Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
xattr.c
Go to the documentation of this file.
1 /*
2  * linux/fs/ext3/xattr.c
3  *
4  * Copyright (C) 2001-2003 Andreas Gruenbacher, <[email protected]>
5  *
6  * Fix by Harrison Xing <[email protected]>.
7  * Ext3 code with a lot of help from Eric Jarman <[email protected]>.
8  * Extended attributes for symlinks and special files added per
9  * suggestion of Luka Renko <[email protected]>.
10  * xattr consolidation Copyright (c) 2004 James Morris <[email protected]>,
11  * Red Hat Inc.
12  * ea-in-inode support by Alex Tomas <[email protected]> aka bzzz
13  * and Andreas Gruenbacher <[email protected]>.
14  */
15 
16 /*
17  * Extended attributes are stored directly in inodes (on file systems with
18  * inodes bigger than 128 bytes) and on additional disk blocks. The i_file_acl
19  * field contains the block number if an inode uses an additional block. All
20  * attributes must fit in the inode and one additional block. Blocks that
21  * contain the identical set of attributes may be shared among several inodes.
22  * Identical blocks are detected by keeping a cache of blocks that have
23  * recently been accessed.
24  *
25  * The attributes in inodes and on blocks have a different header; the entries
26  * are stored in the same format:
27  *
28  * +------------------+
29  * | header |
30  * | entry 1 | |
31  * | entry 2 | | growing downwards
32  * | entry 3 | v
33  * | four null bytes |
34  * | . . . |
35  * | value 1 | ^
36  * | value 3 | | growing upwards
37  * | value 2 | |
38  * +------------------+
39  *
40  * The header is followed by multiple entry descriptors. In disk blocks, the
41  * entry descriptors are kept sorted. In inodes, they are unsorted. The
42  * attribute values are aligned to the end of the block in no specific order.
43  *
44  * Locking strategy
45  * ----------------
46  * EXT3_I(inode)->i_file_acl is protected by EXT3_I(inode)->xattr_sem.
47  * EA blocks are only changed if they are exclusive to an inode, so
48  * holding xattr_sem also means that nothing but the EA block's reference
49  * count can change. Multiple writers to the same block are synchronized
50  * by the buffer lock.
51  */
52 
53 #include "ext3.h"
54 #include <linux/mbcache.h>
55 #include <linux/quotaops.h>
56 #include "xattr.h"
57 #include "acl.h"
58 
59 #define BHDR(bh) ((struct ext3_xattr_header *)((bh)->b_data))
60 #define ENTRY(ptr) ((struct ext3_xattr_entry *)(ptr))
61 #define BFIRST(bh) ENTRY(BHDR(bh)+1)
62 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
63 
64 #define IHDR(inode, raw_inode) \
65  ((struct ext3_xattr_ibody_header *) \
66  ((void *)raw_inode + \
67  EXT3_GOOD_OLD_INODE_SIZE + \
68  EXT3_I(inode)->i_extra_isize))
69 #define IFIRST(hdr) ((struct ext3_xattr_entry *)((hdr)+1))
70 
71 #ifdef EXT3_XATTR_DEBUG
72 # define ea_idebug(inode, f...) do { \
73  printk(KERN_DEBUG "inode %s:%lu: ", \
74  inode->i_sb->s_id, inode->i_ino); \
75  printk(f); \
76  printk("\n"); \
77  } while (0)
78 # define ea_bdebug(bh, f...) do { \
79  char b[BDEVNAME_SIZE]; \
80  printk(KERN_DEBUG "block %s:%lu: ", \
81  bdevname(bh->b_bdev, b), \
82  (unsigned long) bh->b_blocknr); \
83  printk(f); \
84  printk("\n"); \
85  } while (0)
86 #else
87 # define ea_idebug(f...)
88 # define ea_bdebug(f...)
89 #endif
90 
91 static void ext3_xattr_cache_insert(struct buffer_head *);
92 static struct buffer_head *ext3_xattr_cache_find(struct inode *,
93  struct ext3_xattr_header *,
94  struct mb_cache_entry **);
95 static void ext3_xattr_rehash(struct ext3_xattr_header *,
96  struct ext3_xattr_entry *);
97 static int ext3_xattr_list(struct dentry *dentry, char *buffer,
98  size_t buffer_size);
99 
100 static struct mb_cache *ext3_xattr_cache;
101 
102 static const struct xattr_handler *ext3_xattr_handler_map[] = {
104 #ifdef CONFIG_EXT3_FS_POSIX_ACL
107 #endif
109 #ifdef CONFIG_EXT3_FS_SECURITY
111 #endif
112 };
113 
117 #ifdef CONFIG_EXT3_FS_POSIX_ACL
120 #endif
121 #ifdef CONFIG_EXT3_FS_SECURITY
123 #endif
124  NULL
125 };
126 
127 static inline const struct xattr_handler *
128 ext3_xattr_handler(int name_index)
129 {
130  const struct xattr_handler *handler = NULL;
131 
132  if (name_index > 0 && name_index < ARRAY_SIZE(ext3_xattr_handler_map))
133  handler = ext3_xattr_handler_map[name_index];
134  return handler;
135 }
136 
137 /*
138  * Inode operation listxattr()
139  *
140  * dentry->d_inode->i_mutex: don't care
141  */
142 ssize_t
143 ext3_listxattr(struct dentry *dentry, char *buffer, size_t size)
144 {
145  return ext3_xattr_list(dentry, buffer, size);
146 }
147 
148 static int
