xref: /openbmc/linux/fs/reiserfs/xattr.c (revision a09d2831)
1 /*
2  * linux/fs/reiserfs/xattr.c
3  *
4  * Copyright (c) 2002 by Jeff Mahoney, <jeffm@suse.com>
5  *
6  */
7 
8 /*
9  * In order to implement EA/ACLs in a clean, backwards compatible manner,
10  * they are implemented as files in a "private" directory.
11  * Each EA is in it's own file, with the directory layout like so (/ is assumed
12  * to be relative to fs root). Inside the /.reiserfs_priv/xattrs directory,
13  * directories named using the capital-hex form of the objectid and
14  * generation number are used. Inside each directory are individual files
15  * named with the name of the extended attribute.
16  *
17  * So, for objectid 12648430, we could have:
18  * /.reiserfs_priv/xattrs/C0FFEE.0/system.posix_acl_access
19  * /.reiserfs_priv/xattrs/C0FFEE.0/system.posix_acl_default
20  * /.reiserfs_priv/xattrs/C0FFEE.0/user.Content-Type
21  * .. or similar.
22  *
23  * The file contents are the text of the EA. The size is known based on the
24  * stat data describing the file.
25  *
26  * In the case of system.posix_acl_access and system.posix_acl_default, since
27  * these are special cases for filesystem ACLs, they are interpreted by the
28  * kernel, in addition, they are negatively and positively cached and attached
29  * to the inode so that unnecessary lookups are avoided.
30  *
31  * Locking works like so:
32  * Directory components (xattr root, xattr dir) are protectd by their i_mutex.
33  * The xattrs themselves are protected by the xattr_sem.
34  */
35 
36 #include <linux/reiserfs_fs.h>
37 #include <linux/capability.h>
38 #include <linux/dcache.h>
39 #include <linux/namei.h>
40 #include <linux/errno.h>
41 #include <linux/fs.h>
42 #include <linux/file.h>
43 #include <linux/pagemap.h>
44 #include <linux/xattr.h>
45 #include <linux/reiserfs_xattr.h>
46 #include <linux/reiserfs_acl.h>
47 #include <asm/uaccess.h>
48 #include <net/checksum.h>
49 #include <linux/stat.h>
50 #include <linux/quotaops.h>
51 #include <linux/security.h>
52 
53 #define PRIVROOT_NAME ".reiserfs_priv"
54 #define XAROOT_NAME   "xattrs"
55 
56 
57 /* Helpers for inode ops. We do this so that we don't have all the VFS
58  * overhead and also for proper i_mutex annotation.
59  * dir->i_mutex must be held for all of them. */
60 #ifdef CONFIG_REISERFS_FS_XATTR
61 static int xattr_create(struct inode *dir, struct dentry *dentry, int mode)
62 {
63 	BUG_ON(!mutex_is_locked(&dir->i_mutex));
64 	vfs_dq_init(dir);
65 	return dir->i_op->create(dir, dentry, mode, NULL);
66 }
67 #endif
68 
69 static int xattr_mkdir(struct inode *dir, struct dentry *dentry, int mode)
70 {
71 	BUG_ON(!mutex_is_locked(&dir->i_mutex));
72 	vfs_dq_init(dir);
73 	return dir->i_op->mkdir(dir, dentry, mode);
74 }
75 
76 /* We use I_MUTEX_CHILD here to silence lockdep. It's safe because xattr
77  * mutation ops aren't called during rename or splace, which are the
78  * only other users of I_MUTEX_CHILD. It violates the ordering, but that's
79  * better than allocating another subclass just for this code. */
80 static int xattr_unlink(struct inode *dir, struct dentry *dentry)
81 {
82 	int error;
83 	BUG_ON(!mutex_is_locked(&dir->i_mutex));
84 	vfs_dq_init(dir);
85 
86 	reiserfs_mutex_lock_nested_safe(&dentry->d_inode->i_mutex,
87 					I_MUTEX_CHILD, dir->i_sb);
88 	error = dir->i_op->unlink(dir, dentry);
89 	mutex_unlock(&dentry->d_inode->i_mutex);
90 
91 	if (!error)
92 		d_delete(dentry);
93 	return error;
94 }
95 
96 static int xattr_rmdir(struct inode *dir, struct dentry *dentry)
97 {
98 	int error;
99 	BUG_ON(!mutex_is_locked(&dir->i_mutex));
100 	vfs_dq_init(dir);
101 
102 	reiserfs_mutex_lock_nested_safe(&dentry->d_inode->i_mutex,
103 					I_MUTEX_CHILD, dir->i_sb);
104 	dentry_unhash(dentry);
105 	error = dir->i_op->rmdir(dir, dentry);
106 	if (!error)
107 		dentry->d_inode->i_flags |= S_DEAD;
108 	mutex_unlock(&dentry->d_inode->i_mutex);
109 	if (!error)
110 		d_delete(dentry);
111 	dput(dentry);
112 
113 	return error;
114 }
115 
116 #define xattr_may_create(flags)	(!flags || flags & XATTR_CREATE)
117 
118 static struct dentry *open_xa_root(struct super_block *sb, int flags)
119 {
120 	struct dentry *privroot = REISERFS_SB(sb)->priv_root;
121 	struct dentry *xaroot;
122 	if (!privroot->d_inode)
123 		return ERR_PTR(-ENODATA);
124 
125 	mutex_lock_nested(&privroot->d_inode->i_mutex, I_MUTEX_XATTR);
126 
127 	xaroot = dget(REISERFS_SB(sb)->xattr_root);
128 	if (!xaroot)
129 		xaroot = ERR_PTR(-ENODATA);
