xref: /linux/fs/xfs/libxfs/xfs_attr_remote.c (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * 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 the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_bit.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_alloc.h"
31 #include "xfs_trans.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_bmap.h"
34 #include "xfs_bmap_util.h"
35 #include "xfs_attr.h"
36 #include "xfs_attr_leaf.h"
37 #include "xfs_attr_remote.h"
38 #include "xfs_trans_space.h"
39 #include "xfs_trace.h"
40 #include "xfs_cksum.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_error.h"
43 
44 #define ATTR_RMTVALUE_MAPSIZE	1	/* # of map entries at once */
45 
46 /*
47  * Each contiguous block has a header, so it is not just a simple attribute
48  * length to FSB conversion.
49  */
50 int
51 xfs_attr3_rmt_blocks(
52 	struct xfs_mount *mp,
53 	int		attrlen)
54 {
55 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
56 		int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
57 		return (attrlen + buflen - 1) / buflen;
58 	}
59 	return XFS_B_TO_FSB(mp, attrlen);
60 }
61 
62 /*
63  * Checking of the remote attribute header is split into two parts. The verifier
64  * does CRC, location and bounds checking, the unpacking function checks the
65  * attribute parameters and owner.
66  */
67 static bool
68 xfs_attr3_rmt_hdr_ok(
69 	void			*ptr,
70 	xfs_ino_t		ino,
71 	uint32_t		offset,
72 	uint32_t		size,
73 	xfs_daddr_t		bno)
74 {
75 	struct xfs_attr3_rmt_hdr *rmt = ptr;
76 
77 	if (bno != be64_to_cpu(rmt->rm_blkno))
78 		return false;
79 	if (offset != be32_to_cpu(rmt->rm_offset))
80 		return false;
81 	if (size != be32_to_cpu(rmt->rm_bytes))
82 		return false;
83 	if (ino != be64_to_cpu(rmt->rm_owner))
84 		return false;
85 
86 	/* ok */
87 	return true;
88 }
89 
90 static bool
91 xfs_attr3_rmt_verify(
92 	struct xfs_mount	*mp,
93 	void			*ptr,
94 	int			fsbsize,
95 	xfs_daddr_t		bno)
96 {
97 	struct xfs_attr3_rmt_hdr *rmt = ptr;
98 
99 	if (!xfs_sb_version_hascrc(&mp->m_sb))
100 		return false;
101 	if (rmt->rm_magic != cpu_to_be32(XFS_ATTR3_RMT_MAGIC))
102 		return false;
103 	if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid))
104 		return false;
105 	if (be64_to_cpu(rmt->rm_blkno) != bno)
106 		return false;
107 	if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
108 		return false;
109 	if (be32_to_cpu(rmt->rm_offset) +
110 				be32_to_cpu(rmt->rm_bytes) > XATTR_SIZE_MAX)
111 		return false;
112 	if (rmt->rm_owner == 0)
113 		return false;
114 
115 	return true;
116 }
117 
118 static void
119 xfs_attr3_rmt_read_verify(
120 	struct xfs_buf	*bp)
121 {
122 	struct xfs_mount *mp = bp->b_target->bt_mount;
123 	char		*ptr;
124 	int		len;
125 	xfs_daddr_t	bno;
126 	int		blksize = mp->m_attr_geo->blksize;
127 
128 	/* no verification of non-crc buffers */
129 	if (!xfs_sb_version_hascrc(&mp->m_sb))
130 		return;
131 
132 	ptr = bp->b_addr;
133 	bno = bp->b_bn;
134 	len = BBTOB(bp->b_length);
135 	ASSERT(len >= blksize);
136 
