xref: /openbmc/linux/fs/reiserfs/xattr.c (revision 842ed298)
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 
56 #define PRIVROOT_NAME ".reiserfs_priv"
57 #define XAROOT_NAME   "xattrs"
58 
59 
60 /*
61  * Helpers for inode ops. We do this so that we don't have all the VFS
62  * overhead and also for proper i_mutex annotation.
63  * dir->i_mutex must be held for all of them.
64  */
65 #ifdef CONFIG_REISERFS_FS_XATTR
66 static int xattr_create(struct inode *dir, struct dentry *dentry, int mode)
67 {
68 	BUG_ON(!inode_is_locked(dir));
69 	return dir->i_op->create(dir, dentry, mode, true);
70 }
71 #endif
72 
73 static int xattr_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
74 {
75 	BUG_ON(!inode_is_locked(dir));
76 	return dir->i_op->mkdir(dir, dentry, mode);
77 }
78 
79 /*
80  * We use I_MUTEX_CHILD here to silence lockdep. It's safe because xattr
81  * mutation ops aren't called during rename or splace, which are the
82  * only other users of I_MUTEX_CHILD. It violates the ordering, but that's
83  * better than allocating another subclass just for this code.
84  */
85 static int xattr_unlink(struct inode *dir, struct dentry *dentry)
86 {
87 	int error;
88 
89 	BUG_ON(!inode_is_locked(dir));
90 
91 	inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
92 	error = dir->i_op->unlink(dir, dentry);
93 	inode_unlock(d_inode(dentry));
94 
95 	if (!error)
96 		d_delete(dentry);
97 	return error;
98 }
99 
100 static int xattr_rmdir(struct inode *dir, struct dentry *dentry)
101 {
102 	int error;
103 
104 	BUG_ON(!inode_is_locked(dir));
105 
106 	inode_lock_nested(d_inode(dentry), I_MUTEX_CHILD);
107 	error = dir->i_op->rmdir(dir, dentry);
108 	if (!error)
109 		d_inode(dentry)->i_flags |= S_DEAD;
110 	inode_unlock(d_inode(dentry));
111 	if (!error)
112 		d_delete(dentry);
113 
114 	return error;
115 }
116 
117 #define xattr_may_create(flags)	(!flags || flags & XATTR_CREATE)
118 
119 static struct dentry *open_xa_root(struct super_block *sb, int flags)
120 {
121 	struct dentry *privroot = REISERFS_SB(sb)->priv_root;
122 	struct dentry *xaroot;
123 
124 	if (d_really_is_negative(privroot))
125 		return ERR_PTR(-EOPNOTSUPP);
126 
127 	inode_lock_nested(d_inode(privroot), I_MUTEX_XATTR);
128 
129 	xaroot = dget(REISERFS_SB(sb)->xattr_root);
130 	if (!xaroot)
131 		xaroot = ERR_PTR(-EOPNOTSUPP);
132 	else if (d_really_is_negative(xaroot)) {
133 		int err = -ENODATA;
134 
135 		if (xattr_may_create(flags))
136 			err = xattr_mkdir(d_inode(privroot), xaroot, 0700);
137 		if (err) {
138 			dput(xaroot);
139 			xaroot = ERR_PTR(err);
140 		}
141 	}
142 
143 	inode_unlock(d_inode(privroot));
144 	return xaroot;
145 }
146 
147 static struct dentry *open_xa_dir(const struct inode *inode, int flags)
148 {
149 	struct dentry *xaroot, *xadir;
150 	char namebuf[17];
151 
152 	xaroot = open_xa_root(inode->i_sb, flags);
153 	if (IS_ERR(xaroot))
154 		return xaroot;
155 
156 	snprintf(namebuf, sizeof(namebuf), "%X.%X",
157 		 le32_to_cpu(INODE_PKEY(inode)->k_objectid),
158 		 inode->i_generation);
159 
160 	inode_lock_nested(d_inode(xaroot), I_MUTEX_XATTR);
161 
162 	xadir = lookup_one_len(namebuf, xaroot, strlen(namebuf));
163 	if (!IS_ERR(xadir) && d_really_is_negative(xadir)) {
164 		int err = -ENODATA;
165 
166 		if (xattr_may_create(flags))
167 			err = xattr_mkdir(d_inode(xaroot), xadir, 0700);
168 		if (err) {
169 			dput(xadir);
170 			xadir = ERR_PTR(err);
171 		}
172 	}
173 
174 	inode_unlock(d_inode(xaroot));
175 	dput(xaroot);
176 	return xadir;
177 }
178 
179 /*
180  * The following are side effects of other operations that aren't explicitly
181  * modifying extended attributes. This includes operations such as permissions
182  * or ownership changes, object deletions, etc.
