xref: /linux/fs/jfs/xattr.c (revision fd639726bf15fca8ee1a00dce8e0096d0ad9bd18)
1 /*
2  *   Copyright (C) International Business Machines  Corp., 2000-2004
3  *   Copyright (C) Christoph Hellwig, 2002
4  *
5  *   This program is free software;  you can redistribute it and/or modify
6  *   it under the terms of the GNU General Public License as published by
7  *   the Free Software Foundation; either version 2 of the License, or
8  *   (at your option) any later version.
9  *
10  *   This program is distributed in the hope that it will be useful,
11  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
12  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
13  *   the GNU General Public License for more details.
14  *
15  *   You should have received a copy of the GNU General Public License
16  *   along with this program;  if not, write to the Free Software
17  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18  */
19 
20 #include <linux/capability.h>
21 #include <linux/fs.h>
22 #include <linux/xattr.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/slab.h>
25 #include <linux/quotaops.h>
26 #include <linux/security.h>
27 #include "jfs_incore.h"
28 #include "jfs_superblock.h"
29 #include "jfs_dmap.h"
30 #include "jfs_debug.h"
31 #include "jfs_dinode.h"
32 #include "jfs_extent.h"
33 #include "jfs_metapage.h"
34 #include "jfs_xattr.h"
35 #include "jfs_acl.h"
36 
37 /*
38  *	jfs_xattr.c: extended attribute service
39  *
40  * Overall design --
41  *
42  * Format:
43  *
44  *   Extended attribute lists (jfs_ea_list) consist of an overall size (32 bit
45  *   value) and a variable (0 or more) number of extended attribute
46  *   entries.  Each extended attribute entry (jfs_ea) is a <name,value> double
47  *   where <name> is constructed from a null-terminated ascii string
48  *   (1 ... 255 bytes in the name) and <value> is arbitrary 8 bit data
49  *   (1 ... 65535 bytes).  The in-memory format is
50  *
51  *   0       1        2        4                4 + namelen + 1
52  *   +-------+--------+--------+----------------+-------------------+
53  *   | Flags | Name   | Value  | Name String \0 | Data . . . .      |
54  *   |       | Length | Length |                |                   |
55  *   +-------+--------+--------+----------------+-------------------+
56  *
57  *   A jfs_ea_list then is structured as
58  *
59  *   0            4                   4 + EA_SIZE(ea1)
60  *   +------------+-------------------+--------------------+-----
61  *   | Overall EA | First FEA Element | Second FEA Element | .....
62  *   | List Size  |                   |                    |
63  *   +------------+-------------------+--------------------+-----
64  *
65  *   On-disk:
66  *
67  *	FEALISTs are stored on disk using blocks allocated by dbAlloc() and
68  *	written directly. An EA list may be in-lined in the inode if there is
69  *	sufficient room available.
70  */
71 
72 struct ea_buffer {
73 	int flag;		/* Indicates what storage xattr points to */
74 	int max_size;		/* largest xattr that fits in current buffer */
75 	dxd_t new_ea;		/* dxd to replace ea when modifying xattr */
76 	struct metapage *mp;	/* metapage containing ea list */
77 	struct jfs_ea_list *xattr;	/* buffer containing ea list */
78 };
79 
80 /*
81  * ea_buffer.flag values
82  */
83 #define EA_INLINE	0x0001
84 #define EA_EXTENT	0x0002
85 #define EA_NEW		0x0004
86 #define EA_MALLOC	0x0008
87 
88 
89 /*
90  * Mapping of on-disk attribute names: for on-disk attribute names with an
91  * unknown prefix (not "system.", "user.", "security.", or "trusted."), the
92  * prefix "os2." is prepended.  On the way back to disk, "os2." prefixes are
93  * stripped and we make sure that the remaining name does not start with one
94  * of the know prefixes.
