blob: 7913531ec6d5d1c7c750bb13b6912d4c844fa448 [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 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800328 * dentry->d_inode->i_mutex: don't care
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 */
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
Andreas Gruenbachered2908f2006-12-06 20:36:16 -0800345 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 sb->s_dirt = 1;
347 mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
350/*
351 * ext2_xattr_set()
352 *
353 * Create, replace or remove an extended attribute for this inode. Buffer
354 * is NULL to remove an existing extended attribute, and non-NULL to
355 * either replace an existing extended attribute, or create a new extended
356 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
357 * specify that an extended attribute must exist and must not exist
358 * previous to the call, respectively.
359 *
360 * Returns 0, or a negative error number on failure.
361 */
362int
363ext2_xattr_set(struct inode *inode, int name_index, const char *name,
364 const void *value, size_t value_len, int flags)
365{
366 struct super_block *sb = inode->i_sb;
367 struct buffer_head *bh = NULL;
368 struct ext2_xattr_header *header = NULL;
369 struct ext2_xattr_entry *here, *last;
370 size_t name_len, free, min_offs = sb->s_blocksize;
371 int not_found = 1, error;
372 char *end;
373
374 /*
375 * header -- Points either into bh, or to a temporarily
376 * allocated buffer.
377 * here -- The named entry found, or the place for inserting, within
378 * the block pointed to by header.
379 * last -- Points right after the last named entry within the block
380 * pointed to by header.
381 * min_offs -- The offset of the first value (values are aligned
382 * towards the end of the block).
383 * end -- Points right after the block pointed to by header.
384 */
385
386 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
387 name_index, name, value, (long)value_len);
388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 if (value == NULL)
390 value_len = 0;
391 if (name == NULL)
392 return -EINVAL;
393 name_len = strlen(name);
394 if (name_len > 255 || value_len > sb->s_blocksize)
395 return -ERANGE;
396 down_write(&EXT2_I(inode)->xattr_sem);
397 if (EXT2_I(inode)->i_file_acl) {
398 /* The inode already has an extended attribute block. */
399 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
400 error = -EIO;
401 if (!bh)
402 goto cleanup;
403 ea_bdebug(bh, "b_count=%d, refcount=%d",
404 atomic_read(&(bh->b_count)),
405 le32_to_cpu(HDR(bh)->h_refcount));
406 header = HDR(bh);
407 end = bh->b_data + bh->b_size;
408 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
409 header->h_blocks != cpu_to_le32(1)) {
410bad_block: ext2_error(sb, "ext2_xattr_set",
411 "inode %ld: bad block %d", inode->i_ino,
412 EXT2_I(inode)->i_file_acl);
413 error = -EIO;
414 goto cleanup;
415 }
416 /* Find the named attribute. */
417 here = FIRST_ENTRY(bh);
418 while (!IS_LAST_ENTRY(here)) {
419 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
420 if ((char *)next >= end)
421 goto bad_block;
422 if (!here->e_value_block && here->e_value_size) {
423 size_t offs = le16_to_cpu(here->e_value_offs);
424 if (offs < min_offs)
425 min_offs = offs;
426 }
427 not_found = name_index - here->e_name_index;
428 if (!not_found)
429 not_found = name_len - here->e_name_len;
430 if (!not_found)
431 not_found = memcmp(name, here->e_name,name_len);
432 if (not_found <= 0)
433 break;
434 here = next;
435 }
436 last = here;
437 /* We still need to compute min_offs and last. */
438 while (!IS_LAST_ENTRY(last)) {
439 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
440 if ((char *)next >= end)
441 goto bad_block;
442 if (!last->e_value_block && last->e_value_size) {
443 size_t offs = le16_to_cpu(last->e_value_offs);
