GNU Linux-libre 4.19.211-gnu1
[releases.git] / fs / ext2 / xattr.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/ext2/xattr.c
4  *
5  * Copyright (C) 2001-2003 Andreas Gruenbacher <agruen@suse.de>
6  *
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.
12  *
13  */
14
15 /*
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.
22  *
23  *
24  * Extended attribute block layout:
25  *
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  *   +------------------+
37  *
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.
42  *
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.
46  *
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.
55  */
56
57 #include <linux/buffer_head.h>
58 #include <linux/init.h>
59 #include <linux/printk.h>
60 #include <linux/slab.h>
61 #include <linux/mbcache.h>
62 #include <linux/quotaops.h>
63 #include <linux/rwsem.h>
64 #include <linux/security.h>
65 #include "ext2.h"
66 #include "xattr.h"
67 #include "acl.h"
68
69 #define HDR(bh) ((struct ext2_xattr_header *)((bh)->b_data))
70 #define ENTRY(ptr) ((struct ext2_xattr_entry *)(ptr))
71 #define FIRST_ENTRY(bh) ENTRY(HDR(bh)+1)
72 #define IS_LAST_ENTRY(entry) (*(__u32 *)(entry) == 0)
73
74 #ifdef EXT2_XATTR_DEBUG
75 # define ea_idebug(inode, f...) do { \
76                 printk(KERN_DEBUG "inode %s:%ld: ", \
77                         inode->i_sb->s_id, inode->i_ino); \
78                 printk(f); \
79                 printk("\n"); \
80         } while (0)
81 # define ea_bdebug(bh, f...) do { \
82                 printk(KERN_DEBUG "block %pg:%lu: ", \
83                         bh->b_bdev, (unsigned long) bh->b_blocknr); \
84                 printk(f); \
85                 printk("\n"); \
86         } while (0)
87 #else
88 # define ea_idebug(inode, f...) no_printk(f)
89 # define ea_bdebug(bh, f...)    no_printk(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 mb_cache *, 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 const struct xattr_handler *ext2_xattr_handler_map[] = {
102         [EXT2_XATTR_INDEX_USER]              = &ext2_xattr_user_handler,
103 #ifdef CONFIG_EXT2_FS_POSIX_ACL
104         [EXT2_XATTR_INDEX_POSIX_ACL_ACCESS]  = &posix_acl_access_xattr_handler,
105         [EXT2_XATTR_INDEX_POSIX_ACL_DEFAULT] = &posix_acl_default_xattr_handler,
106 #endif
107         [EXT2_XATTR_INDEX_TRUSTED]           = &ext2_xattr_trusted_handler,
108 #ifdef CONFIG_EXT2_FS_SECURITY
109         [EXT2_XATTR_INDEX_SECURITY]          = &ext2_xattr_security_handler,
110 #endif
111 };
112
113 const struct xattr_handler *ext2_xattr_handlers[] = {
114         &ext2_xattr_user_handler,
115         &ext2_xattr_trusted_handler,
116 #ifdef CONFIG_EXT2_FS_POSIX_ACL
117         &posix_acl_access_xattr_handler,
118         &posix_acl_default_xattr_handler,
119 #endif
120 #ifdef CONFIG_EXT2_FS_SECURITY
121         &ext2_xattr_security_handler,
122 #endif
123         NULL
124 };
125
126 #define EA_BLOCK_CACHE(inode)   (EXT2_SB(inode->i_sb)->s_ea_block_cache)
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         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
158
159         ea_idebug(inode, "name=%d.%s, buffer=%p, buffer_size=%ld",
160                   name_index, name, buffer, (long)buffer_size);
161
162         if (name == NULL)
163                 return -EINVAL;
164         name_len = strlen(name);
165         if (name_len > 255)
166                 return -ERANGE;
167
168         down_read(&EXT2_I(inode)->xattr_sem);
169         error = -ENODATA;
170         if (!EXT2_I(inode)->i_file_acl)
171                 goto cleanup;
172         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
173         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
174         error = -EIO;
175         if (!bh)
176                 goto cleanup;
177         ea_bdebug(bh, "b_count=%d, refcount=%d",
178                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
179         end = bh->b_data + bh->b_size;
180         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
181             HDR(bh)->h_blocks != cpu_to_le32(1)) {
182 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_get",
183                         "inode %ld: bad block %d", inode->i_ino,
184                         EXT2_I(inode)->i_file_acl);
185                 error = -EIO;
186                 goto cleanup;
187         }
188
189         /* find named attribute */
190         entry = FIRST_ENTRY(bh);
191         while (!IS_LAST_ENTRY(entry)) {
192                 struct ext2_xattr_entry *next =
193                         EXT2_XATTR_NEXT(entry);
194                 if ((char *)next >= end)
195                         goto bad_block;
196                 if (name_index == entry->e_name_index &&
197                     name_len == entry->e_name_len &&
198                     memcmp(name, entry->e_name, name_len) == 0)
199                         goto found;
200                 entry = next;
201         }
202         if (ext2_xattr_cache_insert(ea_block_cache, bh))
203                 ea_idebug(inode, "cache insert failed");
204         error = -ENODATA;
205         goto cleanup;
206 found:
207         /* check the buffer size */
208         if (entry->e_value_block != 0)
209                 goto bad_block;