130 	else if (!xaroot->d_inode) {
131 		int err = -ENODATA;
132 		if (xattr_may_create(flags))
133 			err = xattr_mkdir(privroot->d_inode, xaroot, 0700);
134 		if (err) {
135 			dput(xaroot);
136 			xaroot = ERR_PTR(err);
137 		}
138 	}
139 
140 	mutex_unlock(&privroot->d_inode->i_mutex);
141 	return xaroot;
142 }
143 
144 static struct dentry *open_xa_dir(const struct inode *inode, int flags)
145 {
146 	struct dentry *xaroot, *xadir;
147 	char namebuf[17];
148 
149 	xaroot = open_xa_root(inode->i_sb, flags);
150 	if (IS_ERR(xaroot))
151 		return xaroot;
152 
153 	snprintf(namebuf, sizeof(namebuf), "%X.%X",
154 		 le32_to_cpu(INODE_PKEY(inode)->k_objectid),
155 		 inode->i_generation);
156 
157 	mutex_lock_nested(&xaroot->d_inode->i_mutex, I_MUTEX_XATTR);
158 
159 	xadir = lookup_one_len(namebuf, xaroot, strlen(namebuf));
160 	if (!IS_ERR(xadir) && !xadir->d_inode) {
161 		int err = -ENODATA;
162 		if (xattr_may_create(flags))
163 			err = xattr_mkdir(xaroot->d_inode, xadir, 0700);
164 		if (err) {
165 			dput(xadir);
166 			xadir = ERR_PTR(err);
167 		}
168 	}
169 
170 	mutex_unlock(&xaroot->d_inode->i_mutex);
171 	dput(xaroot);
172 	return xadir;
173 }
174 
175 /* The following are side effects of other operations that aren't explicitly
176  * modifying extended attributes. This includes operations such as permissions
177  * or ownership changes, object deletions, etc. */
178 struct reiserfs_dentry_buf {
179 	struct dentry *xadir;
180 	int count;
181 	struct dentry *dentries[8];
182 };
183 
184 static int
185 fill_with_dentries(void *buf, const char *name, int namelen, loff_t offset,
186 		    u64 ino, unsigned int d_type)
187 {
188 	struct reiserfs_dentry_buf *dbuf = buf;
189 	struct dentry *dentry;
190 	WARN_ON_ONCE(!mutex_is_locked(&dbuf->xadir->d_inode->i_mutex));
191 
192 	if (dbuf->count == ARRAY_SIZE(dbuf->dentries))
193 		return -ENOSPC;
194 
195 	if (name[0] == '.' && (name[1] == '\0' ||
196 			       (name[1] == '.' && name[2] == '\0')))
197 		return 0;
198 
199 	dentry = lookup_one_len(name, dbuf->xadir, namelen);
200 	if (IS_ERR(dentry)) {
201 		return PTR_ERR(dentry);
202 	} else if (!dentry->d_inode) {
203 		/* A directory entry exists, but no file? */
204 		reiserfs_error(dentry->d_sb, "xattr-20003",
205 			       "Corrupted directory: xattr %s listed but "
206 			       "not found for file %s.\n",
207 			       dentry->d_name.name, dbuf->xadir->d_name.name);
208 		dput(dentry);
209 		return -EIO;
210 	}
211 
212 	dbuf->dentries[dbuf->count++] = dentry;
213 	return 0;
214 }
215 
216 static void
217 cleanup_dentry_buf(struct reiserfs_dentry_buf *buf)
218 {
219 	int i;
220 	for (i = 0; i < buf->count; i++)
221 		if (buf->dentries[i])
222 			dput(buf->dentries[i]);
223 }
224 
225 static int reiserfs_for_each_xattr(struct inode *inode,
226 				   int (*action)(struct dentry *, void *),
227 				   void *data)
228 {
229 	struct dentry *dir;
230 	int i, err = 0;
231 	loff_t pos = 0;
232 	struct reiserfs_dentry_buf buf = {
233 		.count = 0,
234 	};
235 
236 	/* Skip out, an xattr has no xattrs associated with it */
237 	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
238 		return 0;
239 
240 	reiserfs_write_unlock(inode->i_sb);
241 	dir = open_xa_dir(inode, XATTR_REPLACE);
242 	if (IS_ERR(dir)) {
243 		err = PTR_ERR(dir);
244 		reiserfs_write_lock(inode->i_sb);
245 		goto out;
246 	} else if (!dir->d_inode) {
247 		err = 0;
248 		reiserfs_write_lock(inode->i_sb);
249 		goto out_dir;
250 	}
251 
252 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR);
253 
254 	reiserfs_write_lock(inode->i_sb);
255 
256 	buf.xadir = dir;
257 	err = reiserfs_readdir_dentry(dir, &buf, fill_with_dentries, &pos);
258 	while ((err == 0 || err == -ENOSPC) && buf.count) {
259 		err = 0;
260 
261 		for (i = 0; i < buf.count && buf.dentries[i]; i++) {
262 			int lerr = 0;
263 			struct dentry *dentry = buf.dentries[i];
264 
265 			if (err == 0 && !S_ISDIR(dentry->d_inode->i_mode))
266 				lerr = action(dentry, data);
267 
268 			dput(dentry);
269 			buf.dentries[i] = NULL;
270 			err = lerr ?: err;
271 		}
272 		buf.count = 0;
273 		if (!err)
274 			err = reiserfs_readdir_dentry(dir, &buf,
275 						      fill_with_dentries, &pos);
276 	}
277 	mutex_unlock(&dir->d_inode->i_mutex);
278 
279 	/* Clean up after a failed readdir */
280 	cleanup_dentry_buf(&buf);
281 
282 	if (!err) {
283 		/* We start a transaction here to avoid a ABBA situation
284 		 * between the xattr root's i_mutex and the journal lock.
