Linux 5.2
[linux-2.6/linux-2.6-arm.git] / fs / ext2 / xattr.c
blob1e33e0ac8cf1f8dbbaa8e364f5a6ca08a24d84ef
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext2/xattr.c
5 * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
7 * Fix by Harrison Xing <harrison@mountainviewdata.com>.
8 * Extended attributes for symlinks and special files added per
9 * suggestion of Luka Renko <luka.renko@hermes.si>.
10 * xattr consolidation Copyright (c) 2004 James Morris <jmorris@redhat.com>,
11 * Red Hat Inc.
16 * Extended attributes are stored on disk blocks allocated outside of
17 * any inode. The i_file_acl field is then made to point to this allocated
18 * block. If all extended attributes of an inode are identical, these
19 * inodes may share the same extended attribute block. Such situations
20 * are automatically detected by keeping a cache of recent attribute block
21 * numbers and hashes over the block's contents in memory.
24 * Extended attribute block layout:
26 * +------------------+
27 * | header |
28 * | entry 1 | |
29 * | entry 2 | | growing downwards
30 * | entry 3 | v
31 * | four null bytes |
32 * | . . . |
33 * | value 1 | ^
34 * | value 3 | | growing upwards
35 * | value 2 | |
36 * +------------------+
38 * The block header is followed by multiple entry descriptors. These entry
39 * descriptors are variable in size, and aligned to EXT2_XATTR_PAD
40 * byte boundaries. The entry descriptors are sorted by attribute name,
41 * so that two extended attribute blocks can be compared efficiently.
43 * Attribute values are aligned to the end of the block, stored in
44 * no specific order. They are also padded to EXT2_XATTR_PAD byte
45 * boundaries. No additional gaps are left between them.
47 * Locking strategy
48 * ----------------
49 * EXT2_I(inode)->i_file_acl is protected by EXT2_I(inode)->xattr_sem.
50 * EA blocks are only changed if they are exclusive to an inode, so
51 * holding xattr_sem also means that nothing but the EA block's reference
52 * count will change. Multiple writers to an EA block are synchronized
53 * by the bh lock. No more than a single bh lock is held at any time
54 * to avoid deadlocks.
57 #include <linux/buffer_head.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 <linux/security.h>
64 #include "ext2.h"
65 #include "xattr.h"
66 #include "acl.h"
68 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
69 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
70 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
71 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
73 #ifdef EXT2_XATTR_DEBUG
74 # define ea_idebug(inode, f...) do { \
75 printk(KERN_DEBUG "inode %s:%ld: ", \
76 inode->i_sb->s_id, inode->i_ino); \
77 printk(f); \
78 printk("\n"); \
79 } while (0)
80 # define ea_bdebug(bh, f...) do { \
81 printk(KERN_DEBUG "block %pg:%lu: ", \
82 bh->b_bdev, (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
91 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
92 struct ext2_xattr_header *);
94 static int ext2_xattr_cache_insert(struct mb_cache *, struct buffer_head *);
95 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
96 struct ext2_xattr_header *);
97 static void ext2_xattr_rehash(struct ext2_xattr_header *,
98 struct ext2_xattr_entry *);
100 static const struct xattr_handler *ext2_xattr_handler_map[] = {
101 [EXT2_XATTR_INDEX_USER] = &ext2_xattr_user_handler,
102 #ifdef CONFIG_EXT2_FS_POSIX_ACL
103 [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS] = &posix_acl_access_xattr_handler,
104 [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
105 #endif
106 [EXT2_XATTR_INDEX_TRUSTED] = &ext2_xattr_trusted_handler,
107 #ifdef CONFIG_EXT2_FS_SECURITY
108 [EXT2_XATTR_INDEX_SECURITY] = &ext2_xattr_security_handler,
109 #endif
112 const struct xattr_handler *ext2_xattr_handlers[] = {
113 &ext2_xattr_user_handler,
114 &ext2_xattr_trusted_handler,
115 #ifdef CONFIG_EXT2_FS_POSIX_ACL
116 &posix_acl_access_xattr_handler,
117 &posix_acl_default_xattr_handler,
118 #endif
119 #ifdef CONFIG_EXT2_FS_SECURITY
120 &ext2_xattr_security_handler,
121 #endif
122 NULL
125 #define EA_BLOCK_CACHE(inode) (EXT2_SB(inode->i_sb)->s_ea_block_cache)
127 static inline const struct xattr_handler *
128 ext2_xattr_handler(int name_index)
130 const struct xattr_handler *handler = NULL;
132 if (name_index > 0 && name_index < ARRAY_SIZE(ext2_xattr_handler_map))
133 handler = ext2_xattr_handler_map[name_index];
134 return handler;
138 * ext2_xattr_get()
140 * Copy an extended attribute into the buffer
141 * provided, or compute the buffer size required.
