xattr handlers: Simplify list operation
[linux-2.6-block.git] / fs / ext2 / xattr.c
1 /*
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 aligned 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/init.h>
58 #include <linux/slab.h>
59 #include <linux/mbcache.h>
60 #include <linux/quotaops.h>
61 #include <linux/rwsem.h>
62 #include <linux/security.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
92 static int ext2_xattr_set2(struct inode *, struct buffer_head *,
93                            struct ext2_xattr_header *);
94
95 static int ext2_xattr_cache_insert(struct buffer_head *);
96 static struct buffer_head *ext2_xattr_cache_find(struct inode *,
97                                                  struct ext2_xattr_header *);
98 static void ext2_xattr_rehash(struct ext2_xattr_header *,
99                               struct ext2_xattr_entry *);
100
101 static struct mb_cache *ext2_xattr_cache;
102
103 static const 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]  = &posix_acl_access_xattr_handler,
107         [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_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
115 const struct xattr_handler *ext2_xattr_handlers[] = {
116         &ext2_xattr_user_handler,
117         &ext2_xattr_trusted_handler,
118 #ifdef CONFIG_EXT2_FS_POSIX_ACL
119         &posix_acl_access_xattr_handler,
120         &posix_acl_default_xattr_handler,
121 #endif
122 #ifdef CONFIG_EXT2_FS_SECURITY
123         &ext2_xattr_security_handler,
124 #endif
125         NULL
126 };
127
128 static inline const struct xattr_handler *
129 ext2_xattr_handler(int name_index)
130 {
131         const 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  */
148 int
149 ext2_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         name_len = strlen(name);
164         if (name_len > 255)
165                 return -ERANGE;
166
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;
186         }
187
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;
200         }
201         if (ext2_xattr_cache_insert(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;
213
214         if (ext2_xattr_cache_insert(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);
223         }
224         error = size;
225
226 cleanup:
227         brelse(bh);
228         up_read(&EXT2_I(inode)->xattr_sem);
229
230         return error;
231 }
232
233 /*
234  * ext2_xattr_list()
235  *
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.
239  *
240  * Returns a negative error number on failure, or the number of bytes
241  * used / required on success.
242  */
243 static int
244 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
245 {
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
253         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
254                   buffer, (long)buffer_size);
255
256         down_read(&EXT2_I(inode)->xattr_sem);
257         error = 0;
258         if (!EXT2_I(inode)->i_file_acl)
259                 goto cleanup;
260         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
261         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
262         error = -EIO;
263         if (!bh)
264                 goto cleanup;
265         ea_bdebug(bh, "b_count=%d, refcount=%d",
266                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
267         end = bh->b_data + bh->b_size;
268         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
269             HDR(bh)->h_blocks != cpu_to_le32(1)) {
270 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_list",
271                         "inode %ld: bad block %d", inode->i_ino,
272                         EXT2_I(inode)->i_file_acl);
273                 error = -EIO;
274                 goto cleanup;
275         }
276
277         /* check the on-disk data structure */
278         entry = FIRST_ENTRY(bh);
279         while (!IS_LAST_ENTRY(entry)) {
280                 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
281
282                 if ((char *)next >= end)
283                         goto bad_block;
284                 entry = next;
285         }
286         if (ext2_xattr_cache_insert(bh))
287                 ea_idebug(inode, "cache insert failed");
288
289         /* list the attribute names */
290         for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
291              entry = EXT2_XATTR_NEXT(entry)) {
292                 const struct xattr_handler *handler =
293                         ext2_xattr_handler(entry->e_name_index);
294
295                 if (handler && (!handler->list || handler->list(dentry))) {
296                         const char *prefix = handler->prefix ?: handler->name;
297                         size_t prefix_len = strlen(prefix);
298                         size_t size = prefix_len + entry->e_name_len + 1;
299
300                         if (buffer) {
301                                 if (size > rest) {
302                                         error = -ERANGE;
303                                         goto cleanup;
304                                 }
305                                 memcpy(buffer, prefix, prefix_len);
306                                 buffer += prefix_len;
307                                 memcpy(buffer, entry->e_name, entry->e_name_len);
308                                 buffer += entry->e_name_len;
309                                 *buffer++ = 0;
310                         }
311                         rest -= size;
312                 }
313         }
314         error = buffer_size - rest;  /* total size */
315
316 cleanup:
317         brelse(bh);
318         up_read(&EXT2_I(inode)->xattr_sem);
319
320         return error;
321 }
322
323 /*
324  * Inode operation listxattr()
325  *
326  * d_inode(dentry)->i_mutex: don't care
327  */
328 ssize_t
329 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
330 {
331         return ext2_xattr_list(dentry, buffer, size);
332 }
333
334 /*
335  * If the EXT2_FEATURE_COMPAT_EXT_ATTR feature of this file system is
336  * not set, set it.
