ext4_xattr_entry *last, *next;
1555        struct ext4_xattr_entry *here = s->here;
1556        size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1557        int in_inode = i->in_inode;
1558        struct inode *old_ea_inode = NULL;
1559        struct inode *new_ea_inode = NULL;
1560        size_t old_size, new_size;
1561        int ret;
1562
1563        /* Space used by old and new values. */
1564        old_size = (!s->not_found && !here->e_value_inum) ?
1565                        EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1566        new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1567
1568        /*
1569         * Optimization for the simple case when old and new values have the
1570         * same padded sizes. Not applicable if external inodes are involved.
1571         */
1572        if (new_size && new_size == old_size) {
1573                size_t offs = le16_to_cpu(here->e_value_offs);
1574                void *val = s->base + offs;
1575
1576                here->e_value_size = cpu_to_le32(i->value_len);
1577                if (i->value == EXT4_ZERO_XATTR_VALUE) {
1578                        memset(val, 0, new_size);
1579                } else {
1580                        memcpy(val, i->value, i->value_len);
1581                        /* Clear padding bytes. */
1582                        memset(val + i->value_len, 0, new_size - i->value_len);
1583                }
1584                goto update_hash;
1585        }
1586
1587        /* Compute min_offs and last. */
1588        last = s->first;
1589        for (; !IS_LAST_ENTRY(last); last = next) {
1590                next = EXT4_XATTR_NEXT(last);
1591                if ((void *)next >= s->end) {
1592                        EXT4_ERROR_INODE(inode, "corrupted xattr entries");
1593                        ret = -EFSCORRUPTED;
1594                        goto out;
1595                }
1596                if (!last->e_value_inum && last->e_value_size) {
1597                        size_t offs = le16_to_cpu(last->e_value_offs);
1598                        if (offs < min_offs)
1599                                min_offs = offs;
1600                }
1601        }
1602
1603        /* Check whether we have enough space. */
1604        if (i->value) {
1605                size_t free;
1606
1607                free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1608                if (!s->not_found)
1609                        free += EXT4_XATTR_LEN(name_len) + old_size;
1610
1611                if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1612                        ret = -ENOSPC;
1613                        goto out;
1614                }
1615
1616                /*
1617                 * If storing the value in an external inode is an option,
1618                 * reserve space for xattr entries/names in the external
1619                 * attribute block so that a long value does not occupy the
1620                 * whole space and prevent further entries being added.
1621                 */
1622                if (ext4_has_feature_ea_inode(inode->i_sb) &&
1623                    new_size && is_block &&
1624                    (min_offs + old_size - new_size) <
1625                                        EXT4_XATTR_BLOCK_RESERVE(inode)) {
1626                        ret = -ENOSPC;
1627                        goto out;
1628                }
1629        }
1630
1631        /*
1632         * Getting access to old and new ea inodes is subject to failures.
1633         * Finish that work before doing any modifications to the xattr data.
1634         */
1635        if (!s->not_found && here->e_value_inum) {
1636                ret = ext4_xattr_inode_iget(inode,
1637                                            le32_to_cpu(here->e_value_inum),
1638                                            le32_to_cpu(here->e_hash),
1639                                            &old_ea_inode);
1640                if (ret) {
1641                        old_ea_inode = NULL;
1642                        goto out;
1643                }
1644        }
1645        if (i->value && in_inode) {
1646                WARN_ON_ONCE(!i->value_len);
1647
1648                ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1649                if (ret)
1650                        goto out;
1651
1652                ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1653                                                     i->value_len,
1654                                                     &new_ea_inode);
1655                if (ret) {
1656                        new_ea_inode = NULL;
1657                        ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1658                        goto out;
1659                }
1660        }
1661
1662        if (old_ea_inode) {
1663                /* We are ready to release ref count on the old_ea_inode. */
1664                ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1665                if (ret) {
1666                        /* Release newly required ref count on new_ea_inode. */
1667                        if (new_ea_inode) {
1668                                int err;
1669
1670                                err = ext4_xattr_inode_dec_ref(handle,
1671                                                               new_ea_inode);
1672                                if (err)
1673                                        ext4_warning_inode(new_ea_inode,
1674                                                  "dec ref new_ea_inode err=%d",
1675                                                  err);
1676                                ext4_xattr_inode_free_quota(inode, new_ea_inode,
1677                                                            i->value_len);
1678                        }
1679                        goto out;
1680                }
1681
1682                ext4_xattr_inode_free_quota(inode, old_ea_inode,
1683                                            le32_to_cpu(here->e_value_size));
1684        }
1685
1686        /* No failures allowed past this point. */
1687
1688        if (!s->not_found && here->e_value_size && !here->e_value_inum) {
1689                /* Remove the old value. */
1690                void *first_val = s->base + min_offs;
1691                size_t offs = le16_to_cpu(here->e_value_offs);
1692                void *val = s->base + offs;
1693
1694                memmove(first_val + old_size, first_val, val - first_val);
1695                memset(first_val, 0, old_size);
1696                min_offs += old_size;
1697
1698                /* Adjust all value offsets. */
1699                last = s->first;
1700                while (!IS_LAST_ENTRY(last)) {
1701                        size_t o = le16_to_cpu(last->e_value_offs);
1702
1703                        if (!last->e_value_inum &&
1704                            last->e_value_size && o < offs)
1705                                last->e_value_offs = cpu_to_le16(o + old_size);
1706                        last = EXT4_XATTR_NEXT(last);
1707                }
1708        }
1709
1710        if (!i->value) {
1711                /* Remove old name. */
1712                size_t size = EXT4_XATTR_LEN(name_len);
1713
1714                last = ENTRY((void *)last - size);
1715                memmove(here, (void *)here + size,
1716                        (void *)last - (void *)here + sizeof(__u32));
1717                memset(last, 0, size);
1718        } else if (s->not_found) {
1719                /* Insert new name. */
1720                size_t size = EXT4_XATTR_LEN(name_len);
1721                size_t rest = (void *)last - (void *)here + sizeof(__u32);
1722
1723                memmove((void *)here + size, here, rest);
1724                memset(here, 0, size);
1725                here->e_name_index = i->name_index;
1726                here->e_name_len = name_len;
1727                memcpy(here->e_name, i->name, name_len);
1728        } else {
1729                /* This is an update, reset value info. */
1730                here->e_value_inum = 0;
1731                here->e_value_offs = 0;
1732                here->e_value_size = 0;
1733        }
1734
1735        if (i->value) {
1736                /* Insert new value. */
1737                if (in_inode) {
1738                        here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1739                } else if (i->value_len) {
1740                        void *val = s->base + min_offs - new_size;
1741
1742                        here->e_value_offs = cpu_to_le16(min_offs - new_size);
1743                        if (i->value == EXT4_ZERO_XATTR_VALUE) {
1744                                memset(val, 0, new_size);
1745                        } else {
1746                                memcpy(val, i->value, i->value_len);
1747                                /* Clear padding bytes. */
1748                                memset(val + i->value_len, 0,
1749                                       new_size - i->value_len);
1750                        }
1751                }
1752                here->e_value_size = cpu_to_le32(i->value_len);
1753        }
1754
1755update_hash:
1756        if (i->value) {
1757                __le32 hash = 0;
1758
1759                /* Entry hash calculation. */
1760                if (in_inode) {
1761                        __le32 crc32c_hash;
1762
1763                        /*
1764                         * Feed crc32c hash instead of the raw value for entry
1765                         * hash calculation. This is to avoid walking
1766                         * potentially long value buffer again.
