blob: 27982b500e84a972412e2731125b70c30df7a36c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext2/xattr.c
3 *
4 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
5 *
6 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
7 * Extended attributes for symlinks and special files added per
8 * suggestion of Luka Renko <luka.renko@hermes.si>.
9 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
10 * Red Hat Inc.
11 *
12 */
13
14/*
15 * Extended attributes are stored on disk blocks allocated outside of
16 * any inode. The i_file_acl field is then made to point to this allocated
17 * block. If all extended attributes of an inode are identical, these
18 * inodes may share the same extended attribute block. Such situations
19 * are automatically detected by keeping a cache of recent attribute block
20 * numbers and hashes over the block's contents in memory.
21 *
22 *
23 * Extended attribute block layout:
24 *
25 * +------------------+
26 * | header |
27 * | entry 1 | |
28 * | entry 2 | | growing downwards
29 * | entry 3 | v
30 * | four null bytes |
31 * | . . . |
32 * | value 1 | ^
33 * | value 3 | | growing upwards
34 * | value 2 | |
35 * +------------------+
36 *
37 * The block header is followed by multiple entry descriptors. These entry
38 * descriptors are variable in size, and alligned to EXT2_XATTR_PAD
39 * byte boundaries. The entry descriptors are sorted by attribute name,
40 * so that two extended attribute blocks can be compared efficiently.
41 *
42 * Attribute values are aligned to the end of the block, stored in
43 * no specific order. They are also padded to EXT2_XATTR_PAD byte
44 * boundaries. No additional gaps are left between them.
45 *
46 * Locking strategy
47 * ----------------
48 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
49 * EA blocks are only changed if they are exclusive to an inode, so
50 * holding xattr_sem also means that nothing but the EA block's reference
51 * count will change. Multiple writers to an EA block are synchronized
52 * by the bh lock. No more than a single bh lock is held at any time
53 * to avoid deadlocks.
54 */
55
56#include <linux/buffer_head.h>
57#include <linux/module.h>
58#include <linux/init.h>
59#include <linux/slab.h>
60#include <linux/mbcache.h>
61#include <linux/quotaops.h>
62#include <linux/rwsem.h>
63#include "ext2.h"
64#include "xattr.h"
65#include "acl.h"
66
67#define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
68#define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
69#define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
70#define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
71
72#ifdef EXT2_XATTR_DEBUG
73# define ea_idebug(inode, f...) do { \
74 printk(KERN_DEBUG "inode %s:%ld: ", \
75 inode->i_sb->s_id, inode->i_ino); \
76 printk(f); \
77 printk("\n"); \
78 } while (0)
79# define ea_bdebug(bh, f...) do { \
80 char b[BDEVNAME_SIZE]; \
81 printk(KERN_DEBUG "block %s:%lu: ", \
82 bdevname(bh->b_bdev, b), \
83 (unsigned long) bh->b_blocknr); \
84 printk(f); \
85 printk("\n"); \
86 } while (0)
87#else
88# define ea_idebug(f...)
89# define ea_bdebug(f...)
90#endif
91
92static int ext2_xattr_set2(struct inode *, struct buffer_head *,
93 struct ext2_xattr_header *);
94
95static int ext2_xattr_cache_insert(struct buffer_head *);
96static struct buffer_head *ext2_xattr_cache_find(struct inode *,
97 struct ext2_xattr_header *);
98static void ext2_xattr_rehash(struct ext2_xattr_header *,
99 struct ext2_xattr_entry *);
100
101static struct mb_cache *ext2_xattr_cache;
102
103static struct xattr_handler *ext2_xattr_handler_map[] = {
104 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler,
105#ifdef CONFIG_EXT2_FS_POSIX_ACL
106 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &ext2_xattr_acl_access_handler,
107 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &ext2_xattr_acl_default_handler,
108#endif
109 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler,
110#ifdef CONFIG_EXT2_FS_SECURITY
111 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler,
112#endif
113};
114
115struct xattr_handler *ext2_xattr_handlers[] = {
116 &ext2_xattr_user_handler,
117 &ext2_xattr_trusted_handler,
118#ifdef CONFIG_EXT2_FS_POSIX_ACL
119 &ext2_xattr_acl_access_handler,
120 &ext2_xattr_acl_default_handler,
121#endif
122#ifdef CONFIG_EXT2_FS_SECURITY
123 &ext2_xattr_security_handler,
124#endif
125 NULL
126};
127
128static inline struct xattr_handler *
129ext2_xattr_handler(int name_index)
130{
131 struct xattr_handler *handler = NULL;
132
133 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
134 handler = ext2_xattr_handler_map[name_index];
135 return handler;
136}
137
138/*
139 * ext2_xattr_get()
140 *
141 * Copy an extended attribute into the buffer
142 * provided, or compute the buffer size required.
