fs/ext4/xattr.c

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