142 * Buffer is NULL to compute the size of the buffer required.
144 * Returns a negative error number on failure, or the number of bytes
145 * used / required on success.
148 ext2_xattr_get(struct inode *inode, int name_index, const char *name,
149 void *buffer, size_t buffer_size)
151 struct buffer_head *bh = NULL;
152 struct ext2_xattr_entry *entry;
153 size_t name_len, size;
154 char *end;
155 int error;
156 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
158 ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
159 name_index, name, buffer, (long)buffer_size);
161 if (name == NULL)
162 return -EINVAL;
163 name_len = strlen(name);
164 if (name_len > 255)
165 return -ERANGE;
167 down_read(&EXT2_I(inode)->xattr_sem);
168 error = -ENODATA;
169 if (!EXT2_I(inode)->i_file_acl)
170 goto cleanup;
171 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
172 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
173 error = -EIO;
174 if (!bh)
175 goto cleanup;
176 ea_bdebug(bh, "b_count=%d, refcount=%d",
177 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
178 end = bh->b_data + bh->b_size;
179 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
180 HDR(bh)->h_blocks != cpu_to_le32(1)) {
181 bad_block: ext2_error(inode->i_sb, "ext2_xattr_get",
182 "inode %ld: bad block %d", inode->i_ino,
183 EXT2_I(inode)->i_file_acl);
184 error = -EIO;
185 goto cleanup;
188 /* find named attribute */
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;
201 if (ext2_xattr_cache_insert(ea_block_cache, bh))
202 ea_idebug(inode, "cache insert failed");
203 error = -ENODATA;
204 goto cleanup;
205 found:
206 /* check the buffer size */
207 if (entry->e_value_block != 0)
208 goto bad_block;
209 size = le32_to_cpu(entry->e_value_size);
210 if (size > inode->i_sb->s_blocksize ||
211 le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
212 goto bad_block;
214 if (ext2_xattr_cache_insert(ea_block_cache, bh))
215 ea_idebug(inode, "cache insert failed");
216 if (buffer) {
217 error = -ERANGE;
218 if (size > buffer_size)
219 goto cleanup;
220 /* return value of attribute */
221 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
222 size);
224 error = size;
226 cleanup:
227 brelse(bh);
228 up_read(&EXT2_I(inode)->xattr_sem);
230 return error;
234 * ext2_xattr_list()
236 * Copy a list of attribute names into the buffer
237 * provided, or compute the buffer size required.
238 * Buffer is NULL to compute the size of the buffer required.
240 * Returns a negative error number on failure, or the number of bytes
241 * used / required on success.
