xref: /openbmc/linux/fs/jfs/xattr.c (revision c67ce71d)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines  Corp., 2000-2004
4  *   Copyright (C) Christoph Hellwig, 2002
5  */
6 
7 #include <linux/capability.h>
8 #include <linux/fs.h>
9 #include <linux/xattr.h>
10 #include <linux/posix_acl_xattr.h>
11 #include <linux/slab.h>
12 #include <linux/quotaops.h>
13 #include <linux/security.h>
14 #include "jfs_incore.h"
15 #include "jfs_superblock.h"
16 #include "jfs_dmap.h"
17 #include "jfs_debug.h"
18 #include "jfs_dinode.h"
19 #include "jfs_extent.h"
20 #include "jfs_metapage.h"
21 #include "jfs_xattr.h"
22 #include "jfs_acl.h"
23 
24 /*
25  *	jfs_xattr.c: extended attribute service
26  *
27  * Overall design --
28  *
29  * Format:
30  *
31  *   Extended attribute lists (jfs_ea_list) consist of an overall size (32 bit
32  *   value) and a variable (0 or more) number of extended attribute
33  *   entries.  Each extended attribute entry (jfs_ea) is a <name,value> double
34  *   where <name> is constructed from a null-terminated ascii string
35  *   (1 ... 255 bytes in the name) and <value> is arbitrary 8 bit data
36  *   (1 ... 65535 bytes).  The in-memory format is
37  *
38  *   0       1        2        4                4 + namelen + 1
39  *   +-------+--------+--------+----------------+-------------------+
40  *   | Flags | Name   | Value  | Name String \0 | Data . . . .      |
41  *   |       | Length | Length |                |                   |
42  *   +-------+--------+--------+----------------+-------------------+
43  *
44  *   A jfs_ea_list then is structured as
45  *
46  *   0            4                   4 + EA_SIZE(ea1)
47  *   +------------+-------------------+--------------------+-----
48  *   | Overall EA | First FEA Element | Second FEA Element | .....
49  *   | List Size  |                   |                    |
50  *   +------------+-------------------+--------------------+-----
51  *
52  *   On-disk:
53  *
54  *	FEALISTs are stored on disk using blocks allocated by dbAlloc() and
55  *	written directly. An EA list may be in-lined in the inode if there is
56  *	sufficient room available.
57  */
58 
59 struct ea_buffer {
60 	int flag;		/* Indicates what storage xattr points to */
61 	int max_size;		/* largest xattr that fits in current buffer */
62 	dxd_t new_ea;		/* dxd to replace ea when modifying xattr */
63 	struct metapage *mp;	/* metapage containing ea list */
64 	struct jfs_ea_list *xattr;	/* buffer containing ea list */
65 };
66 
67 /*
68  * ea_buffer.flag values
69  */
70 #define EA_INLINE	0x0001
71 #define EA_EXTENT	0x0002
72 #define EA_NEW		0x0004
73 #define EA_MALLOC	0x0008
74 
75 
76 /*
77  * Mapping of on-disk attribute names: for on-disk attribute names with an
78  * unknown prefix (not "system.", "user.", "security.", or "trusted."), the
79  * prefix "os2." is prepended.  On the way back to disk, "os2." prefixes are
80  * stripped and we make sure that the remaining name does not start with one
81  * of the know prefixes.
82  */
83 
84 static int is_known_namespace(const char *name)
85 {
86 	if (strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) &&
87 	    strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN) &&
88 	    strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
89 	    strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN))
90 		return false;
91 
92 	return true;
93 }
94 
95 static inline int name_size(struct jfs_ea *ea)
96 {
97 	if (is_known_namespace(ea->name))
98 		return ea->namelen;
99 	else
100 		return ea->namelen + XATTR_OS2_PREFIX_LEN;
101 }
102 
103 static inline int copy_name(char *buffer, struct jfs_ea *ea)
104 {
105 	int len = ea->namelen;
106 
107 	if (!is_known_namespace(ea->name)) {
108 		memcpy(buffer, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN);
109 		buffer += XATTR_OS2_PREFIX_LEN;
110 		len += XATTR_OS2_PREFIX_LEN;
111 	}
112 	memcpy(buffer, ea->name, ea->namelen);
113 	buffer[ea->namelen] = 0;
114 
115 	return len;
116 }
117 
118 /* Forward references */
119 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf);
120 
121 /*
122  * NAME: ea_write_inline
123  *
124  * FUNCTION: Attempt to write an EA inline if area is available
125  *
126  * PRE CONDITIONS:
127  *	Already verified that the specified EA is small enough to fit inline
128  *
129  * PARAMETERS:
130  *	ip	- Inode pointer
131  *	ealist	- EA list pointer
132  *	size	- size of ealist in bytes
133  *	ea	- dxd_t structure to be filled in with necessary EA information
134  *		  if we successfully copy the EA inline
135  *
136  * NOTES:
137  *	Checks if the inode's inline area is available.  If so, copies EA inline
138  *	and sets <ea> fields appropriately.  Otherwise, returns failure, EA will
139  *	have to be put into an extent.