183  */
184 struct reiserfs_dentry_buf {
185 	struct dir_context ctx;
186 	struct dentry *xadir;
187 	int count;
188 	int err;
189 	struct dentry *dentries[8];
190 };
191 
192 static int
193 fill_with_dentries(struct dir_context *ctx, const char *name, int namelen,
194 		   loff_t offset, u64 ino, unsigned int d_type)
195 {
196 	struct reiserfs_dentry_buf *dbuf =
197 		container_of(ctx, struct reiserfs_dentry_buf, ctx);
198 	struct dentry *dentry;
199 
200 	WARN_ON_ONCE(!inode_is_locked(d_inode(dbuf->xadir)));
201 
202 	if (dbuf->count == ARRAY_SIZE(dbuf->dentries))
203 		return -ENOSPC;
204 
205 	if (name[0] == '.' && (namelen < 2 ||
206 			       (namelen == 2 && name[1] == '.')))
207 		return 0;
208 
209 	dentry = lookup_one_len(name, dbuf->xadir, namelen);
210 	if (IS_ERR(dentry)) {
211 		dbuf->err = PTR_ERR(dentry);
212 		return PTR_ERR(dentry);
213 	} else if (d_really_is_negative(dentry)) {
214 		/* A directory entry exists, but no file? */
215 		reiserfs_error(dentry->d_sb, "xattr-20003",
216 			       "Corrupted directory: xattr %pd listed but "
217 			       "not found for file %pd.\n",
218 			       dentry, dbuf->xadir);
219 		dput(dentry);
220 		dbuf->err = -EIO;
221 		return -EIO;
222 	}
223 
224 	dbuf->dentries[dbuf->count++] = dentry;
225 	return 0;
226 }
227 
228 static void
229 cleanup_dentry_buf(struct reiserfs_dentry_buf *buf)
230 {
231 	int i;
232 
233 	for (i = 0; i < buf->count; i++)
234 		if (buf->dentries[i])
235 			dput(buf->dentries[i]);
236 }
237 
238 static int reiserfs_for_each_xattr(struct inode *inode,
239 				   int (*action)(struct dentry *, void *),
240 				   void *data)
241 {
242 	struct dentry *dir;
243 	int i, err = 0;
244 	struct reiserfs_dentry_buf buf = {
245 		.ctx.actor = fill_with_dentries,
246 	};
247 
248 	/* Skip out, an xattr has no xattrs associated with it */
249 	if (IS_PRIVATE(inode) || get_inode_sd_version(inode) == STAT_DATA_V1)
250 		return 0;
251 
252 	dir = open_xa_dir(inode, XATTR_REPLACE);
253 	if (IS_ERR(dir)) {
254 		err = PTR_ERR(dir);
255 		goto out;
256 	} else if (d_really_is_negative(dir)) {
257 		err = 0;
258 		goto out_dir;
259 	}
260 
261 	inode_lock_nested(d_inode(dir), I_MUTEX_XATTR);
262 
263 	buf.xadir = dir;
264 	while (1) {
265 		err = reiserfs_readdir_inode(d_inode(dir), &buf.ctx);
266 		if (err)
267 			break;
268 		if (buf.err) {
269 			err = buf.err;
270 			break;
271 		}
272 		if (!buf.count)
273 			break;
274 		for (i = 0; !err && i < buf.count && buf.dentries[i]; i++) {
275 			struct dentry *dentry = buf.dentries[i];
276 
277 			if (!d_is_dir(dentry))
278 				err = action(dentry, data);
279 
280 			dput(dentry);
281 			buf.dentries[i] = NULL;
282 		}
283 		if (err)
284 			break;
285 		buf.count = 0;
286 	}
287 	inode_unlock(d_inode(dir));
288 
289 	cleanup_dentry_buf(&buf);
290 
291 	if (!err) {
292 		/*
293 		 * We start a transaction here to avoid a ABBA situation
294 		 * between the xattr root's i_mutex and the journal lock.
295 		 * This doesn't incur much additional overhead since the
296 		 * new transaction will just nest inside the
297 		 * outer transaction.