210         size = le32_to_cpu(entry->e_value_size);
211         if (size > inode->i_sb->s_blocksize ||
212             le16_to_cpu(entry->e_value_offs) + size > inode->i_sb->s_blocksize)
213                 goto bad_block;
214
215         if (ext2_xattr_cache_insert(ea_block_cache, bh))
216                 ea_idebug(inode, "cache insert failed");
217         if (buffer) {
218                 error = -ERANGE;
219                 if (size > buffer_size)
220                         goto cleanup;
221                 /* return value of attribute */
222                 memcpy(buffer, bh->b_data + le16_to_cpu(entry->e_value_offs),
223                         size);
224         }
225         error = size;
226
227 cleanup:
228         brelse(bh);
229         up_read(&EXT2_I(inode)->xattr_sem);
230
231         return error;
232 }
233
234 /*
235  * ext2_xattr_list()
236  *
237  * Copy a list of attribute names into the buffer
238  * provided, or compute the buffer size required.
239  * Buffer is NULL to compute the size of the buffer required.
240  *
241  * Returns a negative error number on failure, or the number of bytes
242  * used / required on success.
243  */
244 static int
245 ext2_xattr_list(struct dentry *dentry, char *buffer, size_t buffer_size)
246 {
247         struct inode *inode = d_inode(dentry);
248         struct buffer_head *bh = NULL;
249         struct ext2_xattr_entry *entry;
250         char *end;
251         size_t rest = buffer_size;
252         int error;
253         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
254
255         ea_idebug(inode, "buffer=%p, buffer_size=%ld",
256                   buffer, (long)buffer_size);
257
258         down_read(&EXT2_I(inode)->xattr_sem);
259         error = 0;
260         if (!EXT2_I(inode)->i_file_acl)
261                 goto cleanup;
262         ea_idebug(inode, "reading block %d", EXT2_I(inode)->i_file_acl);
263         bh = sb_bread(inode->i_sb, EXT2_I(inode)->i_file_acl);
264         error = -EIO;
265         if (!bh)
266                 goto cleanup;
267         ea_bdebug(bh, "b_count=%d, refcount=%d",
268                 atomic_read(&(bh->b_count)), le32_to_cpu(HDR(bh)->h_refcount));
269         end = bh->b_data + bh->b_size;
270         if (HDR(bh)->h_magic != cpu_to_le32(EXT2_XATTR_MAGIC) ||
271             HDR(bh)->h_blocks != cpu_to_le32(1)) {
272 bad_block:      ext2_error(inode->i_sb, "ext2_xattr_list",
273                         "inode %ld: bad block %d", inode->i_ino,
274                         EXT2_I(inode)->i_file_acl);
275                 error = -EIO;
276                 goto cleanup;
277         }
278
279         /* check the on-disk data structure */
280         entry = FIRST_ENTRY(bh);
281         while (!IS_LAST_ENTRY(entry)) {
282                 struct ext2_xattr_entry *next = EXT2_XATTR_NEXT(entry);
283
284                 if ((char *)next >= end)
285                         goto bad_block;
286                 entry = next;
287         }
288         if (ext2_xattr_cache_insert(ea_block_cache, bh))
289                 ea_idebug(inode, "cache insert failed");
290
291         /* list the attribute names */
292         for (entry = FIRST_ENTRY(bh); !IS_LAST_ENTRY(entry);
293              entry = EXT2_XATTR_NEXT(entry)) {
294                 const struct xattr_handler *handler =
295                         ext2_xattr_handler(entry->e_name_index);
296
297                 if (handler && (!handler->list || handler->list(dentry))) {
298                         const char *prefix = handler->prefix ?: handler->name;
299                         size_t prefix_len = strlen(prefix);
300                         size_t size = prefix_len + entry->e_name_len + 1;
301
302                         if (buffer) {
303                                 if (size > rest) {
304                                         error = -ERANGE;
305                                         goto cleanup;
306                                 }
307                                 memcpy(buffer, prefix, prefix_len);
308                                 buffer += prefix_len;
309                                 memcpy(buffer, entry->e_name, entry->e_name_len);
310                                 buffer += entry->e_name_len;
311                                 *buffer++ = 0;
312                         }
313                         rest -= size;
314                 }
315         }
316         error = buffer_size - rest;  /* total size */
317
318 cleanup:
319         brelse(bh);
320         up_read(&EXT2_I(inode)->xattr_sem);
321
322         return error;
323 }
324
325 /*
326  * Inode operation listxattr()
327  *
328  * d_inode(dentry)->i_mutex: don't care
329  */
330 ssize_t
331 ext2_listxattr(struct dentry *dentry, char *buffer, size_t size)
332 {
333         return ext2_xattr_list(dentry, 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  */
340 static void ext2_xattr_update_super_block(struct super_block *sb)
341 {
342         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT2_FEATURE_COMPAT_EXT_ATTR))
343                 return;
344
345         spin_lock(&EXT2_SB(sb)->s_lock);
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);
349 }
350
351 /*
352  * ext2_xattr_set()
353  *
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.