137 	while (len > 0) {
138 		if (!xfs_verify_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF)) {
139 			xfs_buf_ioerror(bp, -EFSBADCRC);
140 			break;
141 		}
142 		if (!xfs_attr3_rmt_verify(mp, ptr, blksize, bno)) {
143 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
144 			break;
145 		}
146 		len -= blksize;
147 		ptr += blksize;
148 		bno += BTOBB(blksize);
149 	}
150 
151 	if (bp->b_error)
152 		xfs_verifier_error(bp);
153 	else
154 		ASSERT(len == 0);
155 }
156 
157 static void
158 xfs_attr3_rmt_write_verify(
159 	struct xfs_buf	*bp)
160 {
161 	struct xfs_mount *mp = bp->b_target->bt_mount;
162 	int		blksize = mp->m_attr_geo->blksize;
163 	char		*ptr;
164 	int		len;
165 	xfs_daddr_t	bno;
166 
167 	/* no verification of non-crc buffers */
168 	if (!xfs_sb_version_hascrc(&mp->m_sb))
169 		return;
170 
171 	ptr = bp->b_addr;
172 	bno = bp->b_bn;
173 	len = BBTOB(bp->b_length);
174 	ASSERT(len >= blksize);
175 
176 	while (len > 0) {
177 		struct xfs_attr3_rmt_hdr *rmt = (struct xfs_attr3_rmt_hdr *)ptr;
178 
179 		if (!xfs_attr3_rmt_verify(mp, ptr, blksize, bno)) {
180 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
181 			xfs_verifier_error(bp);
182 			return;
183 		}
184 
185 		/*
186 		 * Ensure we aren't writing bogus LSNs to disk. See
187 		 * xfs_attr3_rmt_hdr_set() for the explanation.
188 		 */
189 		if (rmt->rm_lsn != cpu_to_be64(NULLCOMMITLSN)) {
190 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
191 			xfs_verifier_error(bp);
192 			return;
193 		}
194 		xfs_update_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF);
195 
196 		len -= blksize;
197 		ptr += blksize;
198 		bno += BTOBB(blksize);
199 	}
200 	ASSERT(len == 0);
201 }
202 
203 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
204 	.verify_read = xfs_attr3_rmt_read_verify,
205 	.verify_write = xfs_attr3_rmt_write_verify,
206 };
207 
208 STATIC int
209 xfs_attr3_rmt_hdr_set(
210 	struct xfs_mount	*mp,
211 	void			*ptr,
212 	xfs_ino_t		ino,
213 	uint32_t		offset,
214 	uint32_t		size,
215 	xfs_daddr_t		bno)
216 {
217 	struct xfs_attr3_rmt_hdr *rmt = ptr;
218 
219 	if (!xfs_sb_version_hascrc(&mp->m_sb))
220 		return 0;
221 
222 	rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
223 	rmt->rm_offset = cpu_to_be32(offset);
224 	rmt->rm_bytes = cpu_to_be32(size);
225 	uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid);
226 	rmt->rm_owner = cpu_to_be64(ino);
227 	rmt->rm_blkno = cpu_to_be64(bno);
228 
229 	/*
230 	 * Remote attribute blocks are written synchronously, so we don't
231 	 * have an LSN that we can stamp in them that makes any sense to log
232 	 * recovery. To ensure that log recovery handles overwrites of these
233 	 * blocks sanely (i.e. once they've been freed and reallocated as some
234 	 * other type of metadata) we need to ensure that the LSN has a value
235 	 * that tells log recovery to ignore the LSN and overwrite the buffer
236 	 * with whatever is in it's log. To do this, we use the magic
237 	 * NULLCOMMITLSN to indicate that the LSN is invalid.