298 		 */
299 		int blocks = JOURNAL_PER_BALANCE_CNT * 2 + 2 +
300 			     4 * REISERFS_QUOTA_TRANS_BLOCKS(inode->i_sb);
301 		struct reiserfs_transaction_handle th;
302 
303 		reiserfs_write_lock(inode->i_sb);
304 		err = journal_begin(&th, inode->i_sb, blocks);
305 		reiserfs_write_unlock(inode->i_sb);
306 		if (!err) {
307 			int jerror;
308 
309 			inode_lock_nested(d_inode(dir->d_parent),
310 					  I_MUTEX_XATTR);
311 			err = action(dir, data);
312 			reiserfs_write_lock(inode->i_sb);
313 			jerror = journal_end(&th);
314 			reiserfs_write_unlock(inode->i_sb);
315 			inode_unlock(d_inode(dir->d_parent));
316 			err = jerror ?: err;
317 		}
318 	}
319 out_dir:
320 	dput(dir);
321 out:
322 	/*
323 	 * -ENODATA: this object doesn't have any xattrs
324 	 * -EOPNOTSUPP: this file system doesn't have xattrs enabled on disk.
325 	 * Neither are errors
326 	 */
327 	if (err == -ENODATA || err == -EOPNOTSUPP)
328 		err = 0;
329 	return err;
330 }
331 
332 static int delete_one_xattr(struct dentry *dentry, void *data)
333 {
334 	struct inode *dir = d_inode(dentry->d_parent);
335 
336 	/* This is the xattr dir, handle specially. */
337 	if (d_is_dir(dentry))
338 		return xattr_rmdir(dir, dentry);
339 
340 	return xattr_unlink(dir, dentry);
341 }
342 
343 static int chown_one_xattr(struct dentry *dentry, void *data)
344 {
345 	struct iattr *attrs = data;
346 	int ia_valid = attrs->ia_valid;
347 	int err;
348 
349 	/*
350 	 * We only want the ownership bits. Otherwise, we'll do
351 	 * things like change a directory to a regular file if
352 	 * ATTR_MODE is set.
353 	 */
354 	attrs->ia_valid &= (ATTR_UID|ATTR_GID);
355 	err = reiserfs_setattr(dentry, attrs);
356 	attrs->ia_valid = ia_valid;
357 
358 	return err;
359 }
360 
361 /* No i_mutex, but the inode is unconnected. */
362 int reiserfs_delete_xattrs(struct inode *inode)
363 {
364 	int err = reiserfs_for_each_xattr(inode, delete_one_xattr, NULL);
365 
366 	if (err)
367 		reiserfs_warning(inode->i_sb, "jdm-20004",
368 				 "Couldn't delete all xattrs (%d)\n", err);
369 	return err;
370 }
371 
372 /* inode->i_mutex: down */
373 int reiserfs_chown_xattrs(struct inode *inode, struct iattr *attrs)
374 {
375 	int err = reiserfs_for_each_xattr(inode, chown_one_xattr, attrs);
376 
377 	if (err)
378 		reiserfs_warning(inode->i_sb, "jdm-20007",
379 				 "Couldn't chown all xattrs (%d)\n", err);
380 	return err;
381 }
382 
383 #ifdef CONFIG_REISERFS_FS_XATTR
384 /*
385  * Returns a dentry corresponding to a specific extended attribute file
386  * for the inode. If flags allow, the file is created. Otherwise, a
387  * valid or negative dentry, or an error is returned.
