xref: /openbmc/linux/fs/jfs/xattr.c (revision 3db55767)
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 	/* When fsck.jfs clears a bad ea, it doesn't clear the size */
438 	if (ji->ea.flag == 0)
439 		ea_size = 0;
440 
441 	if (ea_size == 0) {
442 		if (min_size == 0) {
443 			ea_buf->flag = 0;
444 			ea_buf->max_size = 0;
445 			ea_buf->xattr = NULL;
446 			return 0;
447 		}
448 		if ((min_size <= sizeof (ji->i_inline_ea)) &&
449 		    (ji->mode2 & INLINEEA)) {
450 			ea_buf->flag = EA_INLINE | EA_NEW;
451 			ea_buf->max_size = sizeof (ji->i_inline_ea);
452 			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
453 			DXDlength(&ea_buf->new_ea, 0);
454 			DXDaddress(&ea_buf->new_ea, 0);
455 			ea_buf->new_ea.flag = DXD_INLINE;
456 			DXDsize(&ea_buf->new_ea, min_size);
457 			return 0;
458 		}
459 		current_blocks = 0;
460 	} else if (ji->ea.flag & DXD_INLINE) {
461 		if (min_size <= sizeof (ji->i_inline_ea)) {
462 			ea_buf->flag = EA_INLINE;
463 			ea_buf->max_size = sizeof (ji->i_inline_ea);
464 			ea_buf->xattr = (struct jfs_ea_list *) ji->i_inline_ea;
465 			goto size_check;
466 		}
467 		current_blocks = 0;
468 	} else {
469 		if (!(ji->ea.flag & DXD_EXTENT)) {
470 			jfs_error(sb, "invalid ea.flag\n");
471 			return -EIO;
472 		}
473 		current_blocks = (ea_size + sb->s_blocksize - 1) >>
474 		    sb->s_blocksize_bits;
475 	}
476 	size = max(min_size, ea_size);
477 
478 	if (size > PSIZE) {
479 		/*
480 		 * To keep the rest of the code simple.  Allocate a
481 		 * contiguous buffer to work with. Make the buffer large
482 		 * enough to make use of the whole extent.
483 		 */
484 		ea_buf->max_size = (size + sb->s_blocksize - 1) &
485 		    ~(sb->s_blocksize - 1);
486 
487 		ea_buf->xattr = kmalloc(ea_buf->max_size, GFP_KERNEL);
488 		if (ea_buf->xattr == NULL)
489 			return -ENOMEM;
490 
491 		ea_buf->flag = EA_MALLOC;
492 
493 		if (ea_size == 0)
494 			return 0;
495 
496 		if ((rc = ea_read(inode, ea_buf->xattr))) {
497 			kfree(ea_buf->xattr);
498 			ea_buf->xattr = NULL;
499 			return rc;
500 		}
501 		goto size_check;
502 	}
503 	blocks_needed = (min_size + sb->s_blocksize - 1) >>
504 	    sb->s_blocksize_bits;
505 
506 	if (blocks_needed > current_blocks) {
507 		/* Allocate new blocks to quota. */
508 		rc = dquot_alloc_block(inode, blocks_needed);
509 		if (rc)
510 			return -EDQUOT;
511 
512 		quota_allocation = blocks_needed;
513 
514 		rc = dbAlloc(inode, INOHINT(inode), (s64) blocks_needed,
515 			     &blkno);
516 		if (rc)
517 			goto clean_up;
518 
519 		DXDlength(&ea_buf->new_ea, blocks_needed);
520 		DXDaddress(&ea_buf->new_ea, blkno);
521 		ea_buf->new_ea.flag = DXD_EXTENT;
522 		DXDsize(&ea_buf->new_ea, min_size);
523 
524 		ea_buf->flag = EA_EXTENT | EA_NEW;
525 
526 		ea_buf->mp = get_metapage(inode, blkno,
527 					  blocks_needed << sb->s_blocksize_bits,
528 					  1);
529 		if (ea_buf->mp == NULL) {
530 			dbFree(inode, blkno, (s64) blocks_needed);
531 			rc = -EIO;
532 			goto clean_up;
533 		}
534 		ea_buf->xattr = ea_buf->mp->data;
535 		ea_buf->max_size = (min_size + sb->s_blocksize - 1) &