140  *
141  * RETURNS: 0 for successful copy to inline area; -1 if area not available
142  */
143 static int ea_write_inline(struct inode *ip, struct jfs_ea_list *ealist,
144 			   int size, dxd_t * ea)
145 {
146 	struct jfs_inode_info *ji = JFS_IP(ip);
147 
148 	/*
149 	 * Make sure we have an EA -- the NULL EA list is valid, but you
150 	 * can't copy it!
151 	 */
152 	if (ealist && size > sizeof (struct jfs_ea_list)) {
153 		assert(size <= sizeof (ji->i_inline_ea));
154 
155 		/*
156 		 * See if the space is available or if it is already being
157 		 * used for an inline EA.
158 		 */
159 		if (!(ji->mode2 & INLINEEA) && !(ji->ea.flag & DXD_INLINE))
160 			return -EPERM;
161 
162 		DXDsize(ea, size);
163 		DXDlength(ea, 0);
164 		DXDaddress(ea, 0);
165 		memcpy(ji->i_inline_ea, ealist, size);
166 		ea->flag = DXD_INLINE;
167 		ji->mode2 &= ~INLINEEA;
168 	} else {
169 		ea->flag = 0;
170 		DXDsize(ea, 0);
171 		DXDlength(ea, 0);
172 		DXDaddress(ea, 0);
173 
174 		/* Free up INLINE area */
175 		if (ji->ea.flag & DXD_INLINE)
176 			ji->mode2 |= INLINEEA;
177 	}
178 
179 	return 0;
180 }
181 
182 /*
183  * NAME: ea_write
184  *
185  * FUNCTION: Write an EA for an inode
186  *
187  * PRE CONDITIONS: EA has been verified
188  *
189  * PARAMETERS:
190  *	ip	- Inode pointer
191  *	ealist	- EA list pointer
192  *	size	- size of ealist in bytes
193  *	ea	- dxd_t structure to be filled in appropriately with where the
194  *		  EA was copied
195  *
196  * NOTES: Will write EA inline if able to, otherwise allocates blocks for an
197  *	extent and synchronously writes it to those blocks.
198  *
199  * RETURNS: 0 for success; Anything else indicates failure
200  */
201 static int ea_write(struct inode *ip, struct jfs_ea_list *ealist, int size,
202 		       dxd_t * ea)
203 {
204 	struct super_block *sb = ip->i_sb;
205 	struct jfs_inode_info *ji = JFS_IP(ip);
206 	struct jfs_sb_info *sbi = JFS_SBI(sb);
207 	int nblocks;
208 	s64 blkno;
209 	int rc = 0, i;
210 	char *cp;
211 	s32 nbytes, nb;
212 	s32 bytes_to_write;
213 	struct metapage *mp;
214 
215 	/*
216 	 * Quick check to see if this is an in-linable EA.  Short EAs
217 	 * and empty EAs are all in-linable, provided the space exists.
218 	 */
219 	if (!ealist || size <= sizeof (ji->i_inline_ea)) {
220 		if (!ea_write_inline(ip, ealist, size, ea))
221 			return 0;
222 	}
223 
224 	/* figure out how many blocks we need */
225 	nblocks = (size + (sb->s_blocksize - 1)) >> sb->s_blocksize_bits;
226 
227 	/* Allocate new blocks to quota. */
228 	rc = dquot_alloc_block(ip, nblocks);
229 	if (rc)
230 		return rc;
231 
232 	rc = dbAlloc(ip, INOHINT(ip), nblocks, &blkno);
233 	if (rc) {
234 		/*Rollback quota allocation. */
235 		dquot_free_block(ip, nblocks);
236 		return rc;
237 	}
238 
239 	/*
240 	 * Now have nblocks worth of storage to stuff into the FEALIST.
241 	 * loop over the FEALIST copying data into the buffer one page at
242 	 * a time.