285 		 * This doesn't incur much additional overhead since the
286 		 * new transaction will just nest inside the
287 		 * outer transaction. */
288 		int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 +
289 			     4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
290 		struct reiserfs_transaction_handle th;
291 		err = journal_begin(&th, inode->i_sb, blocks);
292 		if (!err) {
293 			int jerror;
294 			reiserfs_mutex_lock_nested_safe(
295 					  &dir->d_parent->d_inode->i_mutex,
296 					  I_MUTEX_XATTR, inode->i_sb);
297 			err = action(dir, data);
298 			jerror = journal_end(&th, inode->i_sb, blocks);
299 			mutex_unlock(&dir->d_parent->d_inode->i_mutex);
300 			err = jerror ?: err;
301 		}
302 	}
303 out_dir:
304 	dput(dir);
305 out:
306 	/* -ENODATA isn't an error */
307 	if (err == -ENODATA)
308 		err = 0;
309 	return err;
310 }
311 
312 static int delete_one_xattr(struct dentry *dentry, void *data)
313 {
314 	struct inode *dir = dentry->d_parent->d_inode;
315 
316 	/* This is the xattr dir, handle specially. */
317 	if (S_ISDIR(dentry->d_inode->i_mode))
318 		return xattr_rmdir(dir, dentry);
319 
320 	return xattr_unlink(dir, dentry);
321 }
322 
323 static int chown_one_xattr(struct dentry *dentry, void *data)
324 {
325 	struct iattr *attrs = data;
326 	return reiserfs_setattr(dentry, attrs);
327 }
328 
329 /* No i_mutex, but the inode is unconnected. */
330 int reiserfs_delete_xattrs(struct inode *inode)
331 {
332 	int err = reiserfs_for_each_xattr(inode, delete_one_xattr, NULL);
333 	if (err)
334 		reiserfs_warning(inode->i_sb, "jdm-20004",
335 				 "Couldn't delete all xattrs (%d)\n", err);
336 	return err;
337 }
338 
339 /* inode->i_mutex: down */
340 int reiserfs_chown_xattrs(struct inode *inode, struct iattr *attrs)
341 {
342 	int err = reiserfs_for_each_xattr(inode, chown_one_xattr, attrs);
343 	if (err)
344 		reiserfs_warning(inode->i_sb, "jdm-20007",
345 				 "Couldn't chown all xattrs (%d)\n", err);
346 	return err;
347 }
348 
349 #ifdef CONFIG_REISERFS_FS_XATTR
350 /* Returns a dentry corresponding to a specific extended attribute file
351  * for the inode. If flags allow, the file is created. Otherwise, a
352  * valid or negative dentry, or an error is returned. */
353 static struct dentry *xattr_lookup(struct inode *inode, const char *name,
354 				    int flags)
355 {
356 	struct dentry *xadir, *xafile;
357 	int err = 0;
358 
359 	xadir = open_xa_dir(inode, flags);
360 	if (IS_ERR(xadir))
361 		return ERR_CAST(xadir);
362 
363 	mutex_lock_nested(&xadir->d_inode->i_mutex, I_MUTEX_XATTR);
364 	xafile = lookup_one_len(name, xadir, strlen(name));
365 	if (IS_ERR(xafile)) {
366 		err = PTR_ERR(xafile);
367 		goto out;
368 	}
369 
370 	if (xafile->d_inode && (flags & XATTR_CREATE))
371 		err = -EEXIST;
372 
373 	if (!xafile->d_inode) {
374 		err = -ENODATA;
375 		if (xattr_may_create(flags))
376 			err = xattr_create(xadir->d_inode, xafile,
377 					      0700|S_IFREG);
378 	}
379 
380 	if (err)
381 		dput(xafile);
382 out:
383 	mutex_unlock(&xadir->d_inode->i_mutex);
384 	dput(xadir);
385 	if (err)
386 		return ERR_PTR(err);
387 	return xafile;
388 }
389 
390 /* Internal operations on file data */
391 static inline void reiserfs_put_page(struct page *page)
392 {
393 	kunmap(page);
394 	page_cache_release(page);
395 }
396 
397 static struct page *reiserfs_get_page(struct inode *dir, size_t n)
398 {
399 	struct address_space *mapping = dir->i_mapping;
400 	struct page *page;
401 	/* We can deadlock if we try to free dentries,
402 	   and an unlink/rmdir has just occured - GFP_NOFS avoids this */
403 	mapping_set_gfp_mask(mapping, GFP_NOFS);
404 	page = read_mapping_page(mapping, n >> PAGE_CACHE_SHIFT, NULL);
405 	if (!IS_ERR(page)) {
406 		kmap(page);
407 		if (PageError(page))
408 			goto fail;
409 	}
410 	return page;
411 
412       fail:
413 	reiserfs_put_page(page);
414 	return ERR_PTR(-EIO);
415 }
416 
417 static inline __u32 xattr_hash(const char *msg, int len)
418 {
419 	return csum_partial(msg, len, 0);