149 ext3_xattr_check_names(struct ext3_xattr_entry *entry, void *end)
150 {
151  while (!IS_LAST_ENTRY(entry)) {
152  struct ext3_xattr_entry *next = EXT3_XATTR_NEXT(entry);
153  if ((void *)next >= end)
154  return -EIO;
155  entry = next;
156  }
157  return 0;
158 }
159 
160 static inline int
161 ext3_xattr_check_block(struct buffer_head *bh)
162 {
163  int error;
164 
165  if (BHDR(bh)->h_magic != cpu_to_le32(EXT3_XATTR_MAGIC) ||
166  BHDR(bh)->h_blocks != cpu_to_le32(1))
167  return -EIO;
168  error = ext3_xattr_check_names(BFIRST(bh), bh->b_data + bh->b_size);
169  return error;
170 }
171 
172 static inline int
173 ext3_xattr_check_entry(struct ext3_xattr_entry *entry, size_t size)
174 {
175  size_t value_size = le32_to_cpu(entry->e_value_size);
176 
177  if (entry->e_value_block != 0 || value_size > size ||
178  le16_to_cpu(entry->e_value_offs) + value_size > size)
179  return -EIO;
180  return 0;
181 }
182 
183 static int
184 ext3_xattr_find_entry(struct ext3_xattr_entry **pentry, int name_index,
185  const char *name, size_t size, int sorted)
186 {
187  struct ext3_xattr_entry *entry;
188  size_t name_len;
189  int cmp = 1;
190 
191  if (name == NULL)
192  return -EINVAL;
193  name_len = strlen(name);
194  entry = *pentry;
195  for (; !IS_LAST_ENTRY(entry); entry = EXT3_XATTR_NEXT(entry)) {
196  cmp = name_index - entry->e_name_index;
197  if (!cmp)
198  cmp = name_len - entry->e_name_len;
199  if (!cmp)
200  cmp = memcmp(name, entry->e_name, name_len);
201  if (cmp <= 0 && (sorted || cmp == 0))
202  break;
203  }
204  *pentry = entry;
205  if (!cmp && ext3_xattr_check_entry(entry, size))
206  return -EIO;
207  return cmp ? -ENODATA : 0;
208 }
209 
210 static int
211 ext3_xattr_block_get(struct inode *inode, int name_index, const char *name,
212  void *buffer, size_t buffer_size)
213 {
214  struct buffer_head *bh = NULL;
215  struct ext3_xattr_entry *entry;
216  size_t size;
217  int error;
218 
219  ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
220  name_index, name, buffer, (long)buffer_size);
221 
222  error = -ENODATA;
223  if (!EXT3_I(inode)->i_file_acl)
224  goto cleanup;
225  ea_idebug(inode, "reading block %u", EXT3_I(inode)->i_file_acl);
226  bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
227  if (!bh)
228  goto cleanup;
229  ea_bdebug(bh, "b_count=%d, refcount=%d",
230  atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
231  if (ext3_xattr_check_block(bh)) {
232 bad_block: ext3_error(inode->i_sb, __func__,
233  "inode %lu: bad block "E3FSBLK, inode->i_ino,
234  EXT3_I(inode)->i_file_acl);
235  error = -EIO;
236  goto cleanup;
237  }
238  ext3_xattr_cache_insert(bh);
239  entry = BFIRST(bh);
240  error = ext3_xattr_find_entry(&entry, name_index, name, bh->b_size, 1);
241  if (error == -EIO)
242  goto bad_block;
243  if (error)
244  goto cleanup;
245  size = le32_to_cpu(entry->e_value_size);
246  if (buffer) {
247  error = -ERANGE;
248  if (size > buffer_size)
249  goto cleanup;
250  memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
251  size);
252  }
253  error = size;
254 
255 cleanup:
256  brelse(bh);
257  return error;
258 }
259 
260 static int
261 ext3_xattr_ibody_get(struct inode *inode, int name_index, const char *name,
262  void *buffer, size_t buffer_size)
263 {
265  struct ext3_xattr_entry *entry;
266  struct ext3_inode *raw_inode;
267  struct ext3_iloc iloc;
268  size_t size;
269  void *end;
270  int error;
271 
272  if (!ext3_test_inode_state(inode, EXT3_STATE_XATTR))
273  return -ENODATA;
274  error = ext3_get_inode_loc(inode, &iloc);
275  if (error)
276  return error;
277  raw_inode = ext3_raw_inode(&iloc);
278  header = IHDR(inode, raw_inode);
279  entry = IFIRST(header);
280  end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
281  error = ext3_xattr_check_names(entry, end);
282  if (error)
283  goto cleanup;
284  error = ext3_xattr_find_entry(&entry, name_index, name,
285  end - (void *)entry, 0);
286  if (error)
287  goto cleanup;
288  size = le32_to_cpu(entry->e_value_size);
289  if (buffer) {
290  error = -ERANGE;
291  if (size > buffer_size)
292  goto cleanup;
293  memcpy(buffer, (void *)IFIRST(header) +
294  le16_to_cpu(entry->e_value_offs), size);
295  }
296  error = size;
297 
298 cleanup:
299  brelse(iloc.bh);
300  return error;
301 }
302 
303 /*
304  * ext3_xattr_get()
305  *
306  * Copy an extended attribute into the buffer
307  * provided, or compute the buffer size required.
308  * Buffer is NULL to compute the size of the buffer required.
309  *
310  * Returns a negative error number on failure, or the number of bytes
311  * used / required on success.