536 		    ~(sb->s_blocksize - 1);
537 		if (ea_size == 0)
538 			return 0;
539 		if ((rc = ea_read(inode, ea_buf->xattr))) {
540 			discard_metapage(ea_buf->mp);
541 			dbFree(inode, blkno, (s64) blocks_needed);
542 			goto clean_up;
543 		}
544 		goto size_check;
545 	}
546 	ea_buf->flag = EA_EXTENT;
547 	ea_buf->mp = read_metapage(inode, addressDXD(&ji->ea),
548 				   lengthDXD(&ji->ea) << sb->s_blocksize_bits,
549 				   1);
550 	if (ea_buf->mp == NULL) {
551 		rc = -EIO;
552 		goto clean_up;
553 	}
554 	ea_buf->xattr = ea_buf->mp->data;
555 	ea_buf->max_size = (ea_size + sb->s_blocksize - 1) &
556 	    ~(sb->s_blocksize - 1);
557 
558       size_check:
559 	if (EALIST_SIZE(ea_buf->xattr) != ea_size) {
560 		printk(KERN_ERR "ea_get: invalid extended attribute\n");
561 		print_hex_dump(KERN_ERR, "", DUMP_PREFIX_ADDRESS, 16, 1,
562 				     ea_buf->xattr, ea_size, 1);
563 		ea_release(inode, ea_buf);
564 		rc = -EIO;
565 		goto clean_up;
566 	}
567 
568 	return ea_size;
569 
570       clean_up:
571 	/* Rollback quota allocation */
572 	if (quota_allocation)
573 		dquot_free_block(inode, quota_allocation);
574 
575 	return (rc);
576 }
577 
578 static void ea_release(struct inode *inode, struct ea_buffer *ea_buf)
579 {
580 	if (ea_buf->flag & EA_MALLOC)
581 		kfree(ea_buf->xattr);
582 	else if (ea_buf->flag & EA_EXTENT) {
583 		assert(ea_buf->mp);
584 		release_metapage(ea_buf->mp);
585 
586 		if (ea_buf->flag & EA_NEW)
587 			dbFree(inode, addressDXD(&ea_buf->new_ea),
588 			       lengthDXD(&ea_buf->new_ea));
589 	}
590 }
591 
592 static int ea_put(tid_t tid, struct inode *inode, struct ea_buffer *ea_buf,
593 		  int new_size)
594 {
595 	struct jfs_inode_info *ji = JFS_IP(inode);
596 	unsigned long old_blocks, new_blocks;
597 	int rc = 0;
598 
599 	if (new_size == 0) {
600 		ea_release(inode, ea_buf);
601 		ea_buf = NULL;
602 	} else if (ea_buf->flag & EA_INLINE) {
603 		assert(new_size <= sizeof (ji->i_inline_ea));
604 		ji->mode2 &= ~INLINEEA;
605 		ea_buf->new_ea.flag = DXD_INLINE;
606 		DXDsize(&ea_buf->new_ea, new_size);
607 		DXDaddress(&ea_buf->new_ea, 0);
608 		DXDlength(&ea_buf->new_ea, 0);
609 	} else if (ea_buf->flag & EA_MALLOC) {
610 		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
611 		kfree(ea_buf->xattr);
612 	} else if (ea_buf->flag & EA_NEW) {
613 		/* We have already allocated a new dxd */
614 		flush_metapage(ea_buf->mp);
615 	} else {
616 		/* ->xattr must point to original ea's metapage */
617 		rc = ea_write(inode, ea_buf->xattr, new_size, &ea_buf->new_ea);
618 		discard_metapage(ea_buf->mp);
619 	}
620 	if (rc)
621 		return rc;
622 
623 	old_blocks = new_blocks = 0;
624 
625 	if (ji->ea.flag & DXD_EXTENT) {
626 		invalidate_dxd_metapages(inode, ji->ea);
627 		old_blocks = lengthDXD(&ji->ea);
628 	}
629 
630 	if (ea_buf) {
631 		txEA(tid, inode, &ji->ea, &ea_buf->new_ea);
632 		if (ea_buf->new_ea.flag & DXD_EXTENT) {
633 			new_blocks = lengthDXD(&ea_buf->new_ea);
634 			if (ji->ea.flag & DXD_INLINE)
635 				ji->mode2 |= INLINEEA;
636 		}