360  *
361  * Returns 0, or a negative error number on failure.
362  */
363 int
364 ext2_xattr_set(struct inode *inode, int name_index, const char *name,
365                const void *value, size_t value_len, int flags)
366 {
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;
374         
375         /*
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.
385          */
386         
387         ea_idebug(inode, "name=%d.%s, value=%p, value_len=%ld",
388                   name_index, name, value, (long)value_len);
389
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;
416                 }
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;
427                         }
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;
436                 }
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;
447                         }
448                         last = next;
449                 }
450
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. */
458         }
459
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);
475
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);
480                 }
481                 free += EXT2_XATTR_LEN(name_len);
482         }
483         error = -ENOSPC;
484         if (free < EXT2_XATTR_LEN(name_len) + EXT2_XATTR_SIZE(value_len))
485                 goto cleanup;
486
487         /* Here we know that we can set the new attribute. */
488
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);
494
495                         ea_bdebug(bh, "modifying in-place");
496                         /*
497                          * This must happen under buffer lock for
498                          * ext2_xattr_set2() to reliably detect modified block
499                          */
500                         mb_cache_entry_delete(EA_BLOCK_CACHE(inode), hash,
501                                               bh->b_blocknr);
502
503                         /* keep the buffer locked while modifying it. */
504                 } else {
505                         int offset;
506
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);
515
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);
520                 }
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);
531         }
532
533         /* Iff we are modifying the block in-place, bh is locked here. */
534
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));
551
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;
560                         }
561
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;
567
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);
576                         }
577                 }
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);
585                 }
586         }
587
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);
599                 }
600         }
601
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);
613         }
614
615 cleanup:
616         if (!(bh && header == HDR(bh)))
617                 kfree(header);
618         brelse(bh);
619         up_write(&EXT2_I(inode)->xattr_sem);
620
621         return error;
622 }
623
624 /*
625  * Second half of ext2_xattr_set(): Update the file system.
626  */
627 static int
628 ext2_xattr_set2(struct inode *inode, struct buffer_head *old_bh,
629                 struct ext2_xattr_header *header)
630 {
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);
635
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");
646
647                                 error = dquot_alloc_block(inode, 1);
648                                 if (error) {
649                                         unlock_buffer(new_bh);
650                                         goto cleanup;
651                                 }
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));
655                         }
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);
671
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;
678                         }
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);
684                         
685                         ext2_xattr_update_super_block(sb);
686                 }
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;
693                 }
694         }
695
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);
708                         }
709                         goto cleanup;
710                 }
711         } else
712                 mark_inode_dirty(inode);
713
714         error = 0;
715         if (old_bh && old_bh != new_bh) {
716                 /*
717                  * If there was an old block and we are no longer using it,
718                  * release the old block.
719                  */
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);
723
724                         /*
725                          * This must happen under buffer lock for
726                          * ext2_xattr_set2() to reliably detect freed block
727                          */
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));
746                 }
747                 unlock_buffer(old_bh);
748         }
749
750 cleanup:
751         brelse(new_bh);
752
753         return error;
754 }
755
756 /*
757  * ext2_xattr_delete_inode()
758  *
759  * Free extended attribute resources associated with this inode. This
760  * is called immediately before an inode is freed.