243 	 */
244 	cp = (char *) ealist;
245 	nbytes = size;
246 	for (i = 0; i < nblocks; i += sbi->nbperpage) {
247 		/*
248 		 * Determine how many bytes for this request, and round up to
249 		 * the nearest aggregate block size
250 		 */
251 		nb = min(PSIZE, nbytes);
252 		bytes_to_write =
253 		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
254 		    << sb->s_blocksize_bits;
255 
256 		if (!(mp = get_metapage(ip, blkno + i, bytes_to_write, 1))) {
257 			rc = -EIO;
258 			goto failed;
259 		}
260 
261 		memcpy(mp->data, cp, nb);
262 
263 		/*
264 		 * We really need a way to propagate errors for
265 		 * forced writes like this one.  --hch
266 		 *
267 		 * (__write_metapage => release_metapage => flush_metapage)
268 		 */
269 #ifdef _JFS_FIXME
270 		if ((rc = flush_metapage(mp))) {
271 			/*
272 			 * the write failed -- this means that the buffer
273 			 * is still assigned and the blocks are not being
274 			 * used.  this seems like the best error recovery
275 			 * we can get ...
276 			 */
277 			goto failed;
278 		}
279 #else
280 		flush_metapage(mp);
281 #endif
282 
283 		cp += PSIZE;
284 		nbytes -= nb;
285 	}
286 
287 	ea->flag = DXD_EXTENT;
288 	DXDsize(ea, le32_to_cpu(ealist->size));
289 	DXDlength(ea, nblocks);
290 	DXDaddress(ea, blkno);
291 
292 	/* Free up INLINE area */
293 	if (ji->ea.flag & DXD_INLINE)
294 		ji->mode2 |= INLINEEA;
295 
296 	return 0;
297 
298       failed:
299 	/* Rollback quota allocation. */
300 	dquot_free_block(ip, nblocks);
301 
302 	dbFree(ip, blkno, nblocks);
303 	return rc;
304 }
305 
306 /*
307  * NAME: ea_read_inline
308  *
309  * FUNCTION: Read an inlined EA into user's buffer
310  *
311  * PARAMETERS:
312  *	ip	- Inode pointer
313  *	ealist	- Pointer to buffer to fill in with EA
314  *
315  * RETURNS: 0
316  */
317 static int ea_read_inline(struct inode *ip, struct jfs_ea_list *ealist)
318 {
319 	struct jfs_inode_info *ji = JFS_IP(ip);
320 	int ea_size = sizeDXD(&ji->ea);
321 
322 	if (ea_size == 0) {
323 		ealist->size = 0;
324 		return 0;
325 	}
326 
327 	/* Sanity Check */
328 	if ((sizeDXD(&ji->ea) > sizeof (ji->i_inline_ea)))
329 		return -EIO;
330 	if (le32_to_cpu(((struct jfs_ea_list *) &ji->i_inline_ea)->size)
331 	    != ea_size)
332 		return -EIO;
333 
334 	memcpy(ealist, ji->i_inline_ea, ea_size);
335 	return 0;
336 }
337 
338 /*
339  * NAME: ea_read
340  *
341  * FUNCTION: copy EA data into user's buffer
342  *
343  * PARAMETERS:
344  *	ip	- Inode pointer
345  *	ealist	- Pointer to buffer to fill in with EA
346  *
347  * NOTES:  If EA is inline calls ea_read_inline() to copy EA.
348  *
349  * RETURNS: 0 for success; other indicates failure
350  */
351 static int ea_read(struct inode *ip, struct jfs_ea_list *ealist)
352 {
353 	struct super_block *sb = ip->i_sb;
354 	struct jfs_inode_info *ji = JFS_IP(ip);
355 	struct jfs_sb_info *sbi = JFS_SBI(sb);
356 	int nblocks;
357 	s64 blkno;
358 	char *cp = (char *) ealist;
359 	int i;
360 	int nbytes, nb;
361 	s32 bytes_to_read;
362 	struct metapage *mp;
363 
364 	/* quick check for in-line EA */
365 	if (ji->ea.flag & DXD_INLINE)
366 		return ea_read_inline(ip, ealist);
367 
368 	nbytes = sizeDXD(&ji->ea);
369 	if (!nbytes) {
370 		jfs_error(sb, "nbytes is 0\n");
371 		return -EIO;
372 	}
373 
374 	/*
375 	 * Figure out how many blocks were allocated when this EA list was
376 	 * originally written to disk.
377 	 */
378 	nblocks = lengthDXD(&ji->ea) << sbi->l2nbperpage;
379 	blkno = addressDXD(&ji->ea) << sbi->l2nbperpage;
380 
381 	/*
382 	 * I have found the disk blocks which were originally used to store
383 	 * the FEALIST.  now i loop over each contiguous block copying the
384 	 * data into the buffer.
385 	 */
386 	for (i = 0; i < nblocks; i += sbi->nbperpage) {
387 		/*
388 		 * Determine how many bytes for this request, and round up to
389 		 * the nearest aggregate block size
390 		 */
391 		nb = min(PSIZE, nbytes);
392 		bytes_to_read =
393 		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
394 		    << sb->s_blocksize_bits;
395 
396 		if (!(mp = read_metapage(ip, blkno + i, bytes_to_read, 1)))
397 			return -EIO;
398 
399 		memcpy(cp, mp->data, nb);
400 		release_metapage(mp);
401 
402 		cp += PSIZE;
403 		nbytes -= nb;
404 	}
405 
406 	return 0;
407 }
408 
409 /*
410  * NAME: ea_get
411  *
412  * FUNCTION: Returns buffer containing existing extended attributes.