337  */
338 static void ext2_xattr_update_super_block(struct super_block *sb)
339 {
340         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
341                 return;
342
343         spin_lock(&EXT2_SB(sb)->s_lock);
344         EXT2_SET_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR);
345         spin_unlock(&EXT2_SB(sb)->s_lock);
346         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
347 }
348
349 /*
350  * ext2_xattr_set()
351  *
352  * Create, replace or remove an extended attribute for this inode.  Value
353  * is NULL to remove an existing extended attribute, and non-NULL to
354  * either replace an existing extended attribute, or create a new extended
355  * attribute. The flags XATTR_REPLACE and XATTR_CREATE
356  * specify that an extended attribute must exist and must not exist
357  * previous to the call, respectively.
358  *
359  * Returns 0, or a negative error number on failure.
360  */
361 int
362 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
363                const void *value, size_t value_len, int flags)
364 {
365         struct super_block *sb = inode->i_sb;
366         struct buffer_head *bh = NULL;
367         struct ext2_xattr_header *header = NULL;
368         struct ext2_xattr_entry *here, *last;
369         size_t name_len, free, min_offs = sb->s_blocksize;
370         int not_found = 1, error;
371         char *end;
372         
373         /*
374          * header -- Points either into bh, or to a temporarily
375          *           allocated buffer.
376          * here -- The named entry found, or the place for inserting, within
377          *         the block pointed to by header.
378          * last -- Points right after the last named entry within the block
379          *         pointed to by header.
380          * min_offs -- The offset of the first value (values are aligned
381          *             towards the end of the block).
382          * end -- Points right after the block pointed to by header.
383          */
384         
385         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
386                   name_index, name, value, (long)value_len);
387
388         if (value == NULL)
389                 value_len = 0;
390         if (name == NULL)
391                 return -EINVAL;
392         name_len = strlen(name);
393         if (name_len > 255 || value_len > sb->s_blocksize)
394                 return -ERANGE;
395         down_write(&EXT2_I(inode)->xattr_sem);
396         if (EXT2_I(inode)->i_file_acl) {
397                 /* The inode already has an extended attribute block. */
398                 bh = sb_bread(sb, EXT2_I(inode)->i_file_acl);
399                 error = -EIO;
400                 if (!bh)
401                         goto cleanup;
402                 ea_bdebug(bh, "b_count=%d, refcount=%d",
403                         atomic_read(&(bh->b_count)),
404                         le32_to_cpu(HDR(bh)->h_refcount));
405                 header = HDR(bh);
406                 end = bh->b_data + bh->b_size;
407                 if (header->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
408                     header->h_blocks != cpu_to_le32(1)) {
409 bad_block:              ext2_error(sb, "ext2_xattr_set",
410                                 "inode %ld: bad block %d", inode->i_ino, 
411                                    EXT2_I(inode)->i_file_acl);
412                         error = -EIO;
413                         goto cleanup;
414                 }
415                 /* Find the named attribute. */
416                 here = FIRST_ENTRY(bh);
417                 while (!IS_LAST_ENTRY(here)) {
418                         struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(here);
419                         if ((char *)next >= end)
420                                 goto bad_block;
421                         if (!here->e_value_block && here->e_value_size) {
422                                 size_t offs = le16_to_cpu(here->e_value_offs);
423                                 if (offs < min_offs)
424                                         min_offs = offs;
425                         }
426                         not_found = name_index - here->e_name_index;
427                         if (!not_found)
428                                 not_found = name_len - here->e_name_len;
429                         if (!not_found)
430                                 not_found = memcmp(name, here->e_name,name_len);
431                         if (not_found <= 0)