761  */
762 void
763 ext2_xattr_delete_inode(struct inode *inode)
764 {
765         struct buffer_head *bh = NULL;
766         struct ext2_sb_info *sbi = EXT2_SB(inode->i_sb);
767
768         down_write(&EXT2_I(inode)->xattr_sem);
769         if (!EXT2_I(inode)->i_file_acl)
770                 goto cleanup;
771
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;
777         }
778
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;
785         }
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;
793         }
794         lock_buffer(bh);
795         if (HDR(bh)->h_refcount == cpu_to_le32(1)) {
796                 __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
797
798                 /*
799                  * This must happen under buffer lock for ext2_xattr_set2() to
800                  * reliably detect freed block
801                  */
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);
817         }
818         EXT2_I(inode)->i_file_acl = 0;
819
820 cleanup:
821         brelse(bh);
822         up_write(&EXT2_I(inode)->xattr_sem);
823 }
824
825 /*
826  * ext2_xattr_cache_insert()
827  *
828  * Create a new entry in the extended attribute cache, and insert
829  * it unless such an entry is already in the cache.
830  *
831  * Returns 0, or a negative error number on failure.
832  */
833 static int
834 ext2_xattr_cache_insert(struct mb_cache *cache, struct buffer_head *bh)
835 {
836         __u32 hash = le32_to_cpu(HDR(bh)->h_hash);
837         int error;
838
839         error = mb_cache_entry_create(cache, GFP_NOFS, hash, bh->b_blocknr, 1);
840         if (error) {
841                 if (error == -EBUSY) {
842                         ea_bdebug(bh, "already in cache");
843                         error = 0;
844                 }
845         } else
846                 ea_bdebug(bh, "inserting [%x]", (int)hash);
847         return error;
848 }
849
850 /*
851  * ext2_xattr_cmp()
852  *
853  * Compare two extended attribute blocks for equality.
854  *
855  * Returns 0 if the blocks are equal, 1 if they differ, and
856  * a negative error number on errors.
857  */
858 static int
859 ext2_xattr_cmp(struct ext2_xattr_header *header1,
860                struct ext2_xattr_header *header2)
861 {
862         struct ext2_xattr_entry *entry1, *entry2;
863
864         entry1 = ENTRY(header1+1);
865         entry2 = ENTRY(header2+1);
866         while (!IS_LAST_ENTRY(entry1)) {
867                 if (IS_LAST_ENTRY(entry2))
868                         return 1;
869                 if (entry1->e_hash != entry2->e_hash ||
870                     entry1->e_name_index != entry2->e_name_index ||
871                     entry1->e_name_len != entry2->e_name_len ||
872                     entry1->e_value_size != entry2->e_value_size ||
873                     memcmp(entry1->e_name, entry2->e_name, entry1->e_name_len))
874                         return 1;
875                 if (entry1->e_value_block != 0 || entry2->e_value_block != 0)
876                         return -EIO;
877                 if (memcmp((char *)header1 + le16_to_cpu(entry1->e_value_offs),
878                            (char *)header2 + le16_to_cpu(entry2->e_value_offs),
879                            le32_to_cpu(entry1->e_value_size)))
880                         return 1;
881
882                 entry1 = EXT2_XATTR_NEXT(entry1);
883                 entry2 = EXT2_XATTR_NEXT(entry2);
884         }
885         if (!IS_LAST_ENTRY(entry2))
886                 return 1;
887         return 0;
888 }
889
890 /*
891  * ext2_xattr_cache_find()
892  *
893  * Find an identical extended attribute block.
894  *
895  * Returns a locked buffer head to the block found, or NULL if such
896  * a block was not found or an error occurred.
897  */
898 static struct buffer_head *
899 ext2_xattr_cache_find(struct inode *inode, struct ext2_xattr_header *header)
900 {
901         __u32 hash = le32_to_cpu(header->h_hash);
902         struct mb_cache_entry *ce;
903         struct mb_cache *ea_block_cache = EA_BLOCK_CACHE(inode);
904
905         if (!header->h_hash)
906                 return NULL;  /* never share */
907         ea_idebug(inode, "looking for cached blocks [%x]", (int)hash);
908 again:
909         ce = mb_cache_entry_find_first(ea_block_cache, hash);
910         while (ce) {
911                 struct buffer_head *bh;
912
913                 bh = sb_bread(inode->i_sb, ce->e_value);
914                 if (!bh) {
915                         ext2_error(inode->i_sb, "ext2_xattr_cache_find",
916                                 "inode %ld: block %ld read error",
917                                 inode->i_ino, (unsigned long) ce->e_value);
918                 } else {
919                         lock_buffer(bh);
920                         /*
921                          * We have to be careful about races with freeing or
922                          * rehashing of xattr block. Once we hold buffer lock
923                          * xattr block's state is stable so we can check
924                          * whether the block got freed / rehashed or not.