1767                         */
1768                        crc32c_hash = cpu_to_le32(
1769                                       ext4_xattr_inode_get_hash(new_ea_inode));
1770                        hash = ext4_xattr_hash_entry(here->e_name,
1771                                                     here->e_name_len,
1772                                                     &crc32c_hash, 1);
1773                } else if (is_block) {
1774                        __le32 *value = s->base + le16_to_cpu(
1775                                                        here->e_value_offs);
1776
1777                        hash = ext4_xattr_hash_entry(here->e_name,
1778                                                     here->e_name_len, value,
1779                                                     new_size >> 2);
1780                }
1781                here->e_hash = hash;
1782        }
1783
1784        if (is_block)
1785                ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1786
1787        ret = 0;
1788out:
1789        iput(old_ea_inode);
1790        iput(new_ea_inode);
1791        return ret;
1792}
1793
1794struct ext4_xattr_block_find {
1795        struct ext4_xattr_search s;
1796        struct buffer_head *bh;
1797};
1798
1799static int
1800ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1801                      struct ext4_xattr_block_find *bs)
1802{
1803        struct super_block *sb = inode->i_sb;
1804        int error;
1805
1806        ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1807                  i->name_index, i->name, i->value, (long)i->value_len);
1808
1809        if (EXT4_I(inode)->i_file_acl) {
1810                /* The inode already has an extended attribute block. */
1811                bs->bh = ext4_sb_bread(sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
1812                if (IS_ERR(bs->bh)) {
1813                        error = PTR_ERR(bs->bh);
1814                        bs->bh = NULL;
1815                        return error;
1816                }
1817                ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1818                        atomic_read(&(bs->bh->b_count)),
1819                        le32_to_cpu(BHDR(bs->bh)->h_refcount));
1820                error = ext4_xattr_check_block(inode, bs->bh);
1821                if (error)
1822                        return error;
1823                /* Find the named attribute. */
1824                bs->s.base = BHDR(bs->bh);
1825                bs->s.first = BFIRST(bs->bh);
1826                bs->s.end = bs->bh->b_data + bs->bh->b_size;
1827                bs->s.here = bs->s.first;
1828                error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1829                                         i->name_index, i->name, 1);
1830                if (error && error != -ENODATA)
1831                        return error;
1832                bs->s.not_found = error;
1833        }
1834        return 0;
1835}
1836
1837static int
1838ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1839                     struct ext4_xattr_info *i,
1840                     struct ext4_xattr_block_find *bs)
1841{
1842        struct super_block *sb = inode->i_sb;
1843        struct buffer_head *new_bh = NULL;
1844        struct ext4_xattr_search s_copy = bs->s;
1845        struct ext4_xattr_search *s = &s_copy;
1846        struct mb_cache_entry *ce = NULL;
1847        int error = 0;
1848        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1849        struct inode *ea_inode = NULL, *tmp_inode;
1850        size_t old_ea_inode_quota = 0;
1851        unsigned int ea_ino;
1852
1853
1854#define header(x) ((struct ext4_xattr_header *)(x))
1855
1856        if (s->base) {
1857                BUFFER_TRACE(bs->bh, "get_write_access");
1858                error = ext4_journal_get_write_access(handle, bs->bh);
1859                if (error)
1860                        goto cleanup;
1861                lock_buffer(bs->bh);
1862
1863                if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1864                        __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1865
1866                        /*
1867                         * This must happen under buffer lock for
1868                         * ext4_xattr_block_set() to reliably detect modified
1869                         * block
1870                         */
1871                        if (ea_block_cache)
1872                                mb_cache_entry_delete(ea_block_cache, hash,
1873                                                      bs->bh->b_blocknr);
1874                        ea_bdebug(bs->bh, "modifying in-place");
1875                        error = ext4_xattr_set_entry(i, s, handle, inode,
1876                                                     true /* is_block */);
1877                        ext4_xattr_block_csum_set(inode, bs->bh);
1878                        unlock_buffer(bs->bh);
1879                        if (error == -EFSCORRUPTED)
1880                                goto bad_block;
1881                        if (!error)
1882                                error = ext4_handle_dirty_metadata(handle,
1883                                                                   inode,
1884                                                                   bs->bh);
1885                        if (error)
1886                                goto cleanup;
1887                        goto inserted;
1888                } else {
1889                        int offset = (char *)s->here - bs->bh->b_data;
1890
1891                        unlock_buffer(bs->bh);
1892                        ea_bdebug(bs->bh, "cloning");
1893                        s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1894                        error = -ENOMEM;
1895                        if (s->base == NULL)
1896                                goto cleanup;
1897                        memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1898                        s->first = ENTRY(header(s->base)+1);
1899                        header(s->base)->h_refcount = cpu_to_le32(1);
1900                        s->here = ENTRY(s->base + offset);
1901                        s->end = s->base + bs->bh->b_size;
1902
1903                        /*
1904                         * If existing entry points to an xattr inode, we need
1905                         * to prevent ext4_xattr_set_entry() from decrementing
1906                         * ref count on it because the reference belongs to the
1907                         * original block. In this case, make the entry look
1908                         * like it has an empty value.
1909                         */
1910                        if (!s->not_found && s->here->e_value_inum) {
1911                                ea_ino = le32_to_cpu(s->here->e_value_inum);
1912                                error = ext4_xattr_inode_iget(inode, ea_ino,
1913                                              le32_to_cpu(s->here->e_hash),
1914                                              &tmp_inode);
1915                                if (error)
1916                                        goto cleanup;
1917
1918                                if (!ext4_test_inode_state(tmp_inode,
1919                                                EXT4_STATE_LUSTRE_EA_INODE)) {
1920                                        /*
1921                                         * Defer quota free call for previous
1922                                         * inode until success is guaranteed.
