xattr.c revision fb045adb99d9b7c562dc7fef834857f78249daa1
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/gfp.h>
42#include <linux/fs.h>
43#include <linux/file.h>
44#include <linux/pagemap.h>
45#include <linux/xattr.h>
46#include <linux/reiserfs_xattr.h>
47#include <linux/reiserfs_acl.h>
48#include <asm/uaccess.h>
49#include <net/checksum.h>
50#include <linux/stat.h>
51#include <linux/quotaops.h>
52#include <linux/security.h>
53
54#define PRIVROOT_NAME ".reiserfs_priv"
55#define XAROOT_NAME   "xattrs"
56
57
58/* Helpers for inode ops. We do this so that we don't have all the VFS
59 * overhead and also for proper i_mutex annotation.
60 * dir->i_mutex must be held for all of them. */
61#ifdef CONFIG_REISERFS_FS_XATTR
62static int xattr_create(struct inode *dir, struct dentry *dentry, int mode)
63{
64	BUG_ON(!mutex_is_locked(&dir->i_mutex));
65	return dir->i_op->create(dir, dentry, mode, NULL);
66}
67#endif
68
69static int xattr_mkdir(struct inode *dir, struct dentry *dentry, int mode)
70{
71	BUG_ON(!mutex_is_locked(&dir->i_mutex));
72	return dir->i_op->mkdir(dir, dentry, mode);
73}
74
75/* We use I_MUTEX_CHILD here to silence lockdep. It's safe because xattr
76 * mutation ops aren't called during rename or splace, which are the
77 * only other users of I_MUTEX_CHILD. It violates the ordering, but that's
78 * better than allocating another subclass just for this code. */
79static int xattr_unlink(struct inode *dir, struct dentry *dentry)
80{
81	int error;
82	BUG_ON(!mutex_is_locked(&dir->i_mutex));
83
84	reiserfs_mutex_lock_nested_safe(&dentry->d_inode->i_mutex,
85					I_MUTEX_CHILD, dir->i_sb);
86	error = dir->i_op->unlink(dir, dentry);
87	mutex_unlock(&dentry->d_inode->i_mutex);
88
89	if (!error)
90		d_delete(dentry);
91	return error;
92}
93
94static int xattr_rmdir(struct inode *dir, struct dentry *dentry)
95{
96	int error;
97	BUG_ON(!mutex_is_locked(&dir->i_mutex));
98
99	reiserfs_mutex_lock_nested_safe(&dentry->d_inode->i_mutex,
100					I_MUTEX_CHILD, dir->i_sb);
101	dentry_unhash(dentry);
102	error = dir->i_op->rmdir(dir, dentry);
103	if (!error)
104		dentry->d_inode->i_flags |= S_DEAD;
105	mutex_unlock(&dentry->d_inode->i_mutex);
106	if (!error)
107		d_delete(dentry);
108	dput(dentry);
109
110	return error;
111}
112
113#define xattr_may_create(flags)	(!flags || flags & XATTR_CREATE)
114
115static struct dentry *open_xa_root(struct super_block *sb, int flags)
116{
117	struct dentry *privroot = REISERFS_SB(sb)->priv_root;
118	struct dentry *xaroot;
119	if (!privroot->d_inode)
120		return ERR_PTR(-ENODATA);
121
122	mutex_lock_nested(&privroot->d_inode->i_mutex, I_MUTEX_XATTR);
123
124	xaroot = dget(REISERFS_SB(sb)->xattr_root);
125	if (!xaroot)
126		xaroot = ERR_PTR(-ENODATA);
127	else if (!xaroot->d_inode) {
128		int err = -ENODATA;
129		if (xattr_may_create(flags))
130			err = xattr_mkdir(privroot->d_inode, xaroot, 0700);
131		if (err) {
132			dput(xaroot);
133			xaroot = ERR_PTR(err);
134		}
135	}
136
137	mutex_unlock(&privroot->d_inode->i_mutex);
138	return xaroot;
139}
140
141static struct dentry *open_xa_dir(const struct inode *inode, int flags)
142{
143	struct dentry *xaroot, *xadir;
144	char namebuf[17];
145
146	xaroot = open_xa_root(inode->i_sb, flags);
147	if (IS_ERR(xaroot))
148		return xaroot;
149
150	snprintf(namebuf, sizeof(namebuf), "%X.%X",
151		 le32_to_cpu(INODE_PKEY(inode)->k_objectid),
152		 inode->i_generation);
153
154	mutex_lock_nested(&xaroot->d_inode->i_mutex, I_MUTEX_XATTR);
155
156	xadir = lookup_one_len(namebuf, xaroot, strlen(namebuf));