388  */
389 static struct dentry *xattr_lookup(struct inode *inode, const char *name,
390 				    int flags)
391 {
392 	struct dentry *xadir, *xafile;
393 	int err = 0;
394 
395 	xadir = open_xa_dir(inode, flags);
396 	if (IS_ERR(xadir))
397 		return ERR_CAST(xadir);
398 
399 	inode_lock_nested(d_inode(xadir), I_MUTEX_XATTR);
400 	xafile = lookup_one_len(name, xadir, strlen(name));
401 	if (IS_ERR(xafile)) {
402 		err = PTR_ERR(xafile);
403 		goto out;
404 	}
405 
406 	if (d_really_is_positive(xafile) && (flags & XATTR_CREATE))
407 		err = -EEXIST;
408 
409 	if (d_really_is_negative(xafile)) {
410 		err = -ENODATA;
411 		if (xattr_may_create(flags))
412 			err = xattr_create(d_inode(xadir), xafile,
413 					      0700|S_IFREG);
414 	}
415 
416 	if (err)
417 		dput(xafile);
418 out:
419 	inode_unlock(d_inode(xadir));
420 	dput(xadir);
421 	if (err)
422 		return ERR_PTR(err);
423 	return xafile;
424 }
425 
426 /* Internal operations on file data */
427 static inline void reiserfs_put_page(struct page *page)
428 {
429 	kunmap(page);
430 	put_page(page);
431 }
432 
433 static struct page *reiserfs_get_page(struct inode *dir, size_t n)
434 {
435 	struct address_space *mapping = dir->i_mapping;
436 	struct page *page;
437 	/*
438 	 * We can deadlock if we try to free dentries,
439 	 * and an unlink/rmdir has just occurred - GFP_NOFS avoids this
440 	 */
441 	mapping_set_gfp_mask(mapping, GFP_NOFS);
442 	page = read_mapping_page(mapping, n >> PAGE_SHIFT, NULL);
443 	if (!IS_ERR(page)) {
444 		kmap(page);
445 		if (PageError(page))
446 			goto fail;
447 	}
448 	return page;
449 
450 fail:
451 	reiserfs_put_page(page);
452 	return ERR_PTR(-EIO);
453 }
454 
455 static inline __u32 xattr_hash(const char *msg, int len)
456 {
457 	/*
458 	 * csum_partial() gives different results for little-endian and
459 	 * big endian hosts. Images created on little-endian hosts and
460 	 * mounted on big-endian hosts(and vice versa) will see csum mismatches
461 	 * when trying to fetch xattrs. Treating the hash as __wsum_t would
462 	 * lower the frequency of mismatch.  This is an endianness bug in
463 	 * reiserfs.  The return statement would result in a sparse warning. Do
464 	 * not fix the sparse warning so as to not hide a reminder of the bug.
465 	 */
466 	return csum_partial(msg, len, 0);
467 }
468 
469 int reiserfs_commit_write(struct file *f, struct page *page,
470 			  unsigned from, unsigned to);
471 
472 static void update_ctime(struct inode *inode)
473 {
474 	struct timespec64 now = current_time(inode);
475 
476 	if (inode_unhashed(inode) || !inode->i_nlink ||
477 	    timespec64_equal(&inode->i_ctime, &now))
478 		return;
479 
480 	inode->i_ctime = current_time(inode);
481 	mark_inode_dirty(inode);
482 }
483 
484 static int lookup_and_delete_xattr(struct inode *inode, const char *name)
485 {
486 	int err = 0;
487 	struct dentry *dentry, *xadir;
488 
489 	xadir = open_xa_dir(inode, XATTR_REPLACE);
490 	if (IS_ERR(xadir))
491 		return PTR_ERR(xadir);
492 
493 	inode_lock_nested(d_inode(xadir), I_MUTEX_XATTR);
494 	dentry = lookup_one_len(name, xadir, strlen(name));
495 	if (IS_ERR(dentry)) {
496 		err = PTR_ERR(dentry);
497 		goto out_dput;
498 	}
499 
500 	if (d_really_is_positive(dentry)) {
501 		err = xattr_unlink(d_inode(xadir), dentry);
502 		update_ctime(inode);
503 	}
504 
505 	dput(dentry);
506 out_dput:
507 	inode_unlock(d_inode(xadir));
508 	dput(xadir);
509 	return err;
510 }
511 
512 
513 /* Generic extended attribute operations that can be used by xa plugins */
514 
515 /*
516  * inode->i_mutex: down
517  */
518 int
519 reiserfs_xattr_set_handle(struct reiserfs_transaction_handle *th,
520 			  struct inode *inode, const char *name,
521 			  const void *buffer, size_t buffer_size, int flags)
522 {
523 	int err = 0;
524 	struct dentry *dentry;
525 	struct page *page;
526 	char *data;
527 	size_t file_pos = 0;
528 	size_t buffer_pos = 0;
529 	size_t new_size;
530 	__u32 xahash = 0;
531 
532 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
533 		return -EOPNOTSUPP;
534 
535 	if (!buffer) {
536 		err = lookup_and_delete_xattr(inode, name);
