linux/fs/ext4/xattr.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/fs/ext4/xattr.c
   4 *
   5 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
   6 *
   7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
   8 * Ext4 code with a lot of help from Eric Jarman <ejarman@acm.org>.
   9 * Extended attributes for symlinks and special files added per
  10 *  suggestion of Luka Renko <luka.renko@hermes.si>.
  11 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
  12 *  Red Hat Inc.
  13 * ea-in-inode support by Alex Tomas <alex@clusterfs.com> aka bzzz
  14 *  and Andreas Gruenbacher <agruen@suse.de>.
  15 */
  16
  17/*
  18 * Extended attributes are stored directly in inodes (on file systems with
  19 * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
  20 * field contains the block number if an inode uses an additional block. All
  21 * attributes must fit in the inode and one additional block. Blocks that
  22 * contain the identical set of attributes may be shared among several inodes.
  23 * Identical blocks are detected by keeping a cache of blocks that have
  24 * recently been accessed.
  25 *
  26 * The attributes in inodes and on blocks have a different header; the entries
  27 * are stored in the same format:
  28 *
  29 *   +------------------+
  30 *   | header           |
  31 *   | entry 1          | |
  32 *   | entry 2          | | growing downwards
  33 *   | entry 3          | v
  34 *   | four null bytes  |
  35 *   | . . .            |
  36 *   | value 1          | ^
  37 *   | value 3          | | growing upwards
  38 *   | value 2          | |
  39 *   +------------------+
  40 *
  41 * The header is followed by multiple entry descriptors. In disk blocks, the
  42 * entry descriptors are kept sorted. In inodes, they are unsorted. The
  43 * attribute values are aligned to the end of the block in no specific order.
  44 *
  45 * Locking strategy
  46 * ----------------
  47 * EXT4_I(inode)->i_file_acl is protected by EXT4_I(inode)->xattr_sem.
  48 * EA blocks are only changed if they are exclusive to an inode, so
  49 * holding xattr_sem also means that nothing but the EA block's reference
  50 * count can change. Multiple writers to the same block are synchronized
  51 * by the buffer lock.
  52 */
  53
  54#include <linux/init.h>
  55#include <linux/fs.h>
  56#include <linux/slab.h>
  57#include <linux/mbcache.h>
  58#include <linux/quotaops.h>
  59#include <linux/iversion.h>
  60#include "ext4_jbd2.h"
  61#include "ext4.h"
  62#include "xattr.h"
  63#include "acl.h"
  64
  65#ifdef EXT4_XATTR_DEBUG
  66# define ea_idebug(inode, fmt, ...)                                     \
  67        printk(KERN_DEBUG "inode %s:%lu: " fmt "\n",                    \
  68               inode->i_sb->s_id, inode->i_ino, ##__VA_ARGS__)
  69# define ea_bdebug(bh, fmt, ...)                                        \
  70        printk(KERN_DEBUG "block %pg:%lu: " fmt "\n",                   \
  71               bh->b_bdev, (unsigned long)bh->b_blocknr, ##__VA_ARGS__)
  72#else
  73# define ea_idebug(inode, fmt, ...)     no_printk(fmt, ##__VA_ARGS__)
  74# define ea_bdebug(bh, fmt, ...)        no_printk(fmt, ##__VA_ARGS__)
  75#endif
  76
  77static void ext4_xattr_block_cache_insert(struct mb_cache *,
  78                                          struct buffer_head *);
  79static struct buffer_head *
  80ext4_xattr_block_cache_find(struct inode *, struct ext4_xattr_header *,
  81                            struct mb_cache_entry **);
  82static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
  83                                    size_t value_count);
  84static void ext4_xattr_rehash(struct ext4_xattr_header *);
  85
  86static const struct xattr_handler * const ext4_xattr_handler_map[] = {
  87        [EXT4_XATTR_INDEX_USER]              = &ext4_xattr_user_handler,
  88#ifdef CONFIG_EXT4_FS_POSIX_ACL
  89        [EXT4_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
  90        [EXT4_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
  91#endif
  92        [EXT4_XATTR_INDEX_TRUSTED]           = &ext4_xattr_trusted_handler,
  93#ifdef CONFIG_EXT4_FS_SECURITY
  94        [EXT4_XATTR_INDEX_SECURITY]          = &ext4_xattr_security_handler,
  95#endif
  96};
  97
  98const struct xattr_handler *ext4_xattr_handlers[] = {
  99        &ext4_xattr_user_handler,
 100        &ext4_xattr_trusted_handler,
 101#ifdef CONFIG_EXT4_FS_POSIX_ACL
 102        &posix_acl_access_xattr_handler,
 103        &posix_acl_default_xattr_handler,
 104#endif
 105#ifdef CONFIG_EXT4_FS_SECURITY
 106        &ext4_xattr_security_handler,
 107#endif
 108        NULL
 109};
 110
 111#define EA_BLOCK_CACHE(inode)   (((struct ext4_sb_info *) \
 112                                inode->i_sb->s_fs_info)->s_ea_block_cache)
 113
 114#define EA_INODE_CACHE(inode)   (((struct ext4_sb_info *) \
 115                                inode->i_sb->s_fs_info)->s_ea_inode_cache)
 116
 117static int
 118ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
 119                        struct inode *inode);
 120
 121#ifdef CONFIG_LOCKDEP
 122void ext4_xattr_inode_set_class(struct inode *ea_inode)
 123{
 124        lockdep_set_subclass(&ea_inode->i_rwsem, 1);
 125}
 126#endif
 127
 128static __le32 ext4_xattr_block_csum(struct inode *inode,
 129                                    sector_t block_nr,
 130                                    struct ext4_xattr_header *hdr)
 131{
 132        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 133        __u32 csum;
 134        __le64 dsk_block_nr = cpu_to_le64(block_nr);
 135        __u32 dummy_csum = 0;
 136        int offset = offsetof(struct ext4_xattr_header, h_checksum);
 137
 138        csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&dsk_block_nr,
 139                           sizeof(dsk_block_nr));
 140        csum = ext4_chksum(sbi, csum, (__u8 *)hdr, offset);
 141        csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, sizeof(dummy_csum));
 142        offset += sizeof(dummy_csum);
 143        csum = ext4_chksum(sbi, csum, (__u8 *)hdr + offset,
 144                           EXT4_BLOCK_SIZE(inode->i_sb) - offset);
 145
 146        return cpu_to_le32(csum);
 147}
 148
 149static int ext4_xattr_block_csum_verify(struct inode *inode,
 150                                        struct buffer_head *bh)
 151{
 152        struct ext4_xattr_header *hdr = BHDR(bh);
 153        int ret = 1;
 154
 155        if (ext4_has_metadata_csum(inode->i_sb)) {
 156                lock_buffer(bh);
 157                ret = (hdr->h_checksum == ext4_xattr_block_csum(inode,
 158                                                        bh->b_blocknr, hdr));
 159                unlock_buffer(bh);
 160        }
 161        return ret;
 162}
 163
 164static void ext4_xattr_block_csum_set(struct inode *inode,
 165                                      struct buffer_head *bh)
 166{
 167        if (ext4_has_metadata_csum(inode->i_sb))
 168                BHDR(bh)->h_checksum = ext4_xattr_block_csum(inode,
 169                                                bh->b_blocknr, BHDR(bh));
 170}
 171
 172static inline const struct xattr_handler *
 173ext4_xattr_handler(int name_index)
 174{
 175        const struct xattr_handler *handler = NULL;
 176
 177        if (name_index > 0 && name_index < ARRAY_SIZE(ext4_xattr_handler_map))
 178                handler = ext4_xattr_handler_map[name_index];
 179        return handler;
 180}
 181
 182static int
 183ext4_xattr_check_entries(struct ext4_xattr_entry *entry, void *end,
 184                         void *value_start)
 185{
 186        struct ext4_xattr_entry *e = entry;
 187
 188        /* Find the end of the names list */
 189        while (!IS_LAST_ENTRY(e)) {
 190                struct ext4_xattr_entry *next = EXT4_XATTR_NEXT(e);
 191                if ((void *)next >= end)
 192                        return -EFSCORRUPTED;
 193                e = next;
 194        }
 195
 196        /* Check the values */
 197        while (!IS_LAST_ENTRY(entry)) {
 198                u32 size = le32_to_cpu(entry->e_value_size);
 199
 200                if (size > EXT4_XATTR_SIZE_MAX)
 201                        return -EFSCORRUPTED;
 202
 203                if (size != 0 && entry->e_value_inum == 0) {
 204                        u16 offs = le16_to_cpu(entry->e_value_offs);
 205                        void *value;
 206
 207                        /*
 208                         * The value cannot overlap the names, and the value
 209                         * with padding cannot extend beyond 'end'.  Check both
 210                         * the padded and unpadded sizes, since the size may
 211                         * overflow to 0 when adding padding.
 212                         */
 213                        if (offs > end - value_start)
 214                                return -EFSCORRUPTED;
 215                        value = value_start + offs;
 216                        if (value < (void *)e + sizeof(u32) ||
 217                            size > end - value ||
 218                            EXT4_XATTR_SIZE(size) > end - value)
 219                                return -EFSCORRUPTED;
 220                }
 221                entry = EXT4_XATTR_NEXT(entry);
 222        }
 223
 224        return 0;
 225}
 226
 227static inline int
 228__ext4_xattr_check_block(struct inode *inode, struct buffer_head *bh,
 229                         const char *function, unsigned int line)
 230{
 231        int error = -EFSCORRUPTED;
 232
 233        if (buffer_verified(bh))
 234                return 0;
 235
 236        if (BHDR(bh)->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC) ||
 237            BHDR(bh)->h_blocks != cpu_to_le32(1))
 238                goto errout;
 239        error = -EFSBADCRC;
 240        if (!ext4_xattr_block_csum_verify(inode, bh))
 241                goto errout;
 242        error = ext4_xattr_check_entries(BFIRST(bh), bh->b_data + bh->b_size,
 243                                         bh->b_data);
 244errout:
 245        if (error)
 246                __ext4_error_inode(inode, function, line, 0,
 247                                   "corrupted xattr block %llu",
 248                                   (unsigned long long) bh->b_blocknr);
 249        else
 250                set_buffer_verified(bh);
 251        return error;
 252}
 253
 254#define ext4_xattr_check_block(inode, bh) \
 255        __ext4_xattr_check_block((inode), (bh),  __func__, __LINE__)
 256
 257
 258static int
 259__xattr_check_inode(struct inode *inode, struct ext4_xattr_ibody_header *header,
 260                         void *end, const char *function, unsigned int line)
 261{
 262        int error = -EFSCORRUPTED;
 263
 264        if (end - (void *)header < sizeof(*header) + sizeof(u32) ||
 265            (header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)))
 266                goto errout;
 267        error = ext4_xattr_check_entries(IFIRST(header), end, IFIRST(header));
 268errout:
 269        if (error)
 270                __ext4_error_inode(inode, function, line, 0,
 271                                   "corrupted in-inode xattr");
 272        return error;
 273}
 274
 275#define xattr_check_inode(inode, header, end) \
 276        __xattr_check_inode((inode), (header), (end), __func__, __LINE__)
 277
 278static int
 279xattr_find_entry(struct inode *inode, struct ext4_xattr_entry **pentry,
 280                 void *end, int name_index, const char *name, int sorted)
 281{
 282        struct ext4_xattr_entry *entry, *next;
 283        size_t name_len;
 284        int cmp = 1;
 285
 286        if (name == NULL)
 287                return -EINVAL;
 288        name_len = strlen(name);
 289        for (entry = *pentry; !IS_LAST_ENTRY(entry); entry = next) {
 290                next = EXT4_XATTR_NEXT(entry);
 291                if ((void *) next >= end) {
 292                        EXT4_ERROR_INODE(inode, "corrupted xattr entries");
 293                        return -EFSCORRUPTED;
 294                }
 295                cmp = name_index - entry->e_name_index;
 296                if (!cmp)
 297                        cmp = name_len - entry->e_name_len;
 298                if (!cmp)
 299                        cmp = memcmp(name, entry->e_name, name_len);
 300                if (cmp <= 0 && (sorted || cmp == 0))
 301                        break;
 302        }
 303        *pentry = entry;
 304        return cmp ? -ENODATA : 0;
 305}
 306
 307static u32
 308ext4_xattr_inode_hash(struct ext4_sb_info *sbi, const void *buffer, size_t size)
 309{
 310        return ext4_chksum(sbi, sbi->s_csum_seed, buffer, size);
 311}
 312
 313static u64 ext4_xattr_inode_get_ref(struct inode *ea_inode)
 314{
 315        return ((u64)ea_inode->i_ctime.tv_sec << 32) |
 316                (u32) inode_peek_iversion_raw(ea_inode);
 317}
 318
 319static void ext4_xattr_inode_set_ref(struct inode *ea_inode, u64 ref_count)
 320{
 321        ea_inode->i_ctime.tv_sec = (u32)(ref_count >> 32);
 322        inode_set_iversion_raw(ea_inode, ref_count & 0xffffffff);
 323}
 324
 325static u32 ext4_xattr_inode_get_hash(struct inode *ea_inode)
 326{
 327        return (u32)ea_inode->i_atime.tv_sec;
 328}
 329
 330static void ext4_xattr_inode_set_hash(struct inode *ea_inode, u32 hash)
 331{
 332        ea_inode->i_atime.tv_sec = hash;
 333}
 334
 335/*
 336 * Read the EA value from an inode.