95  */
96 
97 static int is_known_namespace(const char *name)
98 {
99 	if (strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN) &&
100 	    strncmp(name, XATTR_USER_PREFIX, XATTR_USER_PREFIX_LEN) &&
101 	    strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN) &&
102 	    strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN))
103 		return false;
104 
105 	return true;
106 }
107 
108 static inline int name_size(struct jfs_ea *ea)
109 {
110 	if (is_known_namespace(ea->name))
111 		return ea->namelen;
112 	else
113 		return ea->namelen + XATTR_OS2_PREFIX_LEN;
114 }
115 
116 static inline int copy_name(char *buffer, struct jfs_ea *ea)
117 {
118 	int len = ea->namelen;
119 
120 	if (!is_known_namespace(ea->name)) {
121 		memcpy(buffer, XATTR_OS2_PREFIX, XATTR_OS2_PREFIX_LEN);
122 		buffer += XATTR_OS2_PREFIX_LEN;
123 		len += XATTR_OS2_PREFIX_LEN;
124 	}
125 	memcpy(buffer, ea->name, ea->namelen);
126 	buffer[ea->namelen] = 0;
127 
128 	return len;
129 }
130 
131 /* Forward references */
132 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf);
133 
134 /*
135  * NAME: ea_write_inline
136  *
137  * FUNCTION: Attempt to write an EA inline if area is available
138  *
139  * PRE CONDITIONS:
140  *	Already verified that the specified EA is small enough to fit inline
141  *
142  * PARAMETERS:
143  *	ip	- Inode pointer
144  *	ealist	- EA list pointer
145  *	size	- size of ealist in bytes
146  *	ea	- dxd_t structure to be filled in with necessary EA information
147  *		  if we successfully copy the EA inline
148  *
149  * NOTES:
150  *	Checks if the inode's inline area is available.  If so, copies EA inline
151  *	and sets <ea> fields appropriately.  Otherwise, returns failure, EA will
152  *	have to be put into an extent.
153  *
154  * RETURNS: 0 for successful copy to inline area; -1 if area not available
155  */
156 static int ea_write_inline(struct inode *ip, struct jfs_ea_list *ealist,
157 			   int size, dxd_t * ea)
158 {
159 	struct jfs_inode_info *ji = JFS_IP(ip);
160 
161 	/*
162 	 * Make sure we have an EA -- the NULL EA list is valid, but you
163 	 * can't copy it!
164 	 */
165 	if (ealist && size > sizeof (struct jfs_ea_list)) {
166 		assert(size <= sizeof (ji->i_inline_ea));
167 
168 		/*
169 		 * See if the space is available or if it is already being
170 		 * used for an inline EA.
171 		 */
172 		if (!(ji->mode2 & INLINEEA) && !(ji->ea.flag & DXD_INLINE))
173 			return -EPERM;
174 
175 		DXDsize(ea, size);
176 		DXDlength(ea, 0);
177 		DXDaddress(ea, 0);
178 		memcpy(ji->i_inline_ea, ealist, size);
179 		ea->flag = DXD_INLINE;
180 		ji->mode2 &= ~INLINEEA;
181 	} else {
182 		ea->flag = 0;
183 		DXDsize(ea, 0);
184 		DXDlength(ea, 0);
185 		DXDaddress(ea, 0);
186 
187 		/* Free up INLINE area */
188 		if (ji->ea.flag & DXD_INLINE)
189 			ji->mode2 |= INLINEEA;
190 	}
191 
192 	return 0;
193 }
194 
195 /*
196  * NAME: ea_write
197  *
198  * FUNCTION: Write an EA for an inode
199  *
200  * PRE CONDITIONS: EA has been verified
201  *
202  * PARAMETERS:
203  *	ip	- Inode pointer
204  *	ealist	- EA list pointer
205  *	size	- size of ealist in bytes
206  *	ea	- dxd_t structure to be filled in appropriately with where the
207  *		  EA was copied
208  *
209  * NOTES: Will write EA inline if able to, otherwise allocates blocks for an
210  *	extent and synchronously writes it to those blocks.
211  *
212  * RETURNS: 0 for success; Anything else indicates failure
213  */
214 static int ea_write(struct inode *ip, struct jfs_ea_list *ealist, int size,
215 		       dxd_t * ea)
216 {
217 	struct super_block *sb = ip->i_sb;
218 	struct jfs_inode_info *ji = JFS_IP(ip);
219 	struct jfs_sb_info *sbi = JFS_SBI(sb);
220 	int nblocks;
221 	s64 blkno;
222 	int rc = 0, i;
223 	char *cp;
224 	s32 nbytes, nb;
225 	s32 bytes_to_write;
226 	struct metapage *mp;
227 
228 	/*
229 	 * Quick check to see if this is an in-linable EA.  Short EAs
230 	 * and empty EAs are all in-linable, provided the space exists.
231 	 */
232 	if (!ealist || size <= sizeof (ji->i_inline_ea)) {
233 		if (!ea_write_inline(ip, ealist, size, ea))
234 			return 0;
235 	}
236 
237 	/* figure out how many blocks we need */
238 	nblocks = (size + (sb->s_blocksize - 1)) >> sb->s_blocksize_bits;
239 
240 	/* Allocate new blocks to quota. */
241 	rc = dquot_alloc_block(ip, nblocks);
242 	if (rc)
243 		return rc;
244 
245 	rc = dbAlloc(ip, INOHINT(ip), nblocks, &blkno);
246 	if (rc) {
247 		/*Rollback quota allocation. */
248 		dquot_free_block(ip, nblocks);
249 		return rc;
250 	}
251 
252 	/*
253 	 * Now have nblocks worth of storage to stuff into the FEALIST.