637 		ji->ea = ea_buf->new_ea;
638 	} else {
639 		txEA(tid, inode, &ji->ea, NULL);
640 		if (ji->ea.flag & DXD_INLINE)
641 			ji->mode2 |= INLINEEA;
642 		ji->ea.flag = 0;
643 		ji->ea.size = 0;
644 	}
645 
646 	/* If old blocks exist, they must be removed from quota allocation. */
647 	if (old_blocks)
648 		dquot_free_block(inode, old_blocks);
649 
650 	inode->i_ctime = current_time(inode);
651 
652 	return 0;
653 }
654 
655 int __jfs_setxattr(tid_t tid, struct inode *inode, const char *name,
656 		   const void *value, size_t value_len, int flags)
657 {
658 	struct jfs_ea_list *ealist;
659 	struct jfs_ea *ea, *old_ea = NULL, *next_ea = NULL;
660 	struct ea_buffer ea_buf;
661 	int old_ea_size = 0;
662 	int xattr_size;
663 	int new_size;
664 	int namelen = strlen(name);
665 	int found = 0;
666 	int rc;
667 	int length;
668 
669 	down_write(&JFS_IP(inode)->xattr_sem);
670 
671 	xattr_size = ea_get(inode, &ea_buf, 0);
672 	if (xattr_size < 0) {
673 		rc = xattr_size;
674 		goto out;
675 	}
676 
677       again:
678 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
679 	new_size = sizeof (struct jfs_ea_list);
680 
681 	if (xattr_size) {
682 		for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist);
683 		     ea = NEXT_EA(ea)) {
684 			if ((namelen == ea->namelen) &&
685 			    (memcmp(name, ea->name, namelen) == 0)) {
686 				found = 1;
687 				if (flags & XATTR_CREATE) {
688 					rc = -EEXIST;
689 					goto release;
690 				}
691 				old_ea = ea;
692 				old_ea_size = EA_SIZE(ea);
693 				next_ea = NEXT_EA(ea);
694 			} else
695 				new_size += EA_SIZE(ea);
696 		}
697 	}
698 
699 	if (!found) {
700 		if (flags & XATTR_REPLACE) {
701 			rc = -ENODATA;
702 			goto release;
703 		}
704 		if (value == NULL) {
705 			rc = 0;
706 			goto release;
707 		}
708 	}
709 	if (value)
710 		new_size += sizeof (struct jfs_ea) + namelen + 1 + value_len;
711 
712 	if (new_size > ea_buf.max_size) {
713 		/*
714 		 * We need to allocate more space for merged ea list.
715 		 * We should only have loop to again: once.
716 		 */
717 		ea_release(inode, &ea_buf);
718 		xattr_size = ea_get(inode, &ea_buf, new_size);
719 		if (xattr_size < 0) {
720 			rc = xattr_size;
721 			goto out;
722 		}
723 		goto again;
724 	}
725 
726 	/* Remove old ea of the same name */
727 	if (found) {
728 		/* number of bytes following target EA */
729 		length = (char *) END_EALIST(ealist) - (char *) next_ea;
730 		if (length > 0)
731 			memmove(old_ea, next_ea, length);
732 		xattr_size -= old_ea_size;
733 	}
734 
735 	/* Add new entry to the end */
736 	if (value) {
737 		if (xattr_size == 0)
738 			/* Completely new ea list */
739 			xattr_size = sizeof (struct jfs_ea_list);
740 
741 		/*
742 		 * The size of EA value is limitted by on-disk format up to
743 		 *  __le16, there would be an overflow if the size is equal
744 		 * to XATTR_SIZE_MAX (65536).  In order to avoid this issue,
745 		 * we can pre-checkup the value size against USHRT_MAX, and
746 		 * return -E2BIG in this case, which is consistent with the
747 		 * VFS setxattr interface.