444 if (offs < min_offs)
445 min_offs = offs;
446 }
447 last = next;
448 }
449
450 /* Check whether we have enough space left. */
451 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
452 } else {
453 /* We will use a new extended attribute block. */
454 free = sb->s_blocksize -
455 sizeof(struct ext2_xattr_header) - sizeof(__u32);
456 here = last = NULL; /* avoid gcc uninitialized warning. */
457 }
458
459 if (not_found) {
460 /* Request to remove a nonexistent attribute? */
461 error = -ENODATA;
462 if (flags & XATTR_REPLACE)
463 goto cleanup;
464 error = 0;
465 if (value == NULL)
466 goto cleanup;
467 } else {
468 /* Request to create an existing attribute? */
469 error = -EEXIST;
470 if (flags & XATTR_CREATE)
471 goto cleanup;
472 if (!here->e_value_block && here->e_value_size) {
473 size_t size = le32_to_cpu(here->e_value_size);
474
475 if (le16_to_cpu(here->e_value_offs) + size >
476 sb->s_blocksize || size > sb->s_blocksize)
477 goto bad_block;
478 free += EXT2_XATTR_SIZE(size);
479 }
480 free += EXT2_XATTR_LEN(name_len);
481 }
482 error = -ENOSPC;
483 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
484 goto cleanup;
485
486 /* Here we know that we can set the new attribute. */
487
488 if (header) {
489 struct mb_cache_entry *ce;
490
491 /* assert(header == HDR(bh)); */
492 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
493 bh->b_blocknr);
494 lock_buffer(bh);
495 if (header->h_refcount == cpu_to_le32(1)) {
496 ea_bdebug(bh, "modifying in-place");
497 if (ce)
498 mb_cache_entry_free(ce);
499 /* keep the buffer locked while modifying it. */
500 } else {
501 int offset;
502
503 if (ce)
504 mb_cache_entry_release(ce);
505 unlock_buffer(bh);
506 ea_bdebug(bh, "cloning");
507 header = kmalloc(bh->b_size, GFP_KERNEL);
508 error = -ENOMEM;
509 if (header == NULL)
510 goto cleanup;
511 memcpy(header, HDR(bh), bh->b_size);
512 header->h_refcount = cpu_to_le32(1);
513
514 offset = (char *)here - bh->b_data;
515 here = ENTRY((char *)header + offset);
516 offset = (char *)last - bh->b_data;
517 last = ENTRY((char *)header + offset);
518 }
519 } else {
520 /* Allocate a buffer where we construct the new block. */
Panagiotis Issarisf8314dc2006-09-27 01:49:37 -0700521 header = kzalloc(sb->s_blocksize, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 error = -ENOMEM;
523 if (header == NULL)
524 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 end = (char *)header + sb->s_blocksize;
526 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
527 header->h_blocks = header->h_refcount = cpu_to_le32(1);
528 last = here = ENTRY(header+1);
529 }
530
531 /* Iff we are modifying the block in-place, bh is locked here. */
532
533 if (not_found) {
534 /* Insert the new name. */
535 size_t size = EXT2_XATTR_LEN(name_len);
536 size_t rest = (char *)last - (char *)here;
537 memmove((char *)here + size, here, rest);
538 memset(here, 0, size);
539 here->e_name_index = name_index;
540 here->e_name_len = name_len;
541 memcpy(here->e_name, name, name_len);
542 } else {
543 if (!here->e_value_block && here->e_value_size) {
544 char *first_val = (char *)header + min_offs;
545 size_t offs = le16_to_cpu(here->e_value_offs);
546 char *val = (char *)header + offs;
547 size_t size = EXT2_XATTR_SIZE(
548 le32_to_cpu(here->e_value_size));
549
550 if (size == EXT2_XATTR_SIZE(value_len)) {
551 /* The old and the new value have the same
552 size. Just replace. */
553 here->e_value_size = cpu_to_le32(value_len);
554 memset(val + size - EXT2_XATTR_PAD, 0,
555 EXT2_XATTR_PAD); /* Clear pad bytes. */
556 memcpy(val, value, value_len);
557 goto skip_replace;
558 }
559
560 /* Remove the old value. */
561 memmove(first_val + size, first_val, val - first_val);
562 memset(first_val, 0, size);
563 here->e_value_offs = 0;
564 min_offs += size;
565
566 /* Adjust all value offsets. */