254 	 * loop over the FEALIST copying data into the buffer one page at
255 	 * a time.
256 	 */
257 	cp = (char *) ealist;
258 	nbytes = size;
259 	for (i = 0; i < nblocks; i += sbi->nbperpage) {
260 		/*
261 		 * Determine how many bytes for this request, and round up to
262 		 * the nearest aggregate block size
263 		 */
264 		nb = min(PSIZE, nbytes);
265 		bytes_to_write =
266 		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
267 		    << sb->s_blocksize_bits;
268 
269 		if (!(mp = get_metapage(ip, blkno + i, bytes_to_write, 1))) {
270 			rc = -EIO;
271 			goto failed;
272 		}
273 
274 		memcpy(mp->data, cp, nb);
275 
276 		/*
277 		 * We really need a way to propagate errors for
278 		 * forced writes like this one.  --hch
279 		 *
280 		 * (__write_metapage => release_metapage => flush_metapage)
281 		 */
282 #ifdef _JFS_FIXME
283 		if ((rc = flush_metapage(mp))) {
284 			/*
285 			 * the write failed -- this means that the buffer
286 			 * is still assigned and the blocks are not being
287 			 * used.  this seems like the best error recovery
288 			 * we can get ...
289 			 */
290 			goto failed;
291 		}
292 #else
293 		flush_metapage(mp);
294 #endif
295 
296 		cp += PSIZE;
297 		nbytes -= nb;
298 	}
299 
300 	ea->flag = DXD_EXTENT;
301 	DXDsize(ea, le32_to_cpu(ealist->size));
302 	DXDlength(ea, nblocks);
303 	DXDaddress(ea, blkno);
304 
305 	/* Free up INLINE area */
306 	if (ji->ea.flag & DXD_INLINE)
307 		ji->mode2 |= INLINEEA;
308 
309 	return 0;
310 
311       failed:
312 	/* Rollback quota allocation. */
313 	dquot_free_block(ip, nblocks);
314 
315 	dbFree(ip, blkno, nblocks);
316 	return rc;
317 }
318 
319 /*
320  * NAME: ea_read_inline
321  *
322  * FUNCTION: Read an inlined EA into user's buffer
323  *
324  * PARAMETERS:
325  *	ip	- Inode pointer
326  *	ealist	- Pointer to buffer to fill in with EA
327  *
328  * RETURNS: 0
329  */
330 static int ea_read_inline(struct inode *ip, struct jfs_ea_list *ealist)
331 {
332 	struct jfs_inode_info *ji = JFS_IP(ip);
333 	int ea_size = sizeDXD(&ji->ea);
334 
335 	if (ea_size == 0) {
336 		ealist->size = 0;
337 		return 0;
338 	}
339 
340 	/* Sanity Check */
341 	if ((sizeDXD(&ji->ea) > sizeof (ji->i_inline_ea)))
342 		return -EIO;
343 	if (le32_to_cpu(((struct jfs_ea_list *) &ji->i_inline_ea)->size)
344 	    != ea_size)
345 		return -EIO;
346 
347 	memcpy(ealist, ji->i_inline_ea, ea_size);
348 	return 0;
349 }
350 
351 /*
352  * NAME: ea_read
353  *
354  * FUNCTION: copy EA data into user's buffer
355  *
356  * PARAMETERS:
357  *	ip	- Inode pointer
358  *	ealist	- Pointer to buffer to fill in with EA
359  *
360  * NOTES:  If EA is inline calls ea_read_inline() to copy EA.
361  *
362  * RETURNS: 0 for success; other indicates failure
363  */
364 static int ea_read(struct inode *ip, struct jfs_ea_list *ealist)
365 {
366 	struct super_block *sb = ip->i_sb;
367 	struct jfs_inode_info *ji = JFS_IP(ip);
368 	struct jfs_sb_info *sbi = JFS_SBI(sb);
369 	int nblocks;
370 	s64 blkno;
371 	char *cp = (char *) ealist;
372 	int i;
373 	int nbytes, nb;
374 	s32 bytes_to_read;
375 	struct metapage *mp;
376 
377 	/* quick check for in-line EA */
378 	if (ji->ea.flag & DXD_INLINE)
379 		return ea_read_inline(ip, ealist);
380 
381 	nbytes = sizeDXD(&ji->ea);
382 	if (!nbytes) {
383 		jfs_error(sb, "nbytes is 0\n");
384 		return -EIO;
385 	}
386 
387 	/*
388 	 * Figure out how many blocks were allocated when this EA list was
389 	 * originally written to disk.