537 		return err;
538 	}
539 
540 	dentry = xattr_lookup(inode, name, flags);
541 	if (IS_ERR(dentry))
542 		return PTR_ERR(dentry);
543 
544 	down_write(&REISERFS_I(inode)->i_xattr_sem);
545 
546 	xahash = xattr_hash(buffer, buffer_size);
547 	while (buffer_pos < buffer_size || buffer_pos == 0) {
548 		size_t chunk;
549 		size_t skip = 0;
550 		size_t page_offset = (file_pos & (PAGE_SIZE - 1));
551 
552 		if (buffer_size - buffer_pos > PAGE_SIZE)
553 			chunk = PAGE_SIZE;
554 		else
555 			chunk = buffer_size - buffer_pos;
556 
557 		page = reiserfs_get_page(d_inode(dentry), file_pos);
558 		if (IS_ERR(page)) {
559 			err = PTR_ERR(page);
560 			goto out_unlock;
561 		}
562 
563 		lock_page(page);
564 		data = page_address(page);
565 
566 		if (file_pos == 0) {
567 			struct reiserfs_xattr_header *rxh;
568 
569 			skip = file_pos = sizeof(struct reiserfs_xattr_header);
570 			if (chunk + skip > PAGE_SIZE)
571 				chunk = PAGE_SIZE - skip;
572 			rxh = (struct reiserfs_xattr_header *)data;
573 			rxh->h_magic = cpu_to_le32(REISERFS_XATTR_MAGIC);
574 			rxh->h_hash = cpu_to_le32(xahash);
575 		}
576 
577 		reiserfs_write_lock(inode->i_sb);
578 		err = __reiserfs_write_begin(page, page_offset, chunk + skip);
579 		if (!err) {
580 			if (buffer)
581 				memcpy(data + skip, buffer + buffer_pos, chunk);
582 			err = reiserfs_commit_write(NULL, page, page_offset,
583 						    page_offset + chunk +
584 						    skip);
585 		}
586 		reiserfs_write_unlock(inode->i_sb);
587 		unlock_page(page);
588 		reiserfs_put_page(page);
589 		buffer_pos += chunk;
590 		file_pos += chunk;
591 		skip = 0;
592 		if (err || buffer_size == 0 || !buffer)
593 			break;
594 	}
595 
596 	new_size = buffer_size + sizeof(struct reiserfs_xattr_header);
597 	if (!err && new_size < i_size_read(d_inode(dentry))) {
598 		struct iattr newattrs = {
599 			.ia_ctime = current_time(inode),
600 			.ia_size = new_size,
601 			.ia_valid = ATTR_SIZE | ATTR_CTIME,
602 		};
603 
604 		inode_lock_nested(d_inode(dentry), I_MUTEX_XATTR);
605 		inode_dio_wait(d_inode(dentry));
606 
607 		err = reiserfs_setattr(dentry, &newattrs);
608 		inode_unlock(d_inode(dentry));
609 	} else
610 		update_ctime(inode);
611 out_unlock:
612 	up_write(&REISERFS_I(inode)->i_xattr_sem);
613 	dput(dentry);
614 	return err;
615 }
616 
617 /* We need to start a transaction to maintain lock ordering */
618 int reiserfs_xattr_set(struct inode *inode, const char *name,
619 		       const void *buffer, size_t buffer_size, int flags)
620 {
621 
622 	struct reiserfs_transaction_handle th;
623 	int error, error2;
624 	size_t jbegin_count = reiserfs_xattr_nblocks(inode, buffer_size);
625 
626 	/* Check before we start a transaction and then do nothing. */
627 	if (!d_really_is_positive(REISERFS_SB(inode->i_sb)->priv_root))
628 		return -EOPNOTSUPP;
629 
630 	if (!(flags & XATTR_REPLACE))
631 		jbegin_count += reiserfs_xattr_jcreate_nblocks(inode);
632 
633 	reiserfs_write_lock(inode->i_sb);
634 	error = journal_begin(&th, inode->i_sb, jbegin_count);
635 	reiserfs_write_unlock(inode->i_sb);
636 	if (error) {
637 		return error;
638 	}
639 
640 	error = reiserfs_xattr_set_handle(&th, inode, name,
641 					  buffer, buffer_size, flags);
642 
643 	reiserfs_write_lock(inode->i_sb);
644 	error2 = journal_end(&th);
645 	reiserfs_write_unlock(inode->i_sb);
646 	if (error == 0)
647 		error = error2;
648 
649 	return error;
650 }
651 
652 /*
653  * inode->i_mutex: down
654  */
655 int
656 reiserfs_xattr_get(struct inode *inode, const char *name, void *buffer,
657 		   size_t buffer_size)
658 {
659 	ssize_t err = 0;
660 	struct dentry *dentry;
661 	size_t isize;
662 	size_t file_pos = 0;
663 	size_t buffer_pos = 0;
664 	struct page *page;
665 	__u32 hash = 0;
666 
667 	if (name == NULL)
668 		return -EINVAL;
669 
670 	/*
671 	 * We can't have xattrs attached to v1 items since they don't have
672 	 * generation numbers
673 	 */
674 	if (get_inode_sd_version(inode) == STAT_DATA_V1)
675 		return -EOPNOTSUPP;
676 
677 	/*
678 	 * priv_root needn't be initialized during mount so allow initial
679 	 * lookups to succeed.
680 	 */
681 	if (!REISERFS_SB(inode->i_sb)->priv_root)
682 		return 0;
683 
684 	dentry = xattr_lookup(inode, name, XATTR_REPLACE);
685 	if (IS_ERR(dentry)) {
686 		err = PTR_ERR(dentry);
687 		goto out;
688 	}
689 
690 	down_read(&REISERFS_I(inode)->i_xattr_sem);
691 
692 	isize = i_size_read(d_inode(dentry));
693 
694 	/* Just return the size needed */
695 	if (buffer == NULL) {
696 		err = isize - sizeof(struct reiserfs_xattr_header);
697 		goto out_unlock;
698 	}
699 
700 	if (buffer_size < isize - sizeof(struct reiserfs_xattr_header)) {
701 		err = -ERANGE;
702 		goto out_unlock;
703 	}
704 
705 	while (file_pos < isize) {
706 		size_t chunk;
707 		char *data;
708 		size_t skip = 0;
709 
710 		if (isize - file_pos > PAGE_SIZE)
711 			chunk = PAGE_SIZE;
712 		else
713 			chunk = isize - file_pos;
714 
715 		page = reiserfs_get_page(d_inode(dentry), file_pos);
716 		if (IS_ERR(page)) {
717 			err = PTR_ERR(page);
718 			goto out_unlock;
719 		}
720 
721 		lock_page(page);
722 		data = page_address(page);
723 		if (file_pos == 0) {
724 			struct reiserfs_xattr_header *rxh =
725 			    (struct reiserfs_xattr_header *)data;
726 			skip = file_pos = sizeof(struct reiserfs_xattr_header);
727 			chunk -= skip;
728 			/* Magic doesn't match up.. */
729 			if (rxh->h_magic != cpu_to_le32(REISERFS_XATTR_MAGIC)) {
730 				unlock_page(page);
731 				reiserfs_put_page(page);
732 				reiserfs_warning(inode->i_sb, "jdm-20001",
733 						 "Invalid magic for xattr (%s) "
734 						 "associated with %k", name,
735 						 INODE_PKEY(inode));
736 				err = -EIO;
737 				goto out_unlock;
738 			}
739 			hash = le32_to_cpu(rxh->h_hash);
740 		}
741 		memcpy(buffer + buffer_pos, data + skip, chunk);
742 		unlock_page(page);
743 		reiserfs_put_page(page);
744 		file_pos += chunk;
745 		buffer_pos += chunk;
746 		skip = 0;
747 	}
748 	err = isize - sizeof(struct reiserfs_xattr_header);
749 
750 	if (xattr_hash(buffer, isize - sizeof(struct reiserfs_xattr_header)) !=
751 	    hash) {
752 		reiserfs_warning(inode->i_sb, "jdm-20002",
753 				 "Invalid hash for xattr (%s) associated "
754 				 "with %k", name, INODE_PKEY(inode));
755 		err = -EIO;
756 	}
757 
758 out_unlock:
759 	up_read(&REISERFS_I(inode)->i_xattr_sem);
760 	dput(dentry);
761 
762 out:
763 	return err;
764 }
765 
766 /*
767  * In order to implement different sets of xattr operations for each xattr
768  * prefix with the generic xattr API, a filesystem should create a
769  * null-terminated array of struct xattr_handler (one for each prefix) and
770  * hang a pointer to it off of the s_xattr field of the superblock.
771  *
772  * The generic_fooxattr() functions will use this list to dispatch xattr
773  * operations to the correct xattr_handler.