413  *	     The size of the buffer will be the larger of the existing
414  *	     attributes size, or min_size.
415  *
416  *	     The buffer, which may be inlined in the inode or in the
417  *	     page cache must be release by calling ea_release or ea_put
418  *
419  * PARAMETERS:
420  *	inode	- Inode pointer
421  *	ea_buf	- Structure to be populated with ealist and its metadata
422  *	min_size- minimum size of buffer to be returned
423  *
424  * RETURNS: 0 for success; Other indicates failure
425  */
426 static int ea_get(struct inode *inode, struct ea_buffer *ea_buf, int min_size)
427 {
428 	struct jfs_inode_info *ji = JFS_IP(inode);
429 	struct super_block *sb = inode->i_sb;
430 	int size;
431 	int ea_size = sizeDXD(&ji->ea);
432 	int blocks_needed, current_blocks;
433 	s64 blkno;
434 	int rc;
435 	int quota_allocation = 0;
436 
437 	memset(&ea_buf->new_ea, 0, sizeof(ea_buf->new_ea));
438 
439 	/* When fsck.jfs clears a bad ea, it doesn't clear the size */
440 	if (ji->ea.flag == 0)
441 		ea_size = 0;
442 
443 	if (ea_size == 0) {
444 		if (min_size == 0) {
445 			ea_buf->flag = 0;
446 			ea_buf->max_size = 0;
447 			ea_buf->xattr = NULL;
448 			return 0;
449 		}
450 		if ((min_size <= sizeof (ji->i_inline_ea)) &&
451 		    (ji->mode2 & INLINEEA)) {
452 			ea_buf->flag = EA_INLINE | EA_NEW;
453 			ea_buf->max_size = sizeof (ji->i_inline_ea);
454 			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
455 			DXDlength(&ea_buf->new_ea, 0);
456 			DXDaddress(&ea_buf->new_ea, 0);
457 			ea_buf->new_ea.flag = DXD_INLINE;
458 			DXDsize(&ea_buf->new_ea, min_size);
459 			return 0;
460 		}
461 		current_blocks = 0;
462 	} else if (ji->ea.flag & DXD_INLINE) {
463 		if (min_size <= sizeof (ji->i_inline_ea)) {
464 			ea_buf->flag = EA_INLINE;
465 			ea_buf->max_size = sizeof (ji->i_inline_ea);
466 			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
467 			goto size_check;
468 		}
469 		current_blocks = 0;
470 	} else {
471 		if (!(ji->ea.flag & DXD_EXTENT)) {
472 			jfs_error(sb, "invalid ea.flag\n");
473 			return -EIO;
474 		}
475 		current_blocks = (ea_size + sb->s_blocksize - 1) >>
476 		    sb->s_blocksize_bits;
477 	}
478 	size = max(min_size, ea_size);
479 
480 	if (size > PSIZE) {
481 		/*
482 		 * To keep the rest of the code simple.  Allocate a
483 		 * contiguous buffer to work with. Make the buffer large
484 		 * enough to make use of the whole extent.