390 	 */
391 	nblocks = lengthDXD(&ji->ea) << sbi->l2nbperpage;
392 	blkno = addressDXD(&ji->ea) << sbi->l2nbperpage;
393 
394 	/*
395 	 * I have found the disk blocks which were originally used to store
396 	 * the FEALIST.  now i loop over each contiguous block copying the
397 	 * data into the buffer.
398 	 */
399 	for (i = 0; i < nblocks; i += sbi->nbperpage) {
400 		/*
401 		 * Determine how many bytes for this request, and round up to
402 		 * the nearest aggregate block size
403 		 */
404 		nb = min(PSIZE, nbytes);
405 		bytes_to_read =
406 		    ((((nb + sb->s_blocksize - 1)) >> sb->s_blocksize_bits))
407 		    << sb->s_blocksize_bits;
408 
409 		if (!(mp = read_metapage(ip, blkno + i, bytes_to_read, 1)))
410 			return -EIO;
411 
412 		memcpy(cp, mp->data, nb);
413 		release_metapage(mp);
414 
415 		cp += PSIZE;
416 		nbytes -= nb;
417 	}
418 
419 	return 0;
420 }
421 
422 /*
423  * NAME: ea_get
424  *
425  * FUNCTION: Returns buffer containing existing extended attributes.
426  *	     The size of the buffer will be the larger of the existing
427  *	     attributes size, or min_size.
428  *
429  *	     The buffer, which may be inlined in the inode or in the
430  *	     page cache must be release by calling ea_release or ea_put
431  *
432  * PARAMETERS:
433  *	inode	- Inode pointer
434  *	ea_buf	- Structure to be populated with ealist and its metadata
435  *	min_size- minimum size of buffer to be returned
436  *
437  * RETURNS: 0 for success; Other indicates failure
438  */
439 static int ea_get(struct inode *inode, struct ea_buffer *ea_buf, int min_size)
440 {
441 	struct jfs_inode_info *ji = JFS_IP(inode);
442 	struct super_block *sb = inode->i_sb;
443 	int size;
444 	int ea_size = sizeDXD(&ji->ea);
445 	int blocks_needed, current_blocks;
446 	s64 blkno;
447 	int rc;
448 	int quota_allocation = 0;
449 
450 	/* When fsck.jfs clears a bad ea, it doesn't clear the size */
451 	if (ji->ea.flag == 0)
452 		ea_size = 0;
453 
454 	if (ea_size == 0) {
455 		if (min_size == 0) {
456 			ea_buf->flag = 0;
457 			ea_buf->max_size = 0;
458 			ea_buf->xattr = NULL;
459 			return 0;
460 		}
461 		if ((min_size <= sizeof (ji->i_inline_ea)) &&
462 		    (ji->mode2 & INLINEEA)) {
463 			ea_buf->flag = EA_INLINE | EA_NEW;
464 			ea_buf->max_size = sizeof (ji->i_inline_ea);
465 			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
466 			DXDlength(&ea_buf->new_ea, 0);
467 			DXDaddress(&ea_buf->new_ea, 0);
468 			ea_buf->new_ea.flag = DXD_INLINE;
469 			DXDsize(&ea_buf->new_ea, min_size);
470 			return 0;
471 		}
472 		current_blocks = 0;
473 	} else if (ji->ea.flag & DXD_INLINE) {
474 		if (min_size <= sizeof (ji->i_inline_ea)) {
475 			ea_buf->flag = EA_INLINE;
476 			ea_buf->max_size = sizeof (ji->i_inline_ea);
477 			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
478 			goto size_check;
479 		}
480 		current_blocks = 0;
481 	} else {
482 		if (!(ji->ea.flag & DXD_EXTENT)) {
483 			jfs_error(sb, "invalid ea.flag\n");
484 			return -EIO;
485 		}
486 		current_blocks = (ea_size + sb->s_blocksize - 1) >>
487 		    sb->s_blocksize_bits;
488 	}
489 	size = max(min_size, ea_size);
490 
491 	if (size > PSIZE) {
492 		/*
493 		 * To keep the rest of the code simple.  Allocate a
494 		 * contiguous buffer to work with
495 		 */
496 		ea_buf->xattr = kmalloc(size, GFP_KERNEL);
497 		if (ea_buf->xattr == NULL)
498 			return -ENOMEM;
499 
500 		ea_buf->flag = EA_MALLOC;
501 		ea_buf->max_size = (size + sb->s_blocksize - 1) &
502 		    ~(sb->s_blocksize - 1);
503 
504 		if (ea_size == 0)
505 			return 0;
506 
507 		if ((rc = ea_read(inode, ea_buf->xattr))) {