748 		 */
749 		if (value_len >= USHRT_MAX) {
750 			rc = -E2BIG;
751 			goto release;
752 		}
753 
754 		ea = (struct jfs_ea *) ((char *) ealist + xattr_size);
755 		ea->flag = 0;
756 		ea->namelen = namelen;
757 		ea->valuelen = (cpu_to_le16(value_len));
758 		memcpy(ea->name, name, namelen);
759 		ea->name[namelen] = 0;
760 		if (value_len)
761 			memcpy(&ea->name[namelen + 1], value, value_len);
762 		xattr_size += EA_SIZE(ea);
763 	}
764 
765 	/* DEBUG - If we did this right, these number match */
766 	if (xattr_size != new_size) {
767 		printk(KERN_ERR
768 		       "__jfs_setxattr: xattr_size = %d, new_size = %d\n",
769 		       xattr_size, new_size);
770 
771 		rc = -EINVAL;
772 		goto release;
773 	}
774 
775 	/*
776 	 * If we're left with an empty list, there's no ea
777 	 */
778 	if (new_size == sizeof (struct jfs_ea_list))
779 		new_size = 0;
780 
781 	ealist->size = cpu_to_le32(new_size);
782 
783 	rc = ea_put(tid, inode, &ea_buf, new_size);
784 
785 	goto out;
786       release:
787 	ea_release(inode, &ea_buf);
788       out:
789 	up_write(&JFS_IP(inode)->xattr_sem);
790 
791 	return rc;
792 }
793 
794 ssize_t __jfs_getxattr(struct inode *inode, const char *name, void *data,
795 		       size_t buf_size)
796 {
797 	struct jfs_ea_list *ealist;
798 	struct jfs_ea *ea;
799 	struct ea_buffer ea_buf;
800 	int xattr_size;
801 	ssize_t size;
802 	int namelen = strlen(name);
803 	char *value;
804 
805 	down_read(&JFS_IP(inode)->xattr_sem);
806 
807 	xattr_size = ea_get(inode, &ea_buf, 0);
808 
809 	if (xattr_size < 0) {
810 		size = xattr_size;
811 		goto out;
812 	}
813 
814 	if (xattr_size == 0)
815 		goto not_found;
816 
817 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
818 
819 	/* Find the named attribute */
820 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea))
821 		if ((namelen == ea->namelen) &&
822 		    memcmp(name, ea->name, namelen) == 0) {
823 			/* Found it */
824 			size = le16_to_cpu(ea->valuelen);
825 			if (!data)
826 				goto release;
827 			else if (size > buf_size) {
828 				size = -ERANGE;
829 				goto release;
830 			}
831 			value = ((char *) &ea->name) + ea->namelen + 1;
832 			memcpy(data, value, size);
833 			goto release;
834 		}
835       not_found:
836 	size = -ENODATA;
837       release:
838 	ea_release(inode, &ea_buf);
839       out:
840 	up_read(&JFS_IP(inode)->xattr_sem);
841 
842 	return size;
843 }
844 
845 /*
846  * No special permissions are needed to list attributes except for trusted.*
847  */
848 static inline int can_list(struct jfs_ea *ea)
849 {
850 	return (strncmp(ea->name, XATTR_TRUSTED_PREFIX,
851 			    XATTR_TRUSTED_PREFIX_LEN) ||
852 		capable(CAP_SYS_ADMIN));
853 }
854 
855 ssize_t jfs_listxattr(struct dentry * dentry, char *data, size_t buf_size)
856 {
857 	struct inode *inode = d_inode(dentry);
858 	char *buffer;
859 	ssize_t size = 0;
860 	int xattr_size;
861 	struct jfs_ea_list *ealist;
862 	struct jfs_ea *ea;
863 	struct ea_buffer ea_buf;
864 
865 	down_read(&JFS_IP(inode)->xattr_sem);
866 
867 	xattr_size = ea_get(inode, &ea_buf, 0);
868 	if (xattr_size < 0) {
869 		size = xattr_size;
870 		goto out;
871 	}
872 
873 	if (xattr_size == 0)
874 		goto release;
875 
876 	ealist = (struct jfs_ea_list *) ea_buf.xattr;
877 
878 	/* compute required size of list */
879 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
880 		if (can_list(ea))
881 			size += name_size(ea) + 1;
882 	}
883 
884 	if (!data)
885 		goto release;
886 
887 	if (size > buf_size) {
888 		size = -ERANGE;
889 		goto release;
890 	}
891 
892 	/* Copy attribute names to buffer */
893 	buffer = data;
894 	for (ea = FIRST_EA(ealist); ea < END_EALIST(ealist); ea = NEXT_EA(ea)) {
895 		if (can_list(ea)) {
896 			int namelen = copy_name(buffer, ea);