157	if (!IS_ERR(xadir) && !xadir->d_inode) {
158		int err = -ENODATA;
159		if (xattr_may_create(flags))
160			err = xattr_mkdir(xaroot->d_inode, xadir, 0700);
161		if (err) {
162			dput(xadir);
163			xadir = ERR_PTR(err);
164		}
165	}
166
167	mutex_unlock(&xaroot->d_inode->i_mutex);
168	dput(xaroot);
169	return xadir;
170}
171
172/* The following are side effects of other operations that aren't explicitly
173 * modifying extended attributes. This includes operations such as permissions
174 * or ownership changes, object deletions, etc. */
175struct reiserfs_dentry_buf {
176	struct dentry *xadir;
177	int count;
178	struct dentry *dentries[8];
179};
180
181static int
182fill_with_dentries(void *buf, const char *name, int namelen, loff_t offset,
183		    u64 ino, unsigned int d_type)
184{
185	struct reiserfs_dentry_buf *dbuf = buf;
186	struct dentry *dentry;
187	WARN_ON_ONCE(!mutex_is_locked(&dbuf->xadir->d_inode->i_mutex));
188
189	if (dbuf->count == ARRAY_SIZE(dbuf->dentries))
190		return -ENOSPC;
191
192	if (name[0] == '.' && (name[1] == '\0' ||
193			       (name[1] == '.' && name[2] == '\0')))
194		return 0;
195
196	dentry = lookup_one_len(name, dbuf->xadir, namelen);
197	if (IS_ERR(dentry)) {
198		return PTR_ERR(dentry);
199	} else if (!dentry->d_inode) {
200		/* A directory entry exists, but no file? */
201		reiserfs_error(dentry->d_sb, "xattr-20003",
202			       "Corrupted directory: xattr %s listed but "
203			       "not found for file %s.\n",
204			       dentry->d_name.name, dbuf->xadir->d_name.name);
205		dput(dentry);
206		return -EIO;
207	}
208
209	dbuf->dentries[dbuf->count++] = dentry;
210	return 0;
211}
212
213static void
214cleanup_dentry_buf(struct reiserfs_dentry_buf *buf)
215{
216	int i;
217	for (i = 0; i < buf->count; i++)
218		if (buf->dentries[i])
219			dput(buf->dentries[i]);
220}
221
222static int reiserfs_for_each_xattr(struct inode *inode,
223				   int (*action)(struct dentry *, void *),
224				   void *data)
225{
226	struct dentry *dir;
227	int i, err = 0;
228	loff_t pos = 0;
229	struct reiserfs_dentry_buf buf = {
230		.count = 0,
231	};
232
233	/* Skip out, an xattr has no xattrs associated with it */
234	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
235		return 0;
236
237	reiserfs_write_unlock(inode->i_sb);
238	dir = open_xa_dir(inode, XATTR_REPLACE);
239	if (IS_ERR(dir)) {
240		err = PTR_ERR(dir);
241		reiserfs_write_lock(inode->i_sb);
242		goto out;
243	} else if (!dir->d_inode) {
244		err = 0;
245		reiserfs_write_lock(inode->i_sb);
246		goto out_dir;
247	}
248
249	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR);
250
251	reiserfs_write_lock(inode->i_sb);
252
253	buf.xadir = dir;
254	err = reiserfs_readdir_dentry(dir, &buf, fill_with_dentries, &pos);
255	while ((err == 0 || err == -ENOSPC) && buf.count) {
256		err = 0;
257
258		for (i = 0; i < buf.count && buf.dentries[i]; i++) {
259			int lerr = 0;
260			struct dentry *dentry = buf.dentries[i];
261
262			if (err == 0 && !S_ISDIR(dentry->d_inode->i_mode))
263				lerr = action(dentry, data);
264
265			dput(dentry);
266			buf.dentries[i] = NULL;
267			err = lerr ?: err;
268		}
269		buf.count = 0;
270		if (!err)
271			err = reiserfs_readdir_dentry(dir, &buf,
272						      fill_with_dentries, &pos);
273	}
274	mutex_unlock(&dir->d_inode->i_mutex);
275
276	/* Clean up after a failed readdir */
277	cleanup_dentry_buf(&buf);
278
279	if (!err) {
280		/* We start a transaction here to avoid a ABBA situation
281		 * between the xattr root's i_mutex and the journal lock.