508 			kfree(ea_buf->xattr);
509 			ea_buf->xattr = NULL;
510 			return rc;
511 		}
512 		goto size_check;
513 	}
514 	blocks_needed = (min_size + sb->s_blocksize - 1) >>
515 	    sb->s_blocksize_bits;
516 
517 	if (blocks_needed > current_blocks) {
518 		/* Allocate new blocks to quota. */
519 		rc = dquot_alloc_block(inode, blocks_needed);
520 		if (rc)
521 			return -EDQUOT;
522 
523 		quota_allocation = blocks_needed;
524 
525 		rc = dbAlloc(inode, INOHINT(inode), (s64) blocks_needed,
526 			     &blkno);
527 		if (rc)
528 			goto clean_up;
529 
530 		DXDlength(&ea_buf->new_ea, blocks_needed);
531 		DXDaddress(&ea_buf->new_ea, blkno);
532 		ea_buf->new_ea.flag = DXD_EXTENT;
533 		DXDsize(&ea_buf->new_ea, min_size);
534 
535 		ea_buf->flag = EA_EXTENT | EA_NEW;
536 
537 		ea_buf->mp = get_metapage(inode, blkno,
538 					  blocks_needed << sb->s_blocksize_bits,
539 					  1);
540 		if (ea_buf->mp == NULL) {
541 			dbFree(inode, blkno, (s64) blocks_needed);
542 			rc = -EIO;
543 			goto clean_up;
544 		}
545 		ea_buf->xattr = ea_buf->mp->data;
546 		ea_buf->max_size = (min_size + sb->s_blocksize - 1) &
547 		    ~(sb->s_blocksize - 1);
548 		if (ea_size == 0)
549 			return 0;
550 		if ((rc = ea_read(inode, ea_buf->xattr))) {
551 			discard_metapage(ea_buf->mp);
552 			dbFree(inode, blkno, (s64) blocks_needed);
553 			goto clean_up;
554 		}
555 		goto size_check;
556 	}
557 	ea_buf->flag = EA_EXTENT;
558 	ea_buf->mp = read_metapage(inode, addressDXD(&ji->ea),
559 				   lengthDXD(&ji->ea) << sb->s_blocksize_bits,
560 				   1);
561 	if (ea_buf->mp == NULL) {
562 		rc = -EIO;
563 		goto clean_up;
564 	}
565 	ea_buf->xattr = ea_buf->mp->data;
566 	ea_buf->max_size = (ea_size + sb->s_blocksize - 1) &
567 	    ~(sb->s_blocksize - 1);
568 
569       size_check:
570 	if (EALIST_SIZE(ea_buf->xattr) != ea_size) {
571 		printk(KERN_ERR "ea_get: invalid extended attribute\n");
572 		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_ADDRESS, 16, 1,
573 				     ea_buf->xattr, ea_size, 1);
574 		ea_release(inode, ea_buf);
575 		rc = -EIO;
576 		goto clean_up;
577 	}
578 
579 	return ea_size;
580 
581       clean_up:
582 	/* Rollback quota allocation */
583 	if (quota_allocation)
584 		dquot_free_block(inode, quota_allocation);
585 
586 	return (rc);
587 }
588 
589 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf)
590 {
591 	if (ea_buf->flag & EA_MALLOC)
592 		kfree(ea_buf->xattr);
593 	else if (ea_buf->flag & EA_EXTENT) {
594 		assert(ea_buf->mp);
595 		release_metapage(ea_buf->mp);
596 
597 		if (ea_buf->flag & EA_NEW)
598 			dbFree(inode, addressDXD(&ea_buf->new_ea),
599 			       lengthDXD(&ea_buf->new_ea));
600 	}
601 }
602 
603 static int ea_put(tid_t tid, struct inode *inode, struct ea_buffer *ea_buf,
604 		  int new_size)
605 {
606 	struct jfs_inode_info *ji = JFS_IP(inode);
607 	unsigned long old_blocks, new_blocks;
608 	int rc = 0;
609 
610 	if (new_size == 0) {
611 		ea_release(inode, ea_buf);
612 		ea_buf = NULL;
613 	} else if (ea_buf->flag & EA_INLINE) {
614 		assert(new_size <= sizeof (ji->i_inline_ea));
615 		ji->mode2 &= ~INLINEEA;
616 		ea_buf->new_ea.flag = DXD_INLINE;
617 		DXDsize(&ea_buf->new_ea, new_size);
618 		DXDaddress(&ea_buf->new_ea, 0);
619 		DXDlength(&ea_buf->new_ea, 0);
620 	} else if (ea_buf->flag & EA_MALLOC) {
621 		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
622 		kfree(ea_buf->xattr);
623 	} else if (ea_buf->flag & EA_NEW) {
624 		/* We have already allocated a new dxd */
625 		flush_metapage(ea_buf->mp);
626 	} else {
627 		/* ->xattr must point to original ea's metapage */