432                                 break;
433                         here = next;
434                 }
435                 last = here;
436                 /* We still need to compute min_offs and last. */
437                 while (!IS_LAST_ENTRY(last)) {
438                         struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(last);
439                         if ((char *)next >= end)
440                                 goto bad_block;
441                         if (!last->e_value_block && last->e_value_size) {
442                                 size_t offs = le16_to_cpu(last->e_value_offs);
443                                 if (offs < min_offs)
444                                         min_offs = offs;
445                         }
446                         last = next;
447                 }
448
449                 /* Check whether we have enough space left. */
450                 free = min_offs - ((char*)last - (char*)header) - sizeof(__u32);
451         } else {
452                 /* We will use a new extended attribute block. */
453                 free = sb->s_blocksize -
454                         sizeof(struct ext2_xattr_header) - sizeof(__u32);
455                 here = last = NULL;  /* avoid gcc uninitialized warning. */
456         }
457
458         if (not_found) {
459                 /* Request to remove a nonexistent attribute? */
460                 error = -ENODATA;
461                 if (flags & XATTR_REPLACE)
462                         goto cleanup;
463                 error = 0;
464                 if (value == NULL)
465                         goto cleanup;
466         } else {
467                 /* Request to create an existing attribute? */
468                 error = -EEXIST;
469                 if (flags & XATTR_CREATE)
470                         goto cleanup;
471                 if (!here->e_value_block && here->e_value_size) {
472                         size_t size = le32_to_cpu(here->e_value_size);
473
474                         if (le16_to_cpu(here->e_value_offs) + size > 
475                             sb->s_blocksize || size > sb->s_blocksize)
476                                 goto bad_block;
477                         free += EXT2_XATTR_SIZE(size);
478                 }
479                 free += EXT2_XATTR_LEN(name_len);
480         }
481         error = -ENOSPC;
482         if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
483                 goto cleanup;
484
485         /* Here we know that we can set the new attribute. */
486
487         if (header) {
488                 struct mb_cache_entry *ce;
489
490                 /* assert(header == HDR(bh)); */
491                 ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev,
492                                         bh->b_blocknr);
493                 lock_buffer(bh);
494                 if (header->h_refcount == cpu_to_le32(1)) {
495                         ea_bdebug(bh, "modifying in-place");
496                         if (ce)
497                                 mb_cache_entry_free(ce);
498                         /* keep the buffer locked while modifying it. */
499                 } else {
500                         int offset;
501
502                         if (ce)
503                                 mb_cache_entry_release(ce);
504                         unlock_buffer(bh);
505                         ea_bdebug(bh, "cloning");
506                         header = kmalloc(bh->b_size, GFP_KERNEL);
507                         error = -ENOMEM;
508                         if (header == NULL)
509                                 goto cleanup;
510                         memcpy(header, HDR(bh), bh->b_size);
511                         header->h_refcount = cpu_to_le32(1);
512
513                         offset = (char *)here - bh->b_data;
514                         here = ENTRY((char *)header + offset);
515                         offset = (char *)last - bh->b_data;
516                         last = ENTRY((char *)header + offset);
517                 }
518         } else {
519                 /* Allocate a buffer where we construct the new block. */
520                 header = kzalloc(sb->s_blocksize, GFP_KERNEL);
521                 error = -ENOMEM;
522                 if (header == NULL)
523                         goto cleanup;
524                 end = (char *)header + sb->s_blocksize;
525                 header->h_magic = cpu_to_le32(EXT2_XATTR_MAGIC);
526                 header->h_blocks = header->h_refcount = cpu_to_le32(1);