282		 * This doesn't incur much additional overhead since the
283		 * new transaction will just nest inside the
284		 * outer transaction. */
285		int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 +
286			     4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
287		struct reiserfs_transaction_handle th;
288		err = journal_begin(&th, inode->i_sb, blocks);
289		if (!err) {
290			int jerror;
291			reiserfs_mutex_lock_nested_safe(
292					  &dir->d_parent->d_inode->i_mutex,
293					  I_MUTEX_XATTR, inode->i_sb);
294			err = action(dir, data);
295			jerror = journal_end(&th, inode->i_sb, blocks);
296			mutex_unlock(&dir->d_parent->d_inode->i_mutex);
297			err = jerror ?: err;
298		}
299	}
300out_dir:
301	dput(dir);
302out:
303	/* -ENODATA isn't an error */
304	if (err == -ENODATA)
305		err = 0;
306	return err;
307}
308
309static int delete_one_xattr(struct dentry *dentry, void *data)
310{
311	struct inode *dir = dentry->d_parent->d_inode;
312
313	/* This is the xattr dir, handle specially. */
314	if (S_ISDIR(dentry->d_inode->i_mode))
315		return xattr_rmdir(dir, dentry);
316
317	return xattr_unlink(dir, dentry);
318}
319
320static int chown_one_xattr(struct dentry *dentry, void *data)
321{
322	struct iattr *attrs = data;
323	return reiserfs_setattr(dentry, attrs);
324}
325
326/* No i_mutex, but the inode is unconnected. */
327int reiserfs_delete_xattrs(struct inode *inode)
328{
329	int err = reiserfs_for_each_xattr(inode, delete_one_xattr, NULL);
330	if (err)
331		reiserfs_warning(inode->i_sb, "jdm-20004",
332				 "Couldn't delete all xattrs (%d)\n", err);
333	return err;
334}
335
336/* inode->i_mutex: down */
337int reiserfs_chown_xattrs(struct inode *inode, struct iattr *attrs)
338{
339	int err = reiserfs_for_each_xattr(inode, chown_one_xattr, attrs);
340	if (err)
341		reiserfs_warning(inode->i_sb, "jdm-20007",
342				 "Couldn't chown all xattrs (%d)\n", err);
343	return err;
344}
345
346#ifdef CONFIG_REISERFS_FS_XATTR
347/* Returns a dentry corresponding to a specific extended attribute file
348 * for the inode. If flags allow, the file is created. Otherwise, a
349 * valid or negative dentry, or an error is returned. */
350static struct dentry *xattr_lookup(struct inode *inode, const char *name,
351				    int flags)
352{
353	struct dentry *xadir, *xafile;
354	int err = 0;
355
356	xadir = open_xa_dir(inode, flags);
357	if (IS_ERR(xadir))
358		return ERR_CAST(xadir);
359
360	mutex_lock_nested(&xadir->d_inode->i_mutex, I_MUTEX_XATTR);
361	xafile = lookup_one_len(name, xadir, strlen(name));
362	if (IS_ERR(xafile)) {
363		err = PTR_ERR(xafile);
364		goto out;
365	}
366
367	if (xafile->d_inode && (flags & XATTR_CREATE))
368		err = -EEXIST;
369
370	if (!xafile->d_inode) {
371		err = -ENODATA;
372		if (xattr_may_create(flags))
373			err = xattr_create(xadir->d_inode, xafile,
374					      0700|S_IFREG);
375	}
376
377	if (err)
378		dput(xafile);
379out:
380	mutex_unlock(&xadir->d_inode->i_mutex);
381	dput(xadir);
382	if (err)
383		return ERR_PTR(err);
384	return xafile;
385}
386
387/* Internal operations on file data */
388static inline void reiserfs_put_page(struct page *page)
389{
390	kunmap(page);
391	page_cache_release(page);
392}
393
394static struct page *reiserfs_get_page(struct inode *dir, size_t n)
395{
396	struct address_space *mapping = dir->i_mapping;
397	struct page *page;
398	/* We can deadlock if we try to free dentries,
399	   and an unlink/rmdir has just occured - GFP_NOFS avoids this */
400	mapping_set_gfp_mask(mapping, GFP_NOFS);
401	page = read_mapping_page(mapping, n >> PAGE_CACHE_SHIFT, NULL);
402	if (!IS_ERR(page)) {
403		kmap(page);
404		if (PageError(page))
405			goto fail;
406	}
407	return page;
408
409      fail:
410	reiserfs_put_page(page);
411	return ERR_PTR(-EIO);
412}
413
414static inline __u32 xattr_hash(const char *msg, int len)
415{
