xattr.c revision 4c05141df57f4ffc1a9a28f1925434924179bfe4
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 "reiserfs.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 "xattr.h"
47#include "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, true);
66}
67#endif
68
69static int xattr_mkdir(struct inode *dir, struct dentry *dentry, umode_t 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	mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD);
85	error = dir->i_op->unlink(dir, dentry);
86	mutex_unlock(&dentry->d_inode->i_mutex);
87
88	if (!error)
89		d_delete(dentry);
90	return error;
91}
92
93static int xattr_rmdir(struct inode *dir, struct dentry *dentry)
94{
95	int error;
96	BUG_ON(!mutex_is_locked(&dir->i_mutex));
97
98	mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD);
99	error = dir->i_op->rmdir(dir, dentry);
100	if (!error)
101		dentry->d_inode->i_flags |= S_DEAD;
102	mutex_unlock(&dentry->d_inode->i_mutex);
103	if (!error)
104		d_delete(dentry);
105
106	return error;
107}
108
109#define xattr_may_create(flags)	(!flags || flags & XATTR_CREATE)
110
111static struct dentry *open_xa_root(struct super_block *sb, int flags)
112{
113	struct dentry *privroot = REISERFS_SB(sb)->priv_root;
114	struct dentry *xaroot;
115	if (!privroot->d_inode)
116		return ERR_PTR(-ENODATA);
117
118	mutex_lock_nested(&privroot->d_inode->i_mutex, I_MUTEX_XATTR);
119
120	xaroot = dget(REISERFS_SB(sb)->xattr_root);
121	if (!xaroot)
122		xaroot = ERR_PTR(-ENODATA);
123	else if (!xaroot->d_inode) {
124		int err = -ENODATA;
125		if (xattr_may_create(flags))
126			err = xattr_mkdir(privroot->d_inode, xaroot, 0700);
127		if (err) {
128			dput(xaroot);
129			xaroot = ERR_PTR(err);
130		}
131	}
132
133	mutex_unlock(&privroot->d_inode->i_mutex);
134	return xaroot;
135}
136
137static struct dentry *open_xa_dir(const struct inode *inode, int flags)
138{
139	struct dentry *xaroot, *xadir;
140	char namebuf[17];
141
142	xaroot = open_xa_root(inode->i_sb, flags);
143	if (IS_ERR(xaroot))
144		return xaroot;
145
146	snprintf(namebuf, sizeof(namebuf), "%X.%X",
147		 le32_to_cpu(INODE_PKEY(inode)->k_objectid),
148		 inode->i_generation);
149
150	mutex_lock_nested(&xaroot->d_inode->i_mutex, I_MUTEX_XATTR);
151
152	xadir = lookup_one_len(namebuf, xaroot, strlen(namebuf));
153	if (!IS_ERR(xadir) && !xadir->d_inode) {
154		int err = -ENODATA;
155		if (xattr_may_create(flags))
156			err = xattr_mkdir(xaroot->d_inode, xadir, 0700);
157		if (err) {
158			dput(xadir);
159			xadir = ERR_PTR(err);
160		}
161	}
162
163	mutex_unlock(&xaroot->d_inode->i_mutex);
164	dput(xaroot);
165	return xadir;
166}
167
168/* The following are side effects of other operations that aren't explicitly
169 * modifying extended attributes. This includes operations such as permissions
170 * or ownership changes, object deletions, etc. */
171struct reiserfs_dentry_buf {
172	struct dir_context ctx;
173	struct dentry *xadir;
174	int count;
175	struct dentry *dentries[8];
176};
177
178static int
179fill_with_dentries(void *buf, const char *name, int namelen, loff_t offset,
180		    u64 ino, unsigned int d_type)
181{
182	struct reiserfs_dentry_buf *dbuf = buf;
183	struct dentry *dentry;
184	WARN_ON_ONCE(!mutex_is_locked(&dbuf->xadir->d_inode->i_mutex));
185
186	if (dbuf->count == ARRAY_SIZE(dbuf->dentries))
187		return -ENOSPC;
188
189	if (name[0] == '.' && (namelen < 2 ||
190			       (namelen == 2 && name[1] == '.')))
191		return 0;
192
193	dentry = lookup_one_len(name, dbuf->xadir, namelen);
194	if (IS_ERR(dentry)) {
195		return PTR_ERR(dentry);
196	} else if (!dentry->d_inode) {
197		/* A directory entry exists, but no file? */
198		reiserfs_error(dentry->d_sb, "xattr-20003",
199			       "Corrupted directory: xattr %s listed but "
200			       "not found for file %s.\n",
201			       dentry->d_name.name, dbuf->xadir->d_name.name);
202		dput(dentry);
203		return -EIO;
204	}
205
206	dbuf->dentries[dbuf->count++] = dentry;
207	return 0;
208}
209
210static void
211cleanup_dentry_buf(struct reiserfs_dentry_buf *buf)
212{
213	int i;
214	for (i = 0; i < buf->count; i++)
215		if (buf->dentries[i])
216			dput(buf->dentries[i]);
217}
218
219static int reiserfs_for_each_xattr(struct inode *inode,
220				   int (*action)(struct dentry *, void *),
221				   void *data)
222{
223	struct dentry *dir;
224	int i, err = 0;
225	struct reiserfs_dentry_buf buf = {
226		.ctx.actor = fill_with_dentries,
227	};
228
229	/* Skip out, an xattr has no xattrs associated with it */
230	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
231		return 0;
232
233	dir = open_xa_dir(inode, XATTR_REPLACE);
234	if (IS_ERR(dir)) {
235		err = PTR_ERR(dir);
236		goto out;
237	} else if (!dir->d_inode) {
238		err = 0;
239		goto out_dir;
240	}
241
242	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR);
243
244	buf.xadir = dir;
245	while (1) {
246		err = reiserfs_readdir_inode(dir->d_inode, &buf.ctx);
247		if (err)
248			break;
249		if (!buf.count)
250			break;
251		for (i = 0; !err && i < buf.count && buf.dentries[i]; i++) {
252			struct dentry *dentry = buf.dentries[i];
253
254			if (!S_ISDIR(dentry->d_inode->i_mode))
255				err = action(dentry, data);
256
257			dput(dentry);
258			buf.dentries[i] = NULL;
259		}
260		if (err)
261			break;
262		buf.count = 0;
263	}
264	mutex_unlock(&dir->d_inode->i_mutex);
265
266	cleanup_dentry_buf(&buf);
267
268	if (!err) {
269		/* We start a transaction here to avoid a ABBA situation
270		 * between the xattr root's i_mutex and the journal lock.
271		 * This doesn't incur much additional overhead since the
272		 * new transaction will just nest inside the
273		 * outer transaction. */
274		int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 +
275			     4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
276		struct reiserfs_transaction_handle th;
277		reiserfs_write_lock(inode->i_sb);
278		err = journal_begin(&th, inode->i_sb, blocks);
279		reiserfs_write_unlock(inode->i_sb);
280		if (!err) {
281			int jerror;
282			mutex_lock_nested(&dir->d_parent->d_inode->i_mutex,
283					  I_MUTEX_XATTR);
284			err = action(dir, data);
285			reiserfs_write_lock(inode->i_sb);
286			jerror = journal_end(&th, inode->i_sb, blocks);
287			reiserfs_write_unlock(inode->i_sb);
288			mutex_unlock(&dir->d_parent->d_inode->i_mutex);
289			err = jerror ?: err;
290		}
291	}
292out_dir:
293	dput(dir);
294out:
295	/* -ENODATA isn't an error */
296	if (err == -ENODATA)
297		err = 0;
298	return err;
299}
300
301static int delete_one_xattr(struct dentry *dentry, void *data)
302{
303	struct inode *dir = dentry->d_parent->d_inode;
304
305	/* This is the xattr dir, handle specially. */
306	if (S_ISDIR(dentry->d_inode->i_mode))
307		return xattr_rmdir(dir, dentry);
308
309	return xattr_unlink(dir, dentry);
310}
311
312static int chown_one_xattr(struct dentry *dentry, void *data)
313{
314	struct iattr *attrs = data;
315	int ia_valid = attrs->ia_valid;
316	int err;
317
318	/*
319	 * We only want the ownership bits. Otherwise, we'll do
320	 * things like change a directory to a regular file if
321	 * ATTR_MODE is set.
322	 */
323	attrs->ia_valid &= (ATTR_UID|ATTR_GID);
324	err = reiserfs_setattr(dentry, attrs);
325	attrs->ia_valid = ia_valid;
326
327	return err;
328}
329
330/* No i_mutex, but the inode is unconnected. */
331int reiserfs_delete_xattrs(struct inode *inode)
332{
333	int err = reiserfs_for_each_xattr(inode, delete_one_xattr, NULL);
334	if (err)
335		reiserfs_warning(inode->i_sb, "jdm-20004",
336				 "Couldn't delete all xattrs (%d)\n", err);
337	return err;
338}
339
340/* inode->i_mutex: down */
341int reiserfs_chown_xattrs(struct inode *inode, struct iattr *attrs)
342{
343	int err = reiserfs_for_each_xattr(inode, chown_one_xattr, attrs);
344	if (err)
345		reiserfs_warning(inode->i_sb, "jdm-20007",
346				 "Couldn't chown all xattrs (%d)\n", err);
347	return err;
348}
349
350#ifdef CONFIG_REISERFS_FS_XATTR
351/* Returns a dentry corresponding to a specific extended attribute file
352 * for the inode. If flags allow, the file is created. Otherwise, a
353 * valid or negative dentry, or an error is returned. */
354static struct dentry *xattr_lookup(struct inode *inode, const char *name,
355				    int flags)
356{
357	struct dentry *xadir, *xafile;
358	int err = 0;
359
360	xadir = open_xa_dir(inode, flags);
361	if (IS_ERR(xadir))
362		return ERR_CAST(xadir);
363
364	mutex_lock_nested(&xadir->d_inode->i_mutex, I_MUTEX_XATTR);
365	xafile = lookup_one_len(name, xadir, strlen(name));
366	if (IS_ERR(xafile)) {
367		err = PTR_ERR(xafile);
368		goto out;
369	}
370
371	if (xafile->d_inode && (flags & XATTR_CREATE))
372		err = -EEXIST;
373
374	if (!xafile->d_inode) {
375		err = -ENODATA;
376		if (xattr_may_create(flags))
377			err = xattr_create(xadir->d_inode, xafile,
378					      0700|S_IFREG);
379	}
380
381	if (err)
382		dput(xafile);
383out:
384	mutex_unlock(&xadir->d_inode->i_mutex);
385	dput(xadir);
386	if (err)
387		return ERR_PTR(err);
388	return xafile;
389}
390
391/* Internal operations on file data */
392static inline void reiserfs_put_page(struct page *page)
393{
394	kunmap(page);
395	page_cache_release(page);
396}
397
398static struct page *reiserfs_get_page(struct inode *dir, size_t n)
399{
400	struct address_space *mapping = dir->i_mapping;
401	struct page *page;
402	/* We can deadlock if we try to free dentries,
403	   and an unlink/rmdir has just occurred - GFP_NOFS avoids this */
404	mapping_set_gfp_mask(mapping, GFP_NOFS);
405	page = read_mapping_page(mapping, n >> PAGE_CACHE_SHIFT, NULL);
406	if (!IS_ERR(page)) {
407		kmap(page);
408		if (PageError(page))
409			goto fail;
410	}
411	return page;
412
413      fail:
414	reiserfs_put_page(page);
415	return ERR_PTR(-EIO);
416}
417
418static inline __u32 xattr_hash(const char *msg, int len)
419{
420	return csum_partial(msg, len, 0);
421}
422
423int reiserfs_commit_write(struct file *f, struct page *page,
424			  unsigned from, unsigned to);
425
426static void update_ctime(struct inode *inode)
427{
428	struct timespec now = current_fs_time(inode->i_sb);
429	if (inode_unhashed(inode) || !inode->i_nlink ||
430	    timespec_equal(&inode->i_ctime, &now))
431		return;
432
433	inode->i_ctime = CURRENT_TIME_SEC;
434	mark_inode_dirty(inode);
435}
436
437static int lookup_and_delete_xattr(struct inode *inode, const char *name)
438{
439	int err = 0;
440	struct dentry *dentry, *xadir;
441
442	xadir = open_xa_dir(inode, XATTR_REPLACE);
443	if (IS_ERR(xadir))
444		return PTR_ERR(xadir);
445
446	mutex_lock_nested(&xadir->d_inode->i_mutex, I_MUTEX_XATTR);
447	dentry = lookup_one_len(name, xadir, strlen(name));
448	if (IS_ERR(dentry)) {
449		err = PTR_ERR(dentry);
450		goto out_dput;
451	}
452
453	if (dentry->d_inode) {
454		err = xattr_unlink(xadir->d_inode, dentry);
455		update_ctime(inode);
456	}
457
458	dput(dentry);
459out_dput:
460	mutex_unlock(&xadir->d_inode->i_mutex);
461	dput(xadir);
462	return err;
463}
464
465
466/* Generic extended attribute operations that can be used by xa plugins */
467
468/*
469 * inode->i_mutex: down
470 */
471int
472reiserfs_xattr_set_handle(struct reiserfs_transaction_handle *th,
473			  struct inode *inode, const char *name,
474			  const void *buffer, size_t buffer_size, int flags)
475{
476	int err = 0;
477	struct dentry *dentry;
478	struct page *page;
479	char *data;
480	size_t file_pos = 0;
481	size_t buffer_pos = 0;
482	size_t new_size;
483	__u32 xahash = 0;
484
485	if (get_inode_sd_version(inode) == STAT_DATA_V1)
486		return -EOPNOTSUPP;
487
488	if (!buffer) {
489		err = lookup_and_delete_xattr(inode, name);
490		return err;
491	}
492
493	dentry = xattr_lookup(inode, name, flags);
494	if (IS_ERR(dentry))
495		return PTR_ERR(dentry);
496
497	down_write(&REISERFS_I(inode)->i_xattr_sem);
498
499	xahash = xattr_hash(buffer, buffer_size);
500	while (buffer_pos < buffer_size || buffer_pos == 0) {
501		size_t chunk;
502		size_t skip = 0;
503		size_t page_offset = (file_pos & (PAGE_CACHE_SIZE - 1));
504		if (buffer_size - buffer_pos > PAGE_CACHE_SIZE)
505			chunk = PAGE_CACHE_SIZE;
506		else
507			chunk = buffer_size - buffer_pos;
508
509		page = reiserfs_get_page(dentry->d_inode, file_pos);
510		if (IS_ERR(page)) {
511			err = PTR_ERR(page);
512			goto out_unlock;
513		}
514
515		lock_page(page);
516		data = page_address(page);
517
518		if (file_pos == 0) {
519			struct reiserfs_xattr_header *rxh;
520			skip = file_pos = sizeof(struct reiserfs_xattr_header);
521			if (chunk + skip > PAGE_CACHE_SIZE)
522				chunk = PAGE_CACHE_SIZE - skip;
523			rxh = (struct reiserfs_xattr_header *)data;
524			rxh->h_magic = cpu_to_le32(REISERFS_XATTR_MAGIC);
525			rxh->h_hash = cpu_to_le32(xahash);
526		}
527
528		reiserfs_write_lock(inode->i_sb);
529		err = __reiserfs_write_begin(page, page_offset, chunk + skip);
530		if (!err) {
531			if (buffer)
532				memcpy(data + skip, buffer + buffer_pos, chunk);
533			err = reiserfs_commit_write(NULL, page, page_offset,
534						    page_offset + chunk +
535						    skip);
536		}
537		reiserfs_write_unlock(inode->i_sb);
538		unlock_page(page);
539		reiserfs_put_page(page);
540		buffer_pos += chunk;
541		file_pos += chunk;
542		skip = 0;
543		if (err || buffer_size == 0 || !buffer)
544			break;
545	}
546
547	new_size = buffer_size + sizeof(struct reiserfs_xattr_header);
548	if (!err && new_size < i_size_read(dentry->d_inode)) {
549		struct iattr newattrs = {
550			.ia_ctime = current_fs_time(inode->i_sb),
551			.ia_size = new_size,
552			.ia_valid = ATTR_SIZE | ATTR_CTIME,
553		};
554
555		mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_XATTR);
556		inode_dio_wait(dentry->d_inode);
557
558		err = reiserfs_setattr(dentry, &newattrs);
559		mutex_unlock(&dentry->d_inode->i_mutex);
560	} else
561		update_ctime(inode);
562out_unlock:
563	up_write(&REISERFS_I(inode)->i_xattr_sem);
564	dput(dentry);
565	return err;
566}
567
568/* We need to start a transaction to maintain lock ordering */
569int reiserfs_xattr_set(struct inode *inode, const char *name,
570		       const void *buffer, size_t buffer_size, int flags)
571{
572
573	struct reiserfs_transaction_handle th;
574	int error, error2;
575	size_t jbegin_count = reiserfs_xattr_nblocks(inode, buffer_size);
576
577	if (!(flags & XATTR_REPLACE))
578		jbegin_count += reiserfs_xattr_jcreate_nblocks(inode);
579
580	reiserfs_write_lock(inode->i_sb);
581	error = journal_begin(&th, inode->i_sb, jbegin_count);
582	reiserfs_write_unlock(inode->i_sb);
583	if (error) {
584		return error;
585	}
586
587	error = reiserfs_xattr_set_handle(&th, inode, name,
588					  buffer, buffer_size, flags);
589
590	reiserfs_write_lock(inode->i_sb);
591	error2 = journal_end(&th, inode->i_sb, jbegin_count);
592	reiserfs_write_unlock(inode->i_sb);
593	if (error == 0)
594		error = error2;
595
596	return error;
597}
598
599/*
600 * inode->i_mutex: down
601 */
602int
603reiserfs_xattr_get(struct inode *inode, const char *name, void *buffer,
604		   size_t buffer_size)
605{
606	ssize_t err = 0;
607	struct dentry *dentry;
608	size_t isize;
609	size_t file_pos = 0;
610	size_t buffer_pos = 0;
611	struct page *page;
612	__u32 hash = 0;
613
614	if (name == NULL)
615		return -EINVAL;
616
617	/* We can't have xattrs attached to v1 items since they don't have
618	 * generation numbers */
619	if (get_inode_sd_version(inode) == STAT_DATA_V1)
620		return -EOPNOTSUPP;
621
622	dentry = xattr_lookup(inode, name, XATTR_REPLACE);
623	if (IS_ERR(dentry)) {
624		err = PTR_ERR(dentry);
625		goto out;
626	}
627
628	down_read(&REISERFS_I(inode)->i_xattr_sem);
629
630	isize = i_size_read(dentry->d_inode);
631
632	/* Just return the size needed */
633	if (buffer == NULL) {
634		err = isize - sizeof(struct reiserfs_xattr_header);
635		goto out_unlock;
636	}
637
638	if (buffer_size < isize - sizeof(struct reiserfs_xattr_header)) {
639		err = -ERANGE;
640		goto out_unlock;
641	}
642
643	while (file_pos < isize) {
644		size_t chunk;
645		char *data;
646		size_t skip = 0;
647		if (isize - file_pos > PAGE_CACHE_SIZE)
648			chunk = PAGE_CACHE_SIZE;
649		else
650			chunk = isize - file_pos;
651
652		page = reiserfs_get_page(dentry->d_inode, file_pos);
653		if (IS_ERR(page)) {
654			err = PTR_ERR(page);
655			goto out_unlock;
656		}
657
658		lock_page(page);
659		data = page_address(page);
660		if (file_pos == 0) {
661			struct reiserfs_xattr_header *rxh =
662			    (struct reiserfs_xattr_header *)data;
663			skip = file_pos = sizeof(struct reiserfs_xattr_header);
664			chunk -= skip;
665			/* Magic doesn't match up.. */
666			if (rxh->h_magic != cpu_to_le32(REISERFS_XATTR_MAGIC)) {
667				unlock_page(page);
668				reiserfs_put_page(page);
669				reiserfs_warning(inode->i_sb, "jdm-20001",
670						 "Invalid magic for xattr (%s) "
671						 "associated with %k", name,
672						 INODE_PKEY(inode));
673				err = -EIO;
674				goto out_unlock;
675			}
676			hash = le32_to_cpu(rxh->h_hash);
677		}
678		memcpy(buffer + buffer_pos, data + skip, chunk);
679		unlock_page(page);
680		reiserfs_put_page(page);
681		file_pos += chunk;
682		buffer_pos += chunk;
683		skip = 0;
684	}
685	err = isize - sizeof(struct reiserfs_xattr_header);
686
687	if (xattr_hash(buffer, isize - sizeof(struct reiserfs_xattr_header)) !=
688	    hash) {
689		reiserfs_warning(inode->i_sb, "jdm-20002",
690				 "Invalid hash for xattr (%s) associated "
691				 "with %k", name, INODE_PKEY(inode));
692		err = -EIO;
693	}
694
695out_unlock:
696	up_read(&REISERFS_I(inode)->i_xattr_sem);
697	dput(dentry);
698
699out:
700	return err;
701}
702
703/*
704 * In order to implement different sets of xattr operations for each xattr
705 * prefix with the generic xattr API, a filesystem should create a
706 * null-terminated array of struct xattr_handler (one for each prefix) and
707 * hang a pointer to it off of the s_xattr field of the superblock.
708 *
709 * The generic_fooxattr() functions will use this list to dispatch xattr
710 * operations to the correct xattr_handler.
711 */
712#define for_each_xattr_handler(handlers, handler)		\
713		for ((handler) = *(handlers)++;			\
714			(handler) != NULL;			\
715			(handler) = *(handlers)++)
716
717/* This is the implementation for the xattr plugin infrastructure */
718static inline const struct xattr_handler *
719find_xattr_handler_prefix(const struct xattr_handler **handlers,
720			   const char *name)
721{
722	const struct xattr_handler *xah;
723
724	if (!handlers)
725		return NULL;
726
727	for_each_xattr_handler(handlers, xah) {
728		if (strncmp(xah->prefix, name, strlen(xah->prefix)) == 0)
729			break;
730	}
731
732	return xah;
733}
734
735
736/*
737 * Inode operation getxattr()
738 */
739ssize_t
740reiserfs_getxattr(struct dentry * dentry, const char *name, void *buffer,
741		  size_t size)
742{
743	const struct xattr_handler *handler;
744
745	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
746
747	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
748		return -EOPNOTSUPP;
749
750	return handler->get(dentry, name, buffer, size, handler->flags);
751}
752
753/*
754 * Inode operation setxattr()
755 *
756 * dentry->d_inode->i_mutex down
757 */
758int
759reiserfs_setxattr(struct dentry *dentry, const char *name, const void *value,
760		  size_t size, int flags)
761{
762	const struct xattr_handler *handler;
763
764	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
765
766	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
767		return -EOPNOTSUPP;
768
769	return handler->set(dentry, name, value, size, flags, handler->flags);
770}
771
772/*
773 * Inode operation removexattr()
774 *
775 * dentry->d_inode->i_mutex down
776 */
777int reiserfs_removexattr(struct dentry *dentry, const char *name)
778{
779	const struct xattr_handler *handler;
780	handler = find_xattr_handler_prefix(dentry->d_sb->s_xattr, name);
781
782	if (!handler || get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
783		return -EOPNOTSUPP;
784
785	return handler->set(dentry, name, NULL, 0, XATTR_REPLACE, handler->flags);
786}
787
788struct listxattr_buf {
789	struct dir_context ctx;
790	size_t size;
791	size_t pos;
792	char *buf;
793	struct dentry *dentry;
794};
795
796static int listxattr_filler(void *buf, const char *name, int namelen,
797			    loff_t offset, u64 ino, unsigned int d_type)
798{
799	struct listxattr_buf *b = (struct listxattr_buf *)buf;
800	size_t size;
801	if (name[0] != '.' ||
802	    (namelen != 1 && (name[1] != '.' || namelen != 2))) {
803		const struct xattr_handler *handler;
804		handler = find_xattr_handler_prefix(b->dentry->d_sb->s_xattr,
805						    name);
806		if (!handler)	/* Unsupported xattr name */
807			return 0;
808		if (b->buf) {
809			size = handler->list(b->dentry, b->buf + b->pos,
810					 b->size, name, namelen,
811					 handler->flags);
812			if (size > b->size)
813				return -ERANGE;
814		} else {
815			size = handler->list(b->dentry, NULL, 0, name,
816					     namelen, handler->flags);
817		}
818
819		b->pos += size;
820	}
821	return 0;
822}
823
824/*
825 * Inode operation listxattr()
826 *
827 * We totally ignore the generic listxattr here because it would be stupid
828 * not to. Since the xattrs are organized in a directory, we can just
829 * readdir to find them.
830 */
831ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size)
832{
833	struct dentry *dir;
834	int err = 0;
835	struct listxattr_buf buf = {
836		.ctx.actor = listxattr_filler,
837		.dentry = dentry,
838		.buf = buffer,
839		.size = buffer ? size : 0,
840	};
841
842	if (!dentry->d_inode)
843		return -EINVAL;
844
845	if (!dentry->d_sb->s_xattr ||
846	    get_inode_sd_version(dentry->d_inode) == STAT_DATA_V1)
847		return -EOPNOTSUPP;
848
849	dir = open_xa_dir(dentry->d_inode, XATTR_REPLACE);
850	if (IS_ERR(dir)) {
851		err = PTR_ERR(dir);
852		if (err == -ENODATA)
853			err = 0;  /* Not an error if there aren't any xattrs */
854		goto out;
855	}
856
857	mutex_lock_nested(&dir->d_inode->i_mutex, I_MUTEX_XATTR);
858	err = reiserfs_readdir_inode(dir->d_inode, &buf.ctx);
859	mutex_unlock(&dir->d_inode->i_mutex);
860
861	if (!err)
862		err = buf.pos;
863
864	dput(dir);
865out:
866	return err;
867}
868
869static int create_privroot(struct dentry *dentry)
870{
871	int err;
872	struct inode *inode = dentry->d_parent->d_inode;
873	WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex));
874
875	err = xattr_mkdir(inode, dentry, 0700);
876	if (err || !dentry->d_inode) {
877		reiserfs_warning(dentry->d_sb, "jdm-20006",
878				 "xattrs/ACLs enabled and couldn't "
879				 "find/create .reiserfs_priv. "
880				 "Failing mount.");
881		return -EOPNOTSUPP;
882	}
883
884	dentry->d_inode->i_flags |= S_PRIVATE;
885	reiserfs_info(dentry->d_sb, "Created %s - reserved for xattr "
886		      "storage.\n", PRIVROOT_NAME);
887
888	return 0;
889}
890
891#else
892int __init reiserfs_xattr_register_handlers(void) { return 0; }
893void reiserfs_xattr_unregister_handlers(void) {}
894static int create_privroot(struct dentry *dentry) { return 0; }
895#endif
896
897/* Actual operations that are exported to VFS-land */
898static const struct xattr_handler *reiserfs_xattr_handlers[] = {
899#ifdef CONFIG_REISERFS_FS_XATTR
900	&reiserfs_xattr_user_handler,
901	&reiserfs_xattr_trusted_handler,
902#endif
903#ifdef CONFIG_REISERFS_FS_SECURITY
904	&reiserfs_xattr_security_handler,
905#endif
906#ifdef CONFIG_REISERFS_FS_POSIX_ACL
907	&reiserfs_posix_acl_access_handler,
908	&reiserfs_posix_acl_default_handler,
909#endif
910	NULL
911};
912
913static int xattr_mount_check(struct super_block *s)
914{
915	/* We need generation numbers to ensure that the oid mapping is correct
916	 * v3.5 filesystems don't have them. */
917	if (old_format_only(s)) {
918		if (reiserfs_xattrs_optional(s)) {
919			/* Old format filesystem, but optional xattrs have
920			 * been enabled. Error out. */
921			reiserfs_warning(s, "jdm-2005",
922					 "xattrs/ACLs not supported "
923					 "on pre-v3.6 format filesystems. "
924					 "Failing mount.");
925			return -EOPNOTSUPP;
926		}
927	}
928
929	return 0;
930}
931
932int reiserfs_permission(struct inode *inode, int mask)
933{
934	/*
935	 * We don't do permission checks on the internal objects.
936	 * Permissions are determined by the "owning" object.
937	 */
938	if (IS_PRIVATE(inode))
939		return 0;
940
941	return generic_permission(inode, mask);
942}
943
944static int xattr_hide_revalidate(struct dentry *dentry, unsigned int flags)
945{
946	return -EPERM;
947}
948
949static const struct dentry_operations xattr_lookup_poison_ops = {
950	.d_revalidate = xattr_hide_revalidate,
951};
952
953int reiserfs_lookup_privroot(struct super_block *s)
954{
955	struct dentry *dentry;
956	int err = 0;
957
958	/* If we don't have the privroot located yet - go find it */
959	mutex_lock(&s->s_root->d_inode->i_mutex);
960	dentry = lookup_one_len(PRIVROOT_NAME, s->s_root,
961				strlen(PRIVROOT_NAME));
962	if (!IS_ERR(dentry)) {
963		REISERFS_SB(s)->priv_root = dentry;
964		d_set_d_op(dentry, &xattr_lookup_poison_ops);
965		if (dentry->d_inode)
966			dentry->d_inode->i_flags |= S_PRIVATE;
967	} else
968		err = PTR_ERR(dentry);
969	mutex_unlock(&s->s_root->d_inode->i_mutex);
970
971	return err;
972}
973
974/* We need to take a copy of the mount flags since things like
975 * MS_RDONLY don't get set until *after* we're called.
976 * mount_flags != mount_options */
977int reiserfs_xattr_init(struct super_block *s, int mount_flags)
978{
979	int err = 0;
980	struct dentry *privroot = REISERFS_SB(s)->priv_root;
981
982	err = xattr_mount_check(s);
983	if (err)
984		goto error;
985
986	if (!privroot->d_inode && !(mount_flags & MS_RDONLY)) {
987		mutex_lock(&s->s_root->d_inode->i_mutex);
988		err = create_privroot(REISERFS_SB(s)->priv_root);
989		mutex_unlock(&s->s_root->d_inode->i_mutex);
990	}
991
992	if (privroot->d_inode) {
993		s->s_xattr = reiserfs_xattr_handlers;
994		mutex_lock(&privroot->d_inode->i_mutex);
995		if (!REISERFS_SB(s)->xattr_root) {
996			struct dentry *dentry;
997			dentry = lookup_one_len(XAROOT_NAME, privroot,
998						strlen(XAROOT_NAME));
999			if (!IS_ERR(dentry))
1000				REISERFS_SB(s)->xattr_root = dentry;
1001			else
1002				err = PTR_ERR(dentry);
1003		}
1004		mutex_unlock(&privroot->d_inode->i_mutex);
1005	}
1006
1007error:
1008	if (err) {
1009		clear_bit(REISERFS_XATTRS_USER, &(REISERFS_SB(s)->s_mount_opt));
1010		clear_bit(REISERFS_POSIXACL, &(REISERFS_SB(s)->s_mount_opt));
1011	}
1012
1013	/* The super_block MS_POSIXACL must mirror the (no)acl mount option. */
1014	if (reiserfs_posixacl(s))
1015		s->s_flags |= MS_POSIXACL;
1016	else
1017		s->s_flags &= ~MS_POSIXACL;
1018
1019	return err;
1020}
1021