238 	 */
239 	rmt->rm_lsn = cpu_to_be64(NULLCOMMITLSN);
240 
241 	return sizeof(struct xfs_attr3_rmt_hdr);
242 }
243 
244 /*
245  * Helper functions to copy attribute data in and out of the one disk extents
246  */
247 STATIC int
248 xfs_attr_rmtval_copyout(
249 	struct xfs_mount *mp,
250 	struct xfs_buf	*bp,
251 	xfs_ino_t	ino,
252 	int		*offset,
253 	int		*valuelen,
254 	__uint8_t	**dst)
255 {
256 	char		*src = bp->b_addr;
257 	xfs_daddr_t	bno = bp->b_bn;
258 	int		len = BBTOB(bp->b_length);
259 	int		blksize = mp->m_attr_geo->blksize;
260 
261 	ASSERT(len >= blksize);
262 
263 	while (len > 0 && *valuelen > 0) {
264 		int hdr_size = 0;
265 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
266 
267 		byte_cnt = min(*valuelen, byte_cnt);
268 
269 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
270 			if (!xfs_attr3_rmt_hdr_ok(src, ino, *offset,
271 						  byte_cnt, bno)) {
272 				xfs_alert(mp,
273 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
274 					bno, *offset, byte_cnt, ino);
275 				return -EFSCORRUPTED;
276 			}
277 			hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
278 		}
279 
280 		memcpy(*dst, src + hdr_size, byte_cnt);
281 
282 		/* roll buffer forwards */
283 		len -= blksize;
284 		src += blksize;
285 		bno += BTOBB(blksize);
286 
287 		/* roll attribute data forwards */
288 		*valuelen -= byte_cnt;
289 		*dst += byte_cnt;
290 		*offset += byte_cnt;
291 	}
292 	return 0;
293 }
294 
295 STATIC void
296 xfs_attr_rmtval_copyin(
297 	struct xfs_mount *mp,
298 	struct xfs_buf	*bp,
299 	xfs_ino_t	ino,
300 	int		*offset,
301 	int		*valuelen,
302 	__uint8_t	**src)
303 {
304 	char		*dst = bp->b_addr;
305 	xfs_daddr_t	bno = bp->b_bn;
306 	int		len = BBTOB(bp->b_length);
307 	int		blksize = mp->m_attr_geo->blksize;
308 
309 	ASSERT(len >= blksize);
310 
311 	while (len > 0 && *valuelen > 0) {
312 		int hdr_size;
313 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
314 
315 		byte_cnt = min(*valuelen, byte_cnt);
316 		hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
317 						 byte_cnt, bno);
318 
319 		memcpy(dst + hdr_size, *src, byte_cnt);
320 
321 		/*
322 		 * If this is the last block, zero the remainder of it.
323 		 * Check that we are actually the last block, too.
324 		 */
325 		if (byte_cnt + hdr_size < blksize) {
326 			ASSERT(*valuelen - byte_cnt == 0);
327 			ASSERT(len == blksize);
328 			memset(dst + hdr_size + byte_cnt, 0,
329 					blksize - hdr_size - byte_cnt);
330 		}
331 
332 		/* roll buffer forwards */
333 		len -= blksize;
334 		dst += blksize;
335 		bno += BTOBB(blksize);
336 
337 		/* roll attribute data forwards */
338 		*valuelen -= byte_cnt;
339 		*src += byte_cnt;
340 		*offset += byte_cnt;
341 	}
342 }
343 
344 /*
345  * Read the value associated with an attribute from the out-of-line buffer
346  * that we stored it in.
347  */
348 int
349 xfs_attr_rmtval_get(
350 	struct xfs_da_args	*args)
351 {
352 	struct xfs_bmbt_irec	map[ATTR_RMTVALUE_MAPSIZE];
353 	struct xfs_mount	*mp = args->dp->i_mount;
354 	struct xfs_buf		*bp;
355 	xfs_dablk_t		lblkno = args->rmtblkno;
356 	__uint8_t		*dst = args->value;
357 	int			valuelen;
358 	int			nmap;
359 	int			error;
360 	int			blkcnt = args->rmtblkcnt;
361 	int			i;
362 	int			offset = 0;
363 
364 	trace_xfs_attr_rmtval_get(args);
365 
366 	ASSERT(!(args->flags & ATTR_KERNOVAL));
367 	ASSERT(args->rmtvaluelen == args->valuelen);
368 
369 	valuelen = args->rmtvaluelen;
370 	while (valuelen > 0) {
371 		nmap = ATTR_RMTVALUE_MAPSIZE;
372 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
373 				       blkcnt, map, &nmap,
374 				       XFS_BMAPI_ATTRFORK);
375 		if (error)
376 			return error;
377 		ASSERT(nmap >= 1);
378 
379 		for (i = 0; (i < nmap) && (valuelen > 0); i++) {
380 			xfs_daddr_t	dblkno;
381 			int		dblkcnt;
382 
383 			ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
384 			       (map[i].br_startblock != HOLESTARTBLOCK));
385 			dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
386 			dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
387 			error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
388 						   dblkno, dblkcnt, 0, &bp,
389 						   &xfs_attr3_rmt_buf_ops);
390 			if (error)
391 				return error;
392 
393 			error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
394 							&offset, &valuelen,
395 							&dst);
396 			xfs_buf_relse(bp);
397 			if (error)
398 				return error;
399 
400 			/* roll attribute extent map forwards */
401 			lblkno += map[i].br_blockcount;
402 			blkcnt -= map[i].br_blockcount;
403 		}
404 	}
405 	ASSERT(valuelen == 0);
406 	return 0;
407 }
408 
409 /*
410  * Write the value associated with an attribute into the out-of-line buffer
411  * that we have defined for it.