416	return csum_partial(msg, len, 0);
417}
418
419int reiserfs_commit_write(struct file *f, struct page *page,
420			  unsigned from, unsigned to);
421
422static void update_ctime(struct inode *inode)
423{
424	struct timespec now = current_fs_time(inode->i_sb);
425	if (inode_unhashed(inode) || !inode->i_nlink ||
426	    timespec_equal(&inode->i_ctime, &now))
427		return;
428
429	inode->i_ctime = CURRENT_TIME_SEC;
430	mark_inode_dirty(inode);
431}
432
433static int lookup_and_delete_xattr(struct inode *inode, const char *name)
434{
435	int err = 0;
436	struct dentry *dentry, *xadir;
437
438	xadir = open_xa_dir(inode, XATTR_REPLACE);
439	if (IS_ERR(xadir))
440		return PTR_ERR(xadir);
441
442	mutex_lock_nested(&xadir->d_inode->i_mutex, I_MUTEX_XATTR);
443	dentry = lookup_one_len(name, xadir, strlen(name));
444	if (IS_ERR(dentry)) {
445		err = PTR_ERR(dentry);
446		goto out_dput;
447	}
448
449	if (dentry->d_inode) {
450		reiserfs_write_lock(inode->i_sb);
451		err = xattr_unlink(xadir->d_inode, dentry);
452		reiserfs_write_unlock(inode->i_sb);
453		update_ctime(inode);
454	}
455
456	dput(dentry);
457out_dput:
458	mutex_unlock(&xadir->d_inode->i_mutex);
459	dput(xadir);
460	return err;
461}
462
463
464/* Generic extended attribute operations that can be used by xa plugins */
465
466/*
467 * inode->i_mutex: down
468 */
469int
470reiserfs_xattr_set_handle(struct reiserfs_transaction_handle *th,
471			  struct inode *inode, const char *name,
472			  const void *buffer, size_t buffer_size, int flags)
473{
474	int err = 0;
475	struct dentry *dentry;
476	struct page *page;
477	char *data;
478	size_t file_pos = 0;
479	size_t buffer_pos = 0;
480	size_t new_size;
481	__u32 xahash = 0;
482
483	if (get_inode_sd_version(inode) == STAT_DATA_V1)
484		return -EOPNOTSUPP;
485
486	reiserfs_write_unlock(inode->i_sb);
487
488	if (!buffer) {
489		err = lookup_and_delete_xattr(inode, name);
490		reiserfs_write_lock(inode->i_sb);
491		return err;
492	}
493
494	dentry = xattr_lookup(inode, name, flags);
495	if (IS_ERR(dentry)) {
496		reiserfs_write_lock(inode->i_sb);
497		return PTR_ERR(dentry);
498	}
499
500	down_write(&REISERFS_I(inode)->i_xattr_sem);
501
502	reiserfs_write_lock(inode->i_sb);
503
504	xahash = xattr_hash(buffer, buffer_size);
505	while (buffer_pos < buffer_size || buffer_pos == 0) {
506		size_t chunk;
507		size_t skip = 0;
508		size_t page_offset = (file_pos & (PAGE_CACHE_SIZE - 1));
509		if (buffer_size - buffer_pos > PAGE_CACHE_SIZE)
510			chunk = PAGE_CACHE_SIZE;
511		else
512			chunk = buffer_size - buffer_pos;
513
514		page = reiserfs_get_page(dentry->d_inode, file_pos);
515		if (IS_ERR(page)) {
516			err = PTR_ERR(page);
517			goto out_unlock;
518		}
519
520		lock_page(page);
521		data = page_address(page);
522
523		if (file_pos == 0) {
524			struct reiserfs_xattr_header *rxh;
525			skip = file_pos = sizeof(struct reiserfs_xattr_header);
526			if (chunk + skip > PAGE_CACHE_SIZE)
527				chunk = PAGE_CACHE_SIZE - skip;
528			rxh = (struct reiserfs_xattr_header *)data;
529			rxh->h_magic = cpu_to_le32(REISERFS_XATTR_MAGIC);
530			rxh->h_hash = cpu_to_le32(xahash);
531		}
532
533		err = __reiserfs_write_begin(page, page_offset, chunk + skip);
534		if (!err) {
535			if (buffer)
536				memcpy(data + skip, buffer + buffer_pos, chunk);
537			err = reiserfs_commit_write(NULL, page, page_offset,
538						    page_offset + chunk +
539						    skip);
540		}
541		unlock_page(page);
542		reiserfs_put_page(page);
543		buffer_pos += chunk;
544		file_pos += chunk;
545		skip = 0;
546		if (err || buffer_size == 0 || !buffer)
547			break;
548	}
549
550	new_size = buffer_size + sizeof(struct reiserfs_xattr_header);
551	if (!err && new_size < i_size_read(dentry->d_inode)) {
552		struct iattr newattrs = {
553			.ia_ctime = current_fs_time(inode->i_sb),
554			.ia_size = new_size,
555			.ia_valid = ATTR_SIZE | ATTR_CTIME,