1923                                         */
1924                                        old_ea_inode_quota = le32_to_cpu(
1925                                                        s->here->e_value_size);
1926                                }
1927                                iput(tmp_inode);
1928
1929                                s->here->e_value_inum = 0;
1930                                s->here->e_value_size = 0;
1931                        }
1932                }
1933        } else {
1934                /* Allocate a buffer where we construct the new block. */
1935                s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1936                error = -ENOMEM;
1937                if (s->base == NULL)
1938                        goto cleanup;
1939                header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1940                header(s->base)->h_blocks = cpu_to_le32(1);
1941                header(s->base)->h_refcount = cpu_to_le32(1);
1942                s->first = ENTRY(header(s->base)+1);
1943                s->here = ENTRY(header(s->base)+1);
1944                s->end = s->base + sb->s_blocksize;
1945        }
1946
1947        error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1948        if (error == -EFSCORRUPTED)
1949                goto bad_block;
1950        if (error)
1951                goto cleanup;
1952
1953        if (i->value && s->here->e_value_inum) {
1954                /*
1955                 * A ref count on ea_inode has been taken as part of the call to
1956                 * ext4_xattr_set_entry() above. We would like to drop this
1957                 * extra ref but we have to wait until the xattr block is
1958                 * initialized and has its own ref count on the ea_inode.
1959                 */
1960                ea_ino = le32_to_cpu(s->here->e_value_inum);
1961                error = ext4_xattr_inode_iget(inode, ea_ino,
1962                                              le32_to_cpu(s->here->e_hash),
1963                                              &ea_inode);
1964                if (error) {
1965                        ea_inode = NULL;
1966                        goto cleanup;
1967                }
1968        }
1969
1970inserted:
1971        if (!IS_LAST_ENTRY(s->first)) {
1972                new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1973                                                     &ce);
1974                if (new_bh) {
1975                        /* We found an identical block in the cache. */
1976                        if (new_bh == bs->bh)
1977                                ea_bdebug(new_bh, "keeping");
1978                        else {
1979                                u32 ref;
1980
1981                                WARN_ON_ONCE(dquot_initialize_needed(inode));
1982
1983                                /* The old block is released after updating
1984                                   the inode. */
1985                                error = dquot_alloc_block(inode,
1986                                                EXT4_C2B(EXT4_SB(sb), 1));
1987                                if (error)
1988                                        goto cleanup;
1989                                BUFFER_TRACE(new_bh, "get_write_access");
1990                                error = ext4_journal_get_write_access(handle,
1991                                                                      new_bh);
1992                                if (error)
1993                                        goto cleanup_dquot;
1994                                lock_buffer(new_bh);
1995                                /*
1996                                 * We have to be careful about races with
1997                                 * freeing, rehashing or adding references to
1998                                 * xattr block. Once we hold buffer lock xattr
1999                                 * block's state is stable so we can check
2000                                 * whether the block got freed / rehashed or
2001                                 * not.  Since we unhash mbcache entry under
2002                                 * buffer lock when freeing / rehashing xattr
2003                                 * block, checking whether entry is still
2004                                 * hashed is reliable. Same rules hold for
2005                                 * e_reusable handling.
2006                                 */
2007                                if (hlist_bl_unhashed(&ce->e_hash_list) ||
2008                                    !ce->e_reusable) {
2009                                        /*
2010                                         * Undo everything and check mbcache
2011                                         * again.
2012                                         */
2013                                        unlock_buffer(new_bh);
2014                                        dquot_free_block(inode,
2015                                                         EXT4_C2B(EXT4_SB(sb),
2016                                                                  1));
2017                                        brelse(new_bh);
2018                                        mb_cache_entry_put(ea_block_cache, ce);
2019                                        ce = NULL;
2020                                        new_bh = NULL;
2021                                        goto inserted;
2022                                }
2023                                ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2024                                BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2025                                if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2026                                        ce->e_reusable = 0;
2027                                ea_bdebug(new_bh, "reusing; refcount now=%d",
2028                                          ref);
2029                                ext4_xattr_block_csum_set(inode, new_bh);
2030                                unlock_buffer(new_bh);
2031                                error = ext4_handle_dirty_metadata(handle,
2032                                                                   inode,
2033                                                                   new_bh);
2034                                if (error)
2035                                        goto cleanup_dquot;
2036                        }
2037                        mb_cache_entry_touch(ea_block_cache, ce);
2038                        mb_cache_entry_put(ea_block_cache, ce);
2039                        ce = NULL;
2040                } else if (bs->bh && s->base == bs->bh->b_data) {
2041                        /* We were modifying this block in-place. */
2042                        ea_bdebug(bs->bh, "keeping this block");
2043                        ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2044                        new_bh = bs->bh;
2045                        get_bh(new_bh);
2046                } else {
2047                        /* We need to allocate a new block */
2048                        ext4_fsblk_t goal, block;
2049
2050                        WARN_ON_ONCE(dquot_initialize_needed(inode));
2051
2052                        goal = ext4_group_first_block_no(sb,
2053                                                EXT4_I(inode)->i_block_group);
2054
2055                        /* non-extent files can't have physical blocks past 2^32 */
2056                        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2057                                goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2058
2059                        block = ext4_new_meta_blocks(handle, inode, goal, 0,
2060                                                     NULL, &error);
2061                        if (error)
2062                                goto cleanup;
2063
2064                        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2065                                BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2066