 337 */
 338static int ext4_xattr_inode_read(struct inode *ea_inode, void *buf, size_t size)
 339{
 340        int blocksize = 1 << ea_inode->i_blkbits;
 341        int bh_count = (size + blocksize - 1) >> ea_inode->i_blkbits;
 342        int tail_size = (size % blocksize) ?: blocksize;
 343        struct buffer_head *bhs_inline[8];
 344        struct buffer_head **bhs = bhs_inline;
 345        int i, ret;
 346
 347        if (bh_count > ARRAY_SIZE(bhs_inline)) {
 348                bhs = kmalloc_array(bh_count, sizeof(*bhs), GFP_NOFS);
 349                if (!bhs)
 350                        return -ENOMEM;
 351        }
 352
 353        ret = ext4_bread_batch(ea_inode, 0 /* block */, bh_count,
 354                               true /* wait */, bhs);
 355        if (ret)
 356                goto free_bhs;
 357
 358        for (i = 0; i < bh_count; i++) {
 359                /* There shouldn't be any holes in ea_inode. */
 360                if (!bhs[i]) {
 361                        ret = -EFSCORRUPTED;
 362                        goto put_bhs;
 363                }
 364                memcpy((char *)buf + blocksize * i, bhs[i]->b_data,
 365                       i < bh_count - 1 ? blocksize : tail_size);
 366        }
 367        ret = 0;
 368put_bhs:
 369        for (i = 0; i < bh_count; i++)
 370                brelse(bhs[i]);
 371free_bhs:
 372        if (bhs != bhs_inline)
 373                kfree(bhs);
 374        return ret;
 375}
 376
 377#define EXT4_XATTR_INODE_GET_PARENT(inode) ((__u32)(inode)->i_mtime.tv_sec)
 378
 379static int ext4_xattr_inode_iget(struct inode *parent, unsigned long ea_ino,
 380                                 u32 ea_inode_hash, struct inode **ea_inode)
 381{
 382        struct inode *inode;
 383        int err;
 384
 385        inode = ext4_iget(parent->i_sb, ea_ino);
 386        if (IS_ERR(inode)) {
 387                err = PTR_ERR(inode);
 388                ext4_error(parent->i_sb,
 389                           "error while reading EA inode %lu err=%d", ea_ino,
 390                           err);
 391                return err;
 392        }
 393
 394        if (is_bad_inode(inode)) {
 395                ext4_error(parent->i_sb,
 396                           "error while reading EA inode %lu is_bad_inode",
 397                           ea_ino);
 398                err = -EIO;
 399                goto error;
 400        }
 401
 402        if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
 403                ext4_error(parent->i_sb,
 404                           "EA inode %lu does not have EXT4_EA_INODE_FL flag",
 405                            ea_ino);
 406                err = -EINVAL;
 407                goto error;
 408        }
 409
 410        ext4_xattr_inode_set_class(inode);
 411
 412        /*
 413         * Check whether this is an old Lustre-style xattr inode. Lustre
 414         * implementation does not have hash validation, rather it has a
 415         * backpointer from ea_inode to the parent inode.
 416         */
 417        if (ea_inode_hash != ext4_xattr_inode_get_hash(inode) &&
 418            EXT4_XATTR_INODE_GET_PARENT(inode) == parent->i_ino &&
 419            inode->i_generation == parent->i_generation) {
 420                ext4_set_inode_state(inode, EXT4_STATE_LUSTRE_EA_INODE);
 421                ext4_xattr_inode_set_ref(inode, 1);
 422        } else {
 423                inode_lock(inode);
 424                inode->i_flags |= S_NOQUOTA;
 425                inode_unlock(inode);
 426        }
 427
 428        *ea_inode = inode;
 429        return 0;
 430error:
 431        iput(inode);
 432        return err;
 433}
 434
 435static int
 436ext4_xattr_inode_verify_hashes(struct inode *ea_inode,
 437                               struct ext4_xattr_entry *entry, void *buffer,
 438                               size_t size)
 439{
 440        u32 hash;
 441
 442        /* Verify stored hash matches calculated hash. */
 443        hash = ext4_xattr_inode_hash(EXT4_SB(ea_inode->i_sb), buffer, size);
 444        if (hash != ext4_xattr_inode_get_hash(ea_inode))
 445                return -EFSCORRUPTED;
 446
 447        if (entry) {
 448                __le32 e_hash, tmp_data;
 449
 450                /* Verify entry hash. */
 451                tmp_data = cpu_to_le32(hash);
 452                e_hash = ext4_xattr_hash_entry(entry->e_name, entry->e_name_len,
 453                                               &tmp_data, 1);
 454                if (e_hash != entry->e_hash)
 455                        return -EFSCORRUPTED;
 456        }
 457        return 0;
 458}
 459
 460/*
 461 * Read xattr value from the EA inode.
 462 */
 463static int
 464ext4_xattr_inode_get(struct inode *inode, struct ext4_xattr_entry *entry,
 465                     void *buffer, size_t size)
 466{
 467        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
 468        struct inode *ea_inode;
 469        int err;
 470
 471        err = ext4_xattr_inode_iget(inode, le32_to_cpu(entry->e_value_inum),
 472                                    le32_to_cpu(entry->e_hash), &ea_inode);
 473        if (err) {
 474                ea_inode = NULL;
 475                goto out;
 476        }
 477
 478        if (i_size_read(ea_inode) != size) {
 479                ext4_warning_inode(ea_inode,
 480                                   "ea_inode file size=%llu entry size=%zu",
 481                                   i_size_read(ea_inode), size);
 482                err = -EFSCORRUPTED;
 483                goto out;
 484        }
 485
 486        err = ext4_xattr_inode_read(ea_inode, buffer, size);
 487        if (err)
 488                goto out;
 489
 490        if (!ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE)) {
 491                err = ext4_xattr_inode_verify_hashes(ea_inode, entry, buffer,
 492                                                     size);
 493                if (err) {
 494                        ext4_warning_inode(ea_inode,
 495                                           "EA inode hash validation failed");
 496                        goto out;
 497                }
 498
 499                if (ea_inode_cache)
 500                        mb_cache_entry_create(ea_inode_cache, GFP_NOFS,
 501                                        ext4_xattr_inode_get_hash(ea_inode),
 502                                        ea_inode->i_ino, true /* reusable */);
 503        }
 504out:
 505        iput(ea_inode);
 506        return err;
 507}
 508
 509static int
 510ext4_xattr_block_get(struct inode *inode, int name_index, const char *name,
 511                     void *buffer, size_t buffer_size)
 512{
 513        struct buffer_head *bh = NULL;
 514        struct ext4_xattr_entry *entry;
 515        size_t size;
 516        void *end;
 517        int error;
 518        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
 519
 520        ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
 521                  name_index, name, buffer, (long)buffer_size);
 522
 523        error = -ENODATA;
 524        if (!EXT4_I(inode)->i_file_acl)
 525                goto cleanup;
 526        ea_idebug(inode, "reading block %llu",
 527                  (unsigned long long)EXT4_I(inode)->i_file_acl);
 528        bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 529        if (!bh)
 530                goto cleanup;
 531        ea_bdebug(bh, "b_count=%d, refcount=%d",
 532                atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 533        error = ext4_xattr_check_block(inode, bh);
 534        if (error)
 535                goto cleanup;
 536        ext4_xattr_block_cache_insert(ea_block_cache, bh);
 537        entry = BFIRST(bh);
 538        end = bh->b_data + bh->b_size;
 539        error = xattr_find_entry(inode, &entry, end, name_index, name, 1);
 540        if (error)
 541                goto cleanup;
 542        size = le32_to_cpu(entry->e_value_size);
 543        error = -ERANGE;
 544        if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 545                goto cleanup;
 546        if (buffer) {
 547                if (size > buffer_size)
 548                        goto cleanup;
 549                if (entry->e_value_inum) {
 550                        error = ext4_xattr_inode_get(inode, entry, buffer,
 551                                                     size);
 552                        if (error)
 553                                goto cleanup;
 554                } else {
 555                        u16 offset = le16_to_cpu(entry->e_value_offs);
 556                        void *p = bh->b_data + offset;
 557
 558                        if (unlikely(p + size > end))
 559                                goto cleanup;
 560                        memcpy(buffer, p, size);
 561                }
 562        }
 563        error = size;
 564
 565cleanup:
 566        brelse(bh);
 567        return error;
 568}
 569
 570int
 571ext4_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
 572                     void *buffer, size_t buffer_size)
 573{
 574        struct ext4_xattr_ibody_header *header;
 575        struct ext4_xattr_entry *entry;
 576        struct ext4_inode *raw_inode;
 577        struct ext4_iloc iloc;
 578        size_t size;
 579        void *end;
 580        int error;
 581
 582        if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR))
 583                return -ENODATA;
 584        error = ext4_get_inode_loc(inode, &iloc);
 585        if (error)
 586                return error;
 587        raw_inode = ext4_raw_inode(&iloc);
 588        header = IHDR(inode, raw_inode);
 589        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 590        error = xattr_check_inode(inode, header, end);
 591        if (error)
 592                goto cleanup;
 593        entry = IFIRST(header);
 594        error = xattr_find_entry(inode, &entry, end, name_index, name, 0);
 595        if (error)
 596                goto cleanup;
 597        size = le32_to_cpu(entry->e_value_size);
 598        error = -ERANGE;
 599        if (unlikely(size > EXT4_XATTR_SIZE_MAX))
 600                goto cleanup;
 601        if (buffer) {
 602                if (size > buffer_size)
 603                        goto cleanup;
 604                if (entry->e_value_inum) {
 605                        error = ext4_xattr_inode_get(inode, entry, buffer,
 606                                                     size);
 607                        if (error)
 608                                goto cleanup;
 609                } else {
 610                        u16 offset = le16_to_cpu(entry->e_value_offs);
 611                        void *p = (void *)IFIRST(header) + offset;
 612
 613                        if (unlikely(p + size > end))
 614                                goto cleanup;
 615                        memcpy(buffer, p, size);
 616                }
 617        }
 618        error = size;
 619
 620cleanup:
 621        brelse(iloc.bh);
 622        return error;
 623}
 624
 625/*
 626 * ext4_xattr_get()
 627 *
 628 * Copy an extended attribute into the buffer
 629 * provided, or compute the buffer size required.
 630 * Buffer is NULL to compute the size of the buffer required.
 631 *
 632 * Returns a negative error number on failure, or the number of bytes
 633 * used / required on success.
 634 */
 635int
 636ext4_xattr_get(struct inode *inode, int name_index, const char *name,
 637               void *buffer, size_t buffer_size)
 638{
 639        int error;
 640
 641        if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 642                return -EIO;
 643
 644        if (strlen(name) > 255)
 645                return -ERANGE;
 646
 647        down_read(&EXT4_I(inode)->xattr_sem);
 648        error = ext4_xattr_ibody_get(inode, name_index, name, buffer,
 649                                     buffer_size);
 650        if (error == -ENODATA)
 651                error = ext4_xattr_block_get(inode, name_index, name, buffer,
 652                                             buffer_size);
 653        up_read(&EXT4_I(inode)->xattr_sem);
 654        return error;
 655}
 656
 657static int
 658ext4_xattr_list_entries(struct dentry *dentry, struct ext4_xattr_entry *entry,
 659                        char *buffer, size_t buffer_size)
 660{
 661        size_t rest = buffer_size;
 662
 663        for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry)) {
 664                const struct xattr_handler *handler =
 665                        ext4_xattr_handler(entry->e_name_index);
 666
 667                if (handler && (!handler->list || handler->list(dentry))) {
 668                        const char *prefix = handler->prefix ?: handler->name;
 669                        size_t prefix_len = strlen(prefix);
 670                        size_t size = prefix_len + entry->e_name_len + 1;
 671
 672                        if (buffer) {
 673                                if (size > rest)
 674                                        return -ERANGE;
 675                                memcpy(buffer, prefix, prefix_len);
 676                                buffer += prefix_len;
 677                                memcpy(buffer, entry->e_name, entry->e_name_len);
 678                                buffer += entry->e_name_len;
 679                                *buffer++ = 0;
 680                        }
 681                        rest -= size;
 682                }
 683        }
 684        return buffer_size - rest;  /* total size */
 685}
 686
 687static int
 688ext4_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
 689{
 690        struct inode *inode = d_inode(dentry);
 691        struct buffer_head *bh = NULL;
 692        int error;
 693
 694        ea_idebug(inode, "buffer=%p, buffer_size=%ld",
 695                  buffer, (long)buffer_size);
 696
 697        error = 0;
 698        if (!EXT4_I(inode)->i_file_acl)
 699                goto cleanup;
 700        ea_idebug(inode, "reading block %llu",
 701                  (unsigned long long)EXT4_I(inode)->i_file_acl);
 702        bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 703        error = -EIO;
 704        if (!bh)
 705                goto cleanup;
 706        ea_bdebug(bh, "b_count=%d, refcount=%d",
 707                atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
 708        error = ext4_xattr_check_block(inode, bh);
 709        if (error)
 710                goto cleanup;
 711        ext4_xattr_block_cache_insert(EA_BLOCK_CACHE(inode), bh);
 712        error = ext4_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
 713
 714cleanup:
 715        brelse(bh);
 716
 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                ext4_set_feature_xattr(sb);
 796                ext4_handle_dirty_super(handle, sb);
 797        }
 798}
 799
 800int ext4_get_inode_usage(struct inode *inode, qsize_t *usage)
 801{
 802        struct ext4_iloc iloc = { .bh = NULL };
 803        struct buffer_head *bh = NULL;
 804        struct ext4_inode *raw_inode;
 805        struct ext4_xattr_ibody_header *header;
 806        struct ext4_xattr_entry *entry;
 807        qsize_t ea_inode_refs = 0;
 808        void *end;
 809        int ret;
 810
 811        lockdep_assert_held_read(&EXT4_I(inode)->xattr_sem);
 812
 813        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
 814                ret = ext4_get_inode_loc(inode, &iloc);
 815                if (ret)
 816                        goto out;
 817                raw_inode = ext4_raw_inode(&iloc);
 818                header = IHDR(inode, raw_inode);
 819                end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
 820                ret = xattr_check_inode(inode, header, end);
 821                if (ret)
 822                        goto out;
 823
 824                for (entry = IFIRST(header); !IS_LAST_ENTRY(entry);
 825                     entry = EXT4_XATTR_NEXT(entry))
 826                        if (entry->e_value_inum)
 827                                ea_inode_refs++;
 828        }
 829
 830        if (EXT4_I(inode)->i_file_acl) {
 831                bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
 832                if (!bh) {
 833                        ret = -EIO;
 834                        goto out;
 835                }
 836
 837                ret = ext4_xattr_check_block(inode, bh);
 838                if (ret)
 839                        goto out;
 840
 841                for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
 842                     entry = EXT4_XATTR_NEXT(entry))
 843                        if (entry->e_value_inum)
 844                                ea_inode_refs++;
 845        }
 846        *usage = ea_inode_refs + 1;
 847        ret = 0;
 848out:
 849        brelse(iloc.bh);
 850        brelse(bh);
 851        return ret;
 852}
 853
 854static inline size_t round_up_cluster(struct inode *inode, size_t length)
 855{
 856        struct super_block *sb = inode->i_sb;
 857        size_t cluster_size = 1 << (EXT4_SB(sb)->s_cluster_bits +
 858                                    inode->i_blkbits);
 859        size_t mask = ~(cluster_size - 1);
 860
 861        return (length + cluster_size - 1) & mask;
 862}
 863
 864static int ext4_xattr_inode_alloc_quota(struct inode *inode, size_t len)
 865{
 866        int err;
 867
 868        err = dquot_alloc_inode(inode);
 869        if (err)
 870                return err;
 871        err = dquot_alloc_space_nodirty(inode, round_up_cluster(inode, len));
 872        if (err)
 873                dquot_free_inode(inode);
 874        return err;
 875}
 876
 877static void ext4_xattr_inode_free_quota(struct inode *parent,
 878                                        struct inode *ea_inode,
 879                                        size_t len)
 880{
 881        if (ea_inode &&
 882            ext4_test_inode_state(ea_inode, EXT4_STATE_LUSTRE_EA_INODE))
 883                return;
 884        dquot_free_space_nodirty(parent, round_up_cluster(parent, len));
 885        dquot_free_inode(parent);
 886}
 887
 888int __ext4_xattr_set_credits(struct super_block *sb, struct inode *inode,
 889                             struct buffer_head *block_bh, size_t value_len,
 890                             bool is_create)
 891{
 892        int credits;
 893        int blocks;
 894
 895        /*
 896         * 1) Owner inode update
 897         * 2) Ref count update on old xattr block
 898         * 3) new xattr block
 899         * 4) block bitmap update for new xattr block
 900         * 5) group descriptor for new xattr block
 901         * 6) block bitmap update for old xattr block
 902         * 7) group descriptor for old block
 903         *
 904         * 6 & 7 can happen if we have two racing threads T_a and T_b
 905         * which are each trying to set an xattr on inodes I_a and I_b
 906         * which were both initially sharing an xattr block.