243 static int
244 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
246 struct inode *inode = d_inode(dentry);
247 struct buffer_head *bh = NULL;
248 struct ext2_xattr_entry *entry;
249 char *end;
250 size_t rest = buffer_size;
251 int error;
252 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
254 ea_idebug(inode, "buffer=%p, buffer_size=%ld",
255 buffer, (long)buffer_size);
257 down_read(&EXT2_I(inode)->xattr_sem);
258 error = 0;
259 if (!EXT2_I(inode)->i_file_acl)
260 goto cleanup;
261 ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
262 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
263 error = -EIO;
264 if (!bh)
265 goto cleanup;
266 ea_bdebug(bh, "b_count=%d, refcount=%d",
267 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
268 end = bh->b_data + bh->b_size;
269 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
270 HDR(bh)->h_blocks != cpu_to_le32(1)) {
271 bad_block: ext2_error(inode->i_sb, "ext2_xattr_list",
272 "inode %ld: bad block %d", inode->i_ino,
273 EXT2_I(inode)->i_file_acl);
274 error = -EIO;
275 goto cleanup;
278 /* check the on-disk data structure */
279 entry = FIRST_ENTRY(bh);
280 while (!IS_LAST_ENTRY(entry)) {
281 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
283 if ((char *)next >= end)
284 goto bad_block;
285 entry = next;
287 if (ext2_xattr_cache_insert(ea_block_cache, bh))
288 ea_idebug(inode, "cache insert failed");
290 /* list the attribute names */
291 for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
292 entry = EXT2_XATTR_NEXT(entry)) {
293 const struct xattr_handler *handler =
294 ext2_xattr_handler(entry->e_name_index);
296 if (handler && (!handler->list || handler->list(dentry))) {
297 const char *prefix = handler->prefix ?: handler->name;
298 size_t prefix_len = strlen(prefix);
299 size_t size = prefix_len + entry->e_name_len + 1;
301 if (buffer) {
302 if (size > rest) {
303 error = -ERANGE;
304 goto cleanup;
306 memcpy(buffer, prefix, prefix_len);
307 buffer += prefix_len;
308 memcpy(buffer, entry->e_name, entry->e_name_len);
309 buffer += entry->e_name_len;
310 *buffer++ = 0;
312 rest -= size;
315 error = buffer_size - rest; /* total size */
317 cleanup:
318 brelse(bh);
319 up_read(&EXT2_I(inode)->xattr_sem);
321 return error;
325 * Inode operation listxattr()
327 * d_inode(dentry)->i_mutex: don't care
329 ssize_t
330 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
332 return ext2_xattr_list(dentry, buffer, size);
336 * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
337 * not set, set it.
339 static void ext2_xattr_update_super_block(struct super_block *sb)
341 if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
342 return;
344 spin_lock(&EXT2_SB(sb)->s_lock);
345 ext2_update_dynamic_rev(sb);
346 EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
347 spin_unlock(&EXT2_SB(sb)->s_lock);
348 mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
352 * ext2_xattr_set()
354 * Create, replace or remove an extended attribute for this inode. Value
355 * is NULL to remove an existing extended attribute, and non-NULL to
356 * either replace an existing extended attribute, or create a new extended
357 * attribute. The flags XATTR_REPLACE and XATTR_CREATE
358 * specify that an extended attribute must exist and must not exist
359 * previous to the call, respectively.
361 * Returns 0, or a negative error number on failure.
364 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
365 const void *value, size_t value_len, int flags)
367 struct super_block *sb = inode->i_sb;
368 struct buffer_head *bh = NULL;
369 struct ext2_xattr_header *header = NULL;
370 struct ext2_xattr_entry *here, *last;
371 size_t name_len, free, min_offs = sb->s_blocksize;
372 int not_found = 1, error;
373 char *end;
376 * header -- Points either into bh, or to a temporarily
377 * allocated buffer.
378 * here -- The named entry found, or the place for inserting, within
379 * the block pointed to by header.
380 * last -- Points right after the last named entry within the block
381 * pointed to by header.
382 * min_offs -- The offset of the first value (values are aligned
383 * towards the end of the block).
384 * end -- Points right after the block pointed to by header.