556		};
557
558		reiserfs_write_unlock(inode->i_sb);
559		mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_XATTR);
560		down_write(&dentry->d_inode->i_alloc_sem);
561		reiserfs_write_lock(inode->i_sb);
562
563		err = reiserfs_setattr(dentry, &newattrs);
564		up_write(&dentry->d_inode->i_alloc_sem);
565		mutex_unlock(&dentry->d_inode->i_mutex);
566	} else
567		update_ctime(inode);
568out_unlock:
569	up_write(&REISERFS_I(inode)->i_xattr_sem);
570	dput(dentry);
571	return err;
572}
573
574/* We need to start a transaction to maintain lock ordering */
575int reiserfs_xattr_set(struct inode *inode, const char *name,
576		       const void *buffer, size_t buffer_size, int flags)
577{
578
579	struct reiserfs_transaction_handle th;
580	int error, error2;
581	size_t jbegin_count = reiserfs_xattr_nblocks(inode, buffer_size);
582
583	if (!(flags & XATTR_REPLACE))
584		jbegin_count += reiserfs_xattr_jcreate_nblocks(inode);
585
586	reiserfs_write_lock(inode->i_sb);
587	error = journal_begin(&th, inode->i_sb, jbegin_count);
588	if (error) {
589		reiserfs_write_unlock(inode->i_sb);
590		return error;
591	}
592
593	error = reiserfs_xattr_set_handle(&th, inode, name,
594					  buffer, buffer_size, flags);
595
596	error2 = journal_end(&th, inode->i_sb, jbegin_count);
597	if (error == 0)
598		error = error2;
599	reiserfs_write_unlock(inode->i_sb);
600
601	return error;
602}
603
604/*
605 * inode->i_mutex: down
606 */
607int
608reiserfs_xattr_get(struct inode *inode, const char *name, void *buffer,
609		   size_t buffer_size)
610{
611	ssize_t err = 0;
612	struct dentry *dentry;
613	size_t isize;
614	size_t file_pos = 0;
615	size_t buffer_pos = 0;
616	struct page *page;
617	__u32 hash = 0;
618
619	if (name == NULL)
620		return -EINVAL;
621
622	/* We can't have xattrs attached to v1 items since they don't have
623	 * generation numbers */
624	if (get_inode_sd_version(inode) == STAT_DATA_V1)
625		return -EOPNOTSUPP;
626
627	dentry = xattr_lookup(inode, name, XATTR_REPLACE);
628	if (IS_ERR(dentry)) {
629		err = PTR_ERR(dentry);
630		goto out;
631	}
632
633	down_read(&REISERFS_I(inode)->i_xattr_sem);
634
635	isize = i_size_read(dentry->d_inode);
636
637	/* Just return the size needed */
638	if (buffer == NULL) {
639		err = isize - sizeof(struct reiserfs_xattr_header);
640		goto out_unlock;
641	}
642
643	if (buffer_size < isize - sizeof(struct reiserfs_xattr_header)) {
644		err = -ERANGE;
645		goto out_unlock;
646	}
647
648	while (file_pos < isize) {
649		size_t chunk;
650		char *data;
651		size_t skip = 0;
652		if (isize - file_pos > PAGE_CACHE_SIZE)
653			chunk = PAGE_CACHE_SIZE;
654		else
655			chunk = isize - file_pos;
656
657		page = reiserfs_get_page(dentry->d_inode, file_pos);
658		if (IS_ERR(page)) {
659			err = PTR_ERR(page);
660			goto out_unlock;
661		}
662
663		lock_page(page);
664		data = page_address(page);
665		if (file_pos == 0) {
666			struct reiserfs_xattr_header *rxh =
667			    (struct reiserfs_xattr_header *)data;
668			skip = file_pos = sizeof(struct reiserfs_xattr_header);
669			chunk -= skip;
670			/* Magic doesn't match up.. */
671			if (rxh->h_magic != cpu_to_le32(REISERFS_XATTR_MAGIC)) {
672				unlock_page(page);
673				reiserfs_put_page(page);
674				reiserfs_warning(inode->i_sb, "jdm-20001",
675						 "Invalid magic for xattr (%s) "
676						 "associated with %k", name,
677						 INODE_PKEY(inode));
678				err = -EIO;
679				goto out_unlock;
680			}
681			hash = le32_to_cpu(rxh->h_hash);
682		}
683		memcpy(buffer + buffer_pos, data + skip, chunk);
684		unlock_page(page);
685		reiserfs_put_page(page);
686		file_pos += chunk;
687		buffer_pos += chunk;
688		skip = 0;
689	}
690	err = isize - sizeof(struct reiserfs_xattr_header);
691
692	if (xattr_hash(buffer, isize - sizeof(struct reiserfs_xattr_header)) !=
693	    hash) {
694		reiserfs_warning(inode->i_sb, "jdm-20002",