143 * Buffer is NULL to compute the size of the buffer required.
144 *
145 * Returns a negative error number on failure, or the number of bytes
146 * used / required on success.
147 */
148int
149ext2_xattr_get(struct inode *inode, int name_index, const char *name,
150 void *buffer, size_t buffer_size)
151{
152 struct buffer_head *bh = NULL;
153 struct ext2_xattr_entry *entry;
154 size_t name_len, size;
155 char *end;
156 int error;
157
158 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
159 name_index, name, buffer, (long)buffer_size);
160
161 if (name == NULL)
162 return -EINVAL;
163 down_read(&EXT2_I(inode)->xattr_sem);
164 error = -ENODATA;
165 if (!EXT2_I(inode)->i_file_acl)
166 goto cleanup;
167 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
168 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
169 error = -EIO;
170 if (!bh)
171 goto cleanup;
172 ea_bdebug(bh, "b_count=%d, refcount=%d",
173 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
174 end = bh->b_data + bh->b_size;
175 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
176 HDR(bh)->h_blocks != cpu_to_le32(1)) {
177bad_block: ext2_error(inode->i_sb, "ext2_xattr_get",
178 "inode %ld: bad block %d", inode->i_ino,
179 EXT2_I(inode)->i_file_acl);
180 error = -EIO;
181 goto cleanup;
182 }
183 /* find named attribute */
184 name_len = strlen(name);
185
186 error = -ERANGE;
187 if (name_len > 255)
188 goto cleanup;
189 entry = FIRST_ENTRY(bh);
190 while (!IS_LAST_ENTRY(entry)) {
191 struct ext2_xattr_entry *next =
192 EXT2_XATTR_NEXT(entry);
193 if ((char *)next >= end)
194 goto bad_block;
195 if (name_index == entry->e_name_index &&
196 name_len == entry->e_name_len &&
197 memcmp(name, entry->e_name, name_len) == 0)
198 goto found;
199 entry = next;
200 }
201 /* Check the remaining name entries */
202 while (!IS_LAST_ENTRY(entry)) {
203 struct ext2_xattr_entry *next =
204 EXT2_XATTR_NEXT(entry);
205 if ((char *)next >= end)
206 goto bad_block;
207 entry = next;
208 }
209 if (ext2_xattr_cache_insert(bh))
210 ea_idebug(inode, "cache insert failed");
211 error = -ENODATA;
212 goto cleanup;
213found:
214 /* check the buffer size */
215 if (entry->e_value_block != 0)
216 goto bad_block;
217 size = le32_to_cpu(entry->e_value_size);
218 if (size > inode->i_sb->s_blocksize ||
219 le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
220 goto bad_block;
221
222 if (ext2_xattr_cache_insert(bh))
223 ea_idebug(inode, "cache insert failed");
224 if (buffer) {
225 error = -ERANGE;
226 if (size > buffer_size)
227 goto cleanup;
228 /* return value of attribute */
229 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
230 size);
231 }
232 error = size;
233
234cleanup:
235 brelse(bh);
236 up_read(&EXT2_I(inode)->xattr_sem);
237
238 return error;
239}
240
241/*
242 * ext2_xattr_list()
243 *
244 * Copy a list of attribute names into the buffer
245 * provided, or compute the buffer size required.
246 * Buffer is NULL to compute the size of the buffer required.
247 *
248 * Returns a negative error number on failure, or the number of bytes
249 * used / required on success.