527                 last = here = ENTRY(header+1);
528         }
529
530         /* Iff we are modifying the block in-place, bh is locked here. */
531
532         if (not_found) {
533                 /* Insert the new name. */
534                 size_t size = EXT2_XATTR_LEN(name_len);
535                 size_t rest = (char *)last - (char *)here;
536                 memmove((char *)here + size, here, rest);
537                 memset(here, 0, size);
538                 here->e_name_index = name_index;
539                 here->e_name_len = name_len;
540                 memcpy(here->e_name, name, name_len);
541         } else {
542                 if (!here->e_value_block && here->e_value_size) {
543                         char *first_val = (char *)header + min_offs;
544                         size_t offs = le16_to_cpu(here->e_value_offs);
545                         char *val = (char *)header + offs;
546                         size_t size = EXT2_XATTR_SIZE(
547                                 le32_to_cpu(here->e_value_size));
548
549                         if (size == EXT2_XATTR_SIZE(value_len)) {
550                                 /* The old and the new value have the same
551                                    size. Just replace. */
552                                 here->e_value_size = cpu_to_le32(value_len);
553                                 memset(val + size - EXT2_XATTR_PAD, 0,
554                                        EXT2_XATTR_PAD); /* Clear pad bytes. */
555                                 memcpy(val, value, value_len);
556                                 goto skip_replace;
557                         }
558
559                         /* Remove the old value. */
560                         memmove(first_val + size, first_val, val - first_val);
561                         memset(first_val, 0, size);
562                         here->e_value_offs = 0;
563                         min_offs += size;
564
565                         /* Adjust all value offsets. */
566                         last = ENTRY(header+1);
567                         while (!IS_LAST_ENTRY(last)) {
568                                 size_t o = le16_to_cpu(last->e_value_offs);
569                                 if (!last->e_value_block && o < offs)
570                                         last->e_value_offs =
571                                                 cpu_to_le16(o + size);
572                                 last = EXT2_XATTR_NEXT(last);
573                         }
574                 }
575                 if (value == NULL) {
576                         /* Remove the old name. */
577                         size_t size = EXT2_XATTR_LEN(name_len);
578                         last = ENTRY((char *)last - size);
579                         memmove(here, (char*)here + size,
580                                 (char*)last - (char*)here);
581                         memset(last, 0, size);
582                 }
583         }
584
585         if (value != NULL) {
586                 /* Insert the new value. */
587                 here->e_value_size = cpu_to_le32(value_len);
588                 if (value_len) {
589                         size_t size = EXT2_XATTR_SIZE(value_len);
590                         char *val = (char *)header + min_offs - size;
591                         here->e_value_offs =
592                                 cpu_to_le16((char *)val - (char *)header);
593                         memset(val + size - EXT2_XATTR_PAD, 0,
594                                EXT2_XATTR_PAD); /* Clear the pad bytes. */
595                         memcpy(val, value, value_len);
596                 }
597         }
598
599 skip_replace:
600         if (IS_LAST_ENTRY(ENTRY(header+1))) {
601                 /* This block is now empty. */
602                 if (bh && header == HDR(bh))
603                         unlock_buffer(bh);  /* we were modifying in-place. */
604                 error = ext2_xattr_set2(inode, bh, NULL);
605         } else {
606                 ext2_xattr_rehash(header, here);
607                 if (bh && header == HDR(bh))
608                         unlock_buffer(bh);  /* we were modifying in-place. */
609                 error = ext2_xattr_set2(inode, bh, header);
610         }
611
612 cleanup:
613         brelse(bh);
614         if (!(bh && header == HDR(bh)))
615                 kfree(header);
616         up_write(&EXT2_I(inode)->xattr_sem);
617
618         return error;
619 }
620
621 /*
622  * Second half of ext2_xattr_set(): Update the file system.