695				 "Invalid hash for xattr (%s) associated "
696				 "with %k", name, INODE_PKEY(inode));
697		err = -EIO;
698	}
699
700out_unlock:
701	up_read(&REISERFS_I(inode)->i_xattr_sem);
702	dput(dentry);
703
704out:
705	return err;
706}
707
708/*
709 * In order to implement different sets of xattr operations for each xattr
710 * prefix with the generic xattr API, a filesystem should create a
711 * null-terminated array of struct xattr_handler (one for each prefix) and
712 * hang a pointer to it off of the s_xattr field of the superblock.
713 *
714 * The generic_fooxattr() functions will use this list to dispatch xattr
715 * operations to the correct xattr_handler.
716 */
717#define for_each_xattr_handler(handlers, handler)		\
718		for ((handler) = *(handlers)++;			\
719			(handler) != NULL;			\
720			(handler) = *(handlers)++)
721
722/* This is the implementation for the xattr plugin infrastructure */
723static inline const struct xattr_handler *
724find_xattr_handler_prefix(const struct xattr_handler **handlers,
725			   const char *name)
726{
727	const struct xattr_handler *xah;
728
729	if (!handlers)
730		return NULL;
731
732	for_each_xattr_handler(handlers, xah) {
733		if (strncmp(xah->prefix, name, strlen(xah->prefix)) == 0)
734			break;
735	}
736
737	return xah;
738}
739
740
741/*
742 * Inode operation getxattr()
743 */
744ssize_t
745reiserfs_getxattr(struct dentry * dentry, const char *name, void *buffer,
746		  size_t size)
747{
748	const struct xattr_handler *handler;
749
750	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
751
752	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
753		return -EOPNOTSUPP;
754
755	return handler->get(dentry, name, buffer, size, handler->flags);
756}
757
758/*
759 * Inode operation setxattr()
760 *
761 * dentry->d_inode->i_mutex down
762 */
763int
764reiserfs_setxattr(struct dentry *dentry, const char *name, const void *value,
765		  size_t size, int flags)
766{
767	const struct xattr_handler *handler;
768
769	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
770
771	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
772		return -EOPNOTSUPP;
773
774	return handler->set(dentry, name, value, size, flags, handler->flags);
775}
776
777/*
778 * Inode operation removexattr()
779 *
780 * dentry->d_inode->i_mutex down
781 */
782int reiserfs_removexattr(struct dentry *dentry, const char *name)
783{
784	const struct xattr_handler *handler;
785	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
786
787	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
788		return -EOPNOTSUPP;
789
790	return handler->set(dentry, name, NULL, 0, XATTR_REPLACE, handler->flags);
791}
792
793struct listxattr_buf {
794	size_t size;
795	size_t pos;
796	char *buf;
797	struct dentry *dentry;
798};
799
800static int listxattr_filler(void *buf, const char *name, int namelen,
801			    loff_t offset, u64 ino, unsigned int d_type)
802{
803	struct listxattr_buf *b = (struct listxattr_buf *)buf;
804	size_t size;
805	if (name[0] != '.' ||
806	    (namelen != 1 && (name[1] != '.' || namelen != 2))) {
807		const struct xattr_handler *handler;
808		handler = find_xattr_handler_prefix(b->dentry->d_sb->s_xattr,
809						    name);
810		if (!handler)	/* Unsupported xattr name */
811			return 0;
812		if (b->buf) {
813			size = handler->list(b->dentry, b->buf + b->pos,
814					 b->size, name, namelen,
815					 handler->flags);
816			if (size > b->size)
817				return -ERANGE;
818		} else {
819			size = handler->list(b->dentry, NULL, 0, name,
820					     namelen, handler->flags);
821		}
822
823		b->pos += size;
824	}
825	return 0;
826}
827
828/*
829 * Inode operation listxattr()
830 *
831 * We totally ignore the generic listxattr here because it would be stupid
832 * not to. Since the xattrs are organized in a directory, we can just
833 * readdir to find them.