412  */
413 int
414 xfs_attr_rmtval_set(
415 	struct xfs_da_args	*args)
416 {
417 	struct xfs_inode	*dp = args->dp;
418 	struct xfs_mount	*mp = dp->i_mount;
419 	struct xfs_bmbt_irec	map;
420 	xfs_dablk_t		lblkno;
421 	xfs_fileoff_t		lfileoff = 0;
422 	__uint8_t		*src = args->value;
423 	int			blkcnt;
424 	int			valuelen;
425 	int			nmap;
426 	int			error;
427 	int			offset = 0;
428 
429 	trace_xfs_attr_rmtval_set(args);
430 
431 	/*
432 	 * Find a "hole" in the attribute address space large enough for
433 	 * us to drop the new attribute's value into. Because CRC enable
434 	 * attributes have headers, we can't just do a straight byte to FSB
435 	 * conversion and have to take the header space into account.
436 	 */
437 	blkcnt = xfs_attr3_rmt_blocks(mp, args->rmtvaluelen);
438 	error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
439 						   XFS_ATTR_FORK);
440 	if (error)
441 		return error;
442 
443 	args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
444 	args->rmtblkcnt = blkcnt;
445 
446 	/*
447 	 * Roll through the "value", allocating blocks on disk as required.
448 	 */
449 	while (blkcnt > 0) {
450 		int	committed;
451 
452 		/*
453 		 * Allocate a single extent, up to the size of the value.
454 		 *
455 		 * Note that we have to consider this a data allocation as we
456 		 * write the remote attribute without logging the contents.
457 		 * Hence we must ensure that we aren't using blocks that are on
458 		 * the busy list so that we don't overwrite blocks which have
459 		 * recently been freed but their transactions are not yet
460 		 * committed to disk. If we overwrite the contents of a busy
461 		 * extent and then crash then the block may not contain the
462 		 * correct metadata after log recovery occurs.
463 		 */
464 		xfs_bmap_init(args->flist, args->firstblock);
465 		nmap = 1;
466 		error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
467 				  blkcnt, XFS_BMAPI_ATTRFORK, args->firstblock,
468 				  args->total, &map, &nmap, args->flist);
469 		if (!error) {
470 			error = xfs_bmap_finish(&args->trans, args->flist,
471 						&committed);
472 		}
473 		if (error) {
474 			ASSERT(committed);
475 			args->trans = NULL;
476 			xfs_bmap_cancel(args->flist);
477 			return error;
478 		}
479 
480 		/*
481 		 * bmap_finish() may have committed the last trans and started
482 		 * a new one.  We need the inode to be in all transactions.
483 		 */
484 		if (committed)
485 			xfs_trans_ijoin(args->trans, dp, 0);
486 
487 		ASSERT(nmap == 1);
488 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
489 		       (map.br_startblock != HOLESTARTBLOCK));
490 		lblkno += map.br_blockcount;
491 		blkcnt -= map.br_blockcount;
492 
493 		/*
494 		 * Start the next trans in the chain.
495 		 */
496 		error = xfs_trans_roll(&args->trans, dp);
497 		if (error)
498 			return error;
499 	}
500 
501 	/*
502 	 * Roll through the "value", copying the attribute value to the
503 	 * already-allocated blocks.  Blocks are written synchronously
504 	 * so that we can know they are all on disk before we turn off
505 	 * the INCOMPLETE flag.