628 		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
629 		discard_metapage(ea_buf->mp);
630 	}
631 	if (rc)
632 		return rc;
633 
634 	old_blocks = new_blocks = 0;
635 
636 	if (ji->ea.flag & DXD_EXTENT) {
637 		invalidate_dxd_metapages(inode, ji->ea);
638 		old_blocks = lengthDXD(&ji->ea);
639 	}
640 
641 	if (ea_buf) {
642 		txEA(tid, inode, &ji->ea, &ea_buf->new_ea);
643 		if (ea_buf->new_ea.flag & DXD_EXTENT) {
644 			new_blocks = lengthDXD(&ea_buf->new_ea);
645 			if (ji->ea.flag & DXD_INLINE)
646 				ji->mode2 |= INLINEEA;
647 		}
648 		ji->ea = ea_buf->new_ea;
649 	} else {
650 		txEA(tid, inode, &ji->ea, NULL);
651 		if (ji->ea.flag & DXD_INLINE)
652 			ji->mode2 |= INLINEEA;
653 		ji->ea.flag = 0;
654 		ji->ea.size = 0;
655 	}
656 
657 	/* If old blocks exist, they must be removed from quota allocation. */
658 	if (old_blocks)
659 		dquot_free_block(inode, old_blocks);
660 
661 	inode->i_ctime = current_time(inode);
662 
663 	return 0;
664 }
665 
666 int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
667 		   const void *value, size_t value_len, int flags)
668 {
669 	struct jfs_ea_list *ealist;
670 	struct jfs_ea *ea, *old_ea = NULL, *next_ea = NULL;
671 	struct ea_buffer ea_buf;
672 	int old_ea_size = 0;
673 	int xattr_size;
674 	int new_size;
675 	int namelen = strlen(name);
676 	int found = 0;
677 	int rc;
678 	int length;
679 
680 	down_write(&JFS_IP(inode)->xattr_sem);
681 
682 	xattr_size = ea_get(inode, &ea_buf, 0);
683 	if (xattr_size < 0) {
684 		rc = xattr_size;
685 		goto out;
686 	}
687 
688       again:
689 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
690 	new_size = sizeof (struct jfs_ea_list);
691 
692 	if (xattr_size) {
693 		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
694 		     ea = NEXT_EA(ea)) {
695 			if ((namelen == ea->namelen) &&
696 			    (memcmp(name, ea->name, namelen) == 0)) {
697 				found = 1;
698 				if (flags & XATTR_CREATE) {
699 					rc = -EEXIST;
700 					goto release;
701 				}
702 				old_ea = ea;
703 				old_ea_size = EA_SIZE(ea);
704 				next_ea = NEXT_EA(ea);
705 			} else
706 				new_size += EA_SIZE(ea);
707 		}
708 	}
709 
710 	if (!found) {
711 		if (flags & XATTR_REPLACE) {
712 			rc = -ENODATA;
713 			goto release;
714 		}
715 		if (value == NULL) {
716 			rc = 0;
717 			goto release;
718 		}
719 	}
720 	if (value)
721 		new_size += sizeof (struct jfs_ea) + namelen + 1 + value_len;
722 
723 	if (new_size > ea_buf.max_size) {
724 		/*
725 		 * We need to allocate more space for merged ea list.
726 		 * We should only have loop to again: once.
727 		 */
728 		ea_release(inode, &ea_buf);
729 		xattr_size = ea_get(inode, &ea_buf, new_size);
730 		if (xattr_size < 0) {
731 			rc = xattr_size;
732 			goto out;
733 		}
734 		goto again;
735 	}
736 
737 	/* Remove old ea of the same name */
738 	if (found) {
739 		/* number of bytes following target EA */
740 		length = (char *) END_EALIST(ealist) - (char *) next_ea;
741 		if (length > 0)
742 			memmove(old_ea, next_ea, length);
743 		xattr_size -= old_ea_size;
744 	}
745 
746 	/* Add new entry to the end */
747 	if (value) {
748 		if (xattr_size == 0)
749 			/* Completely new ea list */
750 			xattr_size = sizeof (struct jfs_ea_list);
751 
752 		/*
753 		 * The size of EA value is limitted by on-disk format up to
754 		 *  __le16, there would be an overflow if the size is equal
755 		 * to XATTR_SIZE_MAX (65536).  In order to avoid this issue,
756 		 * we can pre-checkup the value size against USHRT_MAX, and
757 		 * return -E2BIG in this case, which is consistent with the
758 		 * VFS setxattr interface.