774  */
775 #define for_each_xattr_handler(handlers, handler)		\
776 		for ((handler) = *(handlers)++;			\
777 			(handler) != NULL;			\
778 			(handler) = *(handlers)++)
779 
780 /* This is the implementation for the xattr plugin infrastructure */
781 static inline const struct xattr_handler *
782 find_xattr_handler_prefix(const struct xattr_handler **handlers,
783 			   const char *name)
784 {
785 	const struct xattr_handler *xah;
786 
787 	if (!handlers)
788 		return NULL;
789 
790 	for_each_xattr_handler(handlers, xah) {
791 		const char *prefix = xattr_prefix(xah);
792 		if (strncmp(prefix, name, strlen(prefix)) == 0)
793 			break;
794 	}
795 
796 	return xah;
797 }
798 
799 struct listxattr_buf {
800 	struct dir_context ctx;
801 	size_t size;
802 	size_t pos;
803 	char *buf;
804 	struct dentry *dentry;
805 };
806 
807 static int listxattr_filler(struct dir_context *ctx, const char *name,
808 			    int namelen, loff_t offset, u64 ino,
809 			    unsigned int d_type)
810 {
811 	struct listxattr_buf *b =
812 		container_of(ctx, struct listxattr_buf, ctx);
813 	size_t size;
814 
815 	if (name[0] != '.' ||
816 	    (namelen != 1 && (name[1] != '.' || namelen != 2))) {
817 		const struct xattr_handler *handler;
818 
819 		handler = find_xattr_handler_prefix(b->dentry->d_sb->s_xattr,
820 						    name);
821 		if (!handler /* Unsupported xattr name */ ||
822 		    (handler->list && !handler->list(b->dentry)))
823 			return 0;
824 		size = namelen + 1;
825 		if (b->buf) {
826 			if (b->pos + size > b->size) {
827 				b->pos = -ERANGE;
828 				return -ERANGE;
829 			}
830 			memcpy(b->buf + b->pos, name, namelen);
831 			b->buf[b->pos + namelen] = 0;
832 		}
833 		b->pos += size;
834 	}
835 	return 0;
836 }
837 
838 /*
839  * Inode operation listxattr()
840  *
841  * We totally ignore the generic listxattr here because it would be stupid
842  * not to. Since the xattrs are organized in a directory, we can just
843  * readdir to find them.
844  */
845 ssize_t reiserfs_listxattr(struct dentry * dentry, char *buffer, size_t size)
846 {
847 	struct dentry *dir;
848 	int err = 0;
849 	struct listxattr_buf buf = {
850 		.ctx.actor = listxattr_filler,
851 		.dentry = dentry,
852 		.buf = buffer,
853 		.size = buffer ? size : 0,
854 	};
855 
856 	if (d_really_is_negative(dentry))
857 		return -EINVAL;
858 
859 	if (get_inode_sd_version(d_inode(dentry)) == STAT_DATA_V1)
860 		return -EOPNOTSUPP;
861 
862 	dir = open_xa_dir(d_inode(dentry), XATTR_REPLACE);
863 	if (IS_ERR(dir)) {
864 		err = PTR_ERR(dir);
865 		if (err == -ENODATA)
866 			err = 0;  /* Not an error if there aren't any xattrs */
867 		goto out;
868 	}
869 
870 	inode_lock_nested(d_inode(dir), I_MUTEX_XATTR);
871 	err = reiserfs_readdir_inode(d_inode(dir), &buf.ctx);
872 	inode_unlock(d_inode(dir));
873 
874 	if (!err)
875 		err = buf.pos;
876 
877 	dput(dir);
878 out:
879 	return err;
880 }
881 
882 static int create_privroot(struct dentry *dentry)
883 {
884 	int err;
885 	struct inode *inode = d_inode(dentry->d_parent);
886 
887 	WARN_ON_ONCE(!inode_is_locked(inode));
888 
889 	err = xattr_mkdir(inode, dentry, 0700);
890 	if (err || d_really_is_negative(dentry)) {
891 		reiserfs_warning(dentry->d_sb, "jdm-20006",
892 				 "xattrs/ACLs enabled and couldn't "
893 				 "find/create .reiserfs_priv. "
894 				 "Failing mount.");
895 		return -EOPNOTSUPP;
896 	}
897 
898 	d_inode(dentry)->i_flags |= S_PRIVATE;
899 	d_inode(dentry)->i_opflags &= ~IOP_XATTR;
900 	reiserfs_info(dentry->d_sb, "Created %s - reserved for xattr "
901 		      "storage.\n", PRIVROOT_NAME);
902 
903 	return 0;
904 }
905 
906 #else
907 int __init reiserfs_xattr_register_handlers(void) { return 0; }
908 void reiserfs_xattr_unregister_handlers(void) {}
909 static int create_privroot(struct dentry *dentry) { return 0; }
910 #endif
911 
912 /* Actual operations that are exported to VFS-land */
913 const struct xattr_handler *reiserfs_xattr_handlers[] = {
914 #ifdef CONFIG_REISERFS_FS_XATTR
915 	&reiserfs_xattr_user_handler,
916 	&reiserfs_xattr_trusted_handler,
917 #endif
918 #ifdef CONFIG_REISERFS_FS_SECURITY
919 	&reiserfs_xattr_security_handler,
920 #endif
921 #ifdef CONFIG_REISERFS_FS_POSIX_ACL
922 	&posix_acl_access_xattr_handler,
923 	&posix_acl_default_xattr_handler,
924 #endif
925 	NULL
926 };
927 
928 static int xattr_mount_check(struct super_block *s)
929 {
930 	/*
931 	 * We need generation numbers to ensure that the oid mapping is correct
932 	 * v3.5 filesystems don't have them.