420 }
421 
422 int reiserfs_commit_write(struct file *f, struct page *page,
423 			  unsigned from, unsigned to);
424 int reiserfs_prepare_write(struct file *f, struct page *page,
425 			   unsigned from, unsigned to);
426 
427 static void update_ctime(struct inode *inode)
428 {
429 	struct timespec now = current_fs_time(inode->i_sb);
430 	if (hlist_unhashed(&inode->i_hash) || !inode->i_nlink ||
431 	    timespec_equal(&inode->i_ctime, &now))
432 		return;
433 
434 	inode->i_ctime = CURRENT_TIME_SEC;
435 	mark_inode_dirty(inode);
436 }
437 
438 static int lookup_and_delete_xattr(struct inode *inode, const char *name)
439 {
440 	int err = 0;
441 	struct dentry *dentry, *xadir;
442 
443 	xadir = open_xa_dir(inode, XATTR_REPLACE);
444 	if (IS_ERR(xadir))
445 		return PTR_ERR(xadir);
446 
447 	mutex_lock_nested(&xadir->d_inode->i_mutex, I_MUTEX_XATTR);
448 	dentry = lookup_one_len(name, xadir, strlen(name));
449 	if (IS_ERR(dentry)) {
450 		err = PTR_ERR(dentry);
451 		goto out_dput;
452 	}
453 
454 	if (dentry->d_inode) {
455 		reiserfs_write_lock(inode->i_sb);
456 		err = xattr_unlink(xadir->d_inode, dentry);
457 		reiserfs_write_unlock(inode->i_sb);
458 		update_ctime(inode);
459 	}
460 
461 	dput(dentry);
462 out_dput:
463 	mutex_unlock(&xadir->d_inode->i_mutex);
464 	dput(xadir);
465 	return err;
466 }
467 
468 
469 /* Generic extended attribute operations that can be used by xa plugins */
470 
471 /*
472  * inode->i_mutex: down
473  */
474 int
475 reiserfs_xattr_set_handle(struct reiserfs_transaction_handle *th,
476 			  struct inode *inode, const char *name,
477 			  const void *buffer, size_t buffer_size, int flags)
478 {
479 	int err = 0;
480 	struct dentry *dentry;
481 	struct page *page;
482 	char *data;
483 	size_t file_pos = 0;
484 	size_t buffer_pos = 0;
485 	size_t new_size;
486 	__u32 xahash = 0;
487 
488 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
489 		return -EOPNOTSUPP;
490 
491 	reiserfs_write_unlock(inode->i_sb);
492 
493 	if (!buffer) {
494 		err = lookup_and_delete_xattr(inode, name);
495 		reiserfs_write_lock(inode->i_sb);
496 		return err;
497 	}
498 
499 	dentry = xattr_lookup(inode, name, flags);
500 	if (IS_ERR(dentry)) {
501 		reiserfs_write_lock(inode->i_sb);
502 		return PTR_ERR(dentry);
503 	}
504 
505 	down_write(&REISERFS_I(inode)->i_xattr_sem);
506 
507 	reiserfs_write_lock(inode->i_sb);
508 
509 	xahash = xattr_hash(buffer, buffer_size);
510 	while (buffer_pos < buffer_size || buffer_pos == 0) {
511 		size_t chunk;
512 		size_t skip = 0;
513 		size_t page_offset = (file_pos & (PAGE_CACHE_SIZE - 1));
514 		if (buffer_size - buffer_pos > PAGE_CACHE_SIZE)
515 			chunk = PAGE_CACHE_SIZE;
516 		else
517 			chunk = buffer_size - buffer_pos;
518 
519 		page = reiserfs_get_page(dentry->d_inode, file_pos);
520 		if (IS_ERR(page)) {
521 			err = PTR_ERR(page);
522 			goto out_unlock;
523 		}
524 
525 		lock_page(page);
526 		data = page_address(page);
527 
528 		if (file_pos == 0) {
529 			struct reiserfs_xattr_header *rxh;
530 			skip = file_pos = sizeof(struct reiserfs_xattr_header);
531 			if (chunk + skip > PAGE_CACHE_SIZE)
532 				chunk = PAGE_CACHE_SIZE - skip;
533 			rxh = (struct reiserfs_xattr_header *)data;
534 			rxh->h_magic = cpu_to_le32(REISERFS_XATTR_MAGIC);
535 			rxh->h_hash = cpu_to_le32(xahash);
536 		}
537 
538 		err = reiserfs_prepare_write(NULL, page, page_offset,
539 					    page_offset + chunk + skip);
540 		if (!err) {
541 			if (buffer)
542 				memcpy(data + skip, buffer + buffer_pos, chunk);
543 			err = reiserfs_commit_write(NULL, page, page_offset,
544 						    page_offset + chunk +
545 						    skip);
546 		}
547 		unlock_page(page);
548 		reiserfs_put_page(page);
549 		buffer_pos += chunk;
550 		file_pos += chunk;
551 		skip = 0;
552 		if (err || buffer_size == 0 || !buffer)
553 			break;
554 	}
555 
556 	new_size = buffer_size + sizeof(struct reiserfs_xattr_header);
557 	if (!err && new_size < i_size_read(dentry->d_inode)) {
558 		struct iattr newattrs = {