312  */
313 int
314 ext3_xattr_get(struct inode *inode, int name_index, const char *name,
315  void *buffer, size_t buffer_size)
316 {
317  int error;
318 
319  down_read(&EXT3_I(inode)->xattr_sem);
320  error = ext3_xattr_ibody_get(inode, name_index, name, buffer,
321  buffer_size);
322  if (error == -ENODATA)
323  error = ext3_xattr_block_get(inode, name_index, name, buffer,
324  buffer_size);
325  up_read(&EXT3_I(inode)->xattr_sem);
326  return error;
327 }
328 
329 static int
330 ext3_xattr_list_entries(struct dentry *dentry, struct ext3_xattr_entry *entry,
331  char *buffer, size_t buffer_size)
332 {
333  size_t rest = buffer_size;
334 
335  for (; !IS_LAST_ENTRY(entry); entry = EXT3_XATTR_NEXT(entry)) {
336  const struct xattr_handler *handler =
337  ext3_xattr_handler(entry->e_name_index);
338 
339  if (handler) {
340  size_t size = handler->list(dentry, buffer, rest,
341  entry->e_name,
342  entry->e_name_len,
343  handler->flags);
344  if (buffer) {
345  if (size > rest)
346  return -ERANGE;
347  buffer += size;
348  }
349  rest -= size;
350  }
351  }
352  return buffer_size - rest;
353 }
354 
355 static int
356 ext3_xattr_block_list(struct dentry *dentry, char *buffer, size_t buffer_size)
357 {
358  struct inode *inode = dentry->d_inode;
359  struct buffer_head *bh = NULL;
360  int error;
361 
362  ea_idebug(inode, "buffer=%p, buffer_size=%ld",
363  buffer, (long)buffer_size);
364 
365  error = 0;
366  if (!EXT3_I(inode)->i_file_acl)
367  goto cleanup;
368  ea_idebug(inode, "reading block %u", EXT3_I(inode)->i_file_acl);
369  bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
370  error = -EIO;
371  if (!bh)
372  goto cleanup;
373  ea_bdebug(bh, "b_count=%d, refcount=%d",
374  atomic_read(&(bh->b_count)), le32_to_cpu(BHDR(bh)->h_refcount));
375  if (ext3_xattr_check_block(bh)) {
376  ext3_error(inode->i_sb, __func__,
377  "inode %lu: bad block "E3FSBLK, inode->i_ino,
378  EXT3_I(inode)->i_file_acl);
379  error = -EIO;
380  goto cleanup;
381  }
382  ext3_xattr_cache_insert(bh);
383  error = ext3_xattr_list_entries(dentry, BFIRST(bh), buffer, buffer_size);
384 
385 cleanup:
386  brelse(bh);
387 
388  return error;
389 }
390 
391 static int
392 ext3_xattr_ibody_list(struct dentry *dentry, char *buffer, size_t buffer_size)
393 {
394  struct inode *inode = dentry->d_inode;
395  struct ext3_xattr_ibody_header *header;
396  struct ext3_inode *raw_inode;
397  struct ext3_iloc iloc;
398  void *end;
399  int error;
400 
401  if (!ext3_test_inode_state(inode, EXT3_STATE_XATTR))
402  return 0;
403  error = ext3_get_inode_loc(inode, &iloc);
404  if (error)
405  return error;
406  raw_inode = ext3_raw_inode(&iloc);
407  header = IHDR(inode, raw_inode);
408  end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
409  error = ext3_xattr_check_names(IFIRST(header), end);
410  if (error)
411  goto cleanup;
412  error = ext3_xattr_list_entries(dentry, IFIRST(header),
413  buffer, buffer_size);
414 
415 cleanup:
416  brelse(iloc.bh);
417  return error;
418 }
419 
420 /*
421  * ext3_xattr_list()
422  *
423  * Copy a list of attribute names into the buffer
424  * provided, or compute the buffer size required.
425  * Buffer is NULL to compute the size of the buffer required.
426  *
427  * Returns a negative error number on failure, or the number of bytes
428  * used / required on success.
429  */
430 static int
431 ext3_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
432 {
433  int i_error, b_error;
434 
435  down_read(&EXT3_I(dentry->d_inode)->xattr_sem);
436  i_error = ext3_xattr_ibody_list(dentry, buffer, buffer_size);
437  if (i_error < 0) {
438  b_error = 0;
439  } else {
440  if (buffer) {
441  buffer += i_error;
442  buffer_size -= i_error;
443  }
444  b_error = ext3_xattr_block_list(dentry, buffer, buffer_size);
445  if (b_error < 0)
446  i_error = 0;
447  }
448  up_read(&EXT3_I(dentry->d_inode)->xattr_sem);
449  return i_error + b_error;
450 }
451 
452 /*
453  * If the EXT3_FEATURE_COMPAT_EXT_ATTR feature of this file system is
454  * not set, set it.
455  */
456 static void ext3_xattr_update_super_block(handle_t *handle,
457  struct super_block *sb)
458 {
460  return;
461 
462  if (ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh) == 0) {
464  ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
465  }
466 }
467 
468 /*
469  * Release the xattr block BH: If the reference count is > 1, decrement
470  * it; otherwise free the block.