387 ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
388 name_index, name, value, (long)value_len);
390 if (value == NULL)
391 value_len = 0;
392 if (name == NULL)
393 return -EINVAL;
394 name_len = strlen(name);
395 if (name_len > 255 || value_len > sb->s_blocksize)
396 return -ERANGE;
397 down_write(&EXT2_I(inode)->xattr_sem);
398 if (EXT2_I(inode)->i_file_acl) {
399 /* The inode already has an extended attribute block. */
400 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
401 error = -EIO;
402 if (!bh)
403 goto cleanup;
404 ea_bdebug(bh, "b_count=%d, refcount=%d",
405 atomic_read(&(bh->b_count)),
406 le32_to_cpu(HDR(bh)->h_refcount));
407 header = HDR(bh);
408 end = bh->b_data + bh->b_size;
409 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
410 header->h_blocks != cpu_to_le32(1)) {
411 bad_block: ext2_error(sb, "ext2_xattr_set",
412 "inode %ld: bad block %d", inode->i_ino,
413 EXT2_I(inode)->i_file_acl);
414 error = -EIO;
415 goto cleanup;
417 /* Find the named attribute. */
418 here = FIRST_ENTRY(bh);
419 while (!IS_LAST_ENTRY(here)) {
420 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
421 if ((char *)next >= end)
422 goto bad_block;
423 if (!here->e_value_block && here->e_value_size) {
424 size_t offs = le16_to_cpu(here->e_value_offs);
425 if (offs < min_offs)
426 min_offs = offs;
428 not_found = name_index - here->e_name_index;
429 if (!not_found)
430 not_found = name_len - here->e_name_len;
431 if (!not_found)
432 not_found = memcmp(name, here->e_name,name_len);
433 if (not_found <= 0)
434 break;
435 here = next;
437 last = here;
438 /* We still need to compute min_offs and last. */
439 while (!IS_LAST_ENTRY(last)) {
440 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
441 if ((char *)next >= end)
442 goto bad_block;
443 if (!last->e_value_block && last->e_value_size) {
444 size_t offs = le16_to_cpu(last->e_value_offs);
445 if (offs < min_offs)
446 min_offs = offs;
448 last = next;
451 /* Check whether we have enough space left. */
452 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
453 } else {
454 /* We will use a new extended attribute block. */
455 free = sb->s_blocksize -
456 sizeof(struct ext2_xattr_header) - sizeof(__u32);
457 here = last = NULL; /* avoid gcc uninitialized warning. */
460 if (not_found) {
461 /* Request to remove a nonexistent attribute? */
462 error = -ENODATA;
463 if (flags & XATTR_REPLACE)
464 goto cleanup;
465 error = 0;
466 if (value == NULL)
467 goto cleanup;
468 } else {
469 /* Request to create an existing attribute? */
470 error = -EEXIST;
471 if (flags & XATTR_CREATE)
472 goto cleanup;
473 if (!here->e_value_block && here->e_value_size) {
474 size_t size = le32_to_cpu(here->e_value_size);
476 if (le16_to_cpu(here->e_value_offs) + size >
477 sb->s_blocksize || size > sb->s_blocksize)
478 goto bad_block;
479 free += EXT2_XATTR_SIZE(size);
481 free += EXT2_XATTR_LEN(name_len);
483 error = -ENOSPC;
484 if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
485 goto cleanup;
487 /* Here we know that we can set the new attribute. */
489 if (header) {
490 /* assert(header == HDR(bh)); */
491 lock_buffer(bh);
492 if (header->h_refcount == cpu_to_le32(1)) {
493 __u32 hash = le32_to_cpu(header->h_hash);
495 ea_bdebug(bh, "modifying in-place");
497 * This must happen under buffer lock for
498 * ext2_xattr_set2() to reliably detect modified block
500 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
501 bh->b_blocknr);
503 /* keep the buffer locked while modifying it. */
504 } else {