485 		 */
486 		ea_buf->max_size = (size + sb->s_blocksize - 1) &
487 		    ~(sb->s_blocksize - 1);
488 
489 		ea_buf->xattr = kmalloc(ea_buf->max_size, GFP_KERNEL);
490 		if (ea_buf->xattr == NULL)
491 			return -ENOMEM;
492 
493 		ea_buf->flag = EA_MALLOC;
494 
495 		if (ea_size == 0)
496 			return 0;
497 
498 		if ((rc = ea_read(inode, ea_buf->xattr))) {
499 			kfree(ea_buf->xattr);
500 			ea_buf->xattr = NULL;
501 			return rc;
502 		}
503 		goto size_check;
504 	}
505 	blocks_needed = (min_size + sb->s_blocksize - 1) >>
506 	    sb->s_blocksize_bits;
507 
508 	if (blocks_needed > current_blocks) {
509 		/* Allocate new blocks to quota. */
510 		rc = dquot_alloc_block(inode, blocks_needed);
511 		if (rc)
512 			return -EDQUOT;
513 
514 		quota_allocation = blocks_needed;
515 
516 		rc = dbAlloc(inode, INOHINT(inode), (s64) blocks_needed,
517 			     &blkno);
518 		if (rc)
519 			goto clean_up;
520 
521 		DXDlength(&ea_buf->new_ea, blocks_needed);
522 		DXDaddress(&ea_buf->new_ea, blkno);
523 		ea_buf->new_ea.flag = DXD_EXTENT;
524 		DXDsize(&ea_buf->new_ea, min_size);
525 
526 		ea_buf->flag = EA_EXTENT | EA_NEW;
527 
528 		ea_buf->mp = get_metapage(inode, blkno,
529 					  blocks_needed << sb->s_blocksize_bits,
530 					  1);
531 		if (ea_buf->mp == NULL) {
532 			dbFree(inode, blkno, (s64) blocks_needed);
533 			rc = -EIO;
534 			goto clean_up;
535 		}
536 		ea_buf->xattr = ea_buf->mp->data;
537 		ea_buf->max_size = (min_size + sb->s_blocksize - 1) &
538 		    ~(sb->s_blocksize - 1);
539 		if (ea_size == 0)
540 			return 0;
541 		if ((rc = ea_read(inode, ea_buf->xattr))) {
542 			discard_metapage(ea_buf->mp);
543 			dbFree(inode, blkno, (s64) blocks_needed);
544 			goto clean_up;
545 		}
546 		goto size_check;
547 	}
548 	ea_buf->flag = EA_EXTENT;
549 	ea_buf->mp = read_metapage(inode, addressDXD(&ji->ea),
550 				   lengthDXD(&ji->ea) << sb->s_blocksize_bits,
551 				   1);
552 	if (ea_buf->mp == NULL) {
553 		rc = -EIO;
554 		goto clean_up;
555 	}
556 	ea_buf->xattr = ea_buf->mp->data;
557 	ea_buf->max_size = (ea_size + sb->s_blocksize - 1) &
558 	    ~(sb->s_blocksize - 1);
559 
560       size_check:
561 	if (EALIST_SIZE(ea_buf->xattr) != ea_size) {
562 		int size = min_t(int, EALIST_SIZE(ea_buf->xattr), ea_size);
563 
564 		printk(KERN_ERR "ea_get: invalid extended attribute\n");
565 		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_ADDRESS, 16, 1,
566 				     ea_buf->xattr, size, 1);
567 		ea_release(inode, ea_buf);
568 		rc = -EIO;
569 		goto clean_up;
570 	}
571 
572 	return ea_size;
573 
574       clean_up:
575 	/* Rollback quota allocation */
576 	if (quota_allocation)
577 		dquot_free_block(inode, quota_allocation);
578 
579 	return (rc);
580 }
581 
582 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf)
583 {
584 	if (ea_buf->flag & EA_MALLOC)
585 		kfree(ea_buf->xattr);
586 	else if (ea_buf->flag & EA_EXTENT) {
587 		assert(ea_buf->mp);
588 		release_metapage(ea_buf->mp);
589 
590 		if (ea_buf->flag & EA_NEW)
591 			dbFree(inode, addressDXD(&ea_buf->new_ea),
592 			       lengthDXD(&ea_buf->new_ea));
593 	}
594 }
595 
596 static int ea_put(tid_t tid, struct inode *inode, struct ea_buffer *ea_buf,
597 		  int new_size)
598 {
599 	struct jfs_inode_info *ji = JFS_IP(inode);
600 	unsigned long old_blocks, new_blocks;
601 	int rc = 0;
602 
603 	if (new_size == 0) {
604 		ea_release(inode, ea_buf);
605 		ea_buf = NULL;
606 	} else if (ea_buf->flag & EA_INLINE) {
607 		assert(new_size <= sizeof (ji->i_inline_ea));
608 		ji->mode2 &= ~INLINEEA;
609 		ea_buf->new_ea.flag = DXD_INLINE;
610 		DXDsize(&ea_buf->new_ea, new_size);
611 		DXDaddress(&ea_buf->new_ea, 0);
612 		DXDlength(&ea_buf->new_ea, 0);
613 	} else if (ea_buf->flag & EA_MALLOC) {
614 		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