 907         */
 908        credits = 7;
 909
 910        /* Quota updates. */
 911        credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(sb);
 912
 913        /*
 914         * In case of inline data, we may push out the data to a block,
 915         * so we need to reserve credits for this eventuality
 916         */
 917        if (inode && ext4_has_inline_data(inode))
 918                credits += ext4_writepage_trans_blocks(inode) + 1;
 919
 920        /* We are done if ea_inode feature is not enabled. */
 921        if (!ext4_has_feature_ea_inode(sb))
 922                return credits;
 923
 924        /* New ea_inode, inode map, block bitmap, group descriptor. */
 925        credits += 4;
 926
 927        /* Data blocks. */
 928        blocks = (value_len + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 929
 930        /* Indirection block or one level of extent tree. */
 931        blocks += 1;
 932
 933        /* Block bitmap and group descriptor updates for each block. */
 934        credits += blocks * 2;
 935
 936        /* Blocks themselves. */
 937        credits += blocks;
 938
 939        if (!is_create) {
 940                /* Dereference ea_inode holding old xattr value.
 941                 * Old ea_inode, inode map, block bitmap, group descriptor.
 942                 */
 943                credits += 4;
 944
 945                /* Data blocks for old ea_inode. */
 946                blocks = XATTR_SIZE_MAX >> sb->s_blocksize_bits;
 947
 948                /* Indirection block or one level of extent tree for old
 949                 * ea_inode.
 950                 */
 951                blocks += 1;
 952
 953                /* Block bitmap and group descriptor updates for each block. */
 954                credits += blocks * 2;
 955        }
 956
 957        /* We may need to clone the existing xattr block in which case we need
 958         * to increment ref counts for existing ea_inodes referenced by it.
 959         */
 960        if (block_bh) {
 961                struct ext4_xattr_entry *entry = BFIRST(block_bh);
 962
 963                for (; !IS_LAST_ENTRY(entry); entry = EXT4_XATTR_NEXT(entry))
 964                        if (entry->e_value_inum)
 965                                /* Ref count update on ea_inode. */
 966                                credits += 1;
 967        }
 968        return credits;
 969}
 970
 971static int ext4_xattr_ensure_credits(handle_t *handle, struct inode *inode,
 972                                     int credits, struct buffer_head *bh,
 973                                     bool dirty, bool block_csum)
 974{
 975        int error;
 976
 977        if (!ext4_handle_valid(handle))
 978                return 0;
 979
 980        if (handle->h_buffer_credits >= credits)
 981                return 0;
 982
 983        error = ext4_journal_extend(handle, credits - handle->h_buffer_credits);
 984        if (!error)
 985                return 0;
 986        if (error < 0) {
 987                ext4_warning(inode->i_sb, "Extend journal (error %d)", error);
 988                return error;
 989        }
 990
 991        if (bh && dirty) {
 992                if (block_csum)
 993                        ext4_xattr_block_csum_set(inode, bh);
 994                error = ext4_handle_dirty_metadata(handle, NULL, bh);
 995                if (error) {
 996                        ext4_warning(inode->i_sb, "Handle metadata (error %d)",
 997                                     error);
 998                        return error;
 999                }
1000        }
1001
1002        error = ext4_journal_restart(handle, credits);
1003        if (error) {
1004                ext4_warning(inode->i_sb, "Restart journal (error %d)", error);
1005                return error;
1006        }
1007
1008        if (bh) {
1009                error = ext4_journal_get_write_access(handle, bh);
1010                if (error) {
1011                        ext4_warning(inode->i_sb,
1012                                     "Get write access failed (error %d)",
1013                                     error);
1014                        return error;
1015                }
1016        }
1017        return 0;
1018}
1019
1020static int ext4_xattr_inode_update_ref(handle_t *handle, struct inode *ea_inode,
1021                                       int ref_change)
1022{
1023        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(ea_inode);
1024        struct ext4_iloc iloc;
1025        s64 ref_count;
1026        u32 hash;
1027        int ret;
1028
1029        inode_lock(ea_inode);
1030
1031        ret = ext4_reserve_inode_write(handle, ea_inode, &iloc);
1032        if (ret) {
1033                iloc.bh = NULL;
1034                goto out;
1035        }
1036
1037        ref_count = ext4_xattr_inode_get_ref(ea_inode);
1038        ref_count += ref_change;
1039        ext4_xattr_inode_set_ref(ea_inode, ref_count);
1040
1041        if (ref_change > 0) {
1042                WARN_ONCE(ref_count <= 0, "EA inode %lu ref_count=%lld",
1043                          ea_inode->i_ino, ref_count);
1044
1045                if (ref_count == 1) {
1046                        WARN_ONCE(ea_inode->i_nlink, "EA inode %lu i_nlink=%u",
1047                                  ea_inode->i_ino, ea_inode->i_nlink);
1048
1049                        set_nlink(ea_inode, 1);
1050                        ext4_orphan_del(handle, ea_inode);
1051
1052                        if (ea_inode_cache) {
1053                                hash = ext4_xattr_inode_get_hash(ea_inode);
1054                                mb_cache_entry_create(ea_inode_cache,
1055                                                      GFP_NOFS, hash,
1056                                                      ea_inode->i_ino,
1057                                                      true /* reusable */);
1058                        }
1059                }
1060        } else {
1061                WARN_ONCE(ref_count < 0, "EA inode %lu ref_count=%lld",
1062                          ea_inode->i_ino, ref_count);
1063
1064                if (ref_count == 0) {
1065                        WARN_ONCE(ea_inode->i_nlink != 1,
1066                                  "EA inode %lu i_nlink=%u",
1067                                  ea_inode->i_ino, ea_inode->i_nlink);
1068
1069                        clear_nlink(ea_inode);
1070                        ext4_orphan_add(handle, ea_inode);
1071
1072                        if (ea_inode_cache) {
1073                                hash = ext4_xattr_inode_get_hash(ea_inode);
1074                                mb_cache_entry_delete(ea_inode_cache, hash,
1075                                                      ea_inode->i_ino);
1076                        }
1077                }
1078        }
1079
1080        ret = ext4_mark_iloc_dirty(handle, ea_inode, &iloc);
1081        iloc.bh = NULL;
1082        if (ret)
1083                ext4_warning_inode(ea_inode,
1084                                   "ext4_mark_iloc_dirty() failed ret=%d", ret);
1085out:
1086        brelse(iloc.bh);
1087        inode_unlock(ea_inode);
1088        return ret;
1089}
1090
1091static int ext4_xattr_inode_inc_ref(handle_t *handle, struct inode *ea_inode)
1092{
1093        return ext4_xattr_inode_update_ref(handle, ea_inode, 1);
1094}
1095
1096static int ext4_xattr_inode_dec_ref(handle_t *handle, struct inode *ea_inode)
1097{
1098        return ext4_xattr_inode_update_ref(handle, ea_inode, -1);
1099}
1100
1101static int ext4_xattr_inode_inc_ref_all(handle_t *handle, struct inode *parent,
1102                                        struct ext4_xattr_entry *first)
1103{
1104        struct inode *ea_inode;
1105        struct ext4_xattr_entry *entry;
1106        struct ext4_xattr_entry *failed_entry;
1107        unsigned int ea_ino;
1108        int err, saved_err;
1109
1110        for (entry = first; !IS_LAST_ENTRY(entry);
1111             entry = EXT4_XATTR_NEXT(entry)) {
1112                if (!entry->e_value_inum)
1113                        continue;
1114                ea_ino = le32_to_cpu(entry->e_value_inum);
1115                err = ext4_xattr_inode_iget(parent, ea_ino,
1116                                            le32_to_cpu(entry->e_hash),
1117                                            &ea_inode);
1118                if (err)
1119                        goto cleanup;
1120                err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1121                if (err) {
1122                        ext4_warning_inode(ea_inode, "inc ref error %d", err);
1123                        iput(ea_inode);
1124                        goto cleanup;
1125                }
1126                iput(ea_inode);
1127        }
1128        return 0;
1129
1130cleanup:
1131        saved_err = err;
1132        failed_entry = entry;
1133
1134        for (entry = first; entry != failed_entry;
1135             entry = EXT4_XATTR_NEXT(entry)) {
1136                if (!entry->e_value_inum)
1137                        continue;
1138                ea_ino = le32_to_cpu(entry->e_value_inum);
1139                err = ext4_xattr_inode_iget(parent, ea_ino,
1140                                            le32_to_cpu(entry->e_hash),
1141                                            &ea_inode);
1142                if (err) {
1143                        ext4_warning(parent->i_sb,
1144                                     "cleanup ea_ino %u iget error %d", ea_ino,
1145                                     err);
1146                        continue;
1147                }
1148                err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1149                if (err)
1150                        ext4_warning_inode(ea_inode, "cleanup dec ref error %d",
1151                                           err);
1152                iput(ea_inode);
1153        }
1154        return saved_err;
1155}
1156
1157static void
1158ext4_xattr_inode_dec_ref_all(handle_t *handle, struct inode *parent,
1159                             struct buffer_head *bh,
1160                             struct ext4_xattr_entry *first, bool block_csum,
1161                             struct ext4_xattr_inode_array **ea_inode_array,
1162                             int extra_credits, bool skip_quota)
1163{
1164        struct inode *ea_inode;
1165        struct ext4_xattr_entry *entry;
1166        bool dirty = false;
1167        unsigned int ea_ino;
1168        int err;
1169        int credits;
1170
1171        /* One credit for dec ref on ea_inode, one for orphan list addition, */
1172        credits = 2 + extra_credits;
1173
1174        for (entry = first; !IS_LAST_ENTRY(entry);
1175             entry = EXT4_XATTR_NEXT(entry)) {
1176                if (!entry->e_value_inum)
1177                        continue;
1178                ea_ino = le32_to_cpu(entry->e_value_inum);
1179                err = ext4_xattr_inode_iget(parent, ea_ino,
1180                                            le32_to_cpu(entry->e_hash),
1181                                            &ea_inode);
1182                if (err)
1183                        continue;
1184
1185                err = ext4_expand_inode_array(ea_inode_array, ea_inode);
1186                if (err) {
1187                        ext4_warning_inode(ea_inode,
1188                                           "Expand inode array err=%d", err);
1189                        iput(ea_inode);
1190                        continue;
1191                }
1192
1193                err = ext4_xattr_ensure_credits(handle, parent, credits, bh,
1194                                                dirty, block_csum);
1195                if (err) {
1196                        ext4_warning_inode(ea_inode, "Ensure credits err=%d",
1197                                           err);
1198                        continue;
1199                }
1200
1201                err = ext4_xattr_inode_dec_ref(handle, ea_inode);
1202                if (err) {
1203                        ext4_warning_inode(ea_inode, "ea_inode dec ref err=%d",
1204                                           err);
1205                        continue;
1206                }
1207
1208                if (!skip_quota)
1209                        ext4_xattr_inode_free_quota(parent, ea_inode,
1210                                              le32_to_cpu(entry->e_value_size));
1211
1212                /*
1213                 * Forget about ea_inode within the same transaction that
1214                 * decrements the ref count. This avoids duplicate decrements in
1215                 * case the rest of the work spills over to subsequent
1216                 * transactions.
1217                 */
1218                entry->e_value_inum = 0;
1219                entry->e_value_size = 0;
1220
1221                dirty = true;
1222        }
1223
1224        if (dirty) {
1225                /*
1226                 * Note that we are deliberately skipping csum calculation for
1227                 * the final update because we do not expect any journal
1228                 * restarts until xattr block is freed.
1229                 */
1230
1231                err = ext4_handle_dirty_metadata(handle, NULL, bh);
1232                if (err)
1233                        ext4_warning_inode(parent,
1234                                           "handle dirty metadata err=%d", err);
1235        }
1236}
1237
1238/*
1239 * Release the xattr block BH: If the reference count is > 1, decrement it;
1240 * otherwise free the block.