559 			.ia_ctime = current_fs_time(inode->i_sb),
560 			.ia_size = buffer_size,
561 			.ia_valid = ATTR_SIZE | ATTR_CTIME,
562 		};
563 
564 		reiserfs_write_unlock(inode->i_sb);
565 		mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_XATTR);
566 		down_write(&dentry->d_inode->i_alloc_sem);
567 		reiserfs_write_lock(inode->i_sb);
568 
569 		err = reiserfs_setattr(dentry, &newattrs);
570 		up_write(&dentry->d_inode->i_alloc_sem);
571 		mutex_unlock(&dentry->d_inode->i_mutex);
572 	} else
573 		update_ctime(inode);
574 out_unlock:
575 	up_write(&REISERFS_I(inode)->i_xattr_sem);
576 	dput(dentry);
577 	return err;
578 }
579 
580 /* We need to start a transaction to maintain lock ordering */
581 int reiserfs_xattr_set(struct inode *inode, const char *name,
582 		       const void *buffer, size_t buffer_size, int flags)
583 {
584 
585 	struct reiserfs_transaction_handle th;
586 	int error, error2;
587 	size_t jbegin_count = reiserfs_xattr_nblocks(inode, buffer_size);
588 
589 	if (!(flags & XATTR_REPLACE))
590 		jbegin_count += reiserfs_xattr_jcreate_nblocks(inode);
591 
592 	reiserfs_write_lock(inode->i_sb);
593 	error = journal_begin(&th, inode->i_sb, jbegin_count);
594 	if (error) {
595 		reiserfs_write_unlock(inode->i_sb);
596 		return error;
597 	}
598 
599 	error = reiserfs_xattr_set_handle(&th, inode, name,
600 					  buffer, buffer_size, flags);
601 
602 	error2 = journal_end(&th, inode->i_sb, jbegin_count);
603 	if (error == 0)
604 		error = error2;
605 	reiserfs_write_unlock(inode->i_sb);
606 
607 	return error;
608 }
609 
610 /*
611  * inode->i_mutex: down
612  */
613 int
614 reiserfs_xattr_get(struct inode *inode, const char *name, void *buffer,
615 		   size_t buffer_size)
616 {
617 	ssize_t err = 0;
618 	struct dentry *dentry;
619 	size_t isize;
620 	size_t file_pos = 0;
621 	size_t buffer_pos = 0;
622 	struct page *page;
623 	__u32 hash = 0;
624 
625 	if (name == NULL)
626 		return -EINVAL;
627 
628 	/* We can't have xattrs attached to v1 items since they don't have
629 	 * generation numbers */
630 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
631 		return -EOPNOTSUPP;
632 
633 	dentry = xattr_lookup(inode, name, XATTR_REPLACE);
634 	if (IS_ERR(dentry)) {
635 		err = PTR_ERR(dentry);
636 		goto out;
637 	}
638 
639 	down_read(&REISERFS_I(inode)->i_xattr_sem);
640 
641 	isize = i_size_read(dentry->d_inode);
642 
643 	/* Just return the size needed */
644 	if (buffer == NULL) {
645 		err = isize - sizeof(struct reiserfs_xattr_header);
646 		goto out_unlock;
647 	}
648 
649 	if (buffer_size < isize - sizeof(struct reiserfs_xattr_header)) {
650 		err = -ERANGE;
651 		goto out_unlock;
652 	}
653 
654 	while (file_pos < isize) {
655 		size_t chunk;
656 		char *data;
657 		size_t skip = 0;
658 		if (isize - file_pos > PAGE_CACHE_SIZE)
659 			chunk = PAGE_CACHE_SIZE;
660 		else
661 			chunk = isize - file_pos;
662 
663 		page = reiserfs_get_page(dentry->d_inode, file_pos);
664 		if (IS_ERR(page)) {
665 			err = PTR_ERR(page);
666 			goto out_unlock;
667 		}
668 
669 		lock_page(page);
670 		data = page_address(page);
671 		if (file_pos == 0) {
672 			struct reiserfs_xattr_header *rxh =
673 			    (struct reiserfs_xattr_header *)data;
674 			skip = file_pos = sizeof(struct reiserfs_xattr_header);
675 			chunk -= skip;
676 			/* Magic doesn't match up.. */
677 			if (rxh->h_magic != cpu_to_le32(REISERFS_XATTR_MAGIC)) {
678 				unlock_page(page);
679 				reiserfs_put_page(page);
680 				reiserfs_warning(inode->i_sb, "jdm-20001",
681 						 "Invalid magic for xattr (%s) "
682 						 "associated with %k", name,
683 						 INODE_PKEY(inode));
684 				err = -EIO;
685 				goto out_unlock;
686 			}
687 			hash = le32_to_cpu(rxh->h_hash);
688 		}
689 		memcpy(buffer + buffer_pos, data + skip, chunk);
690 		unlock_page(page);
691 		reiserfs_put_page(page);
692 		file_pos += chunk;
693 		buffer_pos += chunk;
694 		skip = 0;
695 	}
696 	err = isize - sizeof(struct reiserfs_xattr_header);
697 
698 	if (xattr_hash(buffer, isize - sizeof(struct reiserfs_xattr_header)) !=