615 		kfree(ea_buf->xattr);
616 	} else if (ea_buf->flag & EA_NEW) {
617 		/* We have already allocated a new dxd */
618 		flush_metapage(ea_buf->mp);
619 	} else {
620 		/* ->xattr must point to original ea's metapage */
621 		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
622 		discard_metapage(ea_buf->mp);
623 	}
624 	if (rc)
625 		return rc;
626 
627 	old_blocks = new_blocks = 0;
628 
629 	if (ji->ea.flag & DXD_EXTENT) {
630 		invalidate_dxd_metapages(inode, ji->ea);
631 		old_blocks = lengthDXD(&ji->ea);
632 	}
633 
634 	if (ea_buf) {
635 		txEA(tid, inode, &ji->ea, &ea_buf->new_ea);
636 		if (ea_buf->new_ea.flag & DXD_EXTENT) {
637 			new_blocks = lengthDXD(&ea_buf->new_ea);
638 			if (ji->ea.flag & DXD_INLINE)
639 				ji->mode2 |= INLINEEA;
640 		}
641 		ji->ea = ea_buf->new_ea;
642 	} else {
643 		txEA(tid, inode, &ji->ea, NULL);
644 		if (ji->ea.flag & DXD_INLINE)
645 			ji->mode2 |= INLINEEA;
646 		ji->ea.flag = 0;
647 		ji->ea.size = 0;
648 	}
649 
650 	/* If old blocks exist, they must be removed from quota allocation. */
651 	if (old_blocks)
652 		dquot_free_block(inode, old_blocks);
653 
654 	inode_set_ctime_current(inode);
655 
656 	return 0;
657 }
658 
659 int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
660 		   const void *value, size_t value_len, int flags)
661 {
662 	struct jfs_ea_list *ealist;
663 	struct jfs_ea *ea, *old_ea = NULL, *next_ea = NULL;
664 	struct ea_buffer ea_buf;
665 	int old_ea_size = 0;
666 	int xattr_size;
667 	int new_size;
668 	int namelen = strlen(name);
669 	int found = 0;
670 	int rc;
671 	int length;
672 
673 	down_write(&JFS_IP(inode)->xattr_sem);
674 
675 	xattr_size = ea_get(inode, &ea_buf, 0);
676 	if (xattr_size < 0) {
677 		rc = xattr_size;
678 		goto out;
679 	}
680 
681       again:
682 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
683 	new_size = sizeof (struct jfs_ea_list);
684 
685 	if (xattr_size) {
686 		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
687 		     ea = NEXT_EA(ea)) {
688 			if ((namelen == ea->namelen) &&
689 			    (memcmp(name, ea->name, namelen) == 0)) {
690 				found = 1;
691 				if (flags & XATTR_CREATE) {
692 					rc = -EEXIST;
693 					goto release;
694 				}
695 				old_ea = ea;
696 				old_ea_size = EA_SIZE(ea);
697 				next_ea = NEXT_EA(ea);
698 			} else
699 				new_size += EA_SIZE(ea);
700 		}
701 	}
702 
703 	if (!found) {
704 		if (flags & XATTR_REPLACE) {
705 			rc = -ENODATA;
706 			goto release;
707 		}
708 		if (value == NULL) {
709 			rc = 0;
710 			goto release;
711 		}
712 	}
713 	if (value)
714 		new_size += sizeof (struct jfs_ea) + namelen + 1 + value_len;
715 
716 	if (new_size > ea_buf.max_size) {
717 		/*
718 		 * We need to allocate more space for merged ea list.
719 		 * We should only have loop to again: once.
720 		 */
721 		ea_release(inode, &ea_buf);
722 		xattr_size = ea_get(inode, &ea_buf, new_size);
723 		if (xattr_size < 0) {
724 			rc = xattr_size;
725 			goto out;
726 		}
727 		goto again;
728 	}
729 
730 	/* Remove old ea of the same name */
731 	if (found) {
732 		/* number of bytes following target EA */
733 		length = (char *) END_EALIST(ealist) - (char *) next_ea;
734 		if (length > 0)
735 			memmove(old_ea, next_ea, length);
736 		xattr_size -= old_ea_size;
737 	}
738 
739 	/* Add new entry to the end */
740 	if (value) {
741 		if (xattr_size == 0)
742 			/* Completely new ea list */
743 			xattr_size = sizeof (struct jfs_ea_list);
744 
745 		/*
746 		 * The size of EA value is limitted by on-disk format up to
747 		 *  __le16, there would be an overflow if the size is equal
748 		 * to XATTR_SIZE_MAX (65536).  In order to avoid this issue,
749 		 * we can pre-checkup the value size against USHRT_MAX, and
750 		 * return -E2BIG in this case, which is consistent with the
751 		 * VFS setxattr interface.