471  */
472 static void
473 ext3_xattr_release_block(handle_t *handle, struct inode *inode,
474  struct buffer_head *bh)
475 {
476  struct mb_cache_entry *ce = NULL;
477  int error = 0;
478 
479  ce = mb_cache_entry_get(ext3_xattr_cache, bh->b_bdev, bh->b_blocknr);
480  error = ext3_journal_get_write_access(handle, bh);
481  if (error)
482  goto out;
483 
484  lock_buffer(bh);
485 
486  if (BHDR(bh)->h_refcount == cpu_to_le32(1)) {
487  ea_bdebug(bh, "refcount now=0; freeing");
488  if (ce)
490  ext3_free_blocks(handle, inode, bh->b_blocknr, 1);
491  get_bh(bh);
492  ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
493  } else {
494  le32_add_cpu(&BHDR(bh)->h_refcount, -1);
495  error = ext3_journal_dirty_metadata(handle, bh);
496  if (IS_SYNC(inode))
497  handle->h_sync = 1;
498  dquot_free_block(inode, 1);
499  ea_bdebug(bh, "refcount now=%d; releasing",
500  le32_to_cpu(BHDR(bh)->h_refcount));
501  if (ce)
503  }
504  unlock_buffer(bh);
505 out:
506  ext3_std_error(inode->i_sb, error);
507  return;
508 }
509 
512  const char *name;
513  const void *value;
514  size_t value_len;
515 };
516 
519  void *base;
520  void *end;
523 };
524 
525 static int
526 ext3_xattr_set_entry(struct ext3_xattr_info *i, struct ext3_xattr_search *s)
527 {
528  struct ext3_xattr_entry *last;
529  size_t free, min_offs = s->end - s->base, name_len = strlen(i->name);
530 
531  /* Compute min_offs and last. */
532  last = s->first;
533  for (; !IS_LAST_ENTRY(last); last = EXT3_XATTR_NEXT(last)) {
534  if (!last->e_value_block && last->e_value_size) {
535  size_t offs = le16_to_cpu(last->e_value_offs);
536  if (offs < min_offs)
537  min_offs = offs;
538  }
539  }
540  free = min_offs - ((void *)last - s->base) - sizeof(__u32);
541  if (!s->not_found) {
542  if (!s->here->e_value_block && s->here->e_value_size) {
543  size_t size = le32_to_cpu(s->here->e_value_size);
544  free += EXT3_XATTR_SIZE(size);
545  }
546  free += EXT3_XATTR_LEN(name_len);
547  }
548  if (i->value) {
549  if (free < EXT3_XATTR_SIZE(i->value_len) ||
550  free < EXT3_XATTR_LEN(name_len) +
551  EXT3_XATTR_SIZE(i->value_len))
552  return -ENOSPC;
553  }
554 
555  if (i->value && s->not_found) {
556  /* Insert the new name. */
557  size_t size = EXT3_XATTR_LEN(name_len);
558  size_t rest = (void *)last - (void *)s->here + sizeof(__u32);
559  memmove((void *)s->here + size, s->here, rest);
560  memset(s->here, 0, size);
561  s->here->e_name_index = i->name_index;
562  s->here->e_name_len = name_len;
563  memcpy(s->here->e_name, i->name, name_len);
564  } else {
565  if (!s->here->e_value_block && s->here->e_value_size) {
566  void *first_val = s->base + min_offs;
567  size_t offs = le16_to_cpu(s->here->e_value_offs);
568  void *val = s->base + offs;
569  size_t size = EXT3_XATTR_SIZE(
570  le32_to_cpu(s->here->e_value_size));
571 
572  if (i->value && size == EXT3_XATTR_SIZE(i->value_len)) {
573  /* The old and the new value have the same
574  size. Just replace. */
575  s->here->e_value_size =
576  cpu_to_le32(i->value_len);
577  memset(val + size - EXT3_XATTR_PAD, 0,
578  EXT3_XATTR_PAD); /* Clear pad bytes. */
579  memcpy(val, i->value, i->value_len);
580  return 0;
581  }
582 
583  /* Remove the old value. */
584  memmove(first_val + size, first_val, val - first_val);
585  memset(first_val, 0, size);
586  s->here->e_value_size = 0;
587  s->here->e_value_offs = 0;
588  min_offs += size;
589 
590  /* Adjust all value offsets. */
591  last = s->first;
592  while (!IS_LAST_ENTRY(last)) {
593  size_t o = le16_to_cpu(last->e_value_offs);
594  if (!last->e_value_block &&
595  last->e_value_size && o < offs)
596  last->e_value_offs =
597  cpu_to_le16(o + size);
598  last = EXT3_XATTR_NEXT(last);
599  }
600  }
601  if (!i->value) {
602  /* Remove the old name. */
603  size_t size = EXT3_XATTR_LEN(name_len);
604  last = ENTRY((void *)last - size);
605  memmove(s->here, (void *)s->here + size,
606  (void *)last - (void *)s->here + sizeof(__u32));
607  memset(last, 0, size);
608  }
609  }
610 
611  if (i->value) {
612  /* Insert the new value. */
613  s->here->e_value_size = cpu_to_le32(i->value_len);
614  if (i->value_len) {
615  size_t size = EXT3_XATTR_SIZE(i->value_len);
616  void *val = s->base + min_offs - size;
617  s->here->e_value_offs = cpu_to_le16(min_offs - size);
618  memset(val + size - EXT3_XATTR_PAD, 0,
619  EXT3_XATTR_PAD); /* Clear the pad bytes. */
620  memcpy(val, i->value, i->value_len);
621  }
622  }
623  return 0;
624 }
625 
628  struct buffer_head *bh;
629 };
630 
631 static int
632 ext3_xattr_block_find(struct inode *inode, struct ext3_xattr_info *i,
633  struct ext3_xattr_block_find *bs)
634 {
635  struct super_block *sb = inode->i_sb;
636  int error;
637 
638  ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
639  i->name_index, i->name, i->value, (long)i->value_len);
640 
641  if (EXT3_I(inode)->i_file_acl) {