623  */
624 static int
625 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
626                 struct ext2_xattr_header *header)
627 {
628         struct super_block *sb = inode->i_sb;
629         struct buffer_head *new_bh = NULL;
630         int error;
631
632         if (header) {
633                 new_bh = ext2_xattr_cache_find(inode, header);
634                 if (new_bh) {
635                         /* We found an identical block in the cache. */
636                         if (new_bh == old_bh) {
637                                 ea_bdebug(new_bh, "keeping this block");
638                         } else {
639                                 /* The old block is released after updating
640                                    the inode.  */
641                                 ea_bdebug(new_bh, "reusing block");
642
643                                 error = dquot_alloc_block(inode, 1);
644                                 if (error) {
645                                         unlock_buffer(new_bh);
646                                         goto cleanup;
647                                 }
648                                 le32_add_cpu(&HDR(new_bh)->h_refcount, 1);
649                                 ea_bdebug(new_bh, "refcount now=%d",
650                                         le32_to_cpu(HDR(new_bh)->h_refcount));
651                         }
652                         unlock_buffer(new_bh);
653                 } else if (old_bh && header == HDR(old_bh)) {
654                         /* Keep this block. No need to lock the block as we
655                            don't need to change the reference count. */
656                         new_bh = old_bh;
657                         get_bh(new_bh);
658                         ext2_xattr_cache_insert(new_bh);
659                 } else {
660                         /* We need to allocate a new block */
661                         ext2_fsblk_t goal = ext2_group_first_block_no(sb,
662                                                 EXT2_I(inode)->i_block_group);
663                         int block = ext2_new_block(inode, goal, &error);
664                         if (error)
665                                 goto cleanup;
666                         ea_idebug(inode, "creating block %d", block);
667
668                         new_bh = sb_getblk(sb, block);
669                         if (unlikely(!new_bh)) {
670                                 ext2_free_blocks(inode, block, 1);
671                                 mark_inode_dirty(inode);
672                                 error = -ENOMEM;
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 = sync_inode_metadata(inode, 1);
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) {
702                                 dquot_free_block_nodirty(inode, 1);
703                                 mark_inode_dirty(inode);
704                         }
705                         goto cleanup;
706                 }
707         } else
708                 mark_inode_dirty(inode);
709
710         error = 0;
711         if (old_bh && old_bh != new_bh) {
712                 struct mb_cache_entry *ce;
713
714                 /*
715                  * If there was an old block and we are no longer using it,
716                  * release the old block.
717                  */
718                 ce = mb_cache_entry_get(ext2_xattr_cache, old_bh->b_bdev,
719                                         old_bh->b_blocknr);
720                 lock_buffer(old_bh);
721                 if (HDR(old_bh)->h_refcount == cpu_to_le32(1)) {
722                         /* Free the old block. */
723                         if (ce)
724                                 mb_cache_entry_free(ce);
725                         ea_bdebug(old_bh, "freeing");
726                         ext2_free_blocks(inode, old_bh->b_blocknr, 1);
727                         mark_inode_dirty(inode);
728                         /* We let our caller release old_bh, so we
729                          * need to duplicate the buffer before. */
730                         get_bh(old_bh);
731                         bforget(old_bh);
732                 } else {
733                         /* Decrement the refcount only. */
734                         le32_add_cpu(&HDR(old_bh)->h_refcount, -1);
735                         if (ce)
736                                 mb_cache_entry_release(ce);
737                         dquot_free_block_nodirty(inode, 1);
738                         mark_inode_dirty(inode);
739                         mark_buffer_dirty(old_bh);
740                         ea_bdebug(old_bh, "refcount now=%d",
741                                 le32_to_cpu(HDR(old_bh)->h_refcount));
742                 }
743                 unlock_buffer(old_bh);
744         }
745
746 cleanup:
747         brelse(new_bh);
748
749         return error;
750 }
751
752 /*
753  * ext2_xattr_delete_inode()
754  *
755  * Free extended attribute resources associated with this inode. This
756  * is called immediately before an inode is freed.
757  */
758 void
759 ext2_xattr_delete_inode(struct inode *inode)
760 {
761         struct buffer_head *bh = NULL;
762         struct mb_cache_entry *ce;
763
764         down_write(&EXT2_I(inode)->xattr_sem);
765         if (!EXT2_I(inode)->i_file_acl)
766                 goto cleanup;
767         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
768         if (!bh) {
769                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
770                         "inode %ld: block %d read error", inode->i_ino,
771                         EXT2_I(inode)->i_file_acl);
772                 goto cleanup;
773         }
774         ea_bdebug(bh, "b_count=%d", atomic_read(&(bh->b_count)));
775         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
776             HDR(bh)->h_blocks != cpu_to_le32(1)) {
777                 ext2_error(inode->i_sb, "ext2_xattr_delete_inode",
778                         "inode %ld: bad block %d", inode->i_ino,
779                         EXT2_I(inode)->i_file_acl);
780                 goto cleanup;
781         }
782         ce = mb_cache_entry_get(ext2_xattr_cache, bh->b_bdev, bh->b_blocknr);
783         lock_buffer(bh);
784         if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
785                 if (ce)
786                         mb_cache_entry_free(ce);
787                 ext2_free_blocks(inode, EXT2_I(inode)->i_file_acl, 1);
788                 get_bh(bh);
789                 bforget(bh);
790                 unlock_buffer(bh);
791         } else {
792                 le32_add_cpu(&HDR(bh)->h_refcount, -1);
793                 if (ce)
794                         mb_cache_entry_release(ce);
795                 ea_bdebug(bh, "refcount now=%d",
796                         le32_to_cpu(HDR(bh)->h_refcount));
797                 unlock_buffer(bh);
798                 mark_buffer_dirty(bh);
799                 if (IS_SYNC(inode))
800                         sync_dirty_buffer(bh);
801                 dquot_free_block_nodirty(inode, 1);
802         }
803         EXT2_I(inode)->i_file_acl = 0;
804
805 cleanup:
806         brelse(bh);
807         up_write(&EXT2_I(inode)->xattr_sem);
808 }
809
810 /*
811  * ext2_xattr_put_super()
812  *
813  * This is called when a file system is unmounted.