2067                        ea_idebug(inode, "creating block %llu",
2068                                  (unsigned long long)block);
2069
2070                        new_bh = sb_getblk(sb, block);
2071                        if (unlikely(!new_bh)) {
2072                                error = -ENOMEM;
2073getblk_failed:
2074                                ext4_free_blocks(handle, inode, NULL, block, 1,
2075                                                 EXT4_FREE_BLOCKS_METADATA);
2076                                goto cleanup;
2077                        }
2078                        error = ext4_xattr_inode_inc_ref_all(handle, inode,
2079                                                      ENTRY(header(s->base)+1));
2080                        if (error)
2081                                goto getblk_failed;
2082                        if (ea_inode) {
2083                                /* Drop the extra ref on ea_inode. */
2084                                error = ext4_xattr_inode_dec_ref(handle,
2085                                                                 ea_inode);
2086                                if (error)
2087                                        ext4_warning_inode(ea_inode,
2088                                                           "dec ref error=%d",
2089                                                           error);
2090                                iput(ea_inode);
2091                                ea_inode = NULL;
2092                        }
2093
2094                        lock_buffer(new_bh);
2095                        error = ext4_journal_get_create_access(handle, new_bh);
2096                        if (error) {
2097                                unlock_buffer(new_bh);
2098                                error = -EIO;
2099                                goto getblk_failed;
2100                        }
2101                        memcpy(new_bh->b_data, s->base, new_bh->b_size);
2102                        ext4_xattr_block_csum_set(inode, new_bh);
2103                        set_buffer_uptodate(new_bh);
2104                        unlock_buffer(new_bh);
2105                        ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2106                        error = ext4_handle_dirty_metadata(handle, inode,
2107                                                           new_bh);
2108                        if (error)
2109                                goto cleanup;
2110                }
2111        }
2112
2113        if (old_ea_inode_quota)
2114                ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2115
2116        /* Update the inode. */
2117        EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2118
2119        /* Drop the previous xattr block. */
2120        if (bs->bh && bs->bh != new_bh) {
2121                struct ext4_xattr_inode_array *ea_inode_array = NULL;
2122
2123                ext4_xattr_release_block(handle, inode, bs->bh,
2124                                         &ea_inode_array,
2125                                         0 /* extra_credits */);
2126                ext4_xattr_inode_array_free(ea_inode_array);
2127        }
2128        error = 0;
2129
2130cleanup:
2131        if (ea_inode) {
2132                int error2;
2133
2134                error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2135                if (error2)
2136                        ext4_warning_inode(ea_inode, "dec ref error=%d",
2137                                           error2);
2138
2139                /* If there was an error, revert the quota charge. */
2140                if (error)
2141                        ext4_xattr_inode_free_quota(inode, ea_inode,
2142                                                    i_size_read(ea_inode));
2143                iput(ea_inode);
2144        }
2145        if (ce)
2146                mb_cache_entry_put(ea_block_cache, ce);
2147        brelse(new_bh);
2148        if (!(bs->bh && s->base == bs->bh->b_data))
2149                kfree(s->base);
2150
2151        return error;
2152
2153cleanup_dquot:
2154        dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2155        goto cleanup;
2156
2157bad_block:
2158        EXT4_ERROR_INODE(inode, "bad block %llu",
2159                         EXT4_I(inode)->i_file_acl);
2160        goto cleanup;
2161
2162#undef header
2163}
2164
2165int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2166                          struct ext4_xattr_ibody_find *is)
2167{
2168        struct ext4_xattr_ibody_header *header;
2169        struct ext4_inode *raw_inode;
2170        int error;
2171
2172        if (EXT4_I(inode)->i_extra_isize == 0)
2173                return 0;
2174        raw_inode = ext4_raw_inode(&is->iloc);
2175        header = IHDR(inode, raw_inode);
2176        is->s.base = is->s.first = IFIRST(header);
2177        is->s.here = is->s.first;
2178        is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2179        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2180                error = xattr_check_inode(inode, header, is->s.end);
2181                if (error)
2182                        return error;
2183                /* Find the named attribute. */
2184                error = xattr_find_entry(inode, &is->s.here, is->s.end,
2185                                         i->name_index, i->name, 0);
2186                if (error && error != -ENODATA)
2187                        return error;
2188                is->s.not_found = error;
2189        }
2190        return 0;
2191}
2192
2193int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2194                                struct ext4_xattr_info *i,
2195                                struct ext4_xattr_ibody_find *is)
2196{
2197        struct ext4_xattr_ibody_header *header;
2198        struct ext4_xattr_search *s = &is->s;
2199        int error;
2200
2201        if (EXT4_I(inode)->i_extra_isize == 0)
2202                return -ENOSPC;
2203        error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2204        if (error)
2205                return error;
2206        header = IHDR(inode, ext4_raw_inode(&is->iloc));
2207        if (!IS_LAST_ENTRY(s->first)) {
2208                header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2209                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2210        } else {
2211                header->h_magic = cpu_to_le32(0);
2212                ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2213        }
2214        return 0;
2215}
2216
2217static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2218                                 struct ext4_xattr_info *i)
2219{
2220        void *value;
2221
2222        /* When e_value_inum is set the value is stored externally. */
2223        if (s->here->e_value_inum)
2224                return 0;
2225        if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2226                return 0;
2227        value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2228        return !memcmp(value, i->value, i->value_len);
2229}
2230
2231static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2232{
2233        struct buffer_head *bh;
2234        int error;
2235
2236        if (!EXT4_I(inode)->i_file_acl)
2237                return NULL;
2238        bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2239        if (IS_ERR(bh))
2240                return bh;
2241        error = ext4_xattr_check_block(inode, bh);
2242        if (error) {
2243                brelse(bh);
2244                return ERR_PTR(error);
2245        }
2246        return bh;
2247}
2248
2249/*
2250 * ext4_xattr_set_handle()
2251 *
2252 * Create, replace or remove an extended attribute for this inode.  Value
2253 * is NULL to remove an existing extended attribute, and non-NULL to
2254 * either replace an existing extended attribute, or create a new extended
2255 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2256 * specify that an extended attribute must exist and must not exist
2257 * previous to the call, respectively.
2258 *
2259 * Returns 0, or a negative error number on failure.