642  /* The inode already has an extended attribute block. */
643  bs->bh = sb_bread(sb, EXT3_I(inode)->i_file_acl);
644  error = -EIO;
645  if (!bs->bh)
646  goto cleanup;
647  ea_bdebug(bs->bh, "b_count=%d, refcount=%d",
648  atomic_read(&(bs->bh->b_count)),
649  le32_to_cpu(BHDR(bs->bh)->h_refcount));
650  if (ext3_xattr_check_block(bs->bh)) {
651  ext3_error(sb, __func__,
652  "inode %lu: bad block "E3FSBLK, inode->i_ino,
653  EXT3_I(inode)->i_file_acl);
654  error = -EIO;
655  goto cleanup;
656  }
657  /* Find the named attribute. */
658  bs->s.base = BHDR(bs->bh);
659  bs->s.first = BFIRST(bs->bh);
660  bs->s.end = bs->bh->b_data + bs->bh->b_size;
661  bs->s.here = bs->s.first;
662  error = ext3_xattr_find_entry(&bs->s.here, i->name_index,
663  i->name, bs->bh->b_size, 1);
664  if (error && error != -ENODATA)
665  goto cleanup;
666  bs->s.not_found = error;
667  }
668  error = 0;
669 
670 cleanup:
671  return error;
672 }
673 
674 static int
675 ext3_xattr_block_set(handle_t *handle, struct inode *inode,
676  struct ext3_xattr_info *i,
677  struct ext3_xattr_block_find *bs)
678 {
679  struct super_block *sb = inode->i_sb;
680  struct buffer_head *new_bh = NULL;
681  struct ext3_xattr_search *s = &bs->s;
682  struct mb_cache_entry *ce = NULL;
683  int error = 0;
684 
685 #define header(x) ((struct ext3_xattr_header *)(x))
686 
687  if (i->value && i->value_len > sb->s_blocksize)
688  return -ENOSPC;
689  if (s->base) {
690  ce = mb_cache_entry_get(ext3_xattr_cache, bs->bh->b_bdev,
691  bs->bh->b_blocknr);
692  error = ext3_journal_get_write_access(handle, bs->bh);
693  if (error)
694  goto cleanup;
695  lock_buffer(bs->bh);
696 
697  if (header(s->base)->h_refcount == cpu_to_le32(1)) {
698  if (ce) {
700  ce = NULL;
701  }
702  ea_bdebug(bs->bh, "modifying in-place");
703  error = ext3_xattr_set_entry(i, s);
704  if (!error) {
705  if (!IS_LAST_ENTRY(s->first))
706  ext3_xattr_rehash(header(s->base),
707  s->here);
708  ext3_xattr_cache_insert(bs->bh);
709  }
710  unlock_buffer(bs->bh);
711  if (error == -EIO)
712  goto bad_block;
713  if (!error)
714  error = ext3_journal_dirty_metadata(handle,
715  bs->bh);
716  if (error)
717  goto cleanup;
718  goto inserted;
719  } else {
720  int offset = (char *)s->here - bs->bh->b_data;
721 
722  unlock_buffer(bs->bh);
723  journal_release_buffer(handle, bs->bh);
724 
725  if (ce) {
727  ce = NULL;
728  }
729  ea_bdebug(bs->bh, "cloning");
730  s->base = kmalloc(bs->bh->b_size, GFP_NOFS);
731  error = -ENOMEM;
732  if (s->base == NULL)
733  goto cleanup;
734  memcpy(s->base, BHDR(bs->bh), bs->bh->b_size);
735  s->first = ENTRY(header(s->base)+1);
736  header(s->base)->h_refcount = cpu_to_le32(1);
737  s->here = ENTRY(s->base + offset);
738  s->end = s->base + bs->bh->b_size;
739  }
740  } else {
741  /* Allocate a buffer where we construct the new block. */
742  s->base = kzalloc(sb->s_blocksize, GFP_NOFS);
743  /* assert(header == s->base) */
744  error = -ENOMEM;
745  if (s->base == NULL)
746  goto cleanup;
747  header(s->base)->h_magic = cpu_to_le32(EXT3_XATTR_MAGIC);
748  header(s->base)->h_blocks = cpu_to_le32(1);
749  header(s->base)->h_refcount = cpu_to_le32(1);
750  s->first = ENTRY(header(s->base)+1);
751  s->here = ENTRY(header(s->base)+1);
752  s->end = s->base + sb->s_blocksize;
753  }
754 
755  error = ext3_xattr_set_entry(i, s);
756  if (error == -EIO)
757  goto bad_block;
758  if (error)
759  goto cleanup;
760  if (!IS_LAST_ENTRY(s->first))
761  ext3_xattr_rehash(header(s->base), s->here);
762 
763 inserted:
764  if (!IS_LAST_ENTRY(s->first)) {
765  new_bh = ext3_xattr_cache_find(inode, header(s->base), &ce);
766  if (new_bh) {
767  /* We found an identical block in the cache. */
768  if (new_bh == bs->bh)
769  ea_bdebug(new_bh, "keeping");
770  else {
771  /* The old block is released after updating
772  the inode. */
773  error = dquot_alloc_block(inode, 1);
774  if (error)
775  goto cleanup;
776  error = ext3_journal_get_write_access(handle,
777  new_bh);
778  if (error)
779  goto cleanup_dquot;
780  lock_buffer(new_bh);
781  le32_add_cpu(&BHDR(new_bh)->h_refcount, 1);
782  ea_bdebug(new_bh, "reusing; refcount now=%d",
783  le32_to_cpu(BHDR(new_bh)->h_refcount));
784  unlock_buffer(new_bh);
785  error = ext3_journal_dirty_metadata(handle,
786  new_bh);
787  if (error)
788  goto cleanup_dquot;
789  }
791  ce = NULL;
792  } else if (bs->bh && s->base == bs->bh->b_data) {
793  /* We were modifying this block in-place. */
794  ea_bdebug(bs->bh, "keeping this block");
795  new_bh = bs->bh;
796  get_bh(new_bh);
797  } else {
798  /* We need to allocate a new block */
799  ext3_fsblk_t goal = ext3_group_first_block_no(sb,
800  EXT3_I(inode)->i_block_group);
802 
803  /*
804  * Protect us agaist concurrent allocations to the
805  * same inode from ext3_..._writepage(). Reservation
806  * code does not expect racing allocations.