834 */
835ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size)
836{
837	struct dentry *dir;
838	int err = 0;
839	loff_t pos = 0;
840	struct listxattr_buf buf = {
841		.dentry = dentry,
842		.buf = buffer,
843		.size = buffer ? size : 0,
844	};
845
846	if (!dentry->d_inode)
847		return -EINVAL;
848
849	if (!dentry->d_sb->s_xattr ||
850	    get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
851		return -EOPNOTSUPP;
852
853	dir = open_xa_dir(dentry->d_inode, XATTR_REPLACE);
854	if (IS_ERR(dir)) {
855		err = PTR_ERR(dir);
856		if (err == -ENODATA)
857			err = 0;  /* Not an error if there aren't any xattrs */
858		goto out;
859	}
860
861	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR);
862	err = reiserfs_readdir_dentry(dir, &buf, listxattr_filler, &pos);
863	mutex_unlock(&dir->d_inode->i_mutex);
864
865	if (!err)
866		err = buf.pos;
867
868	dput(dir);
869out:
870	return err;
871}
872
873static int reiserfs_check_acl(struct inode *inode, int mask)
874{
875	struct posix_acl *acl;
876	int error = -EAGAIN; /* do regular unix permission checks by default */
877
878	acl = reiserfs_get_acl(inode, ACL_TYPE_ACCESS);
879
880	if (acl) {
881		if (!IS_ERR(acl)) {
882			error = posix_acl_permission(inode, acl, mask);
883			posix_acl_release(acl);
884		} else if (PTR_ERR(acl) != -ENODATA)
885			error = PTR_ERR(acl);
886	}
887
888	return error;
889}
890
891static int create_privroot(struct dentry *dentry)
892{
893	int err;
894	struct inode *inode = dentry->d_parent->d_inode;
895	WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex));
896
897	err = xattr_mkdir(inode, dentry, 0700);
898	if (err || !dentry->d_inode) {
899		reiserfs_warning(dentry->d_sb, "jdm-20006",
900				 "xattrs/ACLs enabled and couldn't "
901				 "find/create .reiserfs_priv. "
902				 "Failing mount.");
903		return -EOPNOTSUPP;
904	}
905
906	dentry->d_inode->i_flags |= S_PRIVATE;
907	reiserfs_info(dentry->d_sb, "Created %s - reserved for xattr "
908		      "storage.\n", PRIVROOT_NAME);
909
910	return 0;
911}
912
913#else
914int __init reiserfs_xattr_register_handlers(void) { return 0; }
915void reiserfs_xattr_unregister_handlers(void) {}
916static int create_privroot(struct dentry *dentry) { return 0; }
917#endif
918
919/* Actual operations that are exported to VFS-land */
920const struct xattr_handler *reiserfs_xattr_handlers[] = {
921#ifdef CONFIG_REISERFS_FS_XATTR
922	&reiserfs_xattr_user_handler,
923	&reiserfs_xattr_trusted_handler,
924#endif
925#ifdef CONFIG_REISERFS_FS_SECURITY
926	&reiserfs_xattr_security_handler,
927#endif
928#ifdef CONFIG_REISERFS_FS_POSIX_ACL
929	&reiserfs_posix_acl_access_handler,
930	&reiserfs_posix_acl_default_handler,
931#endif
932	NULL
933};
934
935static int xattr_mount_check(struct super_block *s)
936{
937	/* We need generation numbers to ensure that the oid mapping is correct
938	 * v3.5 filesystems don't have them. */
939	if (old_format_only(s)) {
940		if (reiserfs_xattrs_optional(s)) {
941			/* Old format filesystem, but optional xattrs have
942			 * been enabled. Error out. */
943			reiserfs_warning(s, "jdm-2005",
944					 "xattrs/ACLs not supported "
945					 "on pre-v3.6 format filesystems. "
946					 "Failing mount.");
947			return -EOPNOTSUPP;
948		}
949	}
950
951	return 0;
952}
953
954int reiserfs_permission(struct inode *inode, int mask)
955{
956	/*
957	 * We don't do permission checks on the internal objects.