1241 */
1242static void
1243ext4_xattr_release_block(handle_t *handle, struct inode *inode,
1244                         struct buffer_head *bh,
1245                         struct ext4_xattr_inode_array **ea_inode_array,
1246                         int extra_credits)
1247{
1248        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1249        u32 hash, ref;
1250        int error = 0;
1251
1252        BUFFER_TRACE(bh, "get_write_access");
1253        error = ext4_journal_get_write_access(handle, bh);
1254        if (error)
1255                goto out;
1256
1257        lock_buffer(bh);
1258        hash = le32_to_cpu(BHDR(bh)->h_hash);
1259        ref = le32_to_cpu(BHDR(bh)->h_refcount);
1260        if (ref == 1) {
1261                ea_bdebug(bh, "refcount now=0; freeing");
1262                /*
1263                 * This must happen under buffer lock for
1264                 * ext4_xattr_block_set() to reliably detect freed block
1265                 */
1266                if (ea_block_cache)
1267                        mb_cache_entry_delete(ea_block_cache, hash,
1268                                              bh->b_blocknr);
1269                get_bh(bh);
1270                unlock_buffer(bh);
1271
1272                if (ext4_has_feature_ea_inode(inode->i_sb))
1273                        ext4_xattr_inode_dec_ref_all(handle, inode, bh,
1274                                                     BFIRST(bh),
1275                                                     true /* block_csum */,
1276                                                     ea_inode_array,
1277                                                     extra_credits,
1278                                                     true /* skip_quota */);
1279                ext4_free_blocks(handle, inode, bh, 0, 1,
1280                                 EXT4_FREE_BLOCKS_METADATA |
1281                                 EXT4_FREE_BLOCKS_FORGET);
1282        } else {
1283                ref--;
1284                BHDR(bh)->h_refcount = cpu_to_le32(ref);
1285                if (ref == EXT4_XATTR_REFCOUNT_MAX - 1) {
1286                        struct mb_cache_entry *ce;
1287
1288                        if (ea_block_cache) {
1289                                ce = mb_cache_entry_get(ea_block_cache, hash,
1290                                                        bh->b_blocknr);
1291                                if (ce) {
1292                                        ce->e_reusable = 1;
1293                                        mb_cache_entry_put(ea_block_cache, ce);
1294                                }
1295                        }
1296                }
1297
1298                ext4_xattr_block_csum_set(inode, bh);
1299                /*
1300                 * Beware of this ugliness: Releasing of xattr block references
1301                 * from different inodes can race and so we have to protect
1302                 * from a race where someone else frees the block (and releases
1303                 * its journal_head) before we are done dirtying the buffer. In
1304                 * nojournal mode this race is harmless and we actually cannot
1305                 * call ext4_handle_dirty_metadata() with locked buffer as
1306                 * that function can call sync_dirty_buffer() so for that case
1307                 * we handle the dirtying after unlocking the buffer.
1308                 */
1309                if (ext4_handle_valid(handle))
1310                        error = ext4_handle_dirty_metadata(handle, inode, bh);
1311                unlock_buffer(bh);
1312                if (!ext4_handle_valid(handle))
1313                        error = ext4_handle_dirty_metadata(handle, inode, bh);
1314                if (IS_SYNC(inode))
1315                        ext4_handle_sync(handle);
1316                dquot_free_block(inode, EXT4_C2B(EXT4_SB(inode->i_sb), 1));
1317                ea_bdebug(bh, "refcount now=%d; releasing",
1318                          le32_to_cpu(BHDR(bh)->h_refcount));
1319        }
1320out:
1321        ext4_std_error(inode->i_sb, error);
1322        return;
1323}
1324
1325/*
1326 * Find the available free space for EAs. This also returns the total number of
1327 * bytes used by EA entries.
1328 */
1329static size_t ext4_xattr_free_space(struct ext4_xattr_entry *last,
1330                                    size_t *min_offs, void *base, int *total)
1331{
1332        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1333                if (!last->e_value_inum && last->e_value_size) {
1334                        size_t offs = le16_to_cpu(last->e_value_offs);
1335                        if (offs < *min_offs)
1336                                *min_offs = offs;
1337                }
1338                if (total)
1339                        *total += EXT4_XATTR_LEN(last->e_name_len);
1340        }
1341        return (*min_offs - ((void *)last - base) - sizeof(__u32));
1342}
1343
1344/*
1345 * Write the value of the EA in an inode.
1346 */
1347static int ext4_xattr_inode_write(handle_t *handle, struct inode *ea_inode,
1348                                  const void *buf, int bufsize)
1349{
1350        struct buffer_head *bh = NULL;
1351        unsigned long block = 0;
1352        int blocksize = ea_inode->i_sb->s_blocksize;
1353        int max_blocks = (bufsize + blocksize - 1) >> ea_inode->i_blkbits;
1354        int csize, wsize = 0;
1355        int ret = 0;
1356        int retries = 0;
1357
1358retry:
1359        while (ret >= 0 && ret < max_blocks) {
1360                struct ext4_map_blocks map;
1361                map.m_lblk = block += ret;
1362                map.m_len = max_blocks -= ret;
1363
1364                ret = ext4_map_blocks(handle, ea_inode, &map,
1365                                      EXT4_GET_BLOCKS_CREATE);
1366                if (ret <= 0) {
1367                        ext4_mark_inode_dirty(handle, ea_inode);
1368                        if (ret == -ENOSPC &&
1369                            ext4_should_retry_alloc(ea_inode->i_sb, &retries)) {
1370                                ret = 0;
1371                                goto retry;
1372                        }
1373                        break;
1374                }
1375        }
1376
1377        if (ret < 0)
1378                return ret;
1379
1380        block = 0;
1381        while (wsize < bufsize) {
1382                if (bh != NULL)
1383                        brelse(bh);
1384                csize = (bufsize - wsize) > blocksize ? blocksize :
1385                                                                bufsize - wsize;
1386                bh = ext4_getblk(handle, ea_inode, block, 0);
1387                if (IS_ERR(bh))
1388                        return PTR_ERR(bh);
1389                ret = ext4_journal_get_write_access(handle, bh);
1390                if (ret)
1391                        goto out;
1392
1393                memcpy(bh->b_data, buf, csize);
1394                set_buffer_uptodate(bh);
1395                ext4_handle_dirty_metadata(handle, ea_inode, bh);
1396
1397                buf += csize;
1398                wsize += csize;
1399                block += 1;
1400        }
1401
1402        inode_lock(ea_inode);
1403        i_size_write(ea_inode, wsize);
1404        ext4_update_i_disksize(ea_inode, wsize);
1405        inode_unlock(ea_inode);
1406
1407        ext4_mark_inode_dirty(handle, ea_inode);
1408
1409out:
1410        brelse(bh);
1411
1412        return ret;
1413}
1414
1415/*
1416 * Create an inode to store the value of a large EA.
1417 */
1418static struct inode *ext4_xattr_inode_create(handle_t *handle,
1419                                             struct inode *inode, u32 hash)
1420{
1421        struct inode *ea_inode = NULL;
1422        uid_t owner[2] = { i_uid_read(inode), i_gid_read(inode) };
1423        int err;
1424
1425        /*
1426         * Let the next inode be the goal, so we try and allocate the EA inode
1427         * in the same group, or nearby one.
1428         */
1429        ea_inode = ext4_new_inode(handle, inode->i_sb->s_root->d_inode,
1430                                  S_IFREG | 0600, NULL, inode->i_ino + 1, owner,
1431                                  EXT4_EA_INODE_FL);
1432        if (!IS_ERR(ea_inode)) {
1433                ea_inode->i_op = &ext4_file_inode_operations;
1434                ea_inode->i_fop = &ext4_file_operations;
1435                ext4_set_aops(ea_inode);
1436                ext4_xattr_inode_set_class(ea_inode);
1437                unlock_new_inode(ea_inode);
1438                ext4_xattr_inode_set_ref(ea_inode, 1);
1439                ext4_xattr_inode_set_hash(ea_inode, hash);
1440                err = ext4_mark_inode_dirty(handle, ea_inode);
1441                if (!err)
1442                        err = ext4_inode_attach_jinode(ea_inode);
1443                if (err) {
1444                        iput(ea_inode);
1445                        return ERR_PTR(err);
1446                }
1447
1448                /*
1449                 * Xattr inodes are shared therefore quota charging is performed
1450                 * at a higher level.
1451                 */
1452                dquot_free_inode(ea_inode);
1453                dquot_drop(ea_inode);
1454                inode_lock(ea_inode);
1455                ea_inode->i_flags |= S_NOQUOTA;
1456                inode_unlock(ea_inode);
1457        }
1458
1459        return ea_inode;
1460}
1461
1462static struct inode *
1463ext4_xattr_inode_cache_find(struct inode *inode, const void *value,
1464                            size_t value_len, u32 hash)
1465{
1466        struct inode *ea_inode;
1467        struct mb_cache_entry *ce;
1468        struct mb_cache *ea_inode_cache = EA_INODE_CACHE(inode);
1469        void *ea_data;
1470
1471        if (!ea_inode_cache)
1472                return NULL;
1473
1474        ce = mb_cache_entry_find_first(ea_inode_cache, hash);
1475        if (!ce)
1476                return NULL;
1477
1478        ea_data = ext4_kvmalloc(value_len, GFP_NOFS);
1479        if (!ea_data) {
1480                mb_cache_entry_put(ea_inode_cache, ce);
1481                return NULL;
1482        }
1483
1484        while (ce) {
1485                ea_inode = ext4_iget(inode->i_sb, ce->e_value);
1486                if (!IS_ERR(ea_inode) &&
1487                    !is_bad_inode(ea_inode) &&
1488                    (EXT4_I(ea_inode)->i_flags & EXT4_EA_INODE_FL) &&
1489                    i_size_read(ea_inode) == value_len &&
1490                    !ext4_xattr_inode_read(ea_inode, ea_data, value_len) &&
1491                    !ext4_xattr_inode_verify_hashes(ea_inode, NULL, ea_data,
1492                                                    value_len) &&
1493                    !memcmp(value, ea_data, value_len)) {
1494                        mb_cache_entry_touch(ea_inode_cache, ce);
1495                        mb_cache_entry_put(ea_inode_cache, ce);
1496                        kvfree(ea_data);
1497                        return ea_inode;
1498                }
1499
1500                if (!IS_ERR(ea_inode))
1501                        iput(ea_inode);
1502                ce = mb_cache_entry_find_next(ea_inode_cache, ce);
1503        }
1504        kvfree(ea_data);
1505        return NULL;
1506}
1507
1508/*
1509 * Add value of the EA in an inode.
1510 */
1511static int ext4_xattr_inode_lookup_create(handle_t *handle, struct inode *inode,
1512                                          const void *value, size_t value_len,
1513                                          struct inode **ret_inode)
1514{
1515        struct inode *ea_inode;
1516        u32 hash;
1517        int err;
1518
1519        hash = ext4_xattr_inode_hash(EXT4_SB(inode->i_sb), value, value_len);
1520        ea_inode = ext4_xattr_inode_cache_find(inode, value, value_len, hash);
1521        if (ea_inode) {
1522                err = ext4_xattr_inode_inc_ref(handle, ea_inode);
1523                if (err) {
1524                        iput(ea_inode);
1525                        return err;
1526                }
1527
1528                *ret_inode = ea_inode;
1529                return 0;
1530        }
1531
1532        /* Create an inode for the EA value */
1533        ea_inode = ext4_xattr_inode_create(handle, inode, hash);
1534        if (IS_ERR(ea_inode))
1535                return PTR_ERR(ea_inode);
1536
1537        err = ext4_xattr_inode_write(handle, ea_inode, value, value_len);
1538        if (err) {
1539                ext4_xattr_inode_dec_ref(handle, ea_inode);
1540                iput(ea_inode);
1541                return err;
1542        }
1543
1544        if (EA_INODE_CACHE(inode))
1545                mb_cache_entry_create(EA_INODE_CACHE(inode), GFP_NOFS, hash,
1546                                      ea_inode->i_ino, true /* reusable */);
1547
1548        *ret_inode = ea_inode;
1549        return 0;
1550}
1551
1552/*
1553 * Reserve min(block_size/8, 1024) bytes for xattr entries/names if ea_inode
1554 * feature is enabled.
1555 */
1556#define EXT4_XATTR_BLOCK_RESERVE(inode) min(i_blocksize(inode)/8, 1024U)
1557
1558static int ext4_xattr_set_entry(struct ext4_xattr_info *i,
1559                                struct ext4_xattr_search *s,
1560                                handle_t *handle, struct inode *inode,
1561                                bool is_block)
1562{
1563        struct ext4_xattr_entry *last;
1564        struct ext4_xattr_entry *here = s->here;
1565        size_t min_offs = s->end - s->base, name_len = strlen(i->name);
1566        int in_inode = i->in_inode;
1567        struct inode *old_ea_inode = NULL;
1568        struct inode *new_ea_inode = NULL;
1569        size_t old_size, new_size;
1570        int ret;
1571
1572        /* Space used by old and new values. */
1573        old_size = (!s->not_found && !here->e_value_inum) ?
1574                        EXT4_XATTR_SIZE(le32_to_cpu(here->e_value_size)) : 0;
1575        new_size = (i->value && !in_inode) ? EXT4_XATTR_SIZE(i->value_len) : 0;
1576
1577        /*
1578         * Optimization for the simple case when old and new values have the
1579         * same padded sizes. Not applicable if external inodes are involved.
1580         */
1581        if (new_size && new_size == old_size) {
1582                size_t offs = le16_to_cpu(here->e_value_offs);
1583                void *val = s->base + offs;
1584
1585                here->e_value_size = cpu_to_le32(i->value_len);
1586                if (i->value == EXT4_ZERO_XATTR_VALUE) {
1587                        memset(val, 0, new_size);
1588                } else {
1589                        memcpy(val, i->value, i->value_len);
1590                        /* Clear padding bytes. */
1591                        memset(val + i->value_len, 0, new_size - i->value_len);
1592                }
1593                goto update_hash;
1594        }
1595
1596        /* Compute min_offs and last. */
1597        last = s->first;
1598        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
1599                if (!last->e_value_inum && last->e_value_size) {
1600                        size_t offs = le16_to_cpu(last->e_value_offs);
1601                        if (offs < min_offs)
1602                                min_offs = offs;
1603                }
1604        }
1605
1606        /* Check whether we have enough space. */
1607        if (i->value) {
1608                size_t free;
1609
1610                free = min_offs - ((void *)last - s->base) - sizeof(__u32);
1611                if (!s->not_found)
1612                        free += EXT4_XATTR_LEN(name_len) + old_size;
1613
1614                if (free < EXT4_XATTR_LEN(name_len) + new_size) {
1615                        ret = -ENOSPC;
1616                        goto out;
1617                }
1618
1619                /*
1620                 * If storing the value in an external inode is an option,
1621                 * reserve space for xattr entries/names in the external
1622                 * attribute block so that a long value does not occupy the
1623                 * whole space and prevent futher entries being added.