759 		 */
760 		if (value_len >= USHRT_MAX) {
761 			rc = -E2BIG;
762 			goto release;
763 		}
764 
765 		ea = (struct jfs_ea *) ((char *) ealist + xattr_size);
766 		ea->flag = 0;
767 		ea->namelen = namelen;
768 		ea->valuelen = (cpu_to_le16(value_len));
769 		memcpy(ea->name, name, namelen);
770 		ea->name[namelen] = 0;
771 		if (value_len)
772 			memcpy(&ea->name[namelen + 1], value, value_len);
773 		xattr_size += EA_SIZE(ea);
774 	}
775 
776 	/* DEBUG - If we did this right, these number match */
777 	if (xattr_size != new_size) {
778 		printk(KERN_ERR
779 		       "__jfs_setxattr: xattr_size = %d, new_size = %d\n",
780 		       xattr_size, new_size);
781 
782 		rc = -EINVAL;
783 		goto release;
784 	}
785 
786 	/*
787 	 * If we're left with an empty list, there's no ea
788 	 */
789 	if (new_size == sizeof (struct jfs_ea_list))
790 		new_size = 0;
791 
792 	ealist->size = cpu_to_le32(new_size);
793 
794 	rc = ea_put(tid, inode, &ea_buf, new_size);
795 
796 	goto out;
797       release:
798 	ea_release(inode, &ea_buf);
799       out:
800 	up_write(&JFS_IP(inode)->xattr_sem);
801 
802 	return rc;
803 }
804 
805 ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,
806 		       size_t buf_size)
807 {
808 	struct jfs_ea_list *ealist;
809 	struct jfs_ea *ea;
810 	struct ea_buffer ea_buf;
811 	int xattr_size;
812 	ssize_t size;
813 	int namelen = strlen(name);
814 	char *value;
815 
816 	down_read(&JFS_IP(inode)->xattr_sem);
817 
818 	xattr_size = ea_get(inode, &ea_buf, 0);
819 
820 	if (xattr_size < 0) {
821 		size = xattr_size;
822 		goto out;
823 	}
824 
825 	if (xattr_size == 0)
826 		goto not_found;
827 
828 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
829 
830 	/* Find the named attribute */
831 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea))
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       not_found:
847 	size = -ENODATA;
848       release:
849 	ea_release(inode, &ea_buf);
850       out:
851 	up_read(&JFS_IP(inode)->xattr_sem);
852 
853 	return size;
854 }
855 
856 /*
857  * No special permissions are needed to list attributes except for trusted.*
858  */
859 static inline int can_list(struct jfs_ea *ea)
860 {
861 	return (strncmp(ea->name, XATTR_TRUSTED_PREFIX,
862 			    XATTR_TRUSTED_PREFIX_LEN) ||
863 		capable(CAP_SYS_ADMIN));
864 }
865 
866 ssize_t jfs_listxattr(struct dentry * dentry, char *data, size_t buf_size)
867 {
868 	struct inode *inode = d_inode(dentry);
869 	char *buffer;
870 	ssize_t size = 0;
871 	int xattr_size;
872 	struct jfs_ea_list *ealist;
873 	struct jfs_ea *ea;
874 	struct ea_buffer ea_buf;
875 
876 	down_read(&JFS_IP(inode)->xattr_sem);
877 
878 	xattr_size = ea_get(inode, &ea_buf, 0);
879 	if (xattr_size < 0) {
880 		size = xattr_size;
881 		goto out;
882 	}
883 
884 	if (xattr_size == 0)
885 		goto release;
886 
887 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
888 
889 	/* compute required size of list */
890 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
891 		if (can_list(ea))
892 			size += name_size(ea) + 1;
893 	}
894 
895 	if (!data)
896 		goto release;
897 
898 	if (size > buf_size) {
899 		size = -ERANGE;
900 		goto release;
901 	}
902 
903 	/* Copy attribute names to buffer */
904 	buffer = data;
905 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
906 		if (can_list(ea)) {
907 			int namelen = copy_name(buffer, ea);
908 			buffer += namelen + 1;
909 		}
910 	}
911 
912       release:
913 	ea_release(inode, &ea_buf);
914       out:
915 	up_read(&JFS_IP(inode)->xattr_sem);
916 	return size;
917 }
918 
919 static int __jfs_xattr_set(struct inode *inode, const char *name,
920 			   const void *value, size_t size, int flags)