506 	 */
507 	lblkno = args->rmtblkno;
508 	blkcnt = args->rmtblkcnt;
509 	valuelen = args->rmtvaluelen;
510 	while (valuelen > 0) {
511 		struct xfs_buf	*bp;
512 		xfs_daddr_t	dblkno;
513 		int		dblkcnt;
514 
515 		ASSERT(blkcnt > 0);
516 
517 		xfs_bmap_init(args->flist, args->firstblock);
518 		nmap = 1;
519 		error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
520 				       blkcnt, &map, &nmap,
521 				       XFS_BMAPI_ATTRFORK);
522 		if (error)
523 			return error;
524 		ASSERT(nmap == 1);
525 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
526 		       (map.br_startblock != HOLESTARTBLOCK));
527 
528 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
529 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
530 
531 		bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
532 		if (!bp)
533 			return -ENOMEM;
534 		bp->b_ops = &xfs_attr3_rmt_buf_ops;
535 
536 		xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
537 				       &valuelen, &src);
538 
539 		error = xfs_bwrite(bp);	/* GROT: NOTE: synchronous write */
540 		xfs_buf_relse(bp);
541 		if (error)
542 			return error;
543 
544 
545 		/* roll attribute extent map forwards */
546 		lblkno += map.br_blockcount;
547 		blkcnt -= map.br_blockcount;
548 	}
549 	ASSERT(valuelen == 0);
550 	return 0;
551 }
552 
553 /*
554  * Remove the value associated with an attribute by deleting the
555  * out-of-line buffer that it is stored on.
556  */
557 int
558 xfs_attr_rmtval_remove(
559 	struct xfs_da_args	*args)
560 {
561 	struct xfs_mount	*mp = args->dp->i_mount;
562 	xfs_dablk_t		lblkno;
563 	int			blkcnt;
564 	int			error;
565 	int			done;
566 
567 	trace_xfs_attr_rmtval_remove(args);
568 
569 	/*
570 	 * Roll through the "value", invalidating the attribute value's blocks.
571 	 */
572 	lblkno = args->rmtblkno;
573 	blkcnt = args->rmtblkcnt;
574 	while (blkcnt > 0) {
575 		struct xfs_bmbt_irec	map;
576 		struct xfs_buf		*bp;
577 		xfs_daddr_t		dblkno;
578 		int			dblkcnt;
579 		int			nmap;
580 
581 		/*
582 		 * Try to remember where we decided to put the value.
583 		 */
584 		nmap = 1;
585 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
586 				       blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
587 		if (error)
588 			return error;
589 		ASSERT(nmap == 1);
590 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
591 		       (map.br_startblock != HOLESTARTBLOCK));
592 
593 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
594 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
595 
596 		/*
597 		 * If the "remote" value is in the cache, remove it.
598 		 */
599 		bp = xfs_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
600 		if (bp) {
601 			xfs_buf_stale(bp);
602 			xfs_buf_relse(bp);
603 			bp = NULL;
604 		}
605 
606 		lblkno += map.br_blockcount;
607 		blkcnt -= map.br_blockcount;
608 	}
609 
610 	/*
611 	 * Keep de-allocating extents until the remote-value region is gone.
612 	 */
613 	lblkno = args->rmtblkno;
614 	blkcnt = args->rmtblkcnt;
615 	done = 0;
616 	while (!done) {
617 		int committed;
618 
619 		xfs_bmap_init(args->flist, args->firstblock);
620 		error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
621 				    XFS_BMAPI_ATTRFORK, 1, args->firstblock,
622 				    args->flist, &done);
623 		if (!error) {
624 			error = xfs_bmap_finish(&args->trans, args->flist,
625 						&committed);
626 		}
627 		if (error) {
628 			ASSERT(committed);
629 			args->trans = NULL;
630 			xfs_bmap_cancel(args->flist);
631 			return error;
632 		}
633 
634 		/*
635 		 * bmap_finish() may have committed the last trans and started
636 		 * a new one.  We need the inode to be in all transactