xref: /openbmc/linux/fs/reiserfs/xattr.c (revision 18afb028)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/fs/reiserfs/xattr.c
4  *
5  * Copyright (c) 2002 by Jeff Mahoney, <jeffm@suse.com>
6  *
7  */
8 
9 /*
10  * In order to implement EA/ACLs in a clean, backwards compatible manner,
11  * they are implemented as files in a "private" directory.
12  * Each EA is in it's own file, with the directory layout like so (/ is assumed
13  * to be relative to fs root). Inside the /.reiserfs_priv/xattrs directory,
14  * directories named using the capital-hex form of the objectid and
15  * generation number are used. Inside each directory are individual files
16  * named with the name of the extended attribute.
17  *
18  * So, for objectid 12648430, we could have:
19  * /.reiserfs_priv/xattrs/C0FFEE.0/system.posix_acl_access
20  * /.reiserfs_priv/xattrs/C0FFEE.0/system.posix_acl_default
21  * /.reiserfs_priv/xattrs/C0FFEE.0/user.Content-Type
22  * .. or similar.
23  *
24  * The file contents are the text of the EA. The size is known based on the
25  * stat data describing the file.
26  *
27  * In the case of system.posix_acl_access and system.posix_acl_default, since
28  * these are special cases for filesystem ACLs, they are interpreted by the
29  * kernel, in addition, they are negatively and positively cached and attached
30  * to the inode so that unnecessary lookups are avoided.
31  *
32  * Locking works like so:
33  * Directory components (xattr root, xattr dir) are protectd by their i_mutex.
34  * The xattrs themselves are protected by the xattr_sem.
35  */
36 
37 #include "reiserfs.h"
38 #include <linux/capability.h>
39 #include <linux/dcache.h>
40 #include <linux/namei.h>
41 #include <linux/errno.h>
42 #include <linux/gfp.h>
43 #include <linux/fs.h>
44 #include <linux/file.h>
45 #include <linux/pagemap.h>
46 #include <linux/xattr.h>
47 #include "xattr.h"
48 #include "acl.h"
49 #include <linux/uaccess.h>
50 #include <net/checksum.h>
51 #include <linux/stat.h>
52 #include <linux/quotaops.h>
53 #include <linux/security.h>
54 #include <linux/posix_acl_xattr.h>
55 #include <linux/xattr.h>
56 
57 #define PRIVROOT_NAME ".reiserfs_priv"
58 #define XAROOT_NAME   "xattrs"
59 
60 
61 /*
62  * Helpers for inode ops. We do this so that we don't have all the VFS
63  * overhead and also for proper i_mutex annotation.
64  * dir->i_mutex must be held for all of them.
65  */
66 #ifdef CONFIG_REISERFS_FS_XATTR
67 static int xattr_create(struct inode *dir, struct dentry *dentry, int mode)
68 {
69 	BUG_ON(!inode_is_locked(dir));
70 	return dir->i_op->create(&nop_mnt_idmap, dir, dentry, mode, true);
71 }
72 #endif
73 
74 static int xattr_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
75 {
76 	BUG_ON(!inode_is_locked(dir));
77 	return dir->i_op->mkdir(&nop_mnt_idmap, dir, dentry, mode);
78 }
79 
80 /*
81  * We use I_MUTEX_CHILD here to silence lockdep. It's safe because xattr
82  * mutation ops aren't called during rename or splace, which are the
83  * only other users of I_MUTEX_CHILD. It violates the ordering, but that's
84  * better than allocating another subclass just for this code.
85  */
86 static int xattr_unlink(struct inode *dir, struct dentry *dentry)
87 {
88 	int error;
89 
90 	BUG_ON(!inode_is_locked(dir));
91 
92 	inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
93 	error = dir->i_op->unlink(dir, dentry);
94 	inode_unlock(d_inode(dentry));
95 
96 	if (!error)
97 		d_delete(dentry);
98 	return error;
99 }
100 
101 static int xattr_rmdir(struct inode *dir, struct dentry *dentry)
102 {
103 	int error;
104 
105 	BUG_ON(!inode_is_locked(dir));
106 
107 	inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
108 	error = dir->i_op->rmdir(dir, dentry);
109 	if (!error)
110 		d_inode(dentry)->i_flags |= S_DEAD;
111 	inode_unlock(d_inode(dentry));
112 	if (!error)
113 		d_delete(dentry);
114 
115 	return error;
116 }
117 
118 #define xattr_may_create(flags)	(!flags || flags & XATTR_CREATE)
119 
120 static struct dentry *open_xa_root(struct super_block *sb, int flags)
121 {
122 	struct dentry *privroot = REISERFS_SB(sb)->priv_root;
123 	struct dentry *xaroot;
124 
125 	if (d_really_is_negative(privroot))
126 		return ERR_PTR(-EOPNOTSUPP);
127 
128 	inode_lock_nested(d_inode(privroot), I_MUTEX_XATTR);
129 
130 	xaroot = dget(REISERFS_SB(sb)->xattr_root);
131 	if (!xaroot)
132 		xaroot = ERR_PTR(-EOPNOTSUPP);
133 	else if (d_really_is_negative(xaroot)) {
134 		int err = -ENODATA;
135 
136 		if (xattr_may_create(flags))
137 			err = xattr_mkdir(d_inode(privroot), xaroot, 0700);
138 		if (err) {
139 			dput(xaroot);
140 			xaroot = ERR_PTR(err);
141 		}
142 	}
143 
144 	inode_unlock(d_inode(privroot));
145 	return xaroot;
146 }
147 
148 static struct dentry *open_xa_dir(const struct inode *inode, int flags)
149 {
150 	struct dentry *xaroot, *xadir;
151 	char namebuf[17];
152 
153 	xaroot = open_xa_root(inode->i_sb, flags);
154 	if (IS_ERR(xaroot))
155 		return xaroot;
156 
157 	snprintf(namebuf, sizeof(namebuf), "%X.%X",
158 		 le32_to_cpu(INODE_PKEY(inode)->k_objectid),
159 		 inode->i_generation);
160 
161 	inode_lock_nested(d_inode(xaroot), I_MUTEX_XATTR);
162 
163 	xadir = lookup_one_len(namebuf, xaroot, strlen(namebuf));
164 	if (!IS_ERR(xadir) && d_really_is_negative(xadir)) {
165 		int err = -ENODATA;
166 
167 		if (xattr_may_create(flags))
168 			err = xattr_mkdir(d_inode(xaroot), xadir, 0700);
169 		if (err) {
170 			dput(xadir);
171 			xadir = ERR_PTR(err);
172 		}
173 	}
174 
175 	inode_unlock(d_inode(xaroot));
176 	dput(xaroot);
177 	return xadir;
178 }
179 
180 /*
181  * The following are side effects of other operations that aren't explicitly
182  * modifying extended attributes. This includes operations such as permissions
183  * or ownership changes, object deletions, etc.
184  */
185 struct reiserfs_dentry_buf {
186 	struct dir_context ctx;
187 	struct dentry *xadir;
188 	int count;
189 	int err;
190 	struct dentry *dentries[8];
191 };
192 
193 static bool
194 fill_with_dentries(struct dir_context *ctx, const char *name, int namelen,
195 		   loff_t offset, u64 ino, unsigned int d_type)
196 {
197 	struct reiserfs_dentry_buf *dbuf =
198 		container_of(ctx, struct reiserfs_dentry_buf, ctx);
199 	struct dentry *dentry;
200 
201 	WARN_ON_ONCE(!inode_is_locked(d_inode(dbuf->xadir)));
202 
203 	if (dbuf->count == ARRAY_SIZE(dbuf->dentries))
204 		return false;
205 
206 	if (name[0] == '.' && (namelen < 2 ||
207 			       (namelen == 2 && name[1] == '.')))
208 		return true;
209 
210 	dentry = lookup_one_len(name, dbuf->xadir, namelen);
211 	if (IS_ERR(dentry)) {
212 		dbuf->err = PTR_ERR(dentry);
213 		return false;
214 	} else if (d_really_is_negative(dentry)) {
215 		/* A directory entry exists, but no file? */
216 		reiserfs_error(dentry->d_sb, "xattr-20003",
217 			       "Corrupted directory: xattr %pd listed but "
218 			       "not found for file %pd.\n",
219 			       dentry, dbuf->xadir);
220 		dput(dentry);
221 		dbuf->err = -EIO;
222 		return false;
223 	}
224 
225 	dbuf->dentries[dbuf->count++] = dentry;
226 	return true;
227 }
228 
229 static void
230 cleanup_dentry_buf(struct reiserfs_dentry_buf *buf)
231 {
232 	int i;
233 
234 	for (i = 0; i < buf->count; i++)
235 		if (buf->dentries[i])
236 			dput(buf->dentries[i]);
237 }
238 
239 static int reiserfs_for_each_xattr(struct inode *inode,
240 				   int (*action)(struct dentry *, void *),
241 				   void *data)
242 {
243 	struct dentry *dir;
244 	int i, err = 0;
245 	struct reiserfs_dentry_buf buf = {
246 		.ctx.actor = fill_with_dentries,
247 	};
248 
249 	/* Skip out, an xattr has no xattrs associated with it */
250 	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
251 		return 0;
252 
253 	dir = open_xa_dir(inode, XATTR_REPLACE);
254 	if (IS_ERR(dir)) {
255 		err = PTR_ERR(dir);
256 		goto out;
257 	} else if (d_really_is_negative(dir)) {
258 		err = 0;
259 		goto out_dir;
260 	}
261 
262 	inode_lock_nested(d_inode(dir), I_MUTEX_XATTR);
263 
264 	buf.xadir = dir;
265 	while (1) {
266 		err = reiserfs_readdir_inode(d_inode(dir), &buf.ctx);
267 		if (err)
268 			break;
269 		if (buf.err) {
270 			err = buf.err;
271 			break;
272 		}
273 		if (!buf.count)
274 			break;
275 		for (i = 0; !err && i < buf.count && buf.dentries[i]; i++) {
276 			struct dentry *dentry = buf.dentries[i];
277 
278 			if (!d_is_dir(dentry))
279 				err = action(dentry, data);
280 
281 			dput(dentry);
282 			buf.dentries[i] = NULL;
283 		}
284 		if (err)
285 			break;
286 		buf.count = 0;
287 	}
288 	inode_unlock(d_inode(dir));
289 
290 	cleanup_dentry_buf(&buf);
291 
292 	if (!err) {
293 		/*
294 		 * We start a transaction here to avoid a ABBA situation
295 		 * between the xattr root's i_mutex and the journal lock.
296 		 * This doesn't incur much additional overhead since the
297 		 * new transaction will just nest inside the
298 		 * outer transaction.
299 		 */
300 		int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 +
301 			     4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
302 		struct reiserfs_transaction_handle th;
303 
304 		reiserfs_write_lock(inode->i_sb);
305 		err = journal_begin(&th, inode->i_sb, blocks);
306 		reiserfs_write_unlock(inode->i_sb);
307 		if (!err) {
308 			int jerror;
309 
310 			inode_lock_nested(d_inode(dir->d_parent),
311 					  I_MUTEX_XATTR);
312 			err = action(dir, data);
313 			reiserfs_write_lock(inode->i_sb);
314 			jerror = journal_end(&th);
315 			reiserfs_write_unlock(inode->i_sb);
316 			inode_unlock(d_inode(dir->d_parent));
317 			err = jerror ?: err;
318 		}
319 	}
320 out_dir:
321 	dput(dir);
322 out:
323 	/*
324 	 * -ENODATA: this object doesn't have any xattrs
325 	 * -EOPNOTSUPP: this file system doesn't have xattrs enabled on disk.
326 	 * Neither are errors
327 	 */
328 	if (err == -ENODATA || err == -EOPNOTSUPP)
329 		err = 0;
330 	return err;
331 }
332 
333 static int delete_one_xattr(struct dentry *dentry, void *data)
334 {
335 	struct inode *dir = d_inode(dentry->d_parent);
336 
337 	/* This is the xattr dir, handle specially. */
338 	if (d_is_dir(dentry))
339 		return xattr_rmdir(dir, dentry);
340 
341 	return xattr_unlink(dir, dentry);
342 }
343 
344 static int chown_one_xattr(struct dentry *dentry, void *data)
345 {
346 	struct iattr *attrs = data;
347 	int ia_valid = attrs->ia_valid;
348 	int err;
349 
350 	/*
351 	 * We only want the ownership bits. Otherwise, we'll do
352 	 * things like change a directory to a regular file if
353 	 * ATTR_MODE is set.
354 	 */
355 	attrs->ia_valid &= (ATTR_UID|ATTR_GID);
356 	err = reiserfs_setattr(&nop_mnt_idmap, dentry, attrs);
357 	attrs->ia_valid = ia_valid;
358 
359 	return err;
360 }
361 
362 /* No i_mutex, but the inode is unconnected. */
363 int reiserfs_delete_xattrs(struct inode *inode)
364 {
365 	int err = reiserfs_for_each_xattr(inode, delete_one_xattr, NULL);
366 
367 	if (err)
368 		reiserfs_warning(inode->i_sb, "jdm-20004",
369 				 "Couldn't delete all xattrs (%d)\n", err);
370 	return err;
371 }
372 
373 /* inode->i_mutex: down */
374 int reiserfs_chown_xattrs(struct inode *inode, struct iattr *attrs)
375 {
376 	int err = reiserfs_for_each_xattr(inode, chown_one_xattr, attrs);
377 
378 	if (err)
379 		reiserfs_warning(inode->i_sb, "jdm-20007",
380 				 "Couldn't chown all xattrs (%d)\n", err);
381 	return err;
382 }
383 
384 #ifdef CONFIG_REISERFS_FS_XATTR
385 /*
386  * Returns a dentry corresponding to a specific extended attribute file
387  * for the inode. If flags allow, the file is created. Otherwise, a
388  * valid or negative dentry, or an error is returned.
389  */
390 static struct dentry *xattr_lookup(struct inode *inode, const char *name,
391 				    int flags)
392 {
393 	struct dentry *xadir, *xafile;
394 	int err = 0;
395 
396 	xadir = open_xa_dir(inode, flags);
397 	if (IS_ERR(xadir))
398 		return ERR_CAST(xadir);
399 
400 	inode_lock_nested(d_inode(xadir), I_MUTEX_XATTR);
401 	xafile = lookup_one_len(name, xadir, strlen(name));
402 	if (IS_ERR(xafile)) {
403 		err = PTR_ERR(xafile);
404 		goto out;
405 	}
406 
407 	if (d_really_is_positive(xafile) && (flags & XATTR_CREATE))
408 		err = -EEXIST;
409 
410 	if (d_really_is_negative(xafile)) {
411 		err = -ENODATA;
412 		if (xattr_may_create(flags))
413 			err = xattr_create(d_inode(xadir), xafile,
414 					      0700|S_IFREG);
415 	}
416 
417 	if (err)
418 		dput(xafile);
419 out:
420 	inode_unlock(d_inode(xadir));
421 	dput(xadir);
422 	if (err)
423 		return ERR_PTR(err);
424 	return xafile;
425 }
426 
427 /* Internal operations on file data */
428 static inline void reiserfs_put_page(struct page *page)
429 {
430 	kunmap(page);
431 	put_page(page);
432 }
433 
434 static struct page *reiserfs_get_page(struct inode *dir, size_t n)
435 {
436 	struct address_space *mapping = dir->i_mapping;
437 	struct page *page;
438 	/*
439 	 * We can deadlock if we try to free dentries,
440 	 * and an unlink/rmdir has just occurred - GFP_NOFS avoids this
441 	 */
442 	mapping_set_gfp_mask(mapping, GFP_NOFS);
443 	page = read_mapping_page(mapping, n >> PAGE_SHIFT, NULL);
444 	if (!IS_ERR(page))
445 		kmap(page);
446 	return page;
447 }
448 
449 static inline __u32 xattr_hash(const char *msg, int len)
450 {
451 	/*
452 	 * csum_partial() gives different results for little-endian and
453 	 * big endian hosts. Images created on little-endian hosts and
454 	 * mounted on big-endian hosts(and vice versa) will see csum mismatches
455 	 * when trying to fetch xattrs. Treating the hash as __wsum_t would
456 	 * lower the frequency of mismatch.  This is an endianness bug in
457 	 * reiserfs.  The return statement would result in a sparse warning. Do
458 	 * not fix the sparse warning so as to not hide a reminder of the bug.
459 	 */
460 	return csum_partial(msg, len, 0);
461 }
462 
463 int reiserfs_commit_write(struct file *f, struct page *page,
464 			  unsigned from, unsigned to);
465 
466 static void update_ctime(struct inode *inode)
467 {
468 	struct timespec64 now = current_time(inode);
469 	struct timespec64 ctime = inode_get_ctime(inode);
470 
471 	if (inode_unhashed(inode) || !inode->i_nlink ||
472 	    timespec64_equal(&ctime, &now))
473 		return;
474 
475 	inode_set_ctime_to_ts(inode, now);
476 	mark_inode_dirty(inode);
477 }
478 
479 static int lookup_and_delete_xattr(struct inode *inode, const char *name)
480 {
481 	int err = 0;
482 	struct dentry *dentry, *xadir;
483 
484 	xadir = open_xa_dir(inode, XATTR_REPLACE);
485 	if (IS_ERR(xadir))
486 		return PTR_ERR(xadir);
487 
488 	inode_lock_nested(d_inode(xadir), I_MUTEX_XATTR);
489 	dentry = lookup_one_len(name, xadir, strlen(name));
490 	if (IS_ERR(dentry)) {
491 		err = PTR_ERR(dentry);
492 		goto out_dput;
493 	}
494 
495 	if (d_really_is_positive(dentry)) {
496 		err = xattr_unlink(d_inode(xadir), dentry);
497 		update_ctime(inode);
498 	}
499 
500 	dput(dentry);
501 out_dput:
502 	inode_unlock(d_inode(xadir));
503 	dput(xadir);
504 	return err;
505 }
506 
507 
508 /* Generic extended attribute operations that can be used by xa plugins */
509 
510 /*
511  * inode->i_mutex: down
512  */
513 int
514 reiserfs_xattr_set_handle(struct reiserfs_transaction_handle *th,
515 			  struct inode *inode, const char *name,
516 			  const void *buffer, size_t buffer_size, int flags)
517 {
518 	int err = 0;
519 	struct dentry *dentry;
520 	struct page *page;
521 	char *data;
522 	size_t file_pos = 0;
523 	size_t buffer_pos = 0;
524 	size_t new_size;
525 	__u32 xahash = 0;
526 
527 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
528 		return -EOPNOTSUPP;
529 
530 	if (!buffer) {
531 		err = lookup_and_delete_xattr(inode, name);
532 		return err;
533 	}
534 
535 	dentry = xattr_lookup(inode, name, flags);
536 	if (IS_ERR(dentry))
537 		return PTR_ERR(dentry);
538 
539 	down_write(&REISERFS_I(inode)->i_xattr_sem);
540 
541 	xahash = xattr_hash(buffer, buffer_size);
542 	while (buffer_pos < buffer_size || buffer_pos == 0) {
543 		size_t chunk;
544 		size_t skip = 0;
545 		size_t page_offset = (file_pos & (PAGE_SIZE - 1));
546 
547 		if (buffer_size - buffer_pos > PAGE_SIZE)
548 			chunk = PAGE_SIZE;
549 		else
550 			chunk = buffer_size - buffer_pos;
551 
552 		page = reiserfs_get_page(d_inode(dentry), file_pos);
553 		if (IS_ERR(page)) {
554 			err = PTR_ERR(page);
555 			goto out_unlock;
556 		}
557 
558 		lock_page(page);
559 		data = page_address(page);
560 
561 		if (file_pos == 0) {
562 			struct reiserfs_xattr_header *rxh;
563 
564 			skip = file_pos = sizeof(struct reiserfs_xattr_header);
565 			if (chunk + skip > PAGE_SIZE)
566 				chunk = PAGE_SIZE - skip;
567 			rxh = (struct reiserfs_xattr_header *)data;
568 			rxh->h_magic = cpu_to_le32(REISERFS_XATTR_MAGIC);
569 			rxh->h_hash = cpu_to_le32(xahash);
570 		}
571 
572 		reiserfs_write_lock(inode->i_sb);
573 		err = __reiserfs_write_begin(page, page_offset, chunk + skip);
574 		if (!err) {
575 			if (buffer)
576 				memcpy(data + skip, buffer + buffer_pos, chunk);
577 			err = reiserfs_commit_write(NULL, page, page_offset,
578 						    page_offset + chunk +
579 						    skip);
580 		}
581 		reiserfs_write_unlock(inode->i_sb);
582 		unlock_page(page);
583 		reiserfs_put_page(page);
584 		buffer_pos += chunk;
585 		file_pos += chunk;
586 		skip = 0;
587 		if (err || buffer_size == 0 || !buffer)
588 			break;
589 	}
590 
591 	new_size = buffer_size + sizeof(struct reiserfs_xattr_header);
592 	if (!err && new_size < i_size_read(d_inode(dentry))) {
593 		struct iattr newattrs = {
594 			.ia_ctime = current_time(inode),
595 			.ia_size = new_size,
596 			.ia_valid = ATTR_SIZE | ATTR_CTIME,
597 		};
598 
599 		inode_lock_nested(d_inode(dentry), I_MUTEX_XATTR);
600 		inode_dio_wait(d_inode(dentry));
601 
602 		err = reiserfs_setattr(&nop_mnt_idmap, dentry, &newattrs);
603 		inode_unlock(d_inode(dentry));
604 	} else
605 		update_ctime(inode);
606 out_unlock:
607 	up_write(&REISERFS_I(inode)->i_xattr_sem);
608 	dput(dentry);
609 	return err;
610 }
611 
612 /* We need to start a transaction to maintain lock ordering */
613 int reiserfs_xattr_set(struct inode *inode, const char *name,
614 		       const void *buffer, size_t buffer_size, int flags)
615 {
616 
617 	struct reiserfs_transaction_handle th;
618 	int error, error2;
619 	size_t jbegin_count = reiserfs_xattr_nblocks(inode, buffer_size);
620 
621 	/* Check before we start a transaction and then do nothing. */
622 	if (!d_really_is_positive(REISERFS_SB(inode->i_sb)->priv_root))
623 		return -EOPNOTSUPP;
624 
625 	if (!(flags & XATTR_REPLACE))
626 		jbegin_count += reiserfs_xattr_jcreate_nblocks(inode);
627 
628 	reiserfs_write_lock(inode->i_sb);
629 	error = journal_begin(&th, inode->i_sb, jbegin_count);
630 	reiserfs_write_unlock(inode->i_sb);
631 	if (error) {
632 		return error;
633 	}
634 
635 	error = reiserfs_xattr_set_handle(&th, inode, name,
636 					  buffer, buffer_size, flags);
637 
638 	reiserfs_write_lock(inode->i_sb);
639 	error2 = journal_end(&th);
640 	reiserfs_write_unlock(inode->i_sb);
641 	if (error == 0)
642 		error = error2;
643 
644 	return error;
645 }
646 
647 /*
648  * inode->i_mutex: down
649  */
650 int
651 reiserfs_xattr_get(struct inode *inode, const char *name, void *buffer,
652 		   size_t buffer_size)
653 {
654 	ssize_t err = 0;
655 	struct dentry *dentry;
656 	size_t isize;
657 	size_t file_pos = 0;
658 	size_t buffer_pos = 0;
659 	struct page *page;
660 	__u32 hash = 0;
661 
662 	if (name == NULL)
663 		return -EINVAL;
664 
665 	/*
666 	 * We can't have xattrs attached to v1 items since they don't have
667 	 * generation numbers
668 	 */
669 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
670 		return -EOPNOTSUPP;
671 
672 	/*
673 	 * priv_root needn't be initialized during mount so allow initial
674 	 * lookups to succeed.
675 	 */
676 	if (!REISERFS_SB(inode->i_sb)->priv_root)
677 		return 0;
678 
679 	dentry = xattr_lookup(inode, name, XATTR_REPLACE);
680 	if (IS_ERR(dentry)) {
681 		err = PTR_ERR(dentry);
682 		goto out;
683 	}
684 
685 	down_read(&REISERFS_I(inode)->i_xattr_sem);
686 
687 	isize = i_size_read(d_inode(dentry));
688 
689 	/* Just return the size needed */
690 	if (buffer == NULL) {
691 		err = isize - sizeof(struct reiserfs_xattr_header);
692 		goto out_unlock;
693 	}
694 
695 	if (buffer_size < isize - sizeof(struct reiserfs_xattr_header)) {
696 		err = -ERANGE;
697 		goto out_unlock;
698 	}
699 
700 	while (file_pos < isize) {
701 		size_t chunk;
702 		char *data;
703 		size_t skip = 0;
704 
705 		if (isize - file_pos > PAGE_SIZE)
706 			chunk = PAGE_SIZE;
707 		else
708 			chunk = isize - file_pos;
709 
710 		page = reiserfs_get_page(d_inode(dentry), file_pos);
711 		if (IS_ERR(page)) {
712 			err = PTR_ERR(page);
713 			goto out_unlock;
714 		}
715 
716 		lock_page(page);
717 		data = page_address(page);
718 		if (file_pos == 0) {
719 			struct reiserfs_xattr_header *rxh =
720 			    (struct reiserfs_xattr_header *)data;
721 			skip = file_pos = sizeof(struct reiserfs_xattr_header);
722 			chunk -= skip;
723 			/* Magic doesn't match up.. */
724 			if (rxh->h_magic != cpu_to_le32(REISERFS_XATTR_MAGIC)) {
725 				unlock_page(page);
726 				reiserfs_put_page(page);
727 				reiserfs_warning(inode->i_sb, "jdm-20001",
728 						 "Invalid magic for xattr (%s) "
729 						 "associated with %k", name,
730 						 INODE_PKEY(inode));
731 				err = -EIO;
732 				goto out_unlock;
733 			}
734 			hash = le32_to_cpu(rxh->h_hash);
735 		}
736 		memcpy(buffer + buffer_pos, data + skip, chunk);
737 		unlock_page(page);
738 		reiserfs_put_page(page);
739 		file_pos += chunk;
740 		buffer_pos += chunk;
741 		skip = 0;
742 	}
743 	err = isize - sizeof(struct reiserfs_xattr_header);
744 
745 	if (xattr_hash(buffer, isize - sizeof(struct reiserfs_xattr_header)) !=
746 	    hash) {
747 		reiserfs_warning(inode->i_sb, "jdm-20002",
748 				 "Invalid hash for xattr (%s) associated "
749 				 "with %k", name, INODE_PKEY(inode));
750 		err = -EIO;
751 	}
752 
753 out_unlock:
754 	up_read(&REISERFS_I(inode)->i_xattr_sem);
755 	dput(dentry);
756 
757 out:
758 	return err;
759 }
760 
761 /*
762  * In order to implement different sets of xattr operations for each xattr
763  * prefix with the generic xattr API, a filesystem should create a
764  * null-terminated array of struct xattr_handler (one for each prefix) and
765  * hang a pointer to it off of the s_xattr field of the superblock.
766  *
767  * The generic_fooxattr() functions will use this list to dispatch xattr
768  * operations to the correct xattr_handler.
769  */
770 #define for_each_xattr_handler(handlers, handler)		\
771 		for ((handler) = *(handlers)++;			\
772 			(handler) != NULL;			\
773 			(handler) = *(handlers)++)
774 
775 static inline bool reiserfs_posix_acl_list(const char *name,
776 					   struct dentry *dentry)
777 {
778 	return (posix_acl_type(name) >= 0) &&
779 	       IS_POSIXACL(d_backing_inode(dentry));
780 }
781 
782 /* This is the implementation for the xattr plugin infrastructure */
783 static inline bool reiserfs_xattr_list(const struct xattr_handler **handlers,
784 				       const char *name, struct dentry *dentry)
785 {
786 	if (handlers) {
787 		const struct xattr_handler *xah = NULL;
788 
789 		for_each_xattr_handler(handlers, xah) {
790 			const char *prefix = xattr_prefix(xah);
791 
792 			if (strncmp(prefix, name, strlen(prefix)))
793 				continue;
794 
795 			if (!xattr_handler_can_list(xah, dentry))
796 				return false;
797 
798 			return true;
799 		}
800 	}
801 
802 	return reiserfs_posix_acl_list(name, dentry);
803 }
804 
805 struct listxattr_buf {
806 	struct dir_context ctx;
807 	size_t size;
808 	size_t pos;
809 	char *buf;
810 	struct dentry *dentry;
811 };
812 
813 static bool listxattr_filler(struct dir_context *ctx, const char *name,
814 			    int namelen, loff_t offset, u64 ino,
815 			    unsigned int d_type)
816 {
817 	struct listxattr_buf *b =
818 		container_of(ctx, struct listxattr_buf, ctx);
819 	size_t size;
820 
821 	if (name[0] != '.' ||
822 	    (namelen != 1 && (name[1] != '.' || namelen != 2))) {
823 		if (!reiserfs_xattr_list(b->dentry->d_sb->s_xattr, name,
824 					 b->dentry))
825 			return true;
826 		size = namelen + 1;
827 		if (b->buf) {
828 			if (b->pos + size > b->size) {
829 				b->pos = -ERANGE;
830 				return false;
831 			}
832 			memcpy(b->buf + b->pos, name, namelen);
833 			b->buf[b->pos + namelen] = 0;
834 		}
835 		b->pos += size;
836 	}
837 	return true;
838 }
839 
840 /*
841  * Inode operation listxattr()
842  *
843  * We totally ignore the generic listxattr here because it would be stupid
844  * not to. Since the xattrs are organized in a directory, we can just
845  * readdir to find them.
846  */
847 ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size)
848 {
849 	struct dentry *dir;
850 	int err = 0;
851 	struct listxattr_buf buf = {
852 		.ctx.actor = listxattr_filler,
853 		.dentry = dentry,
854 		.buf = buffer,
855 		.size = buffer ? size : 0,
856 	};
857 
858 	if (d_really_is_negative(dentry))
859 		return -EINVAL;
860 
861 	if (get_inode_sd_version(d_inode(dentry)) == STAT_DATA_V1)
862 		return -EOPNOTSUPP;
863 
864 	dir = open_xa_dir(d_inode(dentry), XATTR_REPLACE);
865 	if (IS_ERR(dir)) {
866 		err = PTR_ERR(dir);
867 		if (err == -ENODATA)
868 			err = 0;  /* Not an error if there aren't any xattrs */
869 		goto out;
870 	}
871 
872 	inode_lock_nested(d_inode(dir), I_MUTEX_XATTR);
873 	err = reiserfs_readdir_inode(d_inode(dir), &buf.ctx);
874 	inode_unlock(d_inode(dir));
875 
876 	if (!err)
877 		err = buf.pos;
878 
879 	dput(dir);
880 out:
881 	return err;
882 }
883 
884 static int create_privroot(struct dentry *dentry)
885 {
886 	int err;
887 	struct inode *inode = d_inode(dentry->d_parent);
888 
889 	WARN_ON_ONCE(!inode_is_locked(inode));
890 
891 	err = xattr_mkdir(inode, dentry, 0700);
892 	if (err || d_really_is_negative(dentry)) {
893 		reiserfs_warning(dentry->d_sb, "jdm-20006",
894 				 "xattrs/ACLs enabled and couldn't "
895 				 "find/create .reiserfs_priv. "
896 				 "Failing mount.");
897 		return -EOPNOTSUPP;
898 	}
899 
900 	reiserfs_init_priv_inode(d_inode(dentry));
901 	reiserfs_info(dentry->d_sb, "Created %s - reserved for xattr "
902 		      "storage.\n", PRIVROOT_NAME);
903 
904 	return 0;
905 }
906 
907 #else
908 int __init reiserfs_xattr_register_handlers(void) { return 0; }
909 void reiserfs_xattr_unregister_handlers(void) {}
910 static int create_privroot(struct dentry *dentry) { return 0; }
911 #endif
912 
913 /* Actual operations that are exported to VFS-land */
914 const struct xattr_handler *reiserfs_xattr_handlers[] = {
915 #ifdef CONFIG_REISERFS_FS_XATTR
916 	&reiserfs_xattr_user_handler,
917 	&reiserfs_xattr_trusted_handler,
918 #endif
919 #ifdef CONFIG_REISERFS_FS_SECURITY
920 	&reiserfs_xattr_security_handler,
921 #endif
922 	NULL
923 };
924 
925 static int xattr_mount_check(struct super_block *s)
926 {
927 	/*
928 	 * We need generation numbers to ensure that the oid mapping is correct
929 	 * v3.5 filesystems don't have them.
930 	 */
931 	if (old_format_only(s)) {
932 		if (reiserfs_xattrs_optional(s)) {
933 			/*
934 			 * Old format filesystem, but optional xattrs have
935 			 * been enabled. Error out.
936 			 */
937 			reiserfs_warning(s, "jdm-2005",
938 					 "xattrs/ACLs not supported "
939 					 "on pre-v3.6 format filesystems. "
940 					 "Failing mount.");
941 			return -EOPNOTSUPP;
942 		}
943 	}
944 
945 	return 0;
946 }
947 
948 int reiserfs_permission(struct mnt_idmap *idmap, struct inode *inode,
949 			int mask)
950 {
951 	/*
952 	 * We don't do permission checks on the internal objects.
953 	 * Permissions are determined by the "owning" object.
954 	 */
955 	if (IS_PRIVATE(inode))
956 		return 0;
957 
958 	return generic_permission(&nop_mnt_idmap, inode, mask);
959 }
960 
961 static int xattr_hide_revalidate(struct dentry *dentry, unsigned int flags)
962 {
963 	return -EPERM;
964 }
965 
966 static const struct dentry_operations xattr_lookup_poison_ops = {
967 	.d_revalidate = xattr_hide_revalidate,
968 };
969 
970 int reiserfs_lookup_privroot(struct super_block *s)
971 {
972 	struct dentry *dentry;
973 	int err = 0;
974 
975 	/* If we don't have the privroot located yet - go find it */
976 	inode_lock(d_inode(s->s_root));
977 	dentry = lookup_one_len(PRIVROOT_NAME, s->s_root,
978 				strlen(PRIVROOT_NAME));
979 	if (!IS_ERR(dentry)) {
980 		REISERFS_SB(s)->priv_root = dentry;
981 		d_set_d_op(dentry, &xattr_lookup_poison_ops);
982 		if (d_really_is_positive(dentry))
983 			reiserfs_init_priv_inode(d_inode(dentry));
984 	} else
985 		err = PTR_ERR(dentry);
986 	inode_unlock(d_inode(s->s_root));
987 
988 	return err;
989 }
990 
991 /*
992  * We need to take a copy of the mount flags since things like
993  * SB_RDONLY don't get set until *after* we're called.
994  * mount_flags != mount_options
995  */
996 int reiserfs_xattr_init(struct super_block *s, int mount_flags)
997 {
998 	int err = 0;
999 	struct dentry *privroot = REISERFS_SB(s)->priv_root;
1000 
1001 	err = xattr_mount_check(s);
1002 	if (err)
1003 		goto error;
1004 
1005 	if (d_really_is_negative(privroot) && !(mount_flags & SB_RDONLY)) {
1006 		inode_lock(d_inode(s->s_root));
1007 		err = create_privroot(REISERFS_SB(s)->priv_root);
1008 		inode_unlock(d_inode(s->s_root));
1009 	}
1010 
1011 	if (d_really_is_positive(privroot)) {
1012 		inode_lock(d_inode(privroot));
1013 		if (!REISERFS_SB(s)->xattr_root) {
1014 			struct dentry *dentry;
1015 
1016 			dentry = lookup_one_len(XAROOT_NAME, privroot,
1017 						strlen(XAROOT_NAME));
1018 			if (!IS_ERR(dentry))
1019 				REISERFS_SB(s)->xattr_root = dentry;
1020 			else
1021 				err = PTR_ERR(dentry);
1022 		}
1023 		inode_unlock(d_inode(privroot));
1024 	}
1025 
1026 error:
1027 	if (err) {
1028 		clear_bit(REISERFS_XATTRS_USER, &REISERFS_SB(s)->s_mount_opt);
1029 		clear_bit(REISERFS_POSIXACL, &REISERFS_SB(s)->s_mount_opt);
1030 	}
1031 
1032 	/* The super_block SB_POSIXACL must mirror the (no)acl mount option. */
1033 	if (reiserfs_posixacl(s))
1034 		s->s_flags |= SB_POSIXACL;
1035 	else
1036 		s->s_flags &= ~SB_POSIXACL;
1037 
1038 	return err;
1039 }
1040