1624                 */
1625                if (ext4_has_feature_ea_inode(inode->i_sb) &&
1626                    new_size && is_block &&
1627                    (min_offs + old_size - new_size) <
1628                                        EXT4_XATTR_BLOCK_RESERVE(inode)) {
1629                        ret = -ENOSPC;
1630                        goto out;
1631                }
1632        }
1633
1634        /*
1635         * Getting access to old and new ea inodes is subject to failures.
1636         * Finish that work before doing any modifications to the xattr data.
1637         */
1638        if (!s->not_found && here->e_value_inum) {
1639                ret = ext4_xattr_inode_iget(inode,
1640                                            le32_to_cpu(here->e_value_inum),
1641                                            le32_to_cpu(here->e_hash),
1642                                            &old_ea_inode);
1643                if (ret) {
1644                        old_ea_inode = NULL;
1645                        goto out;
1646                }
1647        }
1648        if (i->value && in_inode) {
1649                WARN_ON_ONCE(!i->value_len);
1650
1651                ret = ext4_xattr_inode_alloc_quota(inode, i->value_len);
1652                if (ret)
1653                        goto out;
1654
1655                ret = ext4_xattr_inode_lookup_create(handle, inode, i->value,
1656                                                     i->value_len,
1657                                                     &new_ea_inode);
1658                if (ret) {
1659                        new_ea_inode = NULL;
1660                        ext4_xattr_inode_free_quota(inode, NULL, i->value_len);
1661                        goto out;
1662                }
1663        }
1664
1665        if (old_ea_inode) {
1666                /* We are ready to release ref count on the old_ea_inode. */
1667                ret = ext4_xattr_inode_dec_ref(handle, old_ea_inode);
1668                if (ret) {
1669                        /* Release newly required ref count on new_ea_inode. */
1670                        if (new_ea_inode) {
1671                                int err;
1672
1673                                err = ext4_xattr_inode_dec_ref(handle,
1674                                                               new_ea_inode);
1675                                if (err)
1676                                        ext4_warning_inode(new_ea_inode,
1677                                                  "dec ref new_ea_inode err=%d",
1678                                                  err);
1679                                ext4_xattr_inode_free_quota(inode, new_ea_inode,
1680                                                            i->value_len);
1681                        }
1682                        goto out;
1683                }
1684
1685                ext4_xattr_inode_free_quota(inode, old_ea_inode,
1686                                            le32_to_cpu(here->e_value_size));
1687        }
1688
1689        /* No failures allowed past this point. */
1690
1691        if (!s->not_found && here->e_value_offs) {
1692                /* Remove the old value. */
1693                void *first_val = s->base + min_offs;
1694                size_t offs = le16_to_cpu(here->e_value_offs);
1695                void *val = s->base + offs;
1696
1697                memmove(first_val + old_size, first_val, val - first_val);
1698                memset(first_val, 0, old_size);
1699                min_offs += old_size;
1700
1701                /* Adjust all value offsets. */
1702                last = s->first;
1703                while (!IS_LAST_ENTRY(last)) {
1704                        size_t o = le16_to_cpu(last->e_value_offs);
1705
1706                        if (!last->e_value_inum &&
1707                            last->e_value_size && o < offs)
1708                                last->e_value_offs = cpu_to_le16(o + old_size);
1709                        last = EXT4_XATTR_NEXT(last);
1710                }
1711        }
1712
1713        if (!i->value) {
1714                /* Remove old name. */
1715                size_t size = EXT4_XATTR_LEN(name_len);
1716
1717                last = ENTRY((void *)last - size);
1718                memmove(here, (void *)here + size,
1719                        (void *)last - (void *)here + sizeof(__u32));
1720                memset(last, 0, size);
1721        } else if (s->not_found) {
1722                /* Insert new name. */
1723                size_t size = EXT4_XATTR_LEN(name_len);
1724                size_t rest = (void *)last - (void *)here + sizeof(__u32);
1725
1726                memmove((void *)here + size, here, rest);
1727                memset(here, 0, size);
1728                here->e_name_index = i->name_index;
1729                here->e_name_len = name_len;
1730                memcpy(here->e_name, i->name, name_len);
1731        } else {
1732                /* This is an update, reset value info. */
1733                here->e_value_inum = 0;
1734                here->e_value_offs = 0;
1735                here->e_value_size = 0;
1736        }
1737
1738        if (i->value) {
1739                /* Insert new value. */
1740                if (in_inode) {
1741                        here->e_value_inum = cpu_to_le32(new_ea_inode->i_ino);
1742                } else if (i->value_len) {
1743                        void *val = s->base + min_offs - new_size;
1744
1745                        here->e_value_offs = cpu_to_le16(min_offs - new_size);
1746                        if (i->value == EXT4_ZERO_XATTR_VALUE) {
1747                                memset(val, 0, new_size);
1748                        } else {
1749                                memcpy(val, i->value, i->value_len);
1750                                /* Clear padding bytes. */
1751                                memset(val + i->value_len, 0,
1752                                       new_size - i->value_len);
1753                        }
1754                }
1755                here->e_value_size = cpu_to_le32(i->value_len);
1756        }
1757
1758update_hash:
1759        if (i->value) {
1760                __le32 hash = 0;
1761
1762                /* Entry hash calculation. */
1763                if (in_inode) {
1764                        __le32 crc32c_hash;
1765
1766                        /*
1767                         * Feed crc32c hash instead of the raw value for entry
1768                         * hash calculation. This is to avoid walking
1769                         * potentially long value buffer again.
1770                         */
1771                        crc32c_hash = cpu_to_le32(
1772                                       ext4_xattr_inode_get_hash(new_ea_inode));
1773                        hash = ext4_xattr_hash_entry(here->e_name,
1774                                                     here->e_name_len,
1775                                                     &crc32c_hash, 1);
1776                } else if (is_block) {
1777                        __le32 *value = s->base + le16_to_cpu(
1778                                                        here->e_value_offs);
1779
1780                        hash = ext4_xattr_hash_entry(here->e_name,
1781                                                     here->e_name_len, value,
1782                                                     new_size >> 2);
1783                }
1784                here->e_hash = hash;
1785        }
1786
1787        if (is_block)
1788                ext4_xattr_rehash((struct ext4_xattr_header *)s->base);
1789
1790        ret = 0;
1791out:
1792        iput(old_ea_inode);
1793        iput(new_ea_inode);
1794        return ret;
1795}
1796
1797struct ext4_xattr_block_find {
1798        struct ext4_xattr_search s;
1799        struct buffer_head *bh;
1800};
1801
1802static int
1803ext4_xattr_block_find(struct inode *inode, struct ext4_xattr_info *i,
1804                      struct ext4_xattr_block_find *bs)
1805{
1806        struct super_block *sb = inode->i_sb;
1807        int error;
1808
1809        ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
1810                  i->name_index, i->name, i->value, (long)i->value_len);
1811
1812        if (EXT4_I(inode)->i_file_acl) {
1813                /* The inode already has an extended attribute block. */
1814                bs->bh = sb_bread(sb, EXT4_I(inode)->i_file_acl);
1815                error = -EIO;
1816                if (!bs->bh)
1817                        goto cleanup;
1818                ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
1819                        atomic_read(&(bs->bh->b_count)),
1820                        le32_to_cpu(BHDR(bs->bh)->h_refcount));
1821                error = ext4_xattr_check_block(inode, bs->bh);
1822                if (error)
1823                        goto cleanup;
1824                /* Find the named attribute. */
1825                bs->s.base = BHDR(bs->bh);
1826                bs->s.first = BFIRST(bs->bh);
1827                bs->s.end = bs->bh->b_data + bs->bh->b_size;
1828                bs->s.here = bs->s.first;
1829                error = xattr_find_entry(inode, &bs->s.here, bs->s.end,
1830                                         i->name_index, i->name, 1);
1831                if (error && error != -ENODATA)
1832                        goto cleanup;
1833                bs->s.not_found = error;
1834        }
1835        error = 0;
1836
1837cleanup:
1838        return error;
1839}
1840
1841static int
1842ext4_xattr_block_set(handle_t *handle, struct inode *inode,
1843                     struct ext4_xattr_info *i,
1844                     struct ext4_xattr_block_find *bs)
1845{
1846        struct super_block *sb = inode->i_sb;
1847        struct buffer_head *new_bh = NULL;
1848        struct ext4_xattr_search s_copy = bs->s;
1849        struct ext4_xattr_search *s = &s_copy;
1850        struct mb_cache_entry *ce = NULL;
1851        int error = 0;
1852        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
1853        struct inode *ea_inode = NULL, *tmp_inode;
1854        size_t old_ea_inode_quota = 0;
1855        unsigned int ea_ino;
1856
1857
1858#define header(x) ((struct ext4_xattr_header *)(x))
1859
1860        if (s->base) {
1861                BUFFER_TRACE(bs->bh, "get_write_access");
1862                error = ext4_journal_get_write_access(handle, bs->bh);
1863                if (error)
1864                        goto cleanup;
1865                lock_buffer(bs->bh);
1866
1867                if (header(s->base)->h_refcount == cpu_to_le32(1)) {
1868                        __u32 hash = le32_to_cpu(BHDR(bs->bh)->h_hash);
1869
1870                        /*
1871                         * This must happen under buffer lock for
1872                         * ext4_xattr_block_set() to reliably detect modified
1873                         * block
1874                         */
1875                        if (ea_block_cache)
1876                                mb_cache_entry_delete(ea_block_cache, hash,
1877                                                      bs->bh->b_blocknr);
1878                        ea_bdebug(bs->bh, "modifying in-place");
1879                        error = ext4_xattr_set_entry(i, s, handle, inode,
1880                                                     true /* is_block */);
1881                        ext4_xattr_block_csum_set(inode, bs->bh);
1882                        unlock_buffer(bs->bh);
1883                        if (error == -EFSCORRUPTED)
1884                                goto bad_block;
1885                        if (!error)
1886                                error = ext4_handle_dirty_metadata(handle,
1887                                                                   inode,
1888                                                                   bs->bh);
1889                        if (error)
1890                                goto cleanup;
1891                        goto inserted;
1892                } else {
1893                        int offset = (char *)s->here - bs->bh->b_data;
1894
1895                        unlock_buffer(bs->bh);
1896                        ea_bdebug(bs->bh, "cloning");
1897                        s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
1898                        error = -ENOMEM;
1899                        if (s->base == NULL)
1900                                goto cleanup;
1901                        memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
1902                        s->first = ENTRY(header(s->base)+1);
1903                        header(s->base)->h_refcount = cpu_to_le32(1);
1904                        s->here = ENTRY(s->base + offset);
1905                        s->end = s->base + bs->bh->b_size;
1906
1907                        /*
1908                         * If existing entry points to an xattr inode, we need
1909                         * to prevent ext4_xattr_set_entry() from decrementing
1910                         * ref count on it because the reference belongs to the
1911                         * original block. In this case, make the entry look
1912                         * like it has an empty value.
1913                         */
1914                        if (!s->not_found && s->here->e_value_inum) {
1915                                ea_ino = le32_to_cpu(s->here->e_value_inum);
1916                                error = ext4_xattr_inode_iget(inode, ea_ino,
1917                                              le32_to_cpu(s->here->e_hash),
1918                                              &tmp_inode);
1919                                if (error)
1920                                        goto cleanup;
1921
1922                                if (!ext4_test_inode_state(tmp_inode,
1923                                                EXT4_STATE_LUSTRE_EA_INODE)) {
1924                                        /*
1925                                         * Defer quota free call for previous
1926                                         * inode until success is guaranteed.
1927                                         */
1928                                        old_ea_inode_quota = le32_to_cpu(
1929                                                        s->here->e_value_size);
1930                                }
1931                                iput(tmp_inode);
1932
1933                                s->here->e_value_inum = 0;
1934                                s->here->e_value_size = 0;
1935                        }
1936                }
1937        } else {
1938                /* Allocate a buffer where we construct the new block. */
1939                s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
1940                /* assert(header == s->base) */
1941                error = -ENOMEM;
1942                if (s->base == NULL)
1943                        goto cleanup;
1944                header(s->base)->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
1945                header(s->base)->h_blocks = cpu_to_le32(1);
1946                header(s->base)->h_refcount = cpu_to_le32(1);
1947                s->first = ENTRY(header(s->base)+1);
1948                s->here = ENTRY(header(s->base)+1);
1949                s->end = s->base + sb->s_blocksize;
1950        }
1951
1952        error = ext4_xattr_set_entry(i, s, handle, inode, true /* is_block */);
1953        if (error == -EFSCORRUPTED)
1954                goto bad_block;
1955        if (error)
1956                goto cleanup;
1957
1958        if (i->value && s->here->e_value_inum) {
1959                /*
1960                 * A ref count on ea_inode has been taken as part of the call to
1961                 * ext4_xattr_set_entry() above. We would like to drop this
1962                 * extra ref but we have to wait until the xattr block is
1963                 * initialized and has its own ref count on the ea_inode.