807  */
808  mutex_lock(&EXT3_I(inode)->truncate_mutex);
809  block = ext3_new_block(handle, inode, goal, &error);
810  mutex_unlock(&EXT3_I(inode)->truncate_mutex);
811  if (error)
812  goto cleanup;
813  ea_idebug(inode, "creating block %d", block);
814 
815  new_bh = sb_getblk(sb, block);
816  if (!new_bh) {
817 getblk_failed:
818  ext3_free_blocks(handle, inode, block, 1);
819  error = -EIO;
820  goto cleanup;
821  }
822  lock_buffer(new_bh);
823  error = ext3_journal_get_create_access(handle, new_bh);
824  if (error) {
825  unlock_buffer(new_bh);
826  goto getblk_failed;
827  }
828  memcpy(new_bh->b_data, s->base, new_bh->b_size);
829  set_buffer_uptodate(new_bh);
830  unlock_buffer(new_bh);
831  ext3_xattr_cache_insert(new_bh);
832  error = ext3_journal_dirty_metadata(handle, new_bh);
833  if (error)
834  goto cleanup;
835  }
836  }
837 
838  /* Update the inode. */
839  EXT3_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
840 
841  /* Drop the previous xattr block. */
842  if (bs->bh && bs->bh != new_bh)
843  ext3_xattr_release_block(handle, inode, bs->bh);
844  error = 0;
845 
846 cleanup:
847  if (ce)
849  brelse(new_bh);
850  if (!(bs->bh && s->base == bs->bh->b_data))
851  kfree(s->base);
852 
853  return error;
854 
855 cleanup_dquot:
856  dquot_free_block(inode, 1);
857  goto cleanup;
858 
859 bad_block:
860  ext3_error(inode->i_sb, __func__,
861  "inode %lu: bad block "E3FSBLK, inode->i_ino,
862  EXT3_I(inode)->i_file_acl);
863  goto cleanup;
864 
865 #undef header
866 }
867 
870  struct ext3_iloc iloc;
871 };
872 
873 static int
874 ext3_xattr_ibody_find(struct inode *inode, struct ext3_xattr_info *i,
875  struct ext3_xattr_ibody_find *is)
876 {
877  struct ext3_xattr_ibody_header *header;
878  struct ext3_inode *raw_inode;
879  int error;
880 
881  if (EXT3_I(inode)->i_extra_isize == 0)
882  return 0;
883  raw_inode = ext3_raw_inode(&is->iloc);
884  header = IHDR(inode, raw_inode);
885  is->s.base = is->s.first = IFIRST(header);
886  is->s.here = is->s.first;
887  is->s.end = (void *)raw_inode + EXT3_SB(inode->i_sb)->s_inode_size;
888  if (ext3_test_inode_state(inode, EXT3_STATE_XATTR)) {
889  error = ext3_xattr_check_names(IFIRST(header), is->s.end);
890  if (error)
891  return error;
892  /* Find the named attribute. */
893  error = ext3_xattr_find_entry(&is->s.here, i->name_index,
894  i->name, is->s.end -
895  (void *)is->s.base, 0);
896  if (error && error != -ENODATA)
897  return error;
898  is->s.not_found = error;
899  }
900  return 0;
901 }
902 
903 static int
904 ext3_xattr_ibody_set(handle_t *handle, struct inode *inode,
905  struct ext3_xattr_info *i,
906  struct ext3_xattr_ibody_find *is)
907 {
908  struct ext3_xattr_ibody_header *header;
909  struct ext3_xattr_search *s = &is->s;
910  int error;
911 
912  if (EXT3_I(inode)->i_extra_isize == 0)
913  return -ENOSPC;
914  error = ext3_xattr_set_entry(i, s);
915  if (error)
916  return error;
917  header = IHDR(inode, ext3_raw_inode(&is->iloc));
918  if (!IS_LAST_ENTRY(s->first)) {
920  ext3_set_inode_state(inode, EXT3_STATE_XATTR);
921  } else {
922  header->h_magic = cpu_to_le32(0);
923  ext3_clear_inode_state(inode, EXT3_STATE_XATTR);
924  }
925  return 0;
926 }
927 
928 /*
929  * ext3_xattr_set_handle()
930  *
931  * Create, replace or remove an extended attribute for this inode. Value
932  * is NULL to remove an existing extended attribute, and non-NULL to
933  * either replace an existing extended attribute, or create a new extended
934  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
935  * specify that an extended attribute must exist and must not exist
936  * previous to the call, respectively.
937  *
938  * Returns 0, or a negative error number on failure.
939  */
940 int
941 ext3_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
942  const char *name, const void *value, size_t value_len,
943  int flags)
944 {
945  struct ext3_xattr_info i = {
947  .name = name,
948  .value = value,
949  .value_len = value_len,
950 
951  };
952  struct ext3_xattr_ibody_find is = {
953  .s = { .not_found = -ENODATA, },
954  };
955  struct ext3_xattr_block_find bs = {
956  .s = { .not_found = -ENODATA, },
957  };
958  int error;
959 
960  if (!name)
961  return -EINVAL;
962  if (strlen(name) > 255)
963  return -ERANGE;
964  down_write(&EXT3_I(inode)->xattr_sem);
965  error = ext3_get_inode_loc(inode, &is.iloc);
966  if (error)
967  goto cleanup;
968 
969  error = ext3_journal_get_write_access(handle, is.iloc.bh);
970  if (error)
971  goto cleanup;
972 
973  if (ext3_test_inode_state(inode, EXT3_STATE_NEW)) {
974  struct ext3_inode *raw_inode = ext3_raw_inode(&is.iloc);
975  memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
976  ext3_clear_inode_state(inode, EXT3_STATE_NEW);
977  }
978 
979  error = ext3_xattr_ibody_find(inode, &i, &is);
980  if (error)
981  goto cleanup;
982  if (is.s.not_found)
983  error = ext3_xattr_block_find(inode, &i, &bs);
984  if (error)
985  goto cleanup;
986  if (is.s.not_found && bs.s.not_found) {
987  error = -ENODATA;
988  if (flags & XATTR_REPLACE)
989  goto cleanup;
990  error = 0;
991  if (!value)
992  goto cleanup;
993  } else {
994  error = -EEXIST;
995  if (flags & XATTR_CREATE)
996  goto cleanup;
997  }
998  if (!value) {
999  if (!is.s.not_found)
1000  error = ext3_xattr_ibody_set(handle, inode, &i, &is);
1001  else if (!bs.s.not_found)
1002  error = ext3_xattr_block_set(handle, inode, &i, &bs);
1003  } else {
1004  error = ext3_xattr_ibody_set(handle, inode, &i, &is);
1005  if (!error && !bs.s.not_found) {
1006  i.value = NULL;
1007  error = ext3_xattr_block_set(handle, inode, &i, &bs);
1008  } else if (error == -ENOSPC) {
1009  if (EXT3_I(inode)->i_file_acl && !bs.s.base) {
1010  error = ext3_xattr_block_find(inode, &i, &bs);
1011  if (error)
1012  goto cleanup;
1013  }
1014  error = ext3_xattr_block_set(handle, inode, &i, &bs);
1015  if (error)
1016  goto cleanup;
1017  if (!is.s.not_found) {
1018  i.value = NULL;
1019  error = ext3_xattr_ibody_set(handle, inode, &i,
1020  &is);
1021  }
1022  }
1023  }
1024  if (!error) {
1025  ext3_xattr_update_super_block(handle, inode->i_sb);
1026  inode->i_ctime = CURRENT_TIME_SEC;
1027  error = ext3_mark_iloc_dirty(handle, inode, &is.iloc);
1028  /*
1029  * The bh is consumed by ext3_mark_iloc_dirty, even with
1030  * error != 0.