897 			buffer += namelen + 1;
898 		}
899 	}
900 
901       release:
902 	ea_release(inode, &ea_buf);
903       out:
904 	up_read(&JFS_IP(inode)->xattr_sem);
905 	return size;
906 }
907 
908 static int __jfs_xattr_set(struct inode *inode, const char *name,
909 			   const void *value, size_t size, int flags)
910 {
911 	struct jfs_inode_info *ji = JFS_IP(inode);
912 	tid_t tid;
913 	int rc;
914 
915 	tid = txBegin(inode->i_sb, 0);
916 	mutex_lock(&ji->commit_mutex);
917 	rc = __jfs_setxattr(tid, inode, name, value, size, flags);
918 	if (!rc)
919 		rc = txCommit(tid, 1, &inode, 0);
920 	txEnd(tid);
921 	mutex_unlock(&ji->commit_mutex);
922 
923 	return rc;
924 }
925 
926 static int jfs_xattr_get(const struct xattr_handler *handler,
927 			 struct dentry *unused, struct inode *inode,
928 			 const char *name, void *value, size_t size)
929 {
930 	name = xattr_full_name(handler, name);
931 	return __jfs_getxattr(inode, name, value, size);
932 }
933 
934 static int jfs_xattr_set(const struct xattr_handler *handler,
935 			 struct mnt_idmap *idmap,
936 			 struct dentry *unused, struct inode *inode,
937 			 const char *name, const void *value,
938 			 size_t size, int flags)
939 {
940 	name = xattr_full_name(handler, name);
941 	return __jfs_xattr_set(inode, name, value, size, flags);
942 }
943 
944 static int jfs_xattr_get_os2(const struct xattr_handler *handler,
945 			     struct dentry *unused, struct inode *inode,
946 			     const char *name, void *value, size_t size)
947 {
948 	if (is_known_namespace(name))
949 		return -EOPNOTSUPP;
950 	return __jfs_getxattr(inode, name, value, size);
951 }
952 
953 static int jfs_xattr_set_os2(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 	if (is_known_namespace(name))
960 		return -EOPNOTSUPP;
961 	return __jfs_xattr_set(inode, name, value, size, flags);
962 }
963 
964 static const struct xattr_handler jfs_user_xattr_handler = {
965 	.prefix = XATTR_USER_PREFIX,
966 	.get = jfs_xattr_get,
967 	.set = jfs_xattr_set,
968 };
969 
970 static const struct xattr_handler jfs_os2_xattr_handler = {
971 	.prefix = XATTR_OS2_PREFIX,
972 	.get = jfs_xattr_get_os2,
973 	.set = jfs_xattr_set_os2,
974 };
975 
976 static const struct xattr_handler jfs_security_xattr_handler = {
977 	.prefix = XATTR_SECURITY_PREFIX,
978 	.get = jfs_xattr_get,
979 	.set = jfs_xattr_set,
980 };
981 
982 static const struct xattr_handler jfs_trusted_xattr_handler = {
983 	.prefix = XATTR_TRUSTED_PREFIX,
984 	.get = jfs_xattr_get,
985 	.set = jfs_xattr_set,
986 };
987 
988 const struct xattr_handler *jfs_xattr_handlers[] = {
989 	&jfs_os2_xattr_handler,
990 	&jfs_user_xattr_handler,
991 	&jfs_security_xattr_handler,
992 	&jfs_trusted_xattr_handler,
993 	NULL,
994 };
995 
996 
997 #ifdef CONFIG_JFS_SECURITY
998 static int jfs_initxattrs(struct inode *inode, const struct xattr *xattr_array,
999 			  void *fs_info)
1000 {
1001 	const struct xattr *xattr;
1002 	tid_t *tid = fs_info;
1003 	char *name;
1004 	int err = 0;
1005 
1006 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
1007 		name = kmalloc(XATTR_SECURITY_PREFIX_LEN +
1008 			       strlen(xattr->name) + 1, GFP_NOFS);
1009 		if (!name) {
1010 			err = -ENOMEM;
1011 			break;
1012 		}
1013 		strcpy(name, XATTR_SECURITY_PREFIX);
1014 		strcpy(name + XATTR_SECURITY_PREFIX_LEN, xattr->name);
1015 
1016 		err = __jfs_setxattr(*tid, inode, name,
1017 				     xattr->value, xattr->value_len, 0);
1018 		kfree(name);
1019 		if (err < 0)
1020 			break;
1021 	}
1022 	return err;
1023 }
1024 
1025 int jfs_init_security(tid_t tid, struct inode *inode, struct inode *dir,
1026 		      const struct qstr *qstr)
1027 {
1028 	return security_inode_init_security(inode, dir, qstr,
1029 					    &jfs_initxattrs, &tid);
1030 }
1031 #endif
1032