505 int offset;
507 unlock_buffer(bh);
508 ea_bdebug(bh, "cloning");
509 header = kmalloc(bh->b_size, GFP_KERNEL);
510 error = -ENOMEM;
511 if (header == NULL)
512 goto cleanup;
513 memcpy(header, HDR(bh), bh->b_size);
514 header->h_refcount = cpu_to_le32(1);
516 offset = (char *)here - bh->b_data;
517 here = ENTRY((char *)header + offset);
518 offset = (char *)last - bh->b_data;
519 last = ENTRY((char *)header + offset);
521 } else {
522 /* Allocate a buffer where we construct the new block. */
523 header = kzalloc(sb->s_blocksize, GFP_KERNEL);
524 error = -ENOMEM;
525 if (header == NULL)
526 goto cleanup;
527 end = (char *)header + sb->s_blocksize;
528 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
529 header->h_blocks = header->h_refcount = cpu_to_le32(1);
530 last = here = ENTRY(header+1);
533 /* Iff we are modifying the block in-place, bh is locked here. */
535 if (not_found) {
536 /* Insert the new name. */
537 size_t size = EXT2_XATTR_LEN(name_len);
538 size_t rest = (char *)last - (char *)here;
539 memmove((char *)here + size, here, rest);
540 memset(here, 0, size);
541 here->e_name_index = name_index;
542 here->e_name_len = name_len;
543 memcpy(here->e_name, name, name_len);
544 } else {
545 if (!here->e_value_block && here->e_value_size) {
546 char *first_val = (char *)header + min_offs;
547 size_t offs = le16_to_cpu(here->e_value_offs);
548 char *val = (char *)header + offs;
549 size_t size = EXT2_XATTR_SIZE(
550 le32_to_cpu(here->e_value_size));
552 if (size == EXT2_XATTR_SIZE(value_len)) {
553 /* The old and the new value have the same
554 size. Just replace. */
555 here->e_value_size = cpu_to_le32(value_len);
556 memset(val + size - EXT2_XATTR_PAD, 0,
557 EXT2_XATTR_PAD); /* Clear pad bytes. */
558 memcpy(val, value, value_len);
559 goto skip_replace;
562 /* Remove the old value. */
563 memmove(first_val + size, first_val, val - first_val);
564 memset(first_val, 0, size);
565 here->e_value_offs = 0;
566 min_offs += size;
568 /* Adjust all value offsets. */
569 last = ENTRY(header+1);
570 while (!IS_LAST_ENTRY(last)) {
571 size_t o = le16_to_cpu(last->e_value_offs);
572 if (!last->e_value_block && o < offs)
573 last->e_value_offs =
574 cpu_to_le16(o + size);
575 last = EXT2_XATTR_NEXT(last);
578 if (value == NULL) {
579 /* Remove the old name. */
580 size_t size = EXT2_XATTR_LEN(name_len);
581 last = ENTRY((char *)last - size);
582 memmove(here, (char*)here + size,
583 (char*)last - (char*)here);
584 memset(last, 0, size);
588 if (value != NULL) {
589 /* Insert the new value. */
590 here->e_value_size = cpu_to_le32(value_len);
591 if (value_len) {
592 size_t size = EXT2_XATTR_SIZE(value_len);
593 char *val = (char *)header + min_offs - size;
594 here->e_value_offs =
595 cpu_to_le16((char *)val - (char *)header);
596 memset(val + size - EXT2_XATTR_PAD, 0,
597 EXT2_XATTR_PAD); /* Clear the pad bytes. */
598 memcpy(val, value, value_len);
602 skip_replace:
603 if (IS_LAST_ENTRY(ENTRY(header+1))) {
604 /* This block is now empty. */
605 if (bh && header == HDR(bh))
606 unlock_buffer(bh); /* we were modifying in-place. */
607 error = ext2_xattr_set2(inode, bh, NULL);
608 } else {
609 ext2_xattr_rehash(header, here);
610 if (bh && header == HDR(bh))
611 unlock_buffer(bh); /* we were modifying in-place. */
612 error = ext2_xattr_set2(inode, bh, header);
615 cleanup:
616 if (!(bh && header == HDR(bh)))
617 kfree(header);
618 brelse(bh);
619 up_write(&EXT2_I(inode)->xattr_sem);
621 return error;
625 * Second half of ext2_xattr_set(): Update the file system.