958	 * Permissions are determined by the "owning" object.
959	 */
960	if (IS_PRIVATE(inode))
961		return 0;
962
963#ifdef CONFIG_REISERFS_FS_XATTR
964	/*
965	 * Stat data v1 doesn't support ACLs.
966	 */
967	if (get_inode_sd_version(inode) != STAT_DATA_V1)
968		return generic_permission(inode, mask, reiserfs_check_acl);
969#endif
970	return generic_permission(inode, mask, NULL);
971}
972
973static int xattr_hide_revalidate(struct dentry *dentry, struct nameidata *nd)
974{
975	return -EPERM;
976}
977
978static const struct dentry_operations xattr_lookup_poison_ops = {
979	.d_revalidate = xattr_hide_revalidate,
980};
981
982int reiserfs_lookup_privroot(struct super_block *s)
983{
984	struct dentry *dentry;
985	int err = 0;
986
987	/* If we don't have the privroot located yet - go find it */
988	reiserfs_mutex_lock_safe(&s->s_root->d_inode->i_mutex, s);
989	dentry = lookup_one_len(PRIVROOT_NAME, s->s_root,
990				strlen(PRIVROOT_NAME));
991	if (!IS_ERR(dentry)) {
992		REISERFS_SB(s)->priv_root = dentry;
993		d_set_d_op(dentry, &xattr_lookup_poison_ops);
994		if (dentry->d_inode)
995			dentry->d_inode->i_flags |= S_PRIVATE;
996	} else
997		err = PTR_ERR(dentry);
998	mutex_unlock(&s->s_root->d_inode->i_mutex);
999
1000	return err;
1001}
1002
1003/* We need to take a copy of the mount flags since things like
1004 * MS_RDONLY don't get set until *after* we're called.
1005 * mount_flags != mount_options */
1006int reiserfs_xattr_init(struct super_block *s, int mount_flags)
1007{
1008	int err = 0;
1009	struct dentry *privroot = REISERFS_SB(s)->priv_root;
1010
1011	err = xattr_mount_check(s);
1012	if (err)
1013		goto error;
1014
1015	if (!privroot->d_inode && !(mount_flags & MS_RDONLY)) {
1016		reiserfs_mutex_lock_safe(&s->s_root->d_inode->i_mutex, s);
1017		err = create_privroot(REISERFS_SB(s)->priv_root);
1018		mutex_unlock(&s->s_root->d_inode->i_mutex);
1019	}
1020
1021	if (privroot->d_inode) {
1022		s->s_xattr = reiserfs_xattr_handlers;
1023		reiserfs_mutex_lock_safe(&privroot->d_inode->i_mutex, s);
1024		if (!REISERFS_SB(s)->xattr_root) {
1025			struct dentry *dentry;
1026			dentry = lookup_one_len(XAROOT_NAME, privroot,
1027						strlen(XAROOT_NAME));
1028			if (!IS_ERR(dentry))
1029				REISERFS_SB(s)->xattr_root = dentry;
1030			else
1031				err = PTR_ERR(dentry);
1032		}
1033		mutex_unlock(&privroot->d_inode->i_mutex);
1034	}
1035
1036error:
1037	if (err) {
1038		clear_bit(REISERFS_XATTRS_USER, &(REISERFS_SB(s)->s_mount_opt));
1039		clear_bit(REISERFS_POSIXACL, &(REISERFS_SB(s)->s_mount_opt));
1040	}
1041
1042	/* The super_block MS_POSIXACL must mirror the (no)acl mount option. */
1043	if (reiserfs_posixacl(s))
1044		s->s_flags |= MS_POSIXACL;
1045	else
1046		s->s_flags &= ~MS_POSIXACL;
1047
1048	return err;
1049}
1050