752 		 */
753 		if (value_len >= USHRT_MAX) {
754 			rc = -E2BIG;
755 			goto release;
756 		}
757 
758 		ea = (struct jfs_ea *) ((char *) ealist + xattr_size);
759 		ea->flag = 0;
760 		ea->namelen = namelen;
761 		ea->valuelen = (cpu_to_le16(value_len));
762 		memcpy(ea->name, name, namelen);
763 		ea->name[namelen] = 0;
764 		if (value_len)
765 			memcpy(&ea->name[namelen + 1], value, value_len);
766 		xattr_size += EA_SIZE(ea);
767 	}
768 
769 	/* DEBUG - If we did this right, these number match */
770 	if (xattr_size != new_size) {
771 		printk(KERN_ERR
772 		       "__jfs_setxattr: xattr_size = %d, new_size = %d\n",
773 		       xattr_size, new_size);
774 
775 		rc = -EINVAL;
776 		goto release;
777 	}
778 
779 	/*
780 	 * If we're left with an empty list, there's no ea
781 	 */
782 	if (new_size == sizeof (struct jfs_ea_list))
783 		new_size = 0;
784 
785 	ealist->size = cpu_to_le32(new_size);
786 
787 	rc = ea_put(tid, inode, &ea_buf, new_size);
788 
789 	goto out;
790       release:
791 	ea_release(inode, &ea_buf);
792       out:
793 	up_write(&JFS_IP(inode)->xattr_sem);
794 
795 	return rc;
796 }
797 
798 ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,
799 		       size_t buf_size)
800 {
801 	struct jfs_ea_list *ealist;
802 	struct jfs_ea *ea, *ealist_end;
803 	struct ea_buffer ea_buf;
804 	int xattr_size;
805 	ssize_t size;
806 	int namelen = strlen(name);
807 	char *value;
808 
809 	down_read(&JFS_IP(inode)->xattr_sem);
810 
811 	xattr_size = ea_get(inode, &ea_buf, 0);
812 
813 	if (xattr_size < 0) {
814 		size = xattr_size;
815 		goto out;
816 	}
817 
818 	if (xattr_size == 0)
819 		goto not_found;
820 
821 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
822 	ealist_end = END_EALIST(ealist);
823 
824 	/* Find the named attribute */
825 	for (ea = FIRST_EA(ealist); ea < ealist_end; ea = NEXT_EA(ea)) {
826 		if (unlikely(ea + 1 > ealist_end) ||
827 		    unlikely(NEXT_EA(ea) > ealist_end)) {
828 			size = -EUCLEAN;
829 			goto release;
830 		}
831 
832 		if ((namelen == ea->namelen) &&
833 		    memcmp(name, ea->name, namelen) == 0) {
834 			/* Found it */
835 			size = le16_to_cpu(ea->valuelen);
836 			if (!data)
837 				goto release;
838 			else if (size > buf_size) {
839 				size = -ERANGE;
840 				goto release;
841 			}
842 			value = ((char *) &ea->name) + ea->namelen + 1;
843 			memcpy(data, value, size);
844 			goto release;
845 		}
846 	}
847       not_found:
848 	size = -ENODATA;
849       release:
850 	ea_release(inode, &ea_buf);
851       out:
852 	up_read(&JFS_IP(inode)->xattr_sem);
853 
854 	return size;
855 }
856 
857 /*
858  * No special permissions are needed to list attributes except for trusted.*
859  */
860 static inline int can_list(struct jfs_ea *ea)
861 {
862 	return (strncmp(ea->name, XATTR_TRUSTED_PREFIX,
863 			    XATTR_TRUSTED_PREFIX_LEN) ||
864 		capable(CAP_SYS_ADMIN));
865 }
866 
867 ssize_t jfs_listxattr(struct dentry * dentry, char *data, size_t buf_size)
868 {
869 	struct inode *inode = d_inode(dentry);
870 	char *buffer;
871 	ssize_t size = 0;
872 	int xattr_size;
873 	struct jfs_ea_list *ealist;
874 	struct jfs_ea *ea, *ealist_end;
875 	struct ea_buffer ea_buf;
876 
877 	down_read(&JFS_IP(inode)->xattr_sem);
878 
879 	xattr_size = ea_get(inode, &ea_buf, 0);
880 	if (xattr_size < 0) {
881 		size = xattr_size;
882 		goto out;
883 	}
884 
885 	if (xattr_size == 0)
886 		goto release;
887 
888 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
889 	ealist_end = END_EALIST(ealist);
890 
891 	/* compute required size of list */
892 	for (ea = FIRST_EA(ealist); ea < ealist_end; ea = NEXT_EA(ea)) {
893 		if (unlikely(ea + 1 > ealist_end) ||
894 		    unlikely(NEXT_EA(ea) > ealist_end)) {
895 			size = -EUCLEAN;
896 			goto release;
897 		}
898 
899 		if (can_list(ea))
900 			size += name_size(ea) + 1;
901 	}
902 
903 	if (!data)
904 		goto release;
905 
906 	if (size > buf_size) {
907 		size = -ERANGE;
908 		goto release;
909 	}
910 
911 	/* Copy attribute names to buffer */
912 	buffer = data;