627 static int
628 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
629 struct ext2_xattr_header *header)
631 struct super_block *sb = inode->i_sb;
632 struct buffer_head *new_bh = NULL;
633 int error;
634 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
636 if (header) {
637 new_bh = ext2_xattr_cache_find(inode, header);
638 if (new_bh) {
639 /* We found an identical block in the cache. */
640 if (new_bh == old_bh) {
641 ea_bdebug(new_bh, "keeping this block");
642 } else {
643 /* The old block is released after updating
644 the inode. */
645 ea_bdebug(new_bh, "reusing block");
647 error = dquot_alloc_block(inode, 1);
648 if (error) {
649 unlock_buffer(new_bh);
650 goto cleanup;
652 le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
653 ea_bdebug(new_bh, "refcount now=%d",
654 le32_to_cpu(HDR(new_bh)->h_refcount));
656 unlock_buffer(new_bh);
657 } else if (old_bh && header == HDR(old_bh)) {
658 /* Keep this block. No need to lock the block as we
659 don't need to change the reference count. */
660 new_bh = old_bh;
661 get_bh(new_bh);
662 ext2_xattr_cache_insert(ea_block_cache, new_bh);
663 } else {
664 /* We need to allocate a new block */
665 ext2_fsblk_t goal = ext2_group_first_block_no(sb,
666 EXT2_I(inode)->i_block_group);
667 int block = ext2_new_block(inode, goal, &error);
668 if (error)
669 goto cleanup;
670 ea_idebug(inode, "creating block %d", block);
672 new_bh = sb_getblk(sb, block);
673 if (unlikely(!new_bh)) {
674 ext2_free_blocks(inode, block, 1);
675 mark_inode_dirty(inode);
676 error = -ENOMEM;
677 goto cleanup;
679 lock_buffer(new_bh);
680 memcpy(new_bh->b_data, header, new_bh->b_size);
681 set_buffer_uptodate(new_bh);
682 unlock_buffer(new_bh);
683 ext2_xattr_cache_insert(ea_block_cache, new_bh);
685 ext2_xattr_update_super_block(sb);
687 mark_buffer_dirty(new_bh);
688 if (IS_SYNC(inode)) {
689 sync_dirty_buffer(new_bh);
690 error = -EIO;
691 if (buffer_req(new_bh) && !buffer_uptodate(new_bh))
692 goto cleanup;
696 /* Update the inode. */
697 EXT2_I(inode)->i_file_acl = new_bh ? new_bh->b_blocknr : 0;
698 inode->i_ctime = current_time(inode);
699 if (IS_SYNC(inode)) {
700 error = sync_inode_metadata(inode, 1);
701 /* In case sync failed due to ENOSPC the inode was actually
702 * written (only some dirty data were not) so we just proceed
703 * as if nothing happened and cleanup the unused block */
704 if (error && error != -ENOSPC) {
705 if (new_bh && new_bh != old_bh) {
706 dquot_free_block_nodirty(inode, 1);
707 mark_inode_dirty(inode);
709 goto cleanup;
711 } else
712 mark_inode_dirty(inode);
714 error = 0;
715 if (old_bh && old_bh != new_bh) {
717 * If there was an old block and we are no longer using it,
718 * release the old block.
720 lock_buffer(old_bh);
721 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
722 __u32 hash = le32_to_cpu(HDR(old_bh)->h_hash);
725 * This must happen under buffer lock for
726 * ext2_xattr_set2() to reliably detect freed block
728 mb_cache_entry_delete(ea_block_cache, hash,
729 old_bh->b_blocknr);
730 /* Free the old block. */
731 ea_bdebug(old_bh, "freeing");
732 ext2_free_blocks(inode, old_bh->b_blocknr, 1);
733 mark_inode_dirty(inode);
734 /* We let our caller release old_bh, so we
735 * need to duplicate the buffer before. */
736 get_bh(old_bh);
737 bforget(old_bh);
738 } else {
739 /* Decrement the refcount only. */
740 le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
741 dquot_free_block_nodirty(inode, 1);
742 mark_inode_dirty(inode);
743 mark_buffer_dirty(old_bh);
744 ea_bdebug(old_bh, "refcount now=%d",
745 le32_to_cpu(HDR(old_bh)->h_refcount));
747 unlock_buffer(old_bh);
750 cleanup:
751 brelse(new_bh);
753 return error;
757 * ext2_xattr_delete_inode()
759 * Free extended attribute resources associated with this inode. This
760 * is called immediately before an inode is freed.