567 last = ENTRY(header+1);
568 while (!IS_LAST_ENTRY(last)) {
569 size_t o = le16_to_cpu(last->e_value_offs);
570 if (!last->e_value_block && o < offs)
571 last->e_value_offs =
572 cpu_to_le16(o + size);
573 last = EXT2_XATTR_NEXT(last);
574 }
575 }
576 if (value == NULL) {
577 /* Remove the old name. */
578 size_t size = EXT2_XATTR_LEN(name_len);
579 last = ENTRY((char *)last - size);
580 memmove(here, (char*)here + size,
581 (char*)last - (char*)here);
582 memset(last, 0, size);
583 }
584 }
585
586 if (value != NULL) {
587 /* Insert the new value. */
588 here->e_value_size = cpu_to_le32(value_len);
589 if (value_len) {
590 size_t size = EXT2_XATTR_SIZE(value_len);
591 char *val = (char *)header + min_offs - size;
592 here->e_value_offs =
593 cpu_to_le16((char *)val - (char *)header);
594 memset(val + size - EXT2_XATTR_PAD, 0,
595 EXT2_XATTR_PAD); /* Clear the pad bytes. */
596 memcpy(val, value, value_len);
597 }
598 }
599
600skip_replace:
601 if (IS_LAST_ENTRY(ENTRY(header+1))) {
602 /* This block is now empty. */
603 if (bh && header == HDR(bh))
604 unlock_buffer(bh); /* we were modifying in-place. */
605 error = ext2_xattr_set2(inode, bh, NULL);
606 } else {
607 ext2_xattr_rehash(header, here);
608 if (bh && header == HDR(bh))
609 unlock_buffer(bh); /* we were modifying in-place. */
610 error = ext2_xattr_set2(inode, bh, header);
611 }
612
613cleanup:
614 brelse(bh);
615 if (!(bh && header == HDR(bh)))
616 kfree(header);
617 up_write(&EXT2_I(inode)->xattr_sem);
618
619 return error;
620}
621
622/*
623 * Second half of ext2_xattr_set(): Update the file system.
624 */
625static int
626ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
627 struct ext2_xattr_header *header)
628{
629 struct super_block *sb = inode->i_sb;
630 struct buffer_head *new_bh = NULL;
631 int error;
632
633 if (header) {
634 new_bh = ext2_xattr_cache_find(inode, header);
635 if (new_bh) {
636 /* We found an identical block in the cache. */
637 if (new_bh == old_bh) {
638 ea_bdebug(new_bh, "keeping this block");
639 } else {
640 /* The old block is released after updating
641 the inode. */
642 ea_bdebug(new_bh, "reusing block");
643
644 error = -EDQUOT;
Jan Kara6f90bee2009-01-26 16:52:06 +0100645 if (vfs_dq_alloc_block(inode, 1)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 unlock_buffer(new_bh);
647 goto cleanup;
648 }
Marcin Slusarzfba4d392008-04-28 02:15:59 -0700649 le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 ea_bdebug(new_bh, "refcount now=%d",
651 le32_to_cpu(HDR(new_bh)->h_refcount));
652 }
653 unlock_buffer(new_bh);
654 } else if (old_bh && header == HDR(old_bh)) {
655 /* Keep this block. No need to lock the block as we
656 don't need to change the reference count. */
657 new_bh = old_bh;
658 get_bh(new_bh);
659 ext2_xattr_cache_insert(new_bh);
660 } else {
661 /* We need to allocate a new block */
Akinobu Mita24097d12008-04-28 02:16:01 -0700662 ext2_fsblk_t goal = ext2_group_first_block_no(sb,
663 EXT2_I(inode)->i_block_group);
Martin J. Bligha686cd82007-10-16 23:30:46 -0700664 int block = ext2_new_block(inode, goal, &error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (error)
666 goto cleanup;
667 ea_idebug(inode, "creating block %d", block);
668
669 new_bh = sb_getblk(sb, block);
670 if (!new_bh) {
671 ext2_free_blocks(inode, block, 1);
672 error = -EIO;
673 goto cleanup;
674 }
675 lock_buffer(new_bh);
676 memcpy(new_bh->b_data, header, new_bh->b_size);
677 set_buffer_uptodate(new_bh);
678 unlock_buffer(new_bh);
679 ext2_xattr_cache_insert(new_bh);
680
681 ext2_xattr_update_super_block(sb);
682 }
683 mark_buffer_dirty(new_bh);
684 if (IS_SYNC(inode)) {
685 sync_dirty_buffer(new_bh);
686 error = -EIO;
687 if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
688 goto cleanup;
689 }
690 }
691
692 /* Update the inode. */
693 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