250 */
251static int
252ext2_xattr_list(struct inode *inode, char *buffer, size_t buffer_size)
253{
254 struct buffer_head *bh = NULL;
255 struct ext2_xattr_entry *entry;
256 char *end;
257 size_t rest = buffer_size;
258 int error;
259
260 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
261 buffer, (long)buffer_size);
262
263 down_read(&EXT2_I(inode)->xattr_sem);
264 error = 0;
265 if (!EXT2_I(inode)->i_file_acl)
266 goto cleanup;
267 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
268 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
269 error = -EIO;
270 if (!bh)
271 goto cleanup;
272 ea_bdebug(bh, "b_count=%d, refcount=%d",
273 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
274 end = bh->b_data + bh->b_size;
275 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
276 HDR(bh)->h_blocks != cpu_to_le32(1)) {
277bad_block: ext2_error(inode->i_sb, "ext2_xattr_list",
278 "inode %ld: bad block %d", inode->i_ino,
279 EXT2_I(inode)->i_file_acl);
280 error = -EIO;
281 goto cleanup;
282 }
283
284 /* check the on-disk data structure */
285 entry = FIRST_ENTRY(bh);
286 while (!IS_LAST_ENTRY(entry)) {
287 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
288
289 if ((char *)next >= end)
290 goto bad_block;
291 entry = next;
292 }
293 if (ext2_xattr_cache_insert(bh))
294 ea_idebug(inode, "cache insert failed");
295
296 /* list the attribute names */
297 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
298 entry = EXT2_XATTR_NEXT(entry)) {
299 struct xattr_handler *handler =
300 ext2_xattr_handler(entry->e_name_index);
301
302 if (handler) {
303 size_t size = handler->list(inode, buffer, rest,
304 entry->e_name,
305 entry->e_name_len);
306 if (buffer) {
307 if (size > rest) {
308 error = -ERANGE;
309 goto cleanup;
310 }
311 buffer += size;
312 }
313 rest -= size;
314 }
315 }
316 error = buffer_size - rest; /* total size */
317
318cleanup:
319 brelse(bh);
320 up_read(&EXT2_I(inode)->xattr_sem);
321
322 return error;
323}
324
325/*
326 * Inode operation listxattr()
327 *
328 * dentry->d_inode->i_sem: don't care
329 */
330ssize_t
331ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
332{
333 return ext2_xattr_list(dentry->d_inode, buffer, size);
334}
335
336/*
337 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
338 * not set, set it.
339 */
340static void ext2_xattr_update_super_block(struct super_block *sb)
341{
342 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
343 return;
344
345 lock_super(sb);
346 EXT2_SB(sb)->s_es->s_feature_compat |=
347 cpu_to_le32(EXT2_FEATURE_COMPAT_EXT_ATTR);
348 sb->s_dirt = 1;
349 mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
350 unlock_super(sb);
351}
352
353/*
354 * ext2_xattr_set()
355 *
356 * Create, replace or remove an extended attribute for this inode. Buffer
357 * is NULL to remove an existing extended attribute, and non-NULL to
358 * either replace an existing extended attribute, or create a new extended
359 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
360 * specify that an extended attribute must exist and must not exist
361 * previous to the call, respectively.
362 *
363 * Returns 0, or a negative error number on failure.
364 */
365int
366ext2_xattr_set(struct inode *inode, int name_index, const char *name,
367 const void *value, size_t value_len, int flags)
368{
369 struct super_block *sb = inode->i_sb;
370 struct buffer_head *bh = NULL;
371 struct ext2_xattr_header *header = NULL;
372 struct ext2_xattr_entry *here, *last;
373 size_t name_len, free, min_offs = sb->s_blocksize;
374 int not_found = 1, error;
375 char *end;
376
377 /*
378 * header -- Points either into bh, or to a temporarily
379 * allocated buffer.
380 * here -- The named entry found, or the place for inserting, within
381 * the block pointed to by header.
382 * last -- Points right after the last named entry within the block
383 * pointed to by header.
384 * min_offs -- The offset of the first value (values are aligned
385 * towards the end of the block).
386 * end -- Points right after the block pointed to by header.