699 	    hash) {
700 		reiserfs_warning(inode->i_sb, "jdm-20002",
701 				 "Invalid hash for xattr (%s) associated "
702 				 "with %k", name, INODE_PKEY(inode));
703 		err = -EIO;
704 	}
705 
706 out_unlock:
707 	up_read(&REISERFS_I(inode)->i_xattr_sem);
708 	dput(dentry);
709 
710 out:
711 	return err;
712 }
713 
714 /*
715  * In order to implement different sets of xattr operations for each xattr
716  * prefix with the generic xattr API, a filesystem should create a
717  * null-terminated array of struct xattr_handler (one for each prefix) and
718  * hang a pointer to it off of the s_xattr field of the superblock.
719  *
720  * The generic_fooxattr() functions will use this list to dispatch xattr
721  * operations to the correct xattr_handler.
722  */
723 #define for_each_xattr_handler(handlers, handler)		\
724 		for ((handler) = *(handlers)++;			\
725 			(handler) != NULL;			\
726 			(handler) = *(handlers)++)
727 
728 /* This is the implementation for the xattr plugin infrastructure */
729 static inline struct xattr_handler *
730 find_xattr_handler_prefix(struct xattr_handler **handlers,
731 			   const char *name)
732 {
733 	struct xattr_handler *xah;
734 
735 	if (!handlers)
736 		return NULL;
737 
738 	for_each_xattr_handler(handlers, xah) {
739 		if (strncmp(xah->prefix, name, strlen(xah->prefix)) == 0)
740 			break;
741 	}
742 
743 	return xah;
744 }
745 
746 
747 /*
748  * Inode operation getxattr()
749  */
750 ssize_t
751 reiserfs_getxattr(struct dentry * dentry, const char *name, void *buffer,
752 		  size_t size)
753 {
754 	struct xattr_handler *handler;
755 
756 	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
757 
758 	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
759 		return -EOPNOTSUPP;
760 
761 	return handler->get(dentry, name, buffer, size, handler->flags);
762 }
763 
764 /*
765  * Inode operation setxattr()
766  *
767  * dentry->d_inode->i_mutex down
768  */
769 int
770 reiserfs_setxattr(struct dentry *dentry, const char *name, const void *value,
771 		  size_t size, int flags)
772 {
773 	struct xattr_handler *handler;
774 
775 	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
776 
777 	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
778 		return -EOPNOTSUPP;
779 
780 	return handler->set(dentry, name, value, size, flags, handler->flags);
781 }
782 
783 /*
784  * Inode operation removexattr()
785  *
786  * dentry->d_inode->i_mutex down
787  */
788 int reiserfs_removexattr(struct dentry *dentry, const char *name)
789 {
790 	struct xattr_handler *handler;
791 	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
792 
793 	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
794 		return -EOPNOTSUPP;
795 
796 	return handler->set(dentry, name, NULL, 0, XATTR_REPLACE, handler->flags);
797 }
798 
799 struct listxattr_buf {
800 	size_t size;
801 	size_t pos;
802 	char *buf;
803 	struct dentry *dentry;
804 };
805 
806 static int listxattr_filler(void *buf, const char *name, int namelen,
807 			    loff_t offset, u64 ino, unsigned int d_type)
808 {
809 	struct listxattr_buf *b = (struct listxattr_buf *)buf;
810 	size_t size;
811 	if (name[0] != '.' ||
812 	    (namelen != 1 && (name[1] != '.' || namelen != 2))) {
813 		struct xattr_handler *handler;
814 		handler = find_xattr_handler_prefix(b->dentry->d_sb->s_xattr,
815 						    name);
816 		if (!handler)	/* Unsupported xattr name */
817 			return 0;
818 		if (b->buf) {
819 			size = handler->list(b->dentry, b->buf + b->pos,
820 					 b->size, name, namelen,
821 					 handler->flags);
822 			if (size > b->size)
823 				return -ERANGE;
824 		} else {
825 			size = handler->list(b->dentry, NULL, 0, name,
826 					     namelen, handler->flags);
827 		}
828 
829 		b->pos += size;
830 	}
831 	return 0;
832 }
833 
834 /*
835  * Inode operation listxattr()
836  *
837  * We totally ignore the generic listxattr here because it would be stupid
838  * not to. Since the xattrs are organized in a directory, we can just
839  * readdir to find them.