1964                 */
1965                ea_ino = le32_to_cpu(s->here->e_value_inum);
1966                error = ext4_xattr_inode_iget(inode, ea_ino,
1967                                              le32_to_cpu(s->here->e_hash),
1968                                              &ea_inode);
1969                if (error) {
1970                        ea_inode = NULL;
1971                        goto cleanup;
1972                }
1973        }
1974
1975inserted:
1976        if (!IS_LAST_ENTRY(s->first)) {
1977                new_bh = ext4_xattr_block_cache_find(inode, header(s->base),
1978                                                     &ce);
1979                if (new_bh) {
1980                        /* We found an identical block in the cache. */
1981                        if (new_bh == bs->bh)
1982                                ea_bdebug(new_bh, "keeping");
1983                        else {
1984                                u32 ref;
1985
1986                                WARN_ON_ONCE(dquot_initialize_needed(inode));
1987
1988                                /* The old block is released after updating
1989                                   the inode. */
1990                                error = dquot_alloc_block(inode,
1991                                                EXT4_C2B(EXT4_SB(sb), 1));
1992                                if (error)
1993                                        goto cleanup;
1994                                BUFFER_TRACE(new_bh, "get_write_access");
1995                                error = ext4_journal_get_write_access(handle,
1996                                                                      new_bh);
1997                                if (error)
1998                                        goto cleanup_dquot;
1999                                lock_buffer(new_bh);
2000                                /*
2001                                 * We have to be careful about races with
2002                                 * freeing, rehashing or adding references to
2003                                 * xattr block. Once we hold buffer lock xattr
2004                                 * block's state is stable so we can check
2005                                 * whether the block got freed / rehashed or
2006                                 * not.  Since we unhash mbcache entry under
2007                                 * buffer lock when freeing / rehashing xattr
2008                                 * block, checking whether entry is still
2009                                 * hashed is reliable. Same rules hold for
2010                                 * e_reusable handling.
2011                                 */
2012                                if (hlist_bl_unhashed(&ce->e_hash_list) ||
2013                                    !ce->e_reusable) {
2014                                        /*
2015                                         * Undo everything and check mbcache
2016                                         * again.
2017                                         */
2018                                        unlock_buffer(new_bh);
2019                                        dquot_free_block(inode,
2020                                                         EXT4_C2B(EXT4_SB(sb),
2021                                                                  1));
2022                                        brelse(new_bh);
2023                                        mb_cache_entry_put(ea_block_cache, ce);
2024                                        ce = NULL;
2025                                        new_bh = NULL;
2026                                        goto inserted;
2027                                }
2028                                ref = le32_to_cpu(BHDR(new_bh)->h_refcount) + 1;
2029                                BHDR(new_bh)->h_refcount = cpu_to_le32(ref);
2030                                if (ref >= EXT4_XATTR_REFCOUNT_MAX)
2031                                        ce->e_reusable = 0;
2032                                ea_bdebug(new_bh, "reusing; refcount now=%d",
2033                                          ref);
2034                                ext4_xattr_block_csum_set(inode, new_bh);
2035                                unlock_buffer(new_bh);
2036                                error = ext4_handle_dirty_metadata(handle,
2037                                                                   inode,
2038                                                                   new_bh);
2039                                if (error)
2040                                        goto cleanup_dquot;
2041                        }
2042                        mb_cache_entry_touch(ea_block_cache, ce);
2043                        mb_cache_entry_put(ea_block_cache, ce);
2044                        ce = NULL;
2045                } else if (bs->bh && s->base == bs->bh->b_data) {
2046                        /* We were modifying this block in-place. */
2047                        ea_bdebug(bs->bh, "keeping this block");
2048                        ext4_xattr_block_cache_insert(ea_block_cache, bs->bh);
2049                        new_bh = bs->bh;
2050                        get_bh(new_bh);
2051                } else {
2052                        /* We need to allocate a new block */
2053                        ext4_fsblk_t goal, block;
2054
2055                        WARN_ON_ONCE(dquot_initialize_needed(inode));
2056
2057                        goal = ext4_group_first_block_no(sb,
2058                                                EXT4_I(inode)->i_block_group);
2059
2060                        /* non-extent files can't have physical blocks past 2^32 */
2061                        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2062                                goal = goal & EXT4_MAX_BLOCK_FILE_PHYS;
2063
2064                        block = ext4_new_meta_blocks(handle, inode, goal, 0,
2065                                                     NULL, &error);
2066                        if (error)
2067                                goto cleanup;
2068
2069                        if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
2070                                BUG_ON(block > EXT4_MAX_BLOCK_FILE_PHYS);
2071
2072                        ea_idebug(inode, "creating block %llu",
2073                                  (unsigned long long)block);
2074
2075                        new_bh = sb_getblk(sb, block);
2076                        if (unlikely(!new_bh)) {
2077                                error = -ENOMEM;
2078getblk_failed:
2079                                ext4_free_blocks(handle, inode, NULL, block, 1,
2080                                                 EXT4_FREE_BLOCKS_METADATA);
2081                                goto cleanup;
2082                        }
2083                        error = ext4_xattr_inode_inc_ref_all(handle, inode,
2084                                                      ENTRY(header(s->base)+1));
2085                        if (error)
2086                                goto getblk_failed;
2087                        if (ea_inode) {
2088                                /* Drop the extra ref on ea_inode. */
2089                                error = ext4_xattr_inode_dec_ref(handle,
2090                                                                 ea_inode);
2091                                if (error)
2092                                        ext4_warning_inode(ea_inode,
2093                                                           "dec ref error=%d",
2094                                                           error);
2095                                iput(ea_inode);
2096                                ea_inode = NULL;
2097                        }
2098
2099                        lock_buffer(new_bh);
2100                        error = ext4_journal_get_create_access(handle, new_bh);
2101                        if (error) {
2102                                unlock_buffer(new_bh);
2103                                error = -EIO;
2104                                goto getblk_failed;
2105                        }
2106                        memcpy(new_bh->b_data, s->base, new_bh->b_size);
2107                        ext4_xattr_block_csum_set(inode, new_bh);
2108                        set_buffer_uptodate(new_bh);
2109                        unlock_buffer(new_bh);
2110                        ext4_xattr_block_cache_insert(ea_block_cache, new_bh);
2111                        error = ext4_handle_dirty_metadata(handle, inode,
2112                                                           new_bh);
2113                        if (error)
2114                                goto cleanup;
2115                }
2116        }
2117
2118        if (old_ea_inode_quota)
2119                ext4_xattr_inode_free_quota(inode, NULL, old_ea_inode_quota);
2120
2121        /* Update the inode. */
2122        EXT4_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
2123
2124        /* Drop the previous xattr block. */
2125        if (bs->bh && bs->bh != new_bh) {
2126                struct ext4_xattr_inode_array *ea_inode_array = NULL;
2127
2128                ext4_xattr_release_block(handle, inode, bs->bh,
2129                                         &ea_inode_array,
2130                                         0 /* extra_credits */);
2131                ext4_xattr_inode_array_free(ea_inode_array);
2132        }
2133        error = 0;
2134
2135cleanup:
2136        if (ea_inode) {
2137                int error2;
2138
2139                error2 = ext4_xattr_inode_dec_ref(handle, ea_inode);
2140                if (error2)
2141                        ext4_warning_inode(ea_inode, "dec ref error=%d",
2142                                           error2);
2143
2144                /* If there was an error, revert the quota charge. */
2145                if (error)
2146                        ext4_xattr_inode_free_quota(inode, ea_inode,
2147                                                    i_size_read(ea_inode));
2148                iput(ea_inode);
2149        }
2150        if (ce)
2151                mb_cache_entry_put(ea_block_cache, ce);
2152        brelse(new_bh);
2153        if (!(bs->bh && s->base == bs->bh->b_data))
2154                kfree(s->base);
2155
2156        return error;
2157
2158cleanup_dquot:
2159        dquot_free_block(inode, EXT4_C2B(EXT4_SB(sb), 1));
2160        goto cleanup;
2161
2162bad_block:
2163        EXT4_ERROR_INODE(inode, "bad block %llu",
2164                         EXT4_I(inode)->i_file_acl);
2165        goto cleanup;
2166
2167#undef header
2168}
2169
2170int ext4_xattr_ibody_find(struct inode *inode, struct ext4_xattr_info *i,
2171                          struct ext4_xattr_ibody_find *is)
2172{
2173        struct ext4_xattr_ibody_header *header;
2174        struct ext4_inode *raw_inode;
2175        int error;
2176
2177        if (EXT4_I(inode)->i_extra_isize == 0)
2178                return 0;
2179        raw_inode = ext4_raw_inode(&is->iloc);
2180        header = IHDR(inode, raw_inode);
2181        is->s.base = is->s.first = IFIRST(header);
2182        is->s.here = is->s.first;
2183        is->s.end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2184        if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2185                error = xattr_check_inode(inode, header, is->s.end);
2186                if (error)
2187                        return error;
2188                /* Find the named attribute. */
2189                error = xattr_find_entry(inode, &is->s.here, is->s.end,
2190                                         i->name_index, i->name, 0);
2191                if (error && error != -ENODATA)
2192                        return error;
2193                is->s.not_found = error;
2194        }
2195        return 0;
2196}
2197
2198int ext4_xattr_ibody_inline_set(handle_t *handle, struct inode *inode,
2199                                struct ext4_xattr_info *i,
2200                                struct ext4_xattr_ibody_find *is)
2201{
2202        struct ext4_xattr_ibody_header *header;
2203        struct ext4_xattr_search *s = &is->s;
2204        int error;
2205
2206        if (EXT4_I(inode)->i_extra_isize == 0)
2207                return -ENOSPC;
2208        error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2209        if (error) {
2210                if (error == -ENOSPC &&
2211                    ext4_has_inline_data(inode)) {
2212                        error = ext4_try_to_evict_inline_data(handle, inode,
2213                                        EXT4_XATTR_LEN(strlen(i->name) +
2214                                        EXT4_XATTR_SIZE(i->value_len)));
2215                        if (error)
2216                                return error;
2217                        error = ext4_xattr_ibody_find(inode, i, is);
2218                        if (error)
2219                                return error;
2220                        error = ext4_xattr_set_entry(i, s, handle, inode,
2221                                                     false /* is_block */);
2222                }
2223                if (error)
2224                        return error;
2225        }
2226        header = IHDR(inode, ext4_raw_inode(&is->iloc));
2227        if (!IS_LAST_ENTRY(s->first)) {
2228                header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2229                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2230        } else {
2231                header->h_magic = cpu_to_le32(0);
2232                ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2233        }
2234        return 0;
2235}
2236
2237static int ext4_xattr_ibody_set(handle_t *handle, struct inode *inode,
2238                                struct ext4_xattr_info *i,
2239                                struct ext4_xattr_ibody_find *is)
2240{
2241        struct ext4_xattr_ibody_header *header;
2242        struct ext4_xattr_search *s = &is->s;
2243        int error;
2244
2245        if (EXT4_I(inode)->i_extra_isize == 0)
2246                return -ENOSPC;
2247        error = ext4_xattr_set_entry(i, s, handle, inode, false /* is_block */);
2248        if (error)
2249                return error;
2250        header = IHDR(inode, ext4_raw_inode(&is->iloc));
2251        if (!IS_LAST_ENTRY(s->first)) {
2252                header->h_magic = cpu_to_le32(EXT4_XATTR_MAGIC);
2253                ext4_set_inode_state(inode, EXT4_STATE_XATTR);
2254        } else {
2255                header->h_magic = cpu_to_le32(0);
2256                ext4_clear_inode_state(inode, EXT4_STATE_XATTR);
2257        }
2258        return 0;
2259}
2260
2261static int ext4_xattr_value_same(struct ext4_xattr_search *s,
2262                                 struct ext4_xattr_info *i)
2263{
2264        void *value;
2265
2266        /* When e_value_inum is set the value is stored externally. */
2267        if (s->here->e_value_inum)
2268                return 0;
2269        if (le32_to_cpu(s->here->e_value_size) != i->value_len)
2270                return 0;
2271        value = ((void *)s->base) + le16_to_cpu(s->here->e_value_offs);
2272        return !memcmp(value, i->value, i->value_len);
2273}
2274
2275static struct buffer_head *ext4_xattr_get_block(struct inode *inode)
2276{
2277        struct buffer_head *bh;
2278        int error;
2279
2280        if (!EXT4_I(inode)->i_file_acl)
2281                return NULL;
2282        bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2283        if (!bh)
2284                return ERR_PTR(-EIO);
2285        error = ext4_xattr_check_block(inode, bh);
2286        if (error)
2287                return ERR_PTR(error);
2288        return bh;
2289}
2290
2291/*
2292 * ext4_xattr_set_handle()
2293 *
2294 * Create, replace or remove an extended attribute for this inode.  Value
2295 * is NULL to remove an existing extended attribute, and non-NULL to
2296 * either replace an existing extended attribute, or create a new extended
2297 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
2298 * specify that an extended attribute must exist and must not exist
2299 * previous to the call, respectively.
2300 *
2301 * Returns 0, or a negative error number on failure.
2302 */
2303int
2304ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
2305                      const char *name, const void *value, size_t value_len,
2306                      int flags)
2307{
2308        struct ext4_xattr_info i = {
2309                .name_index = name_index,
2310                .name = name,
2311                .value = value,
2312                .value_len = value_len,
2313                .in_inode = 0,
2314        };
2315        struct ext4_xattr_ibody_find is = {
2316                .s = { .not_found = -ENODATA, },
2317        };
2318        struct ext4_xattr_block_find bs = {
2319                .s = { .not_found = -ENODATA, },
2320        };
2321        int no_expand;
2322        int error;
2323
2324        if (!name)
2325                return -EINVAL;
2326        if (strlen(name) > 255)
2327                return -ERANGE;
2328
2329        ext4_write_lock_xattr(inode, &no_expand);
2330
2331        /* Check journal credits under write lock. */
2332        if (ext4_handle_valid(handle)) {
2333                struct buffer_head *bh;
2334                int credits;
2335
2336                bh = ext4_xattr_get_block(inode);
2337                if (IS_ERR(bh)) {
2338                        error = PTR_ERR(bh);
2339                        goto cleanup;
2340                }
2341
2342                credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2343                                                   value_len,
2344                                                   flags & XATTR_CREATE);
2345                brelse(bh);
2346
2347                if (!ext4_handle_has_enough_credits(handle, credits)) {
2348                        error = -ENOSPC;
2349                        goto cleanup;
2350                }
2351        }
2352
2353        error = ext4_reserve_inode_write(handle, inode, &is.iloc);
2354        if (error)
2355                goto cleanup;
2356
2357        if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) {
2358                struct ext4_inode *raw_inode = ext4_raw_inode(&is.iloc);
2359                memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
2360                ext4_clear_inode_state(inode, EXT4_STATE_NEW);
2361        }
2362
2363        error = ext4_xattr_ibody_find(inode, &i, &is);
2364        if (error)
2365                goto cleanup;
2366        if (is.s.not_found)
2367                error = ext4_xattr_block_find(inode, &i, &bs);
2368        if (error)
2369                goto cleanup;
2370        if (is.s.not_found && bs.s.not_found) {
2371                error = -ENODATA;
2372                if (flags & XATTR_REPLACE)
2373                        goto cleanup;
2374                error = 0;
2375                if (!value)
2376                        goto cleanup;
2377        } else {
2378                error = -EEXIST;
2379                if (flags & XATTR_CREATE)
2380                        goto cleanup;
2381        }
2382
2383        if (!value) {
2384                if (!is.s.not_found)
2385                        error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2386                else if (!bs.s.not_found)
2387                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2388        } else {
2389                error = 0;
2390                /* Xattr value did not change? Save us some work and bail out */
2391                if (!is.s.not_found && ext4_xattr_value_same(&is.s, &i))
2392                        goto cleanup;
2393                if (!bs.s.not_found && ext4_xattr_value_same(&bs.s, &i))
2394                        goto cleanup;
2395
2396                if (ext4_has_feature_ea_inode(inode->i_sb) &&
2397                    (EXT4_XATTR_SIZE(i.value_len) >
2398                        EXT4_XATTR_MIN_LARGE_EA_SIZE(inode->i_sb->s_blocksize)))
2399                        i.in_inode = 1;
2400retry_inode:
2401                error = ext4_xattr_ibody_set(handle, inode, &i, &is);
2402                if (!error && !bs.s.not_found) {
2403                        i.value = NULL;
2404                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2405                } else if (error == -ENOSPC) {
2406                        if (EXT4_I(inode)->i_file_acl && !bs.s.base) {
2407                                error = ext4_xattr_block_find(inode, &i, &bs);
2408                                if (error)
2409                                        goto cleanup;
2410                        }
2411                        error = ext4_xattr_block_set(handle, inode, &i, &bs);
2412                        if (!error && !is.s.not_found) {
2413                                i.value = NULL;
2414                                error = ext4_xattr_ibody_set(handle, inode, &i,
2415                                                             &is);
2416                        } else if (error == -ENOSPC) {
2417                                /*
2418                                 * Xattr does not fit in the block, store at
2419                                 * external inode if possible.
2420                                 */
2421                                if (ext4_has_feature_ea_inode(inode->i_sb) &&
2422                                    !i.in_inode) {
2423                                        i.in_inode = 1;
2424                                        goto retry_inode;
2425                                }
2426                        }
2427                }
2428        }
2429        if (!error) {
2430                ext4_xattr_update_super_block(handle, inode->i_sb);
2431                inode->i_ctime = current_time(inode);
2432                if (!value)
2433                        no_expand = 0;
2434                error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
2435                /*
2436                 * The bh is consumed by ext4_mark_iloc_dirty, even with
2437                 * error != 0.
2438                 */
2439                is.iloc.bh = NULL;
2440                if (IS_SYNC(inode))
2441                        ext4_handle_sync(handle);
2442        }
2443
2444cleanup:
2445        brelse(is.iloc.bh);
2446        brelse(bs.bh);
2447        ext4_write_unlock_xattr(inode, &no_expand);
2448        return error;
2449}
2450
2451int ext4_xattr_set_credits(struct inode *inode, size_t value_len,
2452                           bool is_create, int *credits)
2453{
2454        struct buffer_head *bh;
2455        int err;
2456
2457        *credits = 0;
2458
2459        if (!EXT4_SB(inode->i_sb)->s_journal)
2460                return 0;
2461
2462        down_read(&EXT4_I(inode)->xattr_sem);
2463
2464        bh = ext4_xattr_get_block(inode);
2465        if (IS_ERR(bh)) {
2466                err = PTR_ERR(bh);
2467        } else {
2468                *credits = __ext4_xattr_set_credits(inode->i_sb, inode, bh,
2469                                                    value_len, is_create);
2470                brelse(bh);
2471                err = 0;
2472        }
2473
2474        up_read(&EXT4_I(inode)->xattr_sem);
2475        return err;
2476}
2477
2478/*
2479 * ext4_xattr_set()
2480 *
2481 * Like ext4_xattr_set_handle, but start from an inode. This extended
2482 * attribute modification is a filesystem transaction by itself.
2483 *
2484 * Returns 0, or a negative error number on failure.
2485 */
2486int
2487ext4_xattr_set(struct inode *inode, int name_index, const char *name,
2488               const void *value, size_t value_len, int flags)
2489{
2490        handle_t *handle;
2491        struct super_block *sb = inode->i_sb;
2492        int error, retries = 0;
2493        int credits;
2494
2495        error = dquot_initialize(inode);
2496        if (error)
2497                return error;
2498
2499retry:
2500        error = ext4_xattr_set_credits(inode, value_len, flags & XATTR_CREATE,
2501                                       &credits);
2502        if (error)
2503                return error;
2504
2505        handle = ext4_journal_start(inode, EXT4_HT_XATTR, credits);
2506        if (IS_ERR(handle)) {
2507                error = PTR_ERR(handle);
2508        } else {
2509                int error2;
2510
2511                error = ext4_xattr_set_handle(handle, inode, name_index, name,
2512                                              value, value_len, flags);
2513                error2 = ext4_journal_stop(handle);
2514                if (error == -ENOSPC &&
2515                    ext4_should_retry_alloc(sb, &retries))
2516                        goto retry;
2517                if (error == 0)
2518                        error = error2;
2519        }
2520
2521        return error;
2522}
2523
2524/*
2525 * Shift the EA entries in the inode to create space for the increased
2526 * i_extra_isize.
2527 */
2528static void ext4_xattr_shift_entries(struct ext4_xattr_entry *entry,
2529                                     int value_offs_shift, void *to,
2530                                     void *from, size_t n)
2531{
2532        struct ext4_xattr_entry *last = entry;
2533        int new_offs;
2534
2535        /* We always shift xattr headers further thus offsets get lower */
2536        BUG_ON(value_offs_shift > 0);
2537
2538        /* Adjust the value offsets of the entries */
2539        for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2540                if (!last->e_value_inum && last->e_value_size) {
2541                        new_offs = le16_to_cpu(last->e_value_offs) +
2542                                                        value_offs_shift;
2543                        last->e_value_offs = cpu_to_le16(new_offs);
2544                }
2545        }
2546        /* Shift the entries by n bytes */
2547        memmove(to, from, n);
2548}
2549
2550/*
2551 * Move xattr pointed to by 'entry' from inode into external xattr block
2552 */
2553static int ext4_xattr_move_to_block(handle_t *handle, struct inode *inode,
2554                                    struct ext4_inode *raw_inode,
2555                                    struct ext4_xattr_entry *entry)
2556{
2557        struct ext4_xattr_ibody_find *is = NULL;
2558        struct ext4_xattr_block_find *bs = NULL;
2559        char *buffer = NULL, *b_entry_name = NULL;
2560        size_t value_size = le32_to_cpu(entry->e_value_size);
2561        struct ext4_xattr_info i = {
2562                .value = NULL,
2563                .value_len = 0,
2564                .name_index = entry->e_name_index,
2565                .in_inode = !!entry->e_value_inum,
2566        };
2567        struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2568        int error;
2569
2570        is = kzalloc(sizeof(struct ext4_xattr_ibody_find), GFP_NOFS);
2571        bs = kzalloc(sizeof(struct ext4_xattr_block_find), GFP_NOFS);
2572        buffer = kmalloc(value_size, GFP_NOFS);
2573        b_entry_name = kmalloc(entry->e_name_len + 1, GFP_NOFS);
2574        if (!is || !bs || !buffer || !b_entry_name) {
2575                error = -ENOMEM;
2576                goto out;
2577        }
2578
2579        is->s.not_found = -ENODATA;
2580        bs->s.not_found = -ENODATA;
2581        is->iloc.bh = NULL;
2582        bs->bh = NULL;
2583
2584        /* Save the entry name and the entry value */
2585        if (entry->e_value_inum) {
2586                error = ext4_xattr_inode_get(inode, entry, buffer, value_size);
2587                if (error)
2588                        goto out;
2589        } else {
2590                size_t value_offs = le16_to_cpu(entry->e_value_offs);
2591                memcpy(buffer, (void *)IFIRST(header) + value_offs, value_size);
2592        }
2593
2594        memcpy(b_entry_name, entry->e_name, entry->e_name_len);
2595        b_entry_name[entry->e_name_len] = '\0';
2596        i.name = b_entry_name;
2597
2598        error = ext4_get_inode_loc(inode, &is->iloc);
2599        if (error)
2600                goto out;
2601
2602        error = ext4_xattr_ibody_find(inode, &i, is);
2603        if (error)
2604                goto out;
2605
2606        /* Remove the chosen entry from the inode */
2607        error = ext4_xattr_ibody_set(handle, inode, &i, is);
2608        if (error)
2609                goto out;
2610
2611        i.value = buffer;
2612        i.value_len = value_size;
2613        error = ext4_xattr_block_find(inode, &i, bs);
2614        if (error)
2615                goto out;
2616
2617        /* Add entry which was removed from the inode into the block */
2618        error = ext4_xattr_block_set(handle, inode, &i, bs);
2619        if (error)
2620                goto out;
2621        error = 0;
2622out:
2623        kfree(b_entry_name);
2624        kfree(buffer);
2625        if (is)
2626                brelse(is->iloc.bh);
2627        kfree(is);
2628        kfree(bs);
2629
2630        return error;
2631}
2632
2633static int ext4_xattr_make_inode_space(handle_t *handle, struct inode *inode,
2634                                       struct ext4_inode *raw_inode,
2635                                       int isize_diff, size_t ifree,
2636                                       size_t bfree, int *total_ino)
2637{
2638        struct ext4_xattr_ibody_header *header = IHDR(inode, raw_inode);
2639        struct ext4_xattr_entry *small_entry;
2640        struct ext4_xattr_entry *entry;
2641        struct ext4_xattr_entry *last;
2642        unsigned int entry_size;        /* EA entry size */
2643        unsigned int total_size;        /* EA entry size + value size */
2644        unsigned int min_total_size;
2645        int error;
2646
2647        while (isize_diff > ifree) {
2648                entry = NULL;
2649                small_entry = NULL;
2650                min_total_size = ~0U;
2651                last = IFIRST(header);
2652                /* Find the entry best suited to be pushed into EA block */
2653                for (; !IS_LAST_ENTRY(last); last = EXT4_XATTR_NEXT(last)) {
2654                        total_size = EXT4_XATTR_LEN(last->e_name_len);
2655                        if (!last->e_value_inum)
2656                                total_size += EXT4_XATTR_SIZE(
2657                                               le32_to_cpu(last->e_value_size));
2658                        if (total_size <= bfree &&
2659                            total_size < min_total_size) {
2660                                if (total_size + ifree < isize_diff) {
2661                                        small_entry = last;
2662                                } else {
2663                                        entry = last;
2664                                        min_total_size = total_size;
2665                                }
2666                        }
2667                }
2668
2669                if (entry == NULL) {
2670                        if (small_entry == NULL)
2671                                return -ENOSPC;
2672                        entry = small_entry;
2673                }
2674
2675                entry_size = EXT4_XATTR_LEN(entry->e_name_len);
2676                total_size = entry_size;
2677                if (!entry->e_value_inum)
2678                        total_size += EXT4_XATTR_SIZE(
2679                                              le32_to_cpu(entry->e_value_size));
2680                error = ext4_xattr_move_to_block(handle, inode, raw_inode,
2681                                                 entry);
2682                if (error)
2683                        return error;
2684
2685                *total_ino -= entry_size;
2686                ifree += total_size;
2687                bfree -= total_size;
2688        }
2689
2690        return 0;
2691}
2692
2693/*
2694 * Expand an inode by new_extra_isize bytes when EAs are present.
2695 * Returns 0 on success or negative error number on failure.
2696 */
2697int ext4_expand_extra_isize_ea(struct inode *inode, int new_extra_isize,
2698                               struct ext4_inode *raw_inode, handle_t *handle)
2699{
2700        struct ext4_xattr_ibody_header *header;
2701        struct buffer_head *bh;
2702        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2703        static unsigned int mnt_count;
2704        size_t min_offs;
2705        size_t ifree, bfree;
2706        int total_ino;
2707        void *base, *end;
2708        int error = 0, tried_min_extra_isize = 0;
2709        int s_min_extra_isize = le16_to_cpu(sbi->s_es->s_min_extra_isize);
2710        int isize_diff; /* How much do we need to grow i_extra_isize */
2711
2712retry:
2713        isize_diff = new_extra_isize - EXT4_I(inode)->i_extra_isize;
2714        if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
2715                return 0;
2716
2717        header = IHDR(inode, raw_inode);
2718
2719        /*
2720         * Check if enough free space is available in the inode to shift the
2721         * entries ahead by new_extra_isize.
2722         */
2723
2724        base = IFIRST(header);
2725        end = (void *)raw_inode + EXT4_SB(inode->i_sb)->s_inode_size;
2726        min_offs = end - base;
2727        total_ino = sizeof(struct ext4_xattr_ibody_header);
2728
2729        error = xattr_check_inode(inode, header, end);
2730        if (error)
2731                goto cleanup;
2732
2733        ifree = ext4_xattr_free_space(base, &min_offs, base, &total_ino);
2734        if (ifree >= isize_diff)
2735                goto shift;
2736
2737        /*
2738         * Enough free space isn't available in the inode, check if
2739         * EA block can hold new_extra_isize bytes.