387 */
388
389 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
390 name_index, name, value, (long)value_len);
391
392 if (IS_RDONLY(inode))
393 return -EROFS;
394 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
395 return -EPERM;
396 if (value == NULL)
397 value_len = 0;
398 if (name == NULL)
399 return -EINVAL;
400 name_len = strlen(name);
401 if (name_len > 255 || value_len > sb->s_blocksize)
402 return -ERANGE;
403 down_write(&EXT2_I(inode)->xattr_sem);
404 if (EXT2_I(inode)->i_file_acl) {
405 /* The inode already has an extended attribute block. */
406 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
407 error = -EIO;
408 if (!bh)
409 goto cleanup;
410 ea_bdebug(bh, "b_count=%d, refcount=%d",
411 atomic_read(&(bh->b_count)),
412 le32_to_cpu(HDR(bh)->h_refcount));
413 header = HDR(bh);
414 end = bh->b_data + bh->b_size;
415 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
416 header->h_blocks != cpu_to_le32(1)) {
417bad_block: ext2_error(sb, "ext2_xattr_set",
418 "inode %ld: bad block %d", inode->i_ino,
419 EXT2_I(inode)->i_file_acl);
420 error = -EIO;
421 goto cleanup;
422 }
423 /* Find the named attribute. */
424 here = FIRST_ENTRY(bh);
425 while (!IS_LAST_ENTRY(here)) {
426 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
427 if ((char *)next >= end)
428 goto bad_block;
429 if (!here->e_value_block && here->e_value_size) {
430 size_t offs = le16_to_cpu(here->e_value_offs);
431 if (offs < min_offs)
432 min_offs = offs;
433 }
434 not_found = name_index - here->e_name_index;
435 if (!not_found)
436 not_found = name_len - here->e_name_len;
437 if (!not_found)
438 not_found = memcmp(name, here->e_name,name_len);
439 if (not_found <= 0)
440 break;
441 here = next;
442 }
443 last = here;
444 /* We still need to compute min_offs and last. */
445 while (!IS_LAST_ENTRY(last)) {
446 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
447 if ((char *)next >= end)
448 goto bad_block;
449 if (!last->e_value_block && last->e_value_size) {
450 size_t offs = le16_to_cpu(last->e_value_offs);
451 if (offs < min_offs)
452 min_offs = offs;
453 }
454 last = next;
455 }
456
457 /* Check whether we have enough space left. */
458 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
459 } else {
460 /* We will use a new extended attribute block. */
461 free = sb->s_blocksize -
462 sizeof(struct ext2_xattr_header) - sizeof(__u32);
463 here = last = NULL; /* avoid gcc uninitialized warning. */
464 }
465
466 if (not_found) {
467 /* Request to remove a nonexistent attribute? */
468 error = -ENODATA;
469 if (flags & XATTR_REPLACE)
470 goto cleanup;
471 error = 0;
472 if (value == NULL)
473 goto cleanup;
474 } else {
475 /* Request to create an existing attribute? */
476 error = -EEXIST;
477 if (flags & XATTR_CREATE)
478 goto cleanup;
479 if (!here->e_value_block && here->e_value_size) {
480 size_t size = le32_to_cpu(here->e_value_size);
481
482 if (le16_to_cpu(here->e_value_offs) + size >
483 sb->s_blocksize || size > sb->s_blocksize)
484 goto bad_block;
485 free += EXT2_XATTR_SIZE(size);
486 }
487 free += EXT2_XATTR_LEN(name_len);
488 }
489 error = -ENOSPC;
490 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
491 goto cleanup;
492
493 /* Here we know that we can set the new attribute. */
494
495 if (header) {
496 struct mb_cache_entry *ce;
497
498 /* assert(header == HDR(bh)); */
499 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
500 bh->b_blocknr);
501 lock_buffer(bh);
502 if (header->h_refcount == cpu_to_le32(1)) {
503 ea_bdebug(bh, "modifying in-place");
504 if (ce)
505 mb_cache_entry_free(ce);
506 /* keep the buffer locked while modifying it. */
507 } else {
508 int offset;
509
510 if (ce)