921 {
922 	struct jfs_inode_info *ji = JFS_IP(inode);
923 	tid_t tid;
924 	int rc;
925 
926 	tid = txBegin(inode->i_sb, 0);
927 	mutex_lock(&ji->commit_mutex);
928 	rc = __jfs_setxattr(tid, inode, name, value, size, flags);
929 	if (!rc)
930 		rc = txCommit(tid, 1, &inode, 0);
931 	txEnd(tid);
932 	mutex_unlock(&ji->commit_mutex);
933 
934 	return rc;
935 }
936 
937 static int jfs_xattr_get(const struct xattr_handler *handler,
938 			 struct dentry *unused, struct inode *inode,
939 			 const char *name, void *value, size_t size)
940 {
941 	name = xattr_full_name(handler, name);
942 	return __jfs_getxattr(inode, name, value, size);
943 }
944 
945 static int jfs_xattr_set(const struct xattr_handler *handler,
946 			 struct dentry *unused, struct inode *inode,
947 			 const char *name, const void *value,
948 			 size_t size, int flags)
949 {
950 	name = xattr_full_name(handler, name);
951 	return __jfs_xattr_set(inode, name, value, size, flags);
952 }
953 
954 static int jfs_xattr_get_os2(const struct xattr_handler *handler,
955 			     struct dentry *unused, struct inode *inode,
956 			     const char *name, void *value, size_t size)
957 {
958 	if (is_known_namespace(name))
959 		return -EOPNOTSUPP;
960 	return __jfs_getxattr(inode, name, value, size);
961 }
962 
963 static int jfs_xattr_set_os2(const struct xattr_handler *handler,
964 			     struct dentry *unused, struct inode *inode,
965 			     const char *name, const void *value,
966 			     size_t size, int flags)
967 {
968 	if (is_known_namespace(name))
969 		return -EOPNOTSUPP;
970 	return __jfs_xattr_set(inode, name, value, size, flags);
971 }
972 
973 static const struct xattr_handler jfs_user_xattr_handler = {
974 	.prefix = XATTR_USER_PREFIX,
975 	.get = jfs_xattr_get,
976 	.set = jfs_xattr_set,
977 };
978 
979 static const struct xattr_handler jfs_os2_xattr_handler = {
980 	.prefix = XATTR_OS2_PREFIX,
981 	.get = jfs_xattr_get_os2,
982 	.set = jfs_xattr_set_os2,
983 };
984 
985 static const struct xattr_handler jfs_security_xattr_handler = {
986 	.prefix = XATTR_SECURITY_PREFIX,
987 	.get = jfs_xattr_get,
988 	.set = jfs_xattr_set,
989 };
990 
991 static const struct xattr_handler jfs_trusted_xattr_handler = {
992 	.prefix = XATTR_TRUSTED_PREFIX,
993 	.get = jfs_xattr_get,
994 	.set = jfs_xattr_set,
995 };
996 
997 const struct xattr_handler *jfs_xattr_handlers[] = {
998 #ifdef CONFIG_JFS_POSIX_ACL
999 	&posix_acl_access_xattr_handler,
1000 	&posix_acl_default_xattr_handler,
1001 #endif
1002 	&jfs_os2_xattr_handler,
1003 	&jfs_user_xattr_handler,
1004 	&jfs_security_xattr_handler,
1005 	&jfs_trusted_xattr_handler,
1006 	NULL,
1007 };
1008 
1009 
1010 #ifdef CONFIG_JFS_SECURITY
1011 static int jfs_initxattrs(struct inode *inode, const struct xattr *xattr_array,
1012 			  void *fs_info)
1013 {
1014 	const struct xattr *xattr;
1015 	tid_t *tid = fs_info;
1016 	char *name;
1017 	int err = 0;
1018 
1019 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1020 		name = kmalloc(XATTR_SECURITY_PREFIX_LEN +
1021 			       strlen(xattr->name) + 1, GFP_NOFS);
1022 		if (!name) {
1023 			err = -ENOMEM;
1024 			break;
1025 		}
1026 		strcpy(name, XATTR_SECURITY_PREFIX);
1027 		strcpy(name + XATTR_SECURITY_PREFIX_LEN, xattr->name);
1028 
1029 		err = __jfs_setxattr(*tid, inode, name,
1030 				     xattr->value, xattr->value_len, 0);
1031 		kfree(name);
1032 		if (err < 0)
1033 			break;
1034 	}
1035 	return err;
1036 }
1037 
1038 int jfs_init_security(tid_t tid, struct inode *inode, struct inode *dir,
1039 		      const struct qstr *qstr)
1040 {
1041 	return security_inode_init_security(inode, dir, qstr,
1042 					    &jfs_initxattrs, &tid);
1043 }
1044 #endif
1045