2740         */
2741        if (EXT4_I(inode)->i_file_acl) {
2742                bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2743                error = -EIO;
2744                if (!bh)
2745                        goto cleanup;
2746                error = ext4_xattr_check_block(inode, bh);
2747                if (error)
2748                        goto cleanup;
2749                base = BHDR(bh);
2750                end = bh->b_data + bh->b_size;
2751                min_offs = end - base;
2752                bfree = ext4_xattr_free_space(BFIRST(bh), &min_offs, base,
2753                                              NULL);
2754                brelse(bh);
2755                if (bfree + ifree < isize_diff) {
2756                        if (!tried_min_extra_isize && s_min_extra_isize) {
2757                                tried_min_extra_isize++;
2758                                new_extra_isize = s_min_extra_isize;
2759                                goto retry;
2760                        }
2761                        error = -ENOSPC;
2762                        goto cleanup;
2763                }
2764        } else {
2765                bfree = inode->i_sb->s_blocksize;
2766        }
2767
2768        error = ext4_xattr_make_inode_space(handle, inode, raw_inode,
2769                                            isize_diff, ifree, bfree,
2770                                            &total_ino);
2771        if (error) {
2772                if (error == -ENOSPC && !tried_min_extra_isize &&
2773                    s_min_extra_isize) {
2774                        tried_min_extra_isize++;
2775                        new_extra_isize = s_min_extra_isize;
2776                        goto retry;
2777                }
2778                goto cleanup;
2779        }
2780shift:
2781        /* Adjust the offsets and shift the remaining entries ahead */
2782        ext4_xattr_shift_entries(IFIRST(header), EXT4_I(inode)->i_extra_isize
2783                        - new_extra_isize, (void *)raw_inode +
2784                        EXT4_GOOD_OLD_INODE_SIZE + new_extra_isize,
2785                        (void *)header, total_ino);
2786        EXT4_I(inode)->i_extra_isize = new_extra_isize;
2787
2788cleanup:
2789        if (error && (mnt_count != le16_to_cpu(sbi->s_es->s_mnt_count))) {
2790                ext4_warning(inode->i_sb, "Unable to expand inode %lu. Delete some EAs or run e2fsck.",
2791                             inode->i_ino);
2792                mnt_count = le16_to_cpu(sbi->s_es->s_mnt_count);
2793        }
2794        return error;
2795}
2796
2797#define EIA_INCR 16 /* must be 2^n */
2798#define EIA_MASK (EIA_INCR - 1)
2799
2800/* Add the large xattr @inode into @ea_inode_array for deferred iput().
2801 * If @ea_inode_array is new or full it will be grown and the old
2802 * contents copied over.
2803 */
2804static int
2805ext4_expand_inode_array(struct ext4_xattr_inode_array **ea_inode_array,
2806                        struct inode *inode)
2807{
2808        if (*ea_inode_array == NULL) {
2809                /*
2810                 * Start with 15 inodes, so it fits into a power-of-two size.
2811                 * If *ea_inode_array is NULL, this is essentially offsetof()
2812                 */
2813                (*ea_inode_array) =
2814                        kmalloc(offsetof(struct ext4_xattr_inode_array,
2815                                         inodes[EIA_MASK]),
2816                                GFP_NOFS);
2817                if (*ea_inode_array == NULL)
2818                        return -ENOMEM;
2819                (*ea_inode_array)->count = 0;
2820        } else if (((*ea_inode_array)->count & EIA_MASK) == EIA_MASK) {
2821                /* expand the array once all 15 + n * 16 slots are full */
2822                struct ext4_xattr_inode_array *new_array = NULL;
2823                int count = (*ea_inode_array)->count;
2824
2825                /* if new_array is NULL, this is essentially offsetof() */
2826                new_array = kmalloc(
2827                                offsetof(struct ext4_xattr_inode_array,
2828                                         inodes[count + EIA_INCR]),
2829                                GFP_NOFS);
2830                if (new_array == NULL)
2831                        return -ENOMEM;
2832                memcpy(new_array, *ea_inode_array,
2833                       offsetof(struct ext4_xattr_inode_array, inodes[count]));
2834                kfree(*ea_inode_array);
2835                *ea_inode_array = new_array;
2836        }
2837        (*ea_inode_array)->inodes[(*ea_inode_array)->count++] = inode;
2838        return 0;
2839}
2840
2841/*
2842 * ext4_xattr_delete_inode()
2843 *
2844 * Free extended attribute resources associated with this inode. Traverse
2845 * all entries and decrement reference on any xattr inodes associated with this
2846 * inode. This is called immediately before an inode is freed. We have exclusive
2847 * access to the inode. If an orphan inode is deleted it will also release its
2848 * references on xattr block and xattr inodes.
2849 */
2850int ext4_xattr_delete_inode(handle_t *handle, struct inode *inode,
2851                            struct ext4_xattr_inode_array **ea_inode_array,
2852                            int extra_credits)
2853{
2854        struct buffer_head *bh = NULL;
2855        struct ext4_xattr_ibody_header *header;
2856        struct ext4_iloc iloc = { .bh = NULL };
2857        struct ext4_xattr_entry *entry;
2858        struct inode *ea_inode;
2859        int error;
2860
2861        error = ext4_xattr_ensure_credits(handle, inode, extra_credits,
2862                                          NULL /* bh */,
2863                                          false /* dirty */,
2864                                          false /* block_csum */);
2865        if (error) {
2866                EXT4_ERROR_INODE(inode, "ensure credits (error %d)", error);
2867                goto cleanup;
2868        }
2869
2870        if (ext4_has_feature_ea_inode(inode->i_sb) &&
2871            ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
2872
2873                error = ext4_get_inode_loc(inode, &iloc);
2874                if (error) {
2875                        EXT4_ERROR_INODE(inode, "inode loc (error %d)", error);
2876                        goto cleanup;
2877                }
2878
2879                error = ext4_journal_get_write_access(handle, iloc.bh);
2880                if (error) {
2881                        EXT4_ERROR_INODE(inode, "write access (error %d)",
2882                                         error);
2883                        goto cleanup;
2884                }
2885
2886                header = IHDR(inode, ext4_raw_inode(&iloc));
2887                if (header->h_magic == cpu_to_le32(EXT4_XATTR_MAGIC))
2888                        ext4_xattr_inode_dec_ref_all(handle, inode, iloc.bh,
2889                                                     IFIRST(header),
2890                                                     false /* block_csum */,
2891                                                     ea_inode_array,
2892                                                     extra_credits,
2893                                                     false /* skip_quota */);
2894        }
2895
2896        if (EXT4_I(inode)->i_file_acl) {
2897                bh = sb_bread(inode->i_sb, EXT4_I(inode)->i_file_acl);
2898                if (!bh) {
2899                        EXT4_ERROR_INODE(inode, "block %llu read error",
2900                                         EXT4_I(inode)->i_file_acl);
2901                        error = -EIO;
2902                        goto cleanup;
2903                }
2904                error = ext4_xattr_check_block(inode, bh);
2905                if (error)
2906                        goto cleanup;
2907
2908                if (ext4_has_feature_ea_inode(inode->i_sb)) {
2909                        for (entry = BFIRST(bh); !IS_LAST_ENTRY(entry);
2910                             entry = EXT4_XATTR_NEXT(entry)) {
2911                                if (!entry->e_value_inum)
2912                                        continue;
2913                                error = ext4_xattr_inode_iget(inode,
2914                                              le32_to_cpu(entry->e_value_inum),
2915                                              le32_to_cpu(entry->e_hash),
2916                                              &ea_inode);
2917                                if (error)
2918                                        continue;
2919                                ext4_xattr_inode_free_quota(inode, ea_inode,
2920                                              le32_to_cpu(entry->e_value_size));
2921                                iput(ea_inode);
2922                        }
2923
2924                }
2925
2926                ext4_xattr_release_block(handle, inode, bh, ea_inode_array,
2927                                         extra_credits);
2928                /*
2929                 * Update i_file_acl value in the same transaction that releases
2930                 * block.
2931                 */
2932                EXT4_I(inode)->i_file_acl = 0;
2933                error = ext4_mark_inode_dirty(handle, inode);
2934                if (error) {
2935                        EXT4_ERROR_INODE(inode, "mark inode dirty (error %d)",
2936                                         error);
2937                        goto cleanup;
2938                }
2939        }
2940        error = 0;
2941cleanup:
2942        brelse(iloc.bh);
2943        brelse(bh);
2944        return error;
2945}
2946
2947void ext4_xattr_inode_array_free(struct ext4_xattr_inode_array *ea_inode_array)
2948{
2949        int idx;
2950
2951        if (ea_inode_array == NULL)
2952                return;
2953
2954        for (idx = 0; idx < ea_inode_array->count; ++idx)
2955                iput(ea_inode_array->inodes[idx]);
2956        kfree(ea_inode_array);
2957}
2958
2959/*
2960 * ext4_xattr_block_cache_insert()
2961 *
2962 * Create a new entry in the extended attribute block cache, and insert
2963 * it unless such an entry is already in the cache.
2964 *
2965 * Returns 0, or a negative error number on failure.
2966 */
2967static void
2968ext4_xattr_block_cache_insert(struct mb_cache *ea_block_cache,
2969                              struct buffer_head *bh)
2970{
2971        struct ext4_xattr_header *header = BHDR(bh);
2972        __u32 hash = le32_to_cpu(header->h_hash);
2973        int reusable = le32_to_cpu(header->h_refcount) <
2974                       EXT4_XATTR_REFCOUNT_MAX;
2975        int error;
2976
2977        if (!ea_block_cache)
2978                return;
2979        error = mb_cache_entry_create(ea_block_cache, GFP_NOFS, hash,
2980                                      bh->b_blocknr, reusable);
2981        if (error) {
2982                if (error == -EBUSY)
2983                        ea_bdebug(bh, "already in cache");
2984        } else
2985                ea_bdebug(bh, "inserting [%x]", (int)hash);
2986}
2987
2988/*
2989 * ext4_xattr_cmp()
2990 *
2991 * Compare two extended attribute blocks for equality.
2992 *
2993 * Returns 0 if the blocks are equal, 1 if they differ, and
2994 * a negative error number on errors.
2995 */
2996static int
2997ext4_xattr_cmp(struct ext4_xattr_header *header1,
2998               struct ext4_xattr_header *header2)
2999{
3000        struct ext4_xattr_entry *entry1, *entry2;
3001
3002        entry1 = ENTRY(header1+1);
3003        entry2 = ENTRY(header2+1);
3004        while (!IS_LAST_ENTRY(entry1)) {
3005                if (IS_LAST_ENTRY(entry2))
3006                        return 1;
3007                if (entry1->e_hash != entry2->e_hash ||
3008                    entry1->e_name_index != entry2->e_name_index ||
3009                    entry1->e_name_len != entry2->e_name_len ||
3010                    entry1->e_value_size != entry2->e_value_size ||
3011                    entry1->e_value_inum != entry2->e_value_inum ||
3012                    memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
3013                        return 1;
3014                if (!entry1->e_value_inum &&
3015                    memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
3016                           (char *)header2 + le16_to_cpu(entry2->e_value_offs),
3017                           le32_to_cpu(entry1->e_value_size)))
3018                        return 1;
3019
3020                entry1 = EXT4_XATTR_NEXT(entry1);
3021                entry2 = EXT4_XATTR_NEXT(entry2);
3022        }
3023        if (!IS_LAST_ENTRY(entry2))
3024                return 1;
3025        return 0;
3026}
3027
3028/*
3029 * ext4_xattr_block_cache_find()
3030 *
3031 * Find an identical extended attribute block.
3032 *
3033 * Returns a pointer to the block found, or NULL if such a block was
3034 * not found or an error occurred.
3035 */
3036static struct buffer_head *
3037ext4_xattr_block_cache_find(struct inode *inode,
3038                            struct ext4_xattr_header *header,
3039                            struct mb_cache_entry **pce)
3040{
3041        __u32 hash = le32_to_cpu(header->h_hash);
3042        struct mb_cache_entry *ce;
3043        struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
3044
3045        if (!ea_block_cache)
3046                return NULL;
3047        if (!header->h_hash)
3048                return NULL;  /* never share */
3049        ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
3050        ce = mb_cache_entry_find_first(ea_block_cache, hash);
3051        while (ce) {
3052                struct buffer_head *bh;
3053
3054                bh = sb_bread(inode->i_sb, ce->e_value);
3055                if (!bh) {
3056                        EXT4_ERROR_INODE(inode, "block %lu read error",
3057                                         (unsigned long)ce->e_value);
3058                } else if (ext4_xattr_cmp(header, BHDR(bh)) == 0) {
3059                        *pce = ce;
3060                        return bh;
3061                }
3062                brelse(bh);
3063                ce = mb_cache_entry_find_next(ea_block_cache, ce);
3064        }
3065        return NULL;
3066}
3067
3068#define NAME_HASH_SHIFT 5
3069#define VALUE_HASH_SHIFT 16
3070
3071/*
3072 * ext4_xattr_hash_entry()
3073 *
3074 * Compute the hash of an extended attribute.
3075 */
3076static __le32 ext4_xattr_hash_entry(char *name, size_t name_len, __le32 *value,
3077                                    size_t value_count)
3078{
3079        __u32 hash = 0;
3080
3081        while (name_len--) {
3082                hash = (hash << NAME_HASH_SHIFT) ^
3083                       (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
3084                       *name++;
3085        }
3086        while (value_count--) {
3087                hash = (hash << VALUE_HASH_SHIFT) ^
3088                       (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
3089                       le32_to_cpu(*value++);
3090        }
3091        return cpu_to_le32(hash);
3092}
3093
3094#undef NAME_HASH_SHIFT
3095#undef VALUE_HASH_SHIFT
3096
3097#define BLOCK_HASH_SHIFT 16
3098
3099/*
3100 * ext4_xattr_rehash()
3101 *
3102 * Re-compute the extended attribute hash value after an entry has changed.
3103 */
3104static void ext4_xattr_rehash(struct ext4_xattr_header *header)
3105{
3106        struct ext4_xattr_entry *here;
3107        __u32 hash = 0;
3108
3109        here = ENTRY(header+1);
3110        while (!IS_LAST_ENTRY(here)) {
3111                if (!here->e_hash) {
3112                        /* Block is not shared if an entry's hash value == 0 */
3113                        hash = 0;
3114                        break;
3115                }
3116                hash = (hash << BLOCK_HASH_SHIFT) ^
3117                       (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
3118                       le32_to_cpu(here->e_hash);
3119                here = EXT4_XATTR_NEXT(here);
3120        }
3121        header->h_hash = cpu_to_le32(hash);
3122}
3123
3124#undef BLOCK_HASH_SHIFT
3125
3126#define HASH_BUCKET_BITS        10
3127
3128struct mb_cache *
3129ext4_xattr_create_cache(void)
3130{
3131        return mb_cache_create(HASH_BUCKET_BITS);
3132}
3133
3134void ext4_xattr_destroy_cache(struct mb_cache *cache)
3135{
3136        if (cache)
3137                mb_cache_destroy(cache);
3138}
3139
3140