762 void
763 ext2_xattr_delete_inode(struct inode *inode)
765 struct buffer_head *bh = NULL;
766 struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
768 down_write(&EXT2_I(inode)->xattr_sem);
769 if (!EXT2_I(inode)->i_file_acl)
770 goto cleanup;
772 if (!ext2_data_block_valid(sbi, EXT2_I(inode)->i_file_acl, 0)) {
773 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
774 "inode %ld: xattr block %d is out of data blocks range",
775 inode->i_ino, EXT2_I(inode)->i_file_acl);
776 goto cleanup;
779 bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
780 if (!bh) {
781 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
782 "inode %ld: block %d read error", inode->i_ino,
783 EXT2_I(inode)->i_file_acl);
784 goto cleanup;
786 ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
787 if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
788 HDR(bh)->h_blocks != cpu_to_le32(1)) {
789 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
790 "inode %ld: bad block %d", inode->i_ino,
791 EXT2_I(inode)->i_file_acl);
792 goto cleanup;
794 lock_buffer(bh);
795 if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
796 __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
799 * This must happen under buffer lock for ext2_xattr_set2() to
800 * reliably detect freed block
802 mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
803 bh->b_blocknr);
804 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
805 get_bh(bh);
806 bforget(bh);
807 unlock_buffer(bh);
808 } else {
809 le32_add_cpu(&HDR(bh)->h_refcount, -1);
810 ea_bdebug(bh, "refcount now=%d",
811 le32_to_cpu(HDR(bh)->h_refcount));
812 unlock_buffer(bh);
813 mark_buffer_dirty(bh);
814 if (IS_SYNC(inode))
815 sync_dirty_buffer(bh);
816 dquot_free_block_nodirty(inode, 1);
818 EXT2_I(inode)->i_file_acl = 0;
820 cleanup:
821 brelse(bh);
822 up_write(&EXT2_I(inode)->xattr_sem);
826 * ext2_xattr_cache_insert()
828 * Create a new entry in the extended attribute cache, and insert
829 * it unless such an entry is already in the cache.
831 * Returns 0, or a negative error number on failure.
833 static int
834 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
836 __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
837 int error;
839 error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr,
840 true);
841 if (error) {
842 if (error == -EBUSY) {
843 ea_bdebug(bh, "already in cache (%d cache entries)",
844 atomic_read(&ext2_xattr_cache->c_entry_count));
845 error = 0;
847 } else
848 ea_bdebug(bh, "inserting [%x]", (int)hash);
849 return error;
853 * ext2_xattr_cmp()
855 * Compare two extended attribute blocks for equality.
857 * Returns 0 if the blocks are equal, 1 if they differ, and
858 * a negative error number on errors.
860 static int
861 ext2_xattr_cmp(struct ext2_xattr_header *header1,
862 struct ext2_xattr_header *header2)
864 struct ext2_xattr_entry *entry1, *entry2;
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;
884 entry1 = EXT2_XATTR_NEXT(entry1);
885 entry2 = EXT2_XATTR_NEXT(entry2);
887 if (!IS_LAST_ENTRY(entry2))
888 return 1;
889 return 0;
893 * ext2_xattr_cache_find()
895 * Find an identical extended attribute block.
897 * Returns a locked buffer head to the block found, or NULL if such
898 * a block was not found or an error occurred.