933 	 */
934 	if (old_format_only(s)) {
935 		if (reiserfs_xattrs_optional(s)) {
936 			/*
937 			 * Old format filesystem, but optional xattrs have
938 			 * been enabled. Error out.
939 			 */
940 			reiserfs_warning(s, "jdm-2005",
941 					 "xattrs/ACLs not supported "
942 					 "on pre-v3.6 format filesystems. "
943 					 "Failing mount.");
944 			return -EOPNOTSUPP;
945 		}
946 	}
947 
948 	return 0;
949 }
950 
951 int reiserfs_permission(struct inode *inode, int mask)
952 {
953 	/*
954 	 * We don't do permission checks on the internal objects.
955 	 * Permissions are determined by the "owning" object.
956 	 */
957 	if (IS_PRIVATE(inode))
958 		return 0;
959 
960 	return generic_permission(inode, mask);
961 }
962 
963 static int xattr_hide_revalidate(struct dentry *dentry, unsigned int flags)
964 {
965 	return -EPERM;
966 }
967 
968 static const struct dentry_operations xattr_lookup_poison_ops = {
969 	.d_revalidate = xattr_hide_revalidate,
970 };
971 
972 int reiserfs_lookup_privroot(struct super_block *s)
973 {
974 	struct dentry *dentry;
975 	int err = 0;
976 
977 	/* If we don't have the privroot located yet - go find it */
978 	inode_lock(d_inode(s->s_root));
979 	dentry = lookup_one_len(PRIVROOT_NAME, s->s_root,
980 				strlen(PRIVROOT_NAME));
981 	if (!IS_ERR(dentry)) {
982 		REISERFS_SB(s)->priv_root = dentry;
983 		d_set_d_op(dentry, &xattr_lookup_poison_ops);
984 		if (d_really_is_positive(dentry)) {
985 			d_inode(dentry)->i_flags |= S_PRIVATE;
986 			d_inode(dentry)->i_opflags &= ~IOP_XATTR;
987 		}
988 	} else
989 		err = PTR_ERR(dentry);
990 	inode_unlock(d_inode(s->s_root));
991 
992 	return err;
993 }
994 
995 /*
996  * We need to take a copy of the mount flags since things like
997  * SB_RDONLY don't get set until *after* we're called.
998  * mount_flags != mount_options
999  */
1000 int reiserfs_xattr_init(struct super_block *s, int mount_flags)
1001 {
1002 	int err = 0;
1003 	struct dentry *privroot = REISERFS_SB(s)->priv_root;
1004 
1005 	err = xattr_mount_check(s);
1006 	if (err)
1007 		goto error;
1008 
1009 	if (d_really_is_negative(privroot) && !(mount_flags & SB_RDONLY)) {
1010 		inode_lock(d_inode(s->s_root));
1011 		err = create_privroot(REISERFS_SB(s)->priv_root);
1012 		inode_unlock(d_inode(s->s_root));
1013 	}
1014 
1015 	if (d_really_is_positive(privroot)) {
1016 		inode_lock(d_inode(privroot));
1017 		if (!REISERFS_SB(s)->xattr_root) {
1018 			struct dentry *dentry;
1019 
1020 			dentry = lookup_one_len(XAROOT_NAME, privroot,
1021 						strlen(XAROOT_NAME));
1022 			if (!IS_ERR(dentry))
1023 				REISERFS_SB(s)->xattr_root = dentry;
1024 			else
1025 				err = PTR_ERR(dentry);
1026 		}
1027 		inode_unlock(d_inode(privroot));
1028 	}
1029 
1030 error:
1031 	if (err) {
1032 		clear_bit(REISERFS_XATTRS_USER, &REISERFS_SB(s)->s_mount_opt);
1033 		clear_bit(REISERFS_POSIXACL, &REISERFS_SB(s)->s_mount_opt);
1034 	}
1035 
1036 	/* The super_block SB_POSIXACL must mirror the (no)acl mount option. */
1037 	if (reiserfs_posixacl(s))
1038 		s->s_flags |= SB_POSIXACL;
1039 	else
1040 		s->s_flags &= ~SB_POSIXACL;
1041 
1042 	return err;
1043 }
1044