2260 */
2261int
2262ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2263                      const char *name, const void *value, size_t value_len,
2264                      int flags)
2265{
2266        struct ext4_xattr_info i = {
2267                .name_index = name_index,
2268                .name = name,
2269                .value = value,
2270                .value_len = value_len,
2271                .in_inode = 0,
2272        };
2273        struct ext4_xattr_ibody_find is = {
2274                .s = { .not_found = -ENODATA, },
2275        };
2276        struct ext4_xattr_block_find bs = {
2277                .s = { .not_found = -ENODATA, },
2278        };
2279        int no_expand;
2280        int error;
2281
2282        if (!name)
2283                return -EINVAL;
2284        if (strlen(name) > 255)
2285                return -ERANGE;
2286
2287        ext4_write_lock_xattr(inode, &no_expand);
2288
2289        /* Check journal credits under write lock. */
2290        if (ext4_handle_valid(handle)) {
2291                struct buffer_head *bh;
2292                int credits;
2293
2294                bh = ext4_xattr_get_block(inode);
2295                if (IS_ERR(bh)) {
2296                        error = PTR_ERR(bh);
2297                        goto cleanup;
2298                }
2299
2300                credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2301                                                   value_len,
2302                                                   flags & XATTR_CREATE);
2303                brelse(bh);
2304
2305                if (jbd2_handle_buffer_credits(handle) < credits) {
2306                        error = -ENOSPC;
2307                        goto cleanup;
2308                }
2309                WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
2310        }
2311
2312        error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2313        if (error)
2314                goto cleanup;
2315
2316        if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2317                struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2318                memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2319                ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2320        }
2321
2322        error = ext4_xattr_ibody_find(inode, &i, &is);
2323        if (error)
2324                goto cleanup;
2325        if (is.s.not_found)
2326                error = ext4_xattr_block_find(inode, &i, &bs);
2327        if (error)
2328                goto cleanup;
2329        if (is.s.not_found && bs.s.not_found) {
2330                error = -ENODATA;
2331                if (flags & XATTR_REPLACE)
2332                        goto cleanup;
2333                error = 0;
2334                if (!value)
2335                        goto cleanup;
2336        } else {
2337                error = -EEXIST;
2338                if (flags & XATTR_CREATE)
2339                        goto cleanup;
2340        }
2341
2342        if (!value) {
2343                if (!is.s.not_found)
2344                        error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2345                else if (!bs.s.not_found)
2346                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2347        } else {
2348                error = 0;
2349                /* Xattr value did not change? Save us some work and bail out */
2350                if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2351                        goto cleanup;
2352                if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2353                        goto cleanup;
2354
2355                if (ext4_has_feature_ea_inode(inode->i_sb) &&
2356                    (EXT4_XATTR_SIZE(i.value_len) >
2357                        EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2358                        i.in_inode = 1;
2359retry_inode:
2360                error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2361                if (!error && !bs.s.not_found) {
2362                        i.value = NULL;
2363                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2364                } else if (error == -ENOSPC) {
2365                        if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2366                                brelse(bs.bh);
2367                                bs.bh = NULL;
2368                                error = ext4_xattr_block_find(inode, &i, &bs);
2369                                if (error)
2370                                        goto cleanup;
2371                        }
2372                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2373                        if (!error && !is.s.not_found) {
2374                                i.value = NULL;
2375                                error = ext4_xattr_ibody_set(handle, inode, &i,
2376                                                             &is);
2377                        } else if (error == -ENOSPC) {
2378                                /*
2379                                 * Xattr does not fit in the block, store at
2380                                 * external inode if possible.
2381                                 */
2382                                if (ext4_has_feature_ea_inode(inode->i_sb) &&
2383                                    i.value_len && !i.in_inode) {
2384                                        i.in_inode = 1;
2385                                        goto retry_inode;
2386                                }
2387                        }
2388                }
2389        }
2390        if (!error) {
2391                ext4_xattr_update_super_block(handle, inode->i_sb);
2392                inode->i_ctime = current_time(inode);
2393                if (!value)
2394                        no_expand = 0;
2395                error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2396                /*
2397                 * The bh is consumed by ext4_mark_iloc_dirty, even with
2398                 * error != 0.
2399                 */
2400                is.iloc.bh = NULL;
2401                if (IS_SYNC(inode))
2402                        ext4_handle_sync(handle);
2403        }
2404        ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR);
2405
2406cleanup:
2407        brelse(is.iloc.bh);
2408        brelse(bs.bh);
2409        ext4_write_unlock_xattr(inode, &no_expand);
2410        return error;
2411}
2412
2413int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2414                           bool is_create, int *credits)
2415{
2416        struct buffer_head *bh;
2417        int err;
2418
2419        *credits = 0;
2420
2421        if (!EXT4_SB(inode->i_sb)->s_journal)
2422                return 0;
2423
2424        down_read(&EXT4_I(inode)->xattr_sem);
2425
2426        bh = ext4_xattr_get_block(inode);
2427        if (IS_ERR(bh)) {
2428                err = PTR_ERR(bh);
2429        } else {
2430                *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2431                                                    value_len, is_create);
2432                brelse(bh);
2433                err = 0;
2434        }
2435
2436        up_read(&EXT4_I(inode)->xattr_sem);
2437        return err;
2438}
2439
2440/*
2441 * ext4_xattr_set()
2442 *
2443 * Like ext4_xattr_set_handle, but start from an inode. This extended
2444 * attribute modification is a filesystem transaction by itself.
2445 *
2446 * Returns 0, or a negative error number on failure.
2447 */
2448int
2449ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2450               const void *value, size_t value_len, int flags)
2451{
2452        handle_t *handle;
2453        struct super_block *sb = inode->i_sb;
2454        int error, retries = 0;
2455        int credits;
2456
2457        error = dquot_initialize(inode);
2458        if (error)
2459                return error;
2460
2461retry:
2462        error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2463                                       &credits);
2464        if (error)
2465                return error;
2466
2467        handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2468        if (IS_ERR(handle)) {
2469                error = PTR_ERR(handle);
2470        } else {
2471                int error2;
2472
2473                error = ext4_xattr_set_handle(handle, inode, name_index, name,
2474                                              value, value_len, flags);
2475                error2 = ext4_journal_stop(handle);
2476                if (error == -ENOSPC &&
2477                    ext4_should_retry_alloc(sb, &retries))
2478                        goto retry;
2479                if (error == 0)
2480                        error = error2;
2481        }
2482        ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR);
2483
2484        return error;
2485}
2486
2487/*
2488 * Shift the EA entries in the inode to create space for the increased
2489 * i_extra_isize.