814  */
815 void
816 ext2_xattr_put_super(struct super_block *sb)
817 {
818         mb_cache_shrink(sb->s_bdev);
819 }
820
821
822 /*
823  * ext2_xattr_cache_insert()
824  *
825  * Create a new entry in the extended attribute cache, and insert
826  * it unless such an entry is already in the cache.
827  *
828  * Returns 0, or a negative error number on failure.
829  */
830 static int
831 ext2_xattr_cache_insert(struct buffer_head *bh)
832 {
833         __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
834         struct mb_cache_entry *ce;
835         int error;
836
837         ce = mb_cache_entry_alloc(ext2_xattr_cache, GFP_NOFS);
838         if (!ce)
839                 return -ENOMEM;
840         error = mb_cache_entry_insert(ce, bh->b_bdev, bh->b_blocknr, hash);
841         if (error) {
842                 mb_cache_entry_free(ce);
843                 if (error == -EBUSY) {
844                         ea_bdebug(bh, "already in cache (%d cache entries)",
845                                 atomic_read(&ext2_xattr_cache->c_entry_count));
846                         error = 0;
847                 }
848         } else {
849                 ea_bdebug(bh, "inserting [%x] (%d cache entries)", (int)hash,
850                           atomic_read(&ext2_xattr_cache->c_entry_count));
851                 mb_cache_entry_release(ce);
852         }
853         return error;
854 }
855
856 /*
857  * ext2_xattr_cmp()
858  *
859  * Compare two extended attribute blocks for equality.
860  *
861  * Returns 0 if the blocks are equal, 1 if they differ, and
862  * a negative error number on errors.
863  */
864 static int
865 ext2_xattr_cmp(struct ext2_xattr_header *header1,
866                struct ext2_xattr_header *header2)
867 {
868         struct ext2_xattr_entry *entry1, *entry2;
869
870         entry1 = ENTRY(header1+1);
871         entry2 = ENTRY(header2+1);
872         while (!IS_LAST_ENTRY(entry1)) {
873                 if (IS_LAST_ENTRY(entry2))
874                         return 1;
875                 if (entry1->e_hash != entry2->e_hash ||
876                     entry1->e_name_index != entry2->e_name_index ||
877                     entry1->e_name_len != entry2->e_name_len ||
878                     entry1->e_value_size != entry2->e_value_size ||
879                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
880                         return 1;
881                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
882                         return -EIO;
883                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
884                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
885                            le32_to_cpu(entry1->e_value_size)))
886                         return 1;
887
888                 entry1 = EXT2_XATTR_NEXT(entry1);
889                 entry2 = EXT2_XATTR_NEXT(entry2);
890         }
891         if (!IS_LAST_ENTRY(entry2))
892                 return 1;
893         return 0;
894 }
895
896 /*
897  * ext2_xattr_cache_find()
898  *
899  * Find an identical extended attribute block.
900  *
901  * Returns a locked buffer head to the block found, or NULL if such
902  * a block was not found or an error occurred.