840  */
841 ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size)
842 {
843 	struct dentry *dir;
844 	int err = 0;
845 	loff_t pos = 0;
846 	struct listxattr_buf buf = {
847 		.dentry = dentry,
848 		.buf = buffer,
849 		.size = buffer ? size : 0,
850 	};
851 
852 	if (!dentry->d_inode)
853 		return -EINVAL;
854 
855 	if (!dentry->d_sb->s_xattr ||
856 	    get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
857 		return -EOPNOTSUPP;
858 
859 	dir = open_xa_dir(dentry->d_inode, XATTR_REPLACE);
860 	if (IS_ERR(dir)) {
861 		err = PTR_ERR(dir);
862 		if (err == -ENODATA)
863 			err = 0;  /* Not an error if there aren't any xattrs */
864 		goto out;
865 	}
866 
867 	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR);
868 	err = reiserfs_readdir_dentry(dir, &buf, listxattr_filler, &pos);
869 	mutex_unlock(&dir->d_inode->i_mutex);
870 
871 	if (!err)
872 		err = buf.pos;
873 
874 	dput(dir);
875 out:
876 	return err;
877 }
878 
879 static int reiserfs_check_acl(struct inode *inode, int mask)
880 {
881 	struct posix_acl *acl;
882 	int error = -EAGAIN; /* do regular unix permission checks by default */
883 
884 	acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
885 
886 	if (acl) {
887 		if (!IS_ERR(acl)) {
888 			error = posix_acl_permission(inode, acl, mask);
889 			posix_acl_release(acl);
890 		} else if (PTR_ERR(acl) != -ENODATA)
891 			error = PTR_ERR(acl);
892 	}
893 
894 	return error;
895 }
896 
897 static int create_privroot(struct dentry *dentry)
898 {
899 	int err;
900 	struct inode *inode = dentry->d_parent->d_inode;
901 	WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex));
902 
903 	err = xattr_mkdir(inode, dentry, 0700);
904 	if (err || !dentry->d_inode) {
905 		reiserfs_warning(dentry->d_sb, "jdm-20006",
906 				 "xattrs/ACLs enabled and couldn't "
907 				 "find/create .reiserfs_priv. "
908 				 "Failing mount.");
909 		return -EOPNOTSUPP;
910 	}
911 
912 	dentry->d_inode->i_flags |= S_PRIVATE;
913 	reiserfs_info(dentry->d_sb, "Created %s - reserved for xattr "
914 		      "storage.\n", PRIVROOT_NAME);
915 
916 	return 0;
917 }
918 
919 #else
920 int __init reiserfs_xattr_register_handlers(void) { return 0; }
921 void reiserfs_xattr_unregister_handlers(void) {}
922 static int create_privroot(struct dentry *dentry) { return 0; }
923 #endif
924 
925 /* Actual operations that are exported to VFS-land */
926 struct xattr_handler *reiserfs_xattr_handlers[] = {
927 #ifdef CONFIG_REISERFS_FS_XATTR
928 	&reiserfs_xattr_user_handler,
929 	&reiserfs_xattr_trusted_handler,
930 #endif
931 #ifdef CONFIG_REISERFS_FS_SECURITY
932 	&reiserfs_xattr_security_handler,
933 #endif
934 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
935 	&reiserfs_posix_acl_access_handler,
936 	&reiserfs_posix_acl_default_handler,
937 #endif
938 	NULL
939 };
940 
941 static int xattr_mount_check(struct super_block *s)
942 {
943 	/* We need generation numbers to ensure that the oid mapping is correct
944 	 * v3.5 filesystems don't have them. */
945 	if (old_format_only(s)) {
946 		if (reiserfs_xattrs_optional(s)) {
947 			/* Old format filesystem, but optional xattrs have
948 			 * been enabled. Error out. */
949 			reiserfs_warning(s, "jdm-2005",
950 					 "xattrs/ACLs not supported "
951 					 "on pre-v3.6 format filesystems. "
952 					 "Failing mount.");
953 			return -EOPNOTSUPP;
954 		}
955 	}
956 
957 	return 0;
958 }
959 
960 int reiserfs_permission(struct inode *inode, int mask)
961 {
962 	/*
963 	 * We don't do permission checks on the internal objects.