2490 */
2491static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2492                                     int value_offs_shift, void *to,
2493                                     void *from, size_t n)
2494{
2495        struct ext4_xattr_entry *last = entry;
2496        int new_offs;
2497
2498        /* We always shift xattr headers further thus offsets get lower */
2499        BUG_ON(value_offs_shift > 0);
2500
2501        /* Adjust the value offsets of the entries */
2502        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2503                if (!last->e_value_inum && last->e_value_size) {
2504                        new_offs = le16_to_cpu(last->e_value_offs) +
2505                                                        value_offs_shift;
2506                        last->e_value_offs = cpu_to_le16(new_offs);
2507                }
2508        }
2509        /* Shift the entries by n bytes */
2510        memmove(to, from, n);
2511}
2512
2513/*
2514 * Move xattr pointed to by 'entry' from inode into external xattr block
2515 */
2516static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2517                                    struct ext4_inode *raw_inode,
2518                                    struct ext4_xattr_entry *entry)
2519{
2520        struct ext4_xattr_ibody_find *is = NULL;
2521        struct ext4_xattr_block_find *bs = NULL;
2522        char *buffer = NULL, *b_entry_name = NULL;
2523        size_t value_size = le32_to_cpu(entry->e_value_size);
2524        struct ext4_xattr_info i = {
2525                .value = NULL,
2526                .value_len = 0,
2527                .name_index = entry->e_name_index,
2528                .in_inode = !!entry->e_value_inum,
2529        };
2530        struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2531        int error;
2532
2533        is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2534        bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2535        buffer = kmalloc(value_size, GFP_NOFS);
2536        b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2537        if (!is || !bs || !buffer || !b_entry_name) {
2538                error = -ENOMEM;
2539                goto out;
2540        }
2541
2542        is->s.not_found = -ENODATA;
2543        bs->s.not_found = -ENODATA;
2544        is->iloc.bh = NULL;
2545        bs->bh = NULL;
2546
2547        /* Save the entry name and the entry value */
2548        if (entry->e_value_inum) {
2549                error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2550                if (error)
2551                        goto out;
2552        } else {
2553                size_t value_offs = le16_to_cpu(entry->e_value_offs);
2554                memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2555        }
2556
2557        memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2558        b_entry_name[entry->e_name_len] = '\0';
2559        i.name = b_entry_name;
2560
2561        error = ext4_get_inode_loc(inode, &is->iloc);
2562        if (error)
2563                goto out;
2564
2565        error = ext4_xattr_ibody_find(inode, &i, is);
2566        if (error)
2567                goto out;
2568
2569        /* Remove the chosen entry from the inode */
2570        error = ext4_xattr_ibody_set(handle, inode, &i, is);
2571        if (error)
2572                goto out;
2573
2574        i.value = buffer;
2575        i.value_len = value_size;
2576        error = ext4_xattr_block_find(inode, &i, bs);
2577        if (error)
2578                goto out;
2579
2580        /* Add entry which was removed from the inode into the block */
2581        error = ext4_xattr_block_set(handle, inode, &i, bs);
2582        if (error)
2583                goto out;
2584        error = 0;
2585out:
2586        kfree(b_entry_name);
2587        kfree(buffer);
2588        if (is)
2589                brelse(is->iloc.bh);
2590        if (bs)
2591                brelse(bs->bh);
2592        kfree(is);
2593        kfree(bs);
2594
2595        return error;
2596}
2597
2598static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2599                                       struct ext4_inode *raw_inode,
2600                                       int isize_diff, size_t ifree,
2601                                       size_t bfree, int *total_ino)
2602{
2603        struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2604        struct ext4_xattr_entry *small_entry;
2605        struct ext4_xattr_entry *entry;
2606        struct ext4_xattr_entry *last;
2607        unsigned int entry_size;        /* EA entry size */
2608        unsigned int total_size;        /* EA entry size + value size */
2609        unsigned int min_total_size;
2610        int error;
2611
2612        while (isize_diff > ifree) {
2613                entry = NULL;
2614                small_entry = NULL;
2615                min_total_size = ~0U;
2616                last = IFIRST(header);
2617                /* Find the entry best suited to be pushed into EA block */
2618                for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2619                        /* never move system.data out of the inode */
2620                        if ((last->e_name_len == 4) &&
2621                            (last->e_name_index == EXT4_XATTR_INDEX_SYSTEM) &&
2622                            !memcmp(last->e_name, "data", 4))
2623                                continue;
2624                        total_size = EXT4_XATTR_LEN(last->e_name_len);
2625                        if (!last->e_value_inum)
2626                                total_size += EXT4_XATTR_SIZE(
2627                                               le32_to_cpu(last->e_value_size));
2628                        if (total_size <= bfree &&
2629                            total_size < min_total_size) {
2630                                if (total_size + ifree < isize_diff) {
2631                                        small_entry = last;
2632                                } else {
2633                                        entry = last;
2634                                        min_total_size = total_size;
2635                                }
2636                        }
2637                }
2638
2639                if (entry == NULL) {
2640                        if (small_entry == NULL)
2641                                return -ENOSPC;
2642                        entry = small_entry;
2643                }
2644
2645                entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2646                total_size = entry_size;
2647                if (!entry->e_value_inum)
2648                        total_size += EXT4_XATTR_SIZE(
2649                                              le32_to_cpu(entry->e_value_size));
2650                error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2651                                                 entry);
2652                if (error)
2653                        return error;
2654
2655                *total_ino -= entry_size;
2656                ifree += total_size;
2657                bfree -= total_size;
2658        }
2659
2660        return 0;
2661}
2662
2663/*
2664 * Expand an inode by new_extra_isize bytes when EAs are present.
2665 * Returns 0 on success or negative error number on failure.
2666 */
2667int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2668                               struct ext4_inode *raw_inode, handle_t *handle)
2669{
2670        struct ext4_xattr_ibody_header *header;
2671        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2672        static unsigned int mnt_count;
2673        size_t min_offs;
2674        size_t ifree, bfree;
2675        int total_ino;
2676        void *base, *end;
2677        int error = 0, tried_min_extra_isize = 0;
2678        int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2679        int isize_diff; /* How much do we need to grow i_extra_isize */
2680
2681retry:
2682        isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2683        if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2684                return 0;
2685
2686        header = IHDR(inode, raw_inode);
2687
2688        /*
2689         * Check if enough free space is available in the inode to shift the
2690         * entries ahead by new_extra_isize.
2691         */
2692
2693        base = IFIRST(header);
2694        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2695        min_offs = end - base;
2696        total_ino = sizeof(struct ext4_xattr_ibody_header) + sizeof(u32);
2697
2698        error = xattr_check_inode(inode, header, end);
2699        if (error)
2700                goto cleanup;
2701
2702        ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2703        if (ifree >= isize_diff)
2704                goto shift;
2705
2706        /*
2707         * Enough free space isn't available in the inode, check if
2708         * EA block can hold new_extra_isize bytes.