903  */
904 static struct buffer_head *
905 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
906 {
907         __u32 hash = le32_to_cpu(header->h_hash);
908         struct mb_cache_entry *ce;
909
910         if (!header->h_hash)
911                 return NULL;  /* never share */
912         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
913 again:
914         ce = mb_cache_entry_find_first(ext2_xattr_cache, inode->i_sb->s_bdev,
915                                        hash);
916         while (ce) {
917                 struct buffer_head *bh;
918
919                 if (IS_ERR(ce)) {
920                         if (PTR_ERR(ce) == -EAGAIN)
921                                 goto again;
922                         break;
923                 }
924
925                 bh = sb_bread(inode->i_sb, ce->e_block);
926                 if (!bh) {
927                         ext2_error(inode->i_sb, "ext2_xattr_cache_find",
928                                 "inode %ld: block %ld read error",
929                                 inode->i_ino, (unsigned long) ce->e_block);
930                 } else {
931                         lock_buffer(bh);
932                         if (le32_to_cpu(HDR(bh)->h_refcount) >
933                                    EXT2_XATTR_REFCOUNT_MAX) {
934                                 ea_idebug(inode, "block %ld refcount %d>%d",
935                                           (unsigned long) ce->e_block,
936                                           le32_to_cpu(HDR(bh)->h_refcount),
937                                           EXT2_XATTR_REFCOUNT_MAX);
938                         } else if (!ext2_xattr_cmp(header, HDR(bh))) {
939                                 ea_bdebug(bh, "b_count=%d",
940                                           atomic_read(&(bh->b_count)));
941                                 mb_cache_entry_release(ce);
942                                 return bh;
943                         }
944                         unlock_buffer(bh);
945                         brelse(bh);
946                 }
947                 ce = mb_cache_entry_find_next(ce, inode->i_sb->s_bdev, hash);
948         }
949         return NULL;
950 }
951
952 #define NAME_HASH_SHIFT 5
953 #define VALUE_HASH_SHIFT 16
954
955 /*
956  * ext2_xattr_hash_entry()
957  *
958  * Compute the hash of an extended attribute.
959  */
960 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
961                                          struct ext2_xattr_entry *entry)
962 {
963         __u32 hash = 0;
964         char *name = entry->e_name;
965         int n;
966
967         for (n=0; n < entry->e_name_len; n++) {
968                 hash = (hash << NAME_HASH_SHIFT) ^
969                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
970                        *name++;
971         }
972
973         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
974                 __le32 *value = (__le32 *)((char *)header +
975                         le16_to_cpu(entry->e_value_offs));
976                 for (n = (le32_to_cpu(entry->e_value_size) +
977                      EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
978                         hash = (hash << VALUE_HASH_SHIFT) ^
979                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
980                                le32_to_cpu(*value++);
981                 }
982         }
983         entry->e_hash = cpu_to_le32(hash);
984 }
985
986 #undef NAME_HASH_SHIFT
987 #undef VALUE_HASH_SHIFT
988
989 #define BLOCK_HASH_SHIFT 16
990
991 /*
992  * ext2_xattr_rehash()
993  *
994  * Re-compute the extended attribute hash value after an entry has changed.
995  */
996 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
997                               struct ext2_xattr_entry *entry)
998 {
999         struct ext2_xattr_entry *here;
1000         __u32 hash = 0;
1001         
1002         ext2_xattr_hash_entry(header, entry);
1003         here = ENTRY(header+1);
1004         while (!IS_LAST_ENTRY(here)) {
1005                 if (!here->e_hash) {
1006                         /* Block is not shared if an entry's hash value == 0 */
1007                         hash = 0;
1008                         break;
1009                 }
1010                 hash = (hash << BLOCK_HASH_SHIFT) ^
1011                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1012                        le32_to_cpu(here->e_hash);
1013                 here = EXT2_XATTR_NEXT(here);
1014         }
1015         header->h_hash = cpu_to_le32(hash);
1016 }
1017
1018 #undef BLOCK_HASH_SHIFT
1019
1020 int __init
1021 init_ext2_xattr(void)
1022 {
1023         ext2_xattr_cache = mb_cache_create("ext2_xattr", 6);
1024         if (!ext2_xattr_cache)
1025                 return -ENOMEM;
1026         return 0;
1027 }
1028
1029 void
1030 exit_ext2_xattr(void)
1031 {
1032         mb_cache_destroy(ext2_xattr_cache);
1033 }