925                          * Since we unhash mbcache entry under buffer lock when
926                          * freeing / rehashing xattr block, checking whether
927                          * entry is still hashed is reliable.
928                          */
929                         if (hlist_bl_unhashed(&ce->e_hash_list)) {
930                                 mb_cache_entry_put(ea_block_cache, ce);
931                                 unlock_buffer(bh);
932                                 brelse(bh);
933                                 goto again;
934                         } else if (le32_to_cpu(HDR(bh)->h_refcount) >
935                                    EXT2_XATTR_REFCOUNT_MAX) {
936                                 ea_idebug(inode, "block %ld refcount %d>%d",
937                                           (unsigned long) ce->e_value,
938                                           le32_to_cpu(HDR(bh)->h_refcount),
939                                           EXT2_XATTR_REFCOUNT_MAX);
940                         } else if (!ext2_xattr_cmp(header, HDR(bh))) {
941                                 ea_bdebug(bh, "b_count=%d",
942                                           atomic_read(&(bh->b_count)));
943                                 mb_cache_entry_touch(ea_block_cache, ce);
944                                 mb_cache_entry_put(ea_block_cache, ce);
945                                 return bh;
946                         }
947                         unlock_buffer(bh);
948                         brelse(bh);
949                 }
950                 ce = mb_cache_entry_find_next(ea_block_cache, ce);
951         }
952         return NULL;
953 }
954
955 #define NAME_HASH_SHIFT 5
956 #define VALUE_HASH_SHIFT 16
957
958 /*
959  * ext2_xattr_hash_entry()
960  *
961  * Compute the hash of an extended attribute.
962  */
963 static inline void ext2_xattr_hash_entry(struct ext2_xattr_header *header,
964                                          struct ext2_xattr_entry *entry)
965 {
966         __u32 hash = 0;
967         char *name = entry->e_name;
968         int n;
969
970         for (n=0; n < entry->e_name_len; n++) {
971                 hash = (hash << NAME_HASH_SHIFT) ^
972                        (hash >> (8*sizeof(hash) - NAME_HASH_SHIFT)) ^
973                        *name++;
974         }
975
976         if (entry->e_value_block == 0 && entry->e_value_size != 0) {
977                 __le32 *value = (__le32 *)((char *)header +
978                         le16_to_cpu(entry->e_value_offs));
979                 for (n = (le32_to_cpu(entry->e_value_size) +
980                      EXT2_XATTR_ROUND) >> EXT2_XATTR_PAD_BITS; n; n--) {
981                         hash = (hash << VALUE_HASH_SHIFT) ^
982                                (hash >> (8*sizeof(hash) - VALUE_HASH_SHIFT)) ^
983                                le32_to_cpu(*value++);
984                 }
985         }
986         entry->e_hash = cpu_to_le32(hash);
987 }
988
989 #undef NAME_HASH_SHIFT
990 #undef VALUE_HASH_SHIFT
991
992 #define BLOCK_HASH_SHIFT 16
993
994 /*
995  * ext2_xattr_rehash()
996  *
997  * Re-compute the extended attribute hash value after an entry has changed.
998  */
999 static void ext2_xattr_rehash(struct ext2_xattr_header *header,
1000                               struct ext2_xattr_entry *entry)
1001 {
1002         struct ext2_xattr_entry *here;
1003         __u32 hash = 0;
1004         
1005         ext2_xattr_hash_entry(header, entry);
1006         here = ENTRY(header+1);
1007         while (!IS_LAST_ENTRY(here)) {
1008                 if (!here->e_hash) {
1009                         /* Block is not shared if an entry's hash value == 0 */
1010                         hash = 0;
1011                         break;
1012                 }
1013                 hash = (hash << BLOCK_HASH_SHIFT) ^
1014                        (hash >> (8*sizeof(hash) - BLOCK_HASH_SHIFT)) ^
1015                        le32_to_cpu(here->e_hash);
1016                 here = EXT2_XATTR_NEXT(here);
1017         }
1018         header->h_hash = cpu_to_le32(hash);
1019 }
1020
1021 #undef BLOCK_HASH_SHIFT
1022
1023 #define HASH_BUCKET_BITS 10
1024
1025 struct mb_cache *ext2_xattr_create_cache(void)
1026 {
1027         return mb_cache_create(HASH_BUCKET_BITS);
1028 }
1029
1030 void ext2_xattr_destroy_cache(struct mb_cache *cache)
1031 {
1032         if (cache)
1033                 mb_cache_destroy(cache);
1034 }