1031  */
1032  is.iloc.bh = NULL;
1033  if (IS_SYNC(inode))
1034  handle->h_sync = 1;
1035  }
1036 
1037 cleanup:
1038  brelse(is.iloc.bh);
1039  brelse(bs.bh);
1040  up_write(&EXT3_I(inode)->xattr_sem);
1041  return error;
1042 }
1043 
1044 /*
1045  * ext3_xattr_set()
1046  *
1047  * Like ext3_xattr_set_handle, but start from an inode. This extended
1048  * attribute modification is a filesystem transaction by itself.
1049  *
1050  * Returns 0, or a negative error number on failure.
1051  */
1052 int
1053 ext3_xattr_set(struct inode *inode, int name_index, const char *name,
1054  const void *value, size_t value_len, int flags)
1055 {
1056  handle_t *handle;
1057  int error, retries = 0;
1058 
1059 retry:
1060  handle = ext3_journal_start(inode, EXT3_DATA_TRANS_BLOCKS(inode->i_sb));
1061  if (IS_ERR(handle)) {
1062  error = PTR_ERR(handle);
1063  } else {
1064  int error2;
1065 
1066  error = ext3_xattr_set_handle(handle, inode, name_index, name,
1067  value, value_len, flags);
1068  error2 = ext3_journal_stop(handle);
1069  if (error == -ENOSPC &&
1070  ext3_should_retry_alloc(inode->i_sb, &retries))
1071  goto retry;
1072  if (error == 0)
1073  error = error2;
1074  }
1075 
1076  return error;
1077 }
1078 
1079 /*
1080  * ext3_xattr_delete_inode()
1081  *
1082  * Free extended attribute resources associated with this inode. This
1083  * is called immediately before an inode is freed. We have exclusive
1084  * access to the inode.
1085  */
1086 void
1087 ext3_xattr_delete_inode(handle_t *handle, struct inode *inode)
1088 {
1089  struct buffer_head *bh = NULL;
1090 
1091  if (!EXT3_I(inode)->i_file_acl)
1092  goto cleanup;
1093  bh = sb_bread(inode->i_sb, EXT3_I(inode)->i_file_acl);
1094  if (!bh) {
1095  ext3_error(inode->i_sb, __func__,
1096  "inode %lu: block "E3FSBLK" read error", inode->i_ino,
1097  EXT3_I(inode)->i_file_acl);
1098  goto cleanup;
1099  }
1100  if (BHDR(bh)->h_magic != cpu_to_le32(EXT3_XATTR_MAGIC) ||
1101  BHDR(bh)->h_blocks != cpu_to_le32(1)) {
1102  ext3_error(inode->i_sb, __func__,
1103  "inode %lu: bad block "E3FSBLK, inode->i_ino,
1104  EXT3_I(inode)->i_file_acl);
1105  goto cleanup;
1106  }
1107  ext3_xattr_release_block(handle, inode, bh);
1108  EXT3_I(inode)->i_file_acl = 0;
1109 
1110 cleanup:
1111  brelse(bh);
1112 }
1113 
1114 /*
1115  * ext3_xattr_put_super()
1116  *
1117  * This is called when a file system is unmounted.
1118  */
1119 void
1121 {
1122  mb_cache_shrink(sb->s_bdev);
1123 }
1124 
1125 /*
1126  * ext3_xattr_cache_insert()
1127  *
1128  * Create a new entry in the extended attribute cache, and insert
1129  * it unless such an entry is already in the cache.
1130  *
1131  * Returns 0, or a negative error number on failure.
1132  */
1133 static void
1134 ext3_xattr_cache_insert(struct buffer_head *bh)
1135 {
1136  __u32 hash = le32_to_cpu(BHDR(bh)->h_hash);
1137  struct mb_cache_entry *ce;
1138  int error;
1139 
1140  ce = mb_cache_entry_alloc(ext3_xattr_cache, GFP_NOFS);
1141  if (!ce) {
1142  ea_bdebug(bh, "out of memory");
1143  return;
1144  }
1145  error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
1146  if (error) {
1147  mb_cache_entry_free(ce);
1148  if (error == -EBUSY) {
1149  ea_bdebug(bh, "already in cache");
1150  error = 0;
1151  }
1152  } else {
1153  ea_bdebug(bh, "inserting [%x]", (int)hash);
1155  }
1156 }
1157 
1158 /*
1159  * ext3_xattr_cmp()
1160  *
1161  * Compare two extended attribute blocks for equality.
1162  *
1163  * Returns 0 if the blocks are equal, 1 if they differ, and
1164  * a negative error number on errors.