511 mb_cache_entry_release(ce);
512 unlock_buffer(bh);
513 ea_bdebug(bh, "cloning");
514 header = kmalloc(bh->b_size, GFP_KERNEL);
515 error = -ENOMEM;
516 if (header == NULL)
517 goto cleanup;
518 memcpy(header, HDR(bh), bh->b_size);
519 header->h_refcount = cpu_to_le32(1);
520
521 offset = (char *)here - bh->b_data;
522 here = ENTRY((char *)header + offset);
523 offset = (char *)last - bh->b_data;
524 last = ENTRY((char *)header + offset);
525 }
526 } else {
527 /* Allocate a buffer where we construct the new block. */
528 header = kmalloc(sb->s_blocksize, GFP_KERNEL);
529 error = -ENOMEM;
530 if (header == NULL)
531 goto cleanup;
532 memset(header, 0, sb->s_blocksize);
533 end = (char *)header + sb->s_blocksize;
534 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
535 header->h_blocks = header->h_refcount = cpu_to_le32(1);
536 last = here = ENTRY(header+1);
537 }
538
539 /* Iff we are modifying the block in-place, bh is locked here. */
540
541 if (not_found) {
542 /* Insert the new name. */
543 size_t size = EXT2_XATTR_LEN(name_len);
544 size_t rest = (char *)last - (char *)here;
545 memmove((char *)here + size, here, rest);
546 memset(here, 0, size);
547 here->e_name_index = name_index;
548 here->e_name_len = name_len;
549 memcpy(here->e_name, name, name_len);
550 } else {
551 if (!here->e_value_block && here->e_value_size) {
552 char *first_val = (char *)header + min_offs;
553 size_t offs = le16_to_cpu(here->e_value_offs);
554 char *val = (char *)header + offs;
555 size_t size = EXT2_XATTR_SIZE(
556 le32_to_cpu(here->e_value_size));
557
558 if (size == EXT2_XATTR_SIZE(value_len)) {
559 /* The old and the new value have the same
560 size. Just replace. */
561 here->e_value_size = cpu_to_le32(value_len);
562 memset(val + size - EXT2_XATTR_PAD, 0,
563 EXT2_XATTR_PAD); /* Clear pad bytes. */
564 memcpy(val, value, value_len);
565 goto skip_replace;
566 }
567
568 /* Remove the old value. */
569 memmove(first_val + size, first_val, val - first_val);
570 memset(first_val, 0, size);
571 here->e_value_offs = 0;
572 min_offs += size;
573
574 /* Adjust all value offsets. */
575 last = ENTRY(header+1);
576 while (!IS_LAST_ENTRY(last)) {
577 size_t o = le16_to_cpu(last->e_value_offs);
578 if (!last->e_value_block && o < offs)
579 last->e_value_offs =
580 cpu_to_le16(o + size);
581 last = EXT2_XATTR_NEXT(last);
582 }
583 }
584 if (value == NULL) {
585 /* Remove the old name. */
586 size_t size = EXT2_XATTR_LEN(name_len);
587 last = ENTRY((char *)last - size);
588 memmove(here, (char*)here + size,
589 (char*)last - (char*)here);
590 memset(last, 0, size);
591 }
592 }
593
594 if (value != NULL) {
595 /* Insert the new value. */
596 here->e_value_size = cpu_to_le32(value_len);
597 if (value_len) {
598 size_t size = EXT2_XATTR_SIZE(value_len);
599 char *val = (char *)header + min_offs - size;
600 here->e_value_offs =
601 cpu_to_le16((char *)val - (char *)header);
602 memset(val + size - EXT2_XATTR_PAD, 0,
603 EXT2_XATTR_PAD); /* Clear the pad bytes. */
604 memcpy(val, value, value_len);
605 }
606 }
607
608skip_replace:
609 if (IS_LAST_ENTRY(ENTRY(header+1))) {
610 /* This block is now empty. */
611 if (bh && header == HDR(bh))
612 unlock_buffer(bh); /* we were modifying in-place. */
613 error = ext2_xattr_set2(inode, bh, NULL);
614 } else {
615 ext2_xattr_rehash(header, here);
616 if (bh && header == HDR(bh))
617 unlock_buffer(bh); /* we were modifying in-place. */
618 error = ext2_xattr_set2(inode, bh, header);
619 }
620
621cleanup:
622 brelse(bh);
623 if (!(bh && header == HDR(bh)))
624 kfree(header);
625 up_write(&EXT2_I(inode)->xattr_sem);
626
627 return error;
628}
629
630/*
631 * Second half of ext2_xattr_set(): Update the file system.