694 inode->i_ctime = CURRENT_TIME_SEC;
695 if (IS_SYNC(inode)) {
696 error = ext2_sync_inode (inode);
697 /* In case sync failed due to ENOSPC the inode was actually
698 * written (only some dirty data were not) so we just proceed
699 * as if nothing happened and cleanup the unused block */
700 if (error && error != -ENOSPC) {
701 if (new_bh && new_bh != old_bh)
Jan Kara6f90bee2009-01-26 16:52:06 +0100702 vfs_dq_free_block(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 goto cleanup;
704 }
705 } else
706 mark_inode_dirty(inode);
707
708 error = 0;
709 if (old_bh && old_bh != new_bh) {
710 struct mb_cache_entry *ce;
711
712 /*
713 * If there was an old block and we are no longer using it,
714 * release the old block.
715 */
716 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
717 old_bh->b_blocknr);
718 lock_buffer(old_bh);
719 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
720 /* Free the old block. */
721 if (ce)
722 mb_cache_entry_free(ce);
723 ea_bdebug(old_bh, "freeing");
724 ext2_free_blocks(inode, old_bh->b_blocknr, 1);
725 /* We let our caller release old_bh, so we
726 * need to duplicate the buffer before. */
727 get_bh(old_bh);
728 bforget(old_bh);
729 } else {
730 /* Decrement the refcount only. */
Marcin Slusarzfba4d392008-04-28 02:15:59 -0700731 le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 if (ce)
733 mb_cache_entry_release(ce);
Jan Kara6f90bee2009-01-26 16:52:06 +0100734 vfs_dq_free_block(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 mark_buffer_dirty(old_bh);
736 ea_bdebug(old_bh, "refcount now=%d",
737 le32_to_cpu(HDR(old_bh)->h_refcount));
738 }
739 unlock_buffer(old_bh);
740 }
741
742cleanup:
743 brelse(new_bh);
744
745 return error;
746}
747
748/*
749 * ext2_xattr_delete_inode()
750 *
751 * Free extended attribute resources associated with this inode. This
752 * is called immediately before an inode is freed.
753 */
754void
755ext2_xattr_delete_inode(struct inode *inode)
756{
757 struct buffer_head *bh = NULL;
758 struct mb_cache_entry *ce;
759
760 down_write(&EXT2_I(inode)->xattr_sem);
761 if (!EXT2_I(inode)->i_file_acl)
762 goto cleanup;
763 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
764 if (!bh) {
765 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
766 "inode %ld: block %d read error", inode->i_ino,
767 EXT2_I(inode)->i_file_acl);
768 goto cleanup;
769 }
770 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
771 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
772 HDR(bh)->h_blocks != cpu_to_le32(1)) {
773 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
774 "inode %ld: bad block %d", inode->i_ino,
775 EXT2_I(inode)->i_file_acl);
776 goto cleanup;
777 }
778 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
779 lock_buffer(bh);
780 if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
781 if (ce)
782 mb_cache_entry_free(ce);
783 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
784 get_bh(bh);
785 bforget(bh);
Peter Staubachb2f49032006-02-17 13:52:36 -0800786 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 } else {
Marcin Slusarzfba4d392008-04-28 02:15:59 -0700788 le32_add_cpu(&HDR(bh)->h_refcount, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 if (ce)
790 mb_cache_entry_release(ce);
Peter Staubachb2f49032006-02-17 13:52:36 -0800791 ea_bdebug(bh, "refcount now=%d",
792 le32_to_cpu(HDR(bh)->h_refcount));
793 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 mark_buffer_dirty(bh);
795 if (IS_SYNC(inode))
796 sync_dirty_buffer(bh);
Jan Kara6f90bee2009-01-26 16:52:06 +0100797 vfs_dq_free_block(inode, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 EXT2_I(inode)->i_file_acl = 0;
800
801cleanup:
802 brelse(bh);
803 up_write(&EXT2_I(inode)->xattr_sem);
804}
805
806/*
807 * ext2_xattr_put_super()
808 *
809 * This is called when a file system is unmounted.