913 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
914 		if (can_list(ea)) {
915 			int namelen = copy_name(buffer, ea);
916 			buffer += namelen + 1;
917 		}
918 	}
919 
920       release:
921 	ea_release(inode, &ea_buf);
922       out:
923 	up_read(&JFS_IP(inode)->xattr_sem);
924 	return size;
925 }
926 
927 static int __jfs_xattr_set(struct inode *inode, const char *name,
928 			   const void *value, size_t size, int flags)
929 {
930 	struct jfs_inode_info *ji = JFS_IP(inode);
931 	tid_t tid;
932 	int rc;
933 
934 	tid = txBegin(inode->i_sb, 0);
935 	mutex_lock(&ji->commit_mutex);
936 	rc = __jfs_setxattr(tid, inode, name, value, size, flags);
937 	if (!rc)
938 		rc = txCommit(tid, 1, &inode, 0);
939 	txEnd(tid);
940 	mutex_unlock(&ji->commit_mutex);
941 
942 	return rc;
943 }
944 
945 static int jfs_xattr_get(const struct xattr_handler *handler,
946 			 struct dentry *unused, struct inode *inode,
947 			 const char *name, void *value, size_t size)
948 {
949 	name = xattr_full_name(handler, name);
950 	return __jfs_getxattr(inode, name, value, size);
951 }
952 
953 static int jfs_xattr_set(const struct xattr_handler *handler,
954 			 struct mnt_idmap *idmap,
955 			 struct dentry *unused, struct inode *inode,
956 			 const char *name, const void *value,
957 			 size_t size, int flags)
958 {
959 	name = xattr_full_name(handler, name);
960 	return __jfs_xattr_set(inode, name, value, size, flags);
961 }
962 
963 static int jfs_xattr_get_os2(const struct xattr_handler *handler,
964 			     struct dentry *unused, struct inode *inode,
965 			     const char *name, void *value, size_t size)
966 {
967 	if (is_known_namespace(name))
968 		return -EOPNOTSUPP;
969 	return __jfs_getxattr(inode, name, value, size);
970 }
971 
972 static int jfs_xattr_set_os2(const struct xattr_handler *handler,
973 			     struct mnt_idmap *idmap,
974 			     struct dentry *unused, struct inode *inode,
975 			     const char *name, const void *value,
976 			     size_t size, int flags)
977 {
978 	if (is_known_namespace(name))
979 		return -EOPNOTSUPP;
980 	return __jfs_xattr_set(inode, name, value, size, flags);
981 }
982 
983 static const struct xattr_handler jfs_user_xattr_handler = {
984 	.prefix = XATTR_USER_PREFIX,
985 	.get = jfs_xattr_get,
986 	.set = jfs_xattr_set,
987 };
988 
989 static const struct xattr_handler jfs_os2_xattr_handler = {
990 	.prefix = XATTR_OS2_PREFIX,
991 	.get = jfs_xattr_get_os2,
992 	.set = jfs_xattr_set_os2,
993 };
994 
995 static const struct xattr_handler jfs_security_xattr_handler = {
996 	.prefix = XATTR_SECURITY_PREFIX,
997 	.get = jfs_xattr_get,
998 	.set = jfs_xattr_set,
999 };
1000 
1001 static const struct xattr_handler jfs_trusted_xattr_handler = {
1002 	.prefix = XATTR_TRUSTED_PREFIX,
1003 	.get = jfs_xattr_get,
1004 	.set = jfs_xattr_set,
1005 };
1006 
1007 const struct xattr_handler *jfs_xattr_handlers[] = {
1008 	&jfs_os2_xattr_handler,
1009 	&jfs_user_xattr_handler,
1010 	&jfs_security_xattr_handler,
1011 	&jfs_trusted_xattr_handler,
1012 	NULL,
1013 };
1014 
1015 
1016 #ifdef CONFIG_JFS_SECURITY
1017 static int jfs_initxattrs(struct inode *inode, const struct xattr *xattr_array,
1018 			  void *fs_info)
1019 {
1020 	const struct xattr *xattr;
1021 	tid_t *tid = fs_info;
1022 	char *name;
1023 	int err = 0;
1024 
1025 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1026 		name = kmalloc(XATTR_SECURITY_PREFIX_LEN +
1027 			       strlen(xattr->name) + 1, GFP_NOFS);
1028 		if (!name) {
1029 			err = -ENOMEM;
1030 			break;
1031 		}
1032 		strcpy(name, XATTR_SECURITY_PREFIX);
1033 		strcpy(name + XATTR_SECURITY_PREFIX_LEN, xattr->name);
1034 
1035 		err = __jfs_setxattr(*tid, inode, name,
1036 				     xattr->value, xattr->value_len, 0);
1037 		kfree(name);
1038 		if (err < 0)
1039 			break;
1040 	}
1041 	return err;
1042 }
1043 
1044 int jfs_init_security(tid_t tid, struct inode *inode, struct inode *dir,
1045 		      const struct qstr *qstr)
1046 {
1047 	return security_inode_init_security(inode, dir, qstr,
1048 					    &jfs_initxattrs, &tid);
1049 }
1050 #endif
1051