1165  */
1166 static int
1167 ext3_xattr_cmp(struct ext3_xattr_header *header1,
1168  struct ext3_xattr_header *header2)
1169 {
1170  struct ext3_xattr_entry *entry1, *entry2;
1171 
1172  entry1 = ENTRY(header1+1);
1173  entry2 = ENTRY(header2+1);
1174  while (!IS_LAST_ENTRY(entry1)) {
1175  if (IS_LAST_ENTRY(entry2))
1176  return 1;
1177  if (entry1->e_hash != entry2->e_hash ||
1178  entry1->e_name_index != entry2->e_name_index ||
1179  entry1->e_name_len != entry2->e_name_len ||
1180  entry1->e_value_size != entry2->e_value_size ||
1181  memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
1182  return 1;
1183  if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
1184  return -EIO;
1185  if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
1186  (char *)header2 + le16_to_cpu(entry2->e_value_offs),
1187  le32_to_cpu(entry1->e_value_size)))
1188  return 1;
1189 
1190  entry1 = EXT3_XATTR_NEXT(entry1);
1191  entry2 = EXT3_XATTR_NEXT(entry2);
1192  }
1193  if (!IS_LAST_ENTRY(entry2))
1194  return 1;
1195  return 0;
1196 }
1197 
1198 /*
1199  * ext3_xattr_cache_find()
1200  *
1201  * Find an identical extended attribute block.
1202  *
1203  * Returns a pointer to the block found, or NULL if such a block was
1204  * not found or an error occurred.
1205  */
1206 static struct buffer_head *
1207 ext3_xattr_cache_find(struct inode *inode, struct ext3_xattr_header *header,
1208  struct mb_cache_entry **pce)
1209 {
1210  __u32 hash = le32_to_cpu(header->h_hash);
1211  struct mb_cache_entry *ce;
1212 
1213  if (!header->h_hash)
1214  return NULL; /* never share */
1215  ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
1216 again:
1217  ce = mb_cache_entry_find_first(ext3_xattr_cache, inode->i_sb->s_bdev,
1218  hash);
1219  while (ce) {
1220  struct buffer_head *bh;
1221 
1222  if (IS_ERR(ce)) {
1223  if (PTR_ERR(ce) == -EAGAIN)
1224  goto again;
1225  break;
1226  }
1227  bh = sb_bread(inode->i_sb, ce->e_block);
1228  if (!bh) {
1229  ext3_error(inode->i_sb, __func__,
1230  "inode %lu: block %lu read error",
1231  inode->i_ino, (unsigned long) ce->e_block);
1232  } else if (le32_to_cpu(BHDR(bh)->h_refcount) >=
1234  ea_idebug(inode, "block %lu refcount %d>=%d",
1235  (unsigned long) ce->e_block,
1236  le32_to_cpu(BHDR(bh)->h_refcount),
1238  } else if (ext3_xattr_cmp(header, BHDR(bh)) == 0) {
1239  *pce = ce;
1240  return bh;
1241  }
1242  brelse(bh);
1243  ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
1244  }
1245  return NULL;
1246 }
1247 
1248 #define NAME_HASH_SHIFT 5
1249 #define VALUE_HASH_SHIFT 16
1250 
1251 /*
1252  * ext3_xattr_hash_entry()
1253  *
1254  * Compute the hash of an extended attribute.
1255  */
1256 static inline void ext3_xattr_hash_entry(struct ext3_xattr_header *header,
1257  struct ext3_xattr_entry *entry)
1258 {
1259  __u32 hash = 0;
1260  char *name = entry->e_name;
1261  int n;
1262 
1263  for (n=0; n < entry->e_name_len; n++) {
1264  hash = (hash << NAME_HASH_SHIFT) ^
1265  (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
1266  *name++;
1267  }
1268 
1269  if (entry->e_value_block == 0 && entry->e_value_size != 0) {
1270  __le32 *value = (__le32 *)((char *)header +
1271  le16_to_cpu(entry->e_value_offs));
1272  for (n = (le32_to_cpu(entry->e_value_size) +
1274  hash = (hash << VALUE_HASH_SHIFT) ^
1275  (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
1276  le32_to_cpu(*value++);
1277  }
1278  }
1279  entry->e_hash = cpu_to_le32(hash);
1280 }
1281 
1282 #undef NAME_HASH_SHIFT
1283 #undef VALUE_HASH_SHIFT
1284 
1285 #define BLOCK_HASH_SHIFT 16
1286 
1287 /*
1288  * ext3_xattr_rehash()
1289  *
1290  * Re-compute the extended attribute hash value after an entry has changed.
1291  */
1292 static void ext3_xattr_rehash(struct ext3_xattr_header *header,
1293  struct ext3_xattr_entry *entry)
1294 {
1295  struct ext3_xattr_entry *here;
1296  __u32 hash = 0;
1297 
1298  ext3_xattr_hash_entry(header, entry);
1299  here = ENTRY(header+1);
1300  while (!IS_LAST_ENTRY(here)) {
1301  if (!here->e_hash) {
1302  /* Block is not shared if an entry's hash value == 0 */
1303  hash = 0;
1304  break;
1305  }
1306  hash = (hash << BLOCK_HASH_SHIFT) ^
1307  (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1308  le32_to_cpu(here->e_hash);
1309  here = EXT3_XATTR_NEXT(here);
1310  }
1311  header->h_hash = cpu_to_le32(hash);
1312 }
1313 
1314 #undef BLOCK_HASH_SHIFT
1315 
1316 int __init
1318 {
1319  ext3_xattr_cache = mb_cache_create("ext3_xattr", 6);
1320  if (!ext3_xattr_cache)
1321  return -ENOMEM;
1322  return 0;
1323 }
1324 
1325 void
1327 {
1328  if (ext3_xattr_cache)
1329  mb_cache_destroy(ext3_xattr_cache);
1330  ext3_xattr_cache = NULL;
1331 }