810 */
811void
812ext2_xattr_put_super(struct super_block *sb)
813{
Andreas Gruenbacher8c52ab42005-07-27 11:45:15 -0700814 mb_cache_shrink(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815}
816
817
818/*
819 * ext2_xattr_cache_insert()
820 *
821 * Create a new entry in the extended attribute cache, and insert
822 * it unless such an entry is already in the cache.
823 *
824 * Returns 0, or a negative error number on failure.
825 */
826static int
827ext2_xattr_cache_insert(struct buffer_head *bh)
828{
829 __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
830 struct mb_cache_entry *ce;
831 int error;
832
Jan Kara335e92e2008-04-15 14:34:43 -0700833 ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 if (!ce)
835 return -ENOMEM;
836 error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, &hash);
837 if (error) {
838 mb_cache_entry_free(ce);
839 if (error == -EBUSY) {
840 ea_bdebug(bh, "already in cache (%d cache entries)",
841 atomic_read(&ext2_xattr_cache->c_entry_count));
842 error = 0;
843 }
844 } else {
845 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
846 atomic_read(&ext2_xattr_cache->c_entry_count));
847 mb_cache_entry_release(ce);
848 }
849 return error;
850}
851
852/*
853 * ext2_xattr_cmp()
854 *
855 * Compare two extended attribute blocks for equality.
856 *
857 * Returns 0 if the blocks are equal, 1 if they differ, and
858 * a negative error number on errors.
859 */
860static int
861ext2_xattr_cmp(struct ext2_xattr_header *header1,
862 struct ext2_xattr_header *header2)
863{
864 struct ext2_xattr_entry *entry1, *entry2;
865
866 entry1 = ENTRY(header1+1);
867 entry2 = ENTRY(header2+1);
868 while (!IS_LAST_ENTRY(entry1)) {
869 if (IS_LAST_ENTRY(entry2))
870 return 1;
871 if (entry1->e_hash != entry2->e_hash ||
872 entry1->e_name_index != entry2->e_name_index ||
873 entry1->e_name_len != entry2->e_name_len ||
874 entry1->e_value_size != entry2->e_value_size ||
875 memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
876 return 1;
877 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
878 return -EIO;
879 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
880 (char *)header2 + le16_to_cpu(entry2->e_value_offs),
881 le32_to_cpu(entry1->e_value_size)))
882 return 1;
883
884 entry1 = EXT2_XATTR_NEXT(entry1);
885 entry2 = EXT2_XATTR_NEXT(entry2);
886 }
887 if (!IS_LAST_ENTRY(entry2))
888 return 1;
889 return 0;
890}
891
892/*
893 * ext2_xattr_cache_find()
894 *
895 * Find an identical extended attribute block.
896 *
897 * Returns a locked buffer head to the block found, or NULL if such
898 * a block was not found or an error occurred.