964 	 * Permissions are determined by the "owning" object.
965 	 */
966 	if (IS_PRIVATE(inode))
967 		return 0;
968 
969 #ifdef CONFIG_REISERFS_FS_XATTR
970 	/*
971 	 * Stat data v1 doesn't support ACLs.
972 	 */
973 	if (get_inode_sd_version(inode) != STAT_DATA_V1)
974 		return generic_permission(inode, mask, reiserfs_check_acl);
975 #endif
976 	return generic_permission(inode, mask, NULL);
977 }
978 
979 /* This will catch lookups from the fs root to .reiserfs_priv */
980 static int
981 xattr_lookup_poison(struct dentry *dentry, struct qstr *q1, struct qstr *name)
982 {
983 	struct dentry *priv_root = REISERFS_SB(dentry->d_sb)->priv_root;
984 	if (container_of(q1, struct dentry, d_name) == priv_root)
985 		return -ENOENT;
986 	if (q1->len == name->len &&
987 		   !memcmp(q1->name, name->name, name->len))
988 		return 0;
989 	return 1;
990 }
991 
992 static const struct dentry_operations xattr_lookup_poison_ops = {
993 	.d_compare = xattr_lookup_poison,
994 };
995 
996 int reiserfs_lookup_privroot(struct super_block *s)
997 {
998 	struct dentry *dentry;
999 	int err = 0;
1000 
1001 	/* If we don't have the privroot located yet - go find it */
1002 	reiserfs_mutex_lock_safe(&s->s_root->d_inode->i_mutex, s);
1003 	dentry = lookup_one_len(PRIVROOT_NAME, s->s_root,
1004 				strlen(PRIVROOT_NAME));
1005 	if (!IS_ERR(dentry)) {
1006 		REISERFS_SB(s)->priv_root = dentry;
1007 		if (!reiserfs_expose_privroot(s))
1008 			s->s_root->d_op = &xattr_lookup_poison_ops;
1009 		if (dentry->d_inode)
1010 			dentry->d_inode->i_flags |= S_PRIVATE;
1011 	} else
1012 		err = PTR_ERR(dentry);
1013 	mutex_unlock(&s->s_root->d_inode->i_mutex);
1014 
1015 	return err;
1016 }
1017 
1018 /* We need to take a copy of the mount flags since things like
1019  * MS_RDONLY don't get set until *after* we're called.
1020  * mount_flags != mount_options */
1021 int reiserfs_xattr_init(struct super_block *s, int mount_flags)
1022 {
1023 	int err = 0;
1024 	struct dentry *privroot = REISERFS_SB(s)->priv_root;
1025 
1026 	err = xattr_mount_check(s);
1027 	if (err)
1028 		goto error;
1029 
1030 	if (!privroot->d_inode && !(mount_flags & MS_RDONLY)) {
1031 		reiserfs_mutex_lock_safe(&s->s_root->d_inode->i_mutex, s);
1032 		err = create_privroot(REISERFS_SB(s)->priv_root);
1033 		mutex_unlock(&s->s_root->d_inode->i_mutex);
1034 	}
1035 
1036 	if (privroot->d_inode) {
1037 		s->s_xattr = reiserfs_xattr_handlers;
1038 		reiserfs_mutex_lock_safe(&privroot->d_inode->i_mutex, s);
1039 		if (!REISERFS_SB(s)->xattr_root) {
1040 			struct dentry *dentry;
1041 			dentry = lookup_one_len(XAROOT_NAME, privroot,
1042 						strlen(XAROOT_NAME));
1043 			if (!IS_ERR(dentry))
1044 				REISERFS_SB(s)->xattr_root = dentry;
1045 			else
1046 				err = PTR_ERR(dentry);
1047 		}
1048 		mutex_unlock(&privroot->d_inode->i_mutex);
1049 	}
1050 
1051 error:
1052 	if (err) {
1053 		clear_bit(REISERFS_XATTRS_USER, &(REISERFS_SB(s)->s_mount_opt));
1054 		clear_bit(REISERFS_POSIXACL, &(REISERFS_SB(s)->s_mount_opt));
1055 	}
1056 
1057 	/* The super_block MS_POSIXACL must mirror the (no)acl mount option. */
1058 	if (reiserfs_posixacl(s))
1059 		s->s_flags |= MS_POSIXACL;
1060 	else
1061 		s->s_flags &= ~MS_POSIXACL;
1062 
1063 	return err;
1064 }
1065