632 */
633static int
634ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
635 struct ext2_xattr_header *header)
636{
637 struct super_block *sb = inode->i_sb;
638 struct buffer_head *new_bh = NULL;
639 int error;
640
641 if (header) {
642 new_bh = ext2_xattr_cache_find(inode, header);
643 if (new_bh) {
644 /* We found an identical block in the cache. */
645 if (new_bh == old_bh) {
646 ea_bdebug(new_bh, "keeping this block");
647 } else {
648 /* The old block is released after updating
649 the inode. */
650 ea_bdebug(new_bh, "reusing block");
651
652 error = -EDQUOT;
653 if (DQUOT_ALLOC_BLOCK(inode, 1)) {
654 unlock_buffer(new_bh);
655 goto cleanup;
656 }
657 HDR(new_bh)->h_refcount = cpu_to_le32(1 +
658 le32_to_cpu(HDR(new_bh)->h_refcount));
659 ea_bdebug(new_bh, "refcount now=%d",
660 le32_to_cpu(HDR(new_bh)->h_refcount));
661 }
662 unlock_buffer(new_bh);
663 } else if (old_bh && header == HDR(old_bh)) {
664 /* Keep this block. No need to lock the block as we
665 don't need to change the reference count. */
666 new_bh = old_bh;
667 get_bh(new_bh);
668 ext2_xattr_cache_insert(new_bh);
669 } else {
670 /* We need to allocate a new block */
671 int goal = le32_to_cpu(EXT2_SB(sb)->s_es->
672 s_first_data_block) +
673 EXT2_I(inode)->i_block_group *
674 EXT2_BLOCKS_PER_GROUP(sb);
675 int block = ext2_new_block(inode, goal,
676 NULL, NULL, &error);
677 if (error)
678 goto cleanup;
679 ea_idebug(inode, "creating block %d", block);
680
681 new_bh = sb_getblk(sb, block);
682 if (!new_bh) {
683 ext2_free_blocks(inode, block, 1);
684 error = -EIO;
685 goto cleanup;
686 }
687 lock_buffer(new_bh);
688 memcpy(new_bh->b_data, header, new_bh->b_size);
689 set_buffer_uptodate(new_bh);
690 unlock_buffer(new_bh);
691 ext2_xattr_cache_insert(new_bh);
692
693 ext2_xattr_update_super_block(sb);
694 }
695 mark_buffer_dirty(new_bh);
696 if (IS_SYNC(inode)) {
697 sync_dirty_buffer(new_bh);
698 error = -EIO;
699 if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
700 goto cleanup;
701 }
702 }
703
704 /* Update the inode. */
705 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
706 inode->i_ctime = CURRENT_TIME_SEC;
707 if (IS_SYNC(inode)) {
708 error = ext2_sync_inode (inode);
709 /* In case sync failed due to ENOSPC the inode was actually
710 * written (only some dirty data were not) so we just proceed
711 * as if nothing happened and cleanup the unused block */
712 if (error && error != -ENOSPC) {
713 if (new_bh && new_bh != old_bh)
714 DQUOT_FREE_BLOCK(inode, 1);
715 goto cleanup;
716 }
717 } else
718 mark_inode_dirty(inode);
719
720 error = 0;
721 if (old_bh && old_bh != new_bh) {
722 struct mb_cache_entry *ce;
723
724 /*
725 * If there was an old block and we are no longer using it,
726 * release the old block.
727 */
728 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
729 old_bh->b_blocknr);
730 lock_buffer(old_bh);
731 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
732 /* Free the old block. */
733 if (ce)
734 mb_cache_entry_free(ce);
735 ea_bdebug(old_bh, "freeing");
736 ext2_free_blocks(inode, old_bh->b_blocknr, 1);
737 /* We let our caller release old_bh, so we
738 * need to duplicate the buffer before. */
739 get_bh(old_bh);
740 bforget(old_bh);
741 } else {
742 /* Decrement the refcount only. */
743 HDR(old_bh)->h_refcount = cpu_to_le32(
744 le32_to_cpu(HDR(old_bh)->h_refcount) - 1);
745 if (ce)
746 mb_cache_entry_release(ce);
747 DQUOT_FREE_BLOCK(inode, 1);
748 mark_buffer_dirty(old_bh);
749 ea_bdebug(old_bh, "refcount now=%d",
750 le32_to_cpu(HDR(old_bh)->h_refcount));
751 }
752 unlock_buffer(old_bh);
753 }
754
755cleanup:
756 brelse(new_bh);
757
758 return error;
759}
760
761/*
762 * ext2_xattr_delete_inode()
763 *
764 * Free extended attribute resources associated with this inode. This
765 * is called immediately before an inode is freed.