899 */
900static struct buffer_head *
901ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
902{
903 __u32 hash = le32_to_cpu(header->h_hash);
904 struct mb_cache_entry *ce;
905
906 if (!header->h_hash)
907 return NULL; /* never share */
908 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
909again:
910 ce = mb_cache_entry_find_first(ext2_xattr_cache, 0,
911 inode->i_sb->s_bdev, hash);
912 while (ce) {
913 struct buffer_head *bh;
914
915 if (IS_ERR(ce)) {
916 if (PTR_ERR(ce) == -EAGAIN)
917 goto again;
918 break;
919 }
920
921 bh = sb_bread(inode->i_sb, ce->e_block);
922 if (!bh) {
923 ext2_error(inode->i_sb, "ext2_xattr_cache_find",
924 "inode %ld: block %ld read error",
925 inode->i_ino, (unsigned long) ce->e_block);
926 } else {
927 lock_buffer(bh);
928 if (le32_to_cpu(HDR(bh)->h_refcount) >
929 EXT2_XATTR_REFCOUNT_MAX) {
930 ea_idebug(inode, "block %ld refcount %d>%d",
931 (unsigned long) ce->e_block,
932 le32_to_cpu(HDR(bh)->h_refcount),
933 EXT2_XATTR_REFCOUNT_MAX);
934 } else if (!ext2_xattr_cmp(header, HDR(bh))) {
935 ea_bdebug(bh, "b_count=%d",
936 atomic_read(&(bh->b_count)));
937 mb_cache_entry_release(ce);
938 return bh;
939 }
940 unlock_buffer(bh);
941 brelse(bh);
942 }
943 ce = mb_cache_entry_find_next(ce, 0, inode->i_sb->s_bdev, hash);
944 }
945 return NULL;
946}
947
948#define NAME_HASH_SHIFT 5
949#define VALUE_HASH_SHIFT 16
950
951/*
952 * ext2_xattr_hash_entry()
953 *
954 * Compute the hash of an extended attribute.
955 */
956static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
957 struct ext2_xattr_entry *entry)
958{
959 __u32 hash = 0;
960 char *name = entry->e_name;
961 int n;
962
963 for (n=0; n < entry->e_name_len; n++) {
964 hash = (hash << NAME_HASH_SHIFT) ^
965 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
966 *name++;
967 }
968
969 if (entry->e_value_block == 0 && entry->e_value_size != 0) {
970 __le32 *value = (__le32 *)((char *)header +
971 le16_to_cpu(entry->e_value_offs));
972 for (n = (le32_to_cpu(entry->e_value_size) +
973 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
974 hash = (hash << VALUE_HASH_SHIFT) ^
975 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
976 le32_to_cpu(*value++);
977 }
978 }
979 entry->e_hash = cpu_to_le32(hash);
980}
981
982#undef NAME_HASH_SHIFT
983#undef VALUE_HASH_SHIFT
984
985#define BLOCK_HASH_SHIFT 16
986
987/*
988 * ext2_xattr_rehash()
989 *
990 * Re-compute the extended attribute hash value after an entry has changed.
991 */
992static void ext2_xattr_rehash(struct ext2_xattr_header *header,
993 struct ext2_xattr_entry *entry)
994{
995 struct ext2_xattr_entry *here;
996 __u32 hash = 0;
997
998 ext2_xattr_hash_entry(header, entry);
999 here = ENTRY(header+1);
1000 while (!IS_LAST_ENTRY(here)) {
1001 if (!here->e_hash) {
1002 /* Block is not shared if an entry's hash value == 0 */
1003 hash = 0;
1004 break;
1005 }
1006 hash = (hash << BLOCK_HASH_SHIFT) ^
1007 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1008 le32_to_cpu(here->e_hash);
1009 here = EXT2_XATTR_NEXT(here);
1010 }
1011 header->h_hash = cpu_to_le32(hash);
1012}
1013
1014#undef BLOCK_HASH_SHIFT
1015
1016int __init
1017init_ext2_xattr(void)
1018{
1019 ext2_xattr_cache = mb_cache_create("ext2_xattr", NULL,
1020 sizeof(struct mb_cache_entry) +
1021 sizeof(((struct mb_cache_entry *) 0)->e_indexes[0]), 1, 6);
1022 if (!ext2_xattr_cache)
1023 return -ENOMEM;
1024 return 0;
1025}
1026
1027void
1028exit_ext2_xattr(void)
1029{
1030 mb_cache_destroy(ext2_xattr_cache);
1031}