900 static struct buffer_head *
901 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
903 __u32 hash = le32_to_cpu(header->h_hash);
904 struct mb_cache_entry *ce;
905 struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
907 if (!header->h_hash)
908 return NULL; /* never share */
909 ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
910 again:
911 ce = mb_cache_entry_find_first(ea_block_cache, hash);
912 while (ce) {
913 struct buffer_head *bh;
915 bh = sb_bread(inode->i_sb, ce->e_value);
916 if (!bh) {
917 ext2_error(inode->i_sb, "ext2_xattr_cache_find",
918 "inode %ld: block %ld read error",
919 inode->i_ino, (unsigned long) ce->e_value);
920 } else {
921 lock_buffer(bh);
923 * We have to be careful about races with freeing or
924 * rehashing of xattr block. Once we hold buffer lock
925 * xattr block's state is stable so we can check
926 * whether the block got freed / rehashed or not.
927 * Since we unhash mbcache entry under buffer lock when
928 * freeing / rehashing xattr block, checking whether
929 * entry is still hashed is reliable.
931 if (hlist_bl_unhashed(&ce->e_hash_list)) {
932 mb_cache_entry_put(ea_block_cache, ce);
933 unlock_buffer(bh);
934 brelse(bh);
935 goto again;
936 } else if (le32_to_cpu(HDR(bh)->h_refcount) >
937 EXT2_XATTR_REFCOUNT_MAX) {
938 ea_idebug(inode, "block %ld refcount %d>%d",
939 (unsigned long) ce->e_value,
940 le32_to_cpu(HDR(bh)->h_refcount),
941 EXT2_XATTR_REFCOUNT_MAX);
942 } else if (!ext2_xattr_cmp(header, HDR(bh))) {
943 ea_bdebug(bh, "b_count=%d",
944 atomic_read(&(bh->b_count)));
945 mb_cache_entry_touch(ea_block_cache, ce);
946 mb_cache_entry_put(ea_block_cache, ce);
947 return bh;
949 unlock_buffer(bh);
950 brelse(bh);
952 ce = mb_cache_entry_find_next(ea_block_cache, ce);
954 return NULL;
957 #define NAME_HASH_SHIFT 5
958 #define VALUE_HASH_SHIFT 16
961 * ext2_xattr_hash_entry()
963 * Compute the hash of an extended attribute.
965 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
966 struct ext2_xattr_entry *entry)
968 __u32 hash = 0;
969 char *name = entry->e_name;
970 int n;
972 for (n=0; n < entry->e_name_len; n++) {
973 hash = (hash << NAME_HASH_SHIFT) ^
974 (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
975 *name++;
978 if (entry->e_value_block == 0 && entry->e_value_size != 0) {
979 __le32 *value = (__le32 *)((char *)header +
980 le16_to_cpu(entry->e_value_offs));
981 for (n = (le32_to_cpu(entry->e_value_size) +
982 EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
983 hash = (hash << VALUE_HASH_SHIFT) ^
984 (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
985 le32_to_cpu(*value++);
988 entry->e_hash = cpu_to_le32(hash);
991 #undef NAME_HASH_SHIFT
992 #undef VALUE_HASH_SHIFT
994 #define BLOCK_HASH_SHIFT 16
997 * ext2_xattr_rehash()
999 * Re-compute the extended attribute hash value after an entry has changed.
1001 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1002 struct ext2_xattr_entry *entry)
1004 struct ext2_xattr_entry *here;
1005 __u32 hash = 0;
1007 ext2_xattr_hash_entry(header, entry);
1008 here = ENTRY(header+1);
1009 while (!IS_LAST_ENTRY(here)) {
1010 if (!here->e_hash) {
1011 /* Block is not shared if an entry's hash value == 0 */
1012 hash = 0;
1013 break;
1015 hash = (hash << BLOCK_HASH_SHIFT) ^
1016 (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1017 le32_to_cpu(here->e_hash);
1018 here = EXT2_XATTR_NEXT(here);
1020 header->h_hash = cpu_to_le32(hash);
1023 #undef BLOCK_HASH_SHIFT
1025 #define HASH_BUCKET_BITS 10
1027 struct mb_cache *ext2_xattr_create_cache(void)
1029 return mb_cache_create(HASH_BUCKET_BITS);
1032 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1034 if (cache)
1035 mb_cache_destroy(cache);