766 */
767void
768ext2_xattr_delete_inode(struct inode *inode)
769{
770 struct buffer_head *bh = NULL;
771 struct mb_cache_entry *ce;
772
773 down_write(&EXT2_I(inode)->xattr_sem);
774 if (!EXT2_I(inode)->i_file_acl)
775 goto cleanup;
776 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
777 if (!bh) {
778 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
779 "inode %ld: block %d read error", inode->i_ino,
780 EXT2_I(inode)->i_file_acl);
781 goto cleanup;
782 }
783 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
784 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
785 HDR(bh)->h_blocks != cpu_to_le32(1)) {
786 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
787 "inode %ld: bad block %d", inode->i_ino,
788 EXT2_I(inode)->i_file_acl);
789 goto cleanup;
790 }
791 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
792 lock_buffer(bh);
793 if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
794 if (ce)
795 mb_cache_entry_free(ce);
796 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
797 get_bh(bh);
798 bforget(bh);
799 } else {
800 HDR(bh)->h_refcount = cpu_to_le32(
801 le32_to_cpu(HDR(bh)->h_refcount) - 1);
802 if (ce)
803 mb_cache_entry_release(ce);
804 mark_buffer_dirty(bh);
805 if (IS_SYNC(inode))
806 sync_dirty_buffer(bh);
807 DQUOT_FREE_BLOCK(inode, 1);
808 }
809 ea_bdebug(bh, "refcount now=%d", le32_to_cpu(HDR(bh)->h_refcount) - 1);
810 unlock_buffer(bh);
811 EXT2_I(inode)->i_file_acl = 0;
812
813cleanup:
814 brelse(bh);
815 up_write(&EXT2_I(inode)->xattr_sem);
816}
817
818/*
819 * ext2_xattr_put_super()
820 *
821 * This is called when a file system is unmounted.
822 */
823void
824ext2_xattr_put_super(struct super_block *sb)
825{
826 mb_cache_shrink(ext2_xattr_cache, sb->s_bdev);
827}
828
829
830/*
831 * ext2_xattr_cache_insert()
832 *
833 * Create a new entry in the extended attribute cache, and insert
834 * it unless such an entry is already in the cache.
835 *
836 * Returns 0, or a negative error number on failure.
837 */
838static int
839ext2_xattr_cache_insert(struct buffer_head *bh)
840{
841 __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
842 struct mb_cache_entry *ce;
843 int error;
844
845 ce = mb_cache_entry_alloc(ext2_xattr_cache);
846 if (!ce)
847 return -ENOMEM;
848 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
849 if (error) {
850 mb_cache_entry_free(ce);
851 if (error == -EBUSY) {
852 ea_bdebug(bh, "already in cache (%d cache entries)",
853 atomic_read(&ext2_xattr_cache->c_entry_count));
854 error = 0;
855 }
856 } else {
857 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
858 atomic_read(&ext2_xattr_cache->c_entry_count));
859 mb_cache_entry_release(ce);
860 }
861 return error;
862}
863
864/*
865 * ext2_xattr_cmp()
866 *
867 * Compare two extended attribute blocks for equality.
868 *
869 * Returns 0 if the blocks are equal, 1 if they differ, and
870 * a negative error number on errors.
871 */
872static int
873ext2_xattr_cmp(struct ext2_xattr_header *header1,
874 struct ext2_xattr_header *header2)
875{
876 struct ext2_xattr_entry *entry1, *entry2;
877
878 entry1 = ENTRY(header1+1);
879 entry2 = ENTRY(header2+1);
880 while (!IS_LAST_ENTRY(entry1)) {
881 if (IS_LAST_ENTRY(entry2))
882 return 1;
883 if (entry1->e_hash != entry2->e_hash ||
884 entry1->e_name_index != entry2->e_name_index ||
885 entry1->e_name_len != entry2->e_name_len ||
886 entry1->e_value_size != entry2->e_value_size ||
887 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
888 return 1;
889 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
890 return -EIO;
891 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
892 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
893 le32_to_cpu(entry1->e_value_size)))
894 return 1;
895
896 entry1 = EXT2_XATTR_NEXT(entry1);
897 entry2 = EXT2_XATTR_NEXT(entry2);
898 }
899 if (!IS_LAST_ENTRY(entry2))
900 return 1;
901 return 0;
902}
903
904/*
905 * ext2_xattr_cache_find()
906 *
907 * Find an identical extended attribute block.
908 *
909 * Returns a locked buffer head to the block found, or NULL if such
910 * a block was not found or an error occurred.