2709         */
2710        if (EXT4_I(inode)->i_file_acl) {
2711                struct buffer_head *bh;
2712
2713                bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2714                if (IS_ERR(bh)) {
2715                        error = PTR_ERR(bh);
2716                        goto cleanup;
2717                }
2718                error = ext4_xattr_check_block(inode, bh);
2719                if (error) {
2720                        brelse(bh);
2721                        goto cleanup;
2722                }
2723                base = BHDR(bh);
2724                end = bh->b_data + bh->b_size;
2725                min_offs = end - base;
2726                bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2727                                              NULL);
2728                brelse(bh);
2729                if (bfree + ifree < isize_diff) {
2730                        if (!tried_min_extra_isize && s_min_extra_isize) {
2731                                tried_min_extra_isize++;
2732                                new_extra_isize = s_min_extra_isize;
2733                                goto retry;
2734                        }
2735                        error = -ENOSPC;
2736                        goto cleanup;
2737                }
2738        } else {
2739                bfree = inode->i_sb->s_blocksize;
2740        }
2741
2742        error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2743                                            isize_diff, ifree, bfree,
2744                                            &total_ino);
2745        if (error) {
2746                if (error == -ENOSPC && !tried_min_extra_isize &&
2747                    s_min_extra_isize) {
2748                        tried_min_extra_isize++;
2749                        new_extra_isize = s_min_extra_isize;
2750                        goto retry;
2751                }
2752                goto cleanup;
2753        }
2754shift:
2755        /* Adjust the offsets and shift the remaining entries ahead */
2756        ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2757                        - new_extra_isize, (void *)raw_inode +
2758                        EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2759                        (void *)header, total_ino);
2760        EXT4_I(inode)->i_extra_isize = new_extra_isize;
2761
2762cleanup:
2763        if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2764                ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2765                             inode->i_ino);
2766                mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2767        }
2768        return error;
2769}
2770
2771#define EIA_INCR 16 /* must be 2^n */
2772#define EIA_MASK (EIA_INCR - 1)
2773
2774/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2775 * If @ea_inode_array is new or full it will be grown and the old
2776 * contents copied over.
2777 */
2778static int
2779ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2780                        struct inode *inode)
2781{
2782        if (*ea_inode_array == NULL) {
2783                /*
2784                 * Start with 15 inodes, so it fits into a power-of-two size.
2785                 * If *ea_inode_array is NULL, this is essentially offsetof()
2786                 */
2787                (*ea_inode_array) =
2788                        kmalloc(offsetof(struct ext4_xattr_inode_array,
2789                                         inodes[EIA_MASK]),
2790                                GFP_NOFS);
2791                if (*ea_inode_array == NULL)
2792                        return -ENOMEM;
2793                (*ea_inode_array)->count = 0;
2794        } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2795                /* expand the array once all 15 + n * 16 slots are full */
2796                struct ext4_xattr_inode_array *new_array = NULL;
2797                int count = (*ea_inode_array)->count;
2798
2799                /* if new_array is NULL, this is essentially offsetof() */
2800                new_array = kmalloc(
2801                                offsetof(struct ext4_xattr_inode_array,
2802                                         inodes[count + EIA_INCR]),
2803                                GFP_NOFS);
2804                if (new_array == NULL)
2805                        return -ENOMEM;
2806                memcpy(new_array, *ea_inode_array,
2807                       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2808                kfree(*ea_inode_array);
2809                *ea_inode_array = new_array;
2810        }
2811        (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2812        return 0;
2813}
2814
2815/*
2816 * ext4_xattr_delete_inode()
2817 *
2818 * Free extended attribute resources associated with this inode. Traverse
2819 * all entries and decrement reference on any xattr inodes associated with this
2820 * inode. This is called immediately before an inode is freed. We have exclusive
2821 * access to the inode. If an orphan inode is deleted it will also release its
2822 * references on xattr block and xattr inodes.
2823 */
2824int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2825                            struct ext4_xattr_inode_array **ea_inode_array,
2826                            int extra_credits)
2827{
2828        struct buffer_head *bh = NULL;
2829        struct ext4_xattr_ibody_header *header;
2830        struct ext4_iloc iloc = { .bh = NULL };
2831        struct ext4_xattr_entry *entry;
2832        struct inode *ea_inode;
2833        int error;
2834
2835        error = ext4_journal_ensure_credits(handle, extra_credits,
2836                        ext4_free_metadata_revoke_credits(inode->i_sb, 1));
2837        if (error < 0) {
2838                EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2839                goto cleanup;
2840        }
2841
2842        if (ext4_has_feature_ea_inode(inode->i_sb) &&
2843            ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2844
2845                error = ext4_get_inode_loc(inode, &iloc);
2846                if (error) {
2847                        EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2848                        goto cleanup;
2849                }
2850
2851                error = ext4_journal_get_write_access(handle, iloc.bh);
2852                if (error) {
2853                        EXT4_ERROR_INODE(inode, "write access (error %d)",
2854                                         error);
2855                        goto cleanup;
2856                }
2857
2858                header = IHDR(inode, ext4_raw_inode(&iloc));
2859                if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2860                        ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2861                                                     IFIRST(header),
2862                                                     false /* block_csum */,
2863                                                     ea_inode_array,
2864                                                     extra_credits,
2865                                                     false /* skip_quota */);
2866        }
2867
2868        if (EXT4_I(inode)->i_file_acl) {
2869                bh = ext4_sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl, REQ_PRIO);
2870                if (IS_ERR(bh)) {
2871                        error = PTR_ERR(bh);
2872                        if (error == -EIO) {
2873                                EXT4_ERROR_INODE_ERR(inode, EIO,
2874                                                     "block %llu read error",
2875                                                     EXT4_I(inode)->i_file_acl);
2876                        }
2877                        bh = NULL;
2878                        goto cleanup;
2879                }
2880                error = ext4_xattr_check_block(inode, bh);
2881                if (error)
2882                        goto cleanup;
2883
2884                if (ext4_has_feature_ea_inode(inode->i_sb)) {
2885                        for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2886                             entry = EXT4_XATTR_NEXT(entry)) {
2887                                if (!entry->e_value_inum)
2888                                        continue;
2889                                error = ext4_xattr_inode_iget(inode,
2890                                              le32_to_cpu(entry->e_value_inum),
2891                                              le32_to_cpu(entry->e_hash),
2892                                              &ea_inode);
2893                                if (error)
2894                                        continue;
2895                                ext4_xattr_inode_free_quota(inode, ea_inode,
2896                                              le32_to_cpu(entry->e_value_size));
2897                                iput(ea_inode);
2898                        }
2899
2900                }
2901
2902                ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2903                                         extra_credits);
2904                /*
2905                 * Update i_file_acl value in the same transaction that releases
2906                 * block.
2907                 */
2908                EXT4_I(inode)->i_file_acl = 0;
2909                error = ext4_mark_inode_dirty(handle, inode);
2910                if (error) {
2911                        EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2912                                         error);
2913                        goto cleanup;
2914                }
2915                ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_XATTR);
2916        }
2917        error = 0;
2918cleanup:
2919        brelse(iloc.bh);
2920        brelse(bh);
2921        return error;
2922}
2923
2924void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2925{
2926        int idx;
2927
2928        if (ea_inode_array == NULL)
2929                return;
2930
2931        for (idx = 0; idx < ea_inode_array->count; ++idx)
2932                iput(ea_inode_array->inodes[idx]);
2933        kfree(ea_inode_array);
2934}
2935
2936/*
2937 * ext4_xattr_block_cache_insert()
2938 *
2939 * Create a new entry in the extended attribute block cache, and insert
2940 * it unless such an entry is already in the cache.
2941 *
2942 * Returns 0, or a negative error number on failure.
2943 */
2944static void
2945ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2946                              struct buffer_head *bh)
2947{
2948        struct ext4_xattr_header *header = BHDR(bh);
2949        __u32 hash = le32_to_cpu(header->h_hash);
2950        int reusable = le32_to_cpu(header->h_refcount) <
2951                       EXT4_XATTR_REFCOUNT_MAX;
2952        int error;
2953
2954        if (!ea_block_cache)
2955                return;
2956        error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2957                                      bh->b_blocknr, reusable);
2958        if (error) {
2959                if (error == -EBUSY)
2960                        ea_bdebug(bh, "already in cache");
2961        } else
2962                ea_bdebug(bh, "inserting [%x]", (int)hash);
2963}
2964
2965/*
2966 * ext4_xattr_cmp()
2967 *
2968 * Compare two extended attribute blocks for equality.
2969 *
2970 * Returns 0 if the blocks are equal, 1 if they differ, and
2971 * a negative error number on errors.
2972 */
2973static int
2974ext4_xattr_cmp(struct ext4_xattr_header *header1,
2975               struct ext4_xattr_header *header2)
2976{
2977        struct ext4_xattr_entry *entry1, *entry2;
2978
2979        entry1 = ENTRY(header1+1);
2980        entry2 = ENTRY(header2+1);
2981        while (!IS_LAST_ENTRY(entry1)) {
2982                if (IS_LAST_ENTRY(entry2))
2983                        return 1;
2984                if (entry1->e_hash != entry2->e_hash ||
2985                    entry1->e_name_index != entry2->e_name_index ||
2986                    entry1->e_name_len != entry2->e_name_len ||
2987                    entry1->e_value_size != entry2->e_value_size ||
2988                    entry1->e_value_inum != entry2->e_value_inum ||
2989                    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
2990                        return 1;
2991                if (!entry1->e_value_inum &&
2992                    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
2993                           (char *)header2 + le16_to_cpu(entry2->e_value_offs),
2994                           le32_to_cpu(entry1->e_value_size)))
2995                        return 1;
2996
2997                entry1 = EXT4_XATTR_NEXT(entry1);
2998                entry2 = EXT4_XATTR_NEXT(entry2);
2999        }
3000        if (!IS_LAST_ENTRY(entry2))
3001                return 1;
3002        return 0;
3003}
3004
3005/*
3006 * ext4_xattr_block_cache_find()
3007 *
3008 * Find an identical extended attribute block.
3009 *
3010 * Returns a pointer to the block found, or NULL if such a block was
3011 * not found or an error occurred.
3012 */
3013static struct buffer_head *
3014ext4_xattr_block_cache_find(struct inode *inode,
3015                            struct ext4_xattr_header *header,
3016                            struct mb_cache_entry **pce)
3017{
3018        __u32 hash = le32_to_cpu(header->h_hash);
3019        struct mb_cache_entry *ce;
3020        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3021
3022        if (!ea_block_cache)
3023                return NULL;
3024        if (!header->h_hash)
3025                return NULL;  /* never share */
3026        ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3027        ce = mb_cache_entry_find_first(ea_block_cache, hash);
3028        while (ce) {
3029                struct buffer_head *bh;
3030
3031                bh = ext4_sb_bread(inode->i_sb, ce->e_value, REQ_PRIO);
3032                if (IS_ERR(bh)) {
3033                        if (PTR_ERR(bh) == -ENOMEM)
3034                                return NULL;
3035                        bh = NULL;
3036                        EXT4_ERROR_INODE(inode, "block %lu read error",
3037                                         (unsigned long)ce->e_value);
3038                } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3039                        *pce = ce;
3040                        return bh;
3041                }
3042                brelse(bh);
3043                ce = mb_cache_entry_find_next(ea_block_cache, ce);
3044        }
3045        return NULL;
3046}
3047
3048#define NAME_HASH_SHIFT 5
3049#define VALUE_HASH_SHIFT 16
3050
3051/*
3052 * ext4_xattr_hash_entry()
3053 *
3054 * Compute the hash of an extended attribute.
3055 */
3056static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3057                                    size_t value_count)
3058{
3059        __u32 hash = 0;
3060
3061        while (name_len--) {
3062                hash = (hash << NAME_HASH_SHIFT) ^
3063                       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3064                       *name++;
3065        }
3066        while (value_count--) {
3067                hash = (hash << VALUE_HASH_SHIFT) ^
3068                       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3069                       le32_to_cpu(*value++);
3070        }
3071        return cpu_to_le32(hash);
3072}
3073
3074#undef NAME_HASH_SHIFT
3075#undef VALUE_HASH_SHIFT
3076
3077#define BLOCK_HASH_SHIFT 16
3078
3079/*
3080 * ext4_xattr_rehash()
3081 *
3082 * Re-compute the extended attribute hash value after an entry has changed.
3083 */
3084static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3085{
3086        struct ext4_xattr_entry *here;
3087        __u32 hash = 0;
3088
3089        here = ENTRY(header+1);
3090        while (!IS_LAST_ENTRY(here)) {
3091                if (!here->e_hash) {
3092                        /* Block is not shared if an entry's hash value == 0 */
3093                        hash = 0;
3094                        break;
3095                }
3096                hash = (hash << BLOCK_HASH_SHIFT) ^
3097                       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3098                       le32_to_cpu(here->e_hash);
3099                here = EXT4_XATTR_NEXT(here);
3100        }
3101        header->h_hash = cpu_to_le32(hash);
3102}
3103
3104#undef BLOCK_HASH_SHIFT
3105
3106#define HASH_BUCKET_BITS        10
3107
3108struct mb_cache *
3109ext4_xattr_create_cache(void)
3110{
3111        return mb_cache_create(HASH_BUCKET_BITS);
3112}
3113
3114void ext4_xattr_destroy_cache(struct mb_cache *cache)
3115{
3116        if (cache)
3117                mb_cache_destroy(cache);
3118}
3119
3120