911 */
912static struct buffer_head *
913ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
914{
915 __u32 hash = le32_to_cpu(header->h_hash);
916 struct mb_cache_entry *ce;
917
918 if (!header->h_hash)
919 return NULL; /* never share */
920 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
921again:
922 ce = mb_cache_entry_find_first(ext2_xattr_cache, 0,
923 inode->i_sb->s_bdev, hash);
924 while (ce) {
925 struct buffer_head *bh;
926
927 if (IS_ERR(ce)) {
928 if (PTR_ERR(ce) == -EAGAIN)
929 goto again;
930 break;
931 }
932
933 bh = sb_bread(inode->i_sb, ce->e_block);
934 if (!bh) {
935 ext2_error(inode->i_sb, "ext2_xattr_cache_find",
936 "inode %ld: block %ld read error",
937 inode->i_ino, (unsigned long) ce->e_block);
938 } else {
939 lock_buffer(bh);
940 if (le32_to_cpu(HDR(bh)->h_refcount) >
941 EXT2_XATTR_REFCOUNT_MAX) {
942 ea_idebug(inode, "block %ld refcount %d>%d",
943 (unsigned long) ce->e_block,
944 le32_to_cpu(HDR(bh)->h_refcount),
945 EXT2_XATTR_REFCOUNT_MAX);
946 } else if (!ext2_xattr_cmp(header, HDR(bh))) {
947 ea_bdebug(bh, "b_count=%d",
948 atomic_read(&(bh->b_count)));
949 mb_cache_entry_release(ce);
950 return bh;
951 }
952 unlock_buffer(bh);
953 brelse(bh);
954 }
955 ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
956 }
957 return NULL;
958}
959
960#define NAME_HASH_SHIFT 5
961#define VALUE_HASH_SHIFT 16
962
963/*
964 * ext2_xattr_hash_entry()
965 *
966 * Compute the hash of an extended attribute.
967 */
968static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
969 struct ext2_xattr_entry *entry)
970{
971 __u32 hash = 0;
972 char *name = entry->e_name;
973 int n;
974
975 for (n=0; n < entry->e_name_len; n++) {
976 hash = (hash << NAME_HASH_SHIFT) ^
977 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
978 *name++;
979 }
980
981 if (entry->e_value_block == 0 && entry->e_value_size != 0) {
982 __le32 *value = (__le32 *)((char *)header +
983 le16_to_cpu(entry->e_value_offs));
984 for (n = (le32_to_cpu(entry->e_value_size) +
985 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
986 hash = (hash << VALUE_HASH_SHIFT) ^
987 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
988 le32_to_cpu(*value++);
989 }
990 }
991 entry->e_hash = cpu_to_le32(hash);
992}
993
994#undef NAME_HASH_SHIFT
995#undef VALUE_HASH_SHIFT
996
997#define BLOCK_HASH_SHIFT 16
998
999/*
1000 * ext2_xattr_rehash()
1001 *
1002 * Re-compute the extended attribute hash value after an entry has changed.
1003 */
1004static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1005 struct ext2_xattr_entry *entry)
1006{
1007 struct ext2_xattr_entry *here;
1008 __u32 hash = 0;
1009
1010 ext2_xattr_hash_entry(header, entry);
1011 here = ENTRY(header+1);
1012 while (!IS_LAST_ENTRY(here)) {
1013 if (!here->e_hash) {
1014 /* Block is not shared if an entry's hash value == 0 */
1015 hash = 0;
1016 break;
1017 }
1018 hash = (hash << BLOCK_HASH_SHIFT) ^
1019 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1020 le32_to_cpu(here->e_hash);
1021 here = EXT2_XATTR_NEXT(here);
1022 }
1023 header->h_hash = cpu_to_le32(hash);
1024}
1025
1026#undef BLOCK_HASH_SHIFT
1027
1028int __init
1029init_ext2_xattr(void)
1030{
1031 ext2_xattr_cache = mb_cache_create("ext2_xattr", NULL,
1032 sizeof(struct mb_cache_entry) +
1033 sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1034 if (!ext2_xattr_cache)
1035 return -ENOMEM;
1036 return 0;
1037}
1038
1039void
1040exit_ext2_xattr(void)
1041{
1042 mb_cache_destroy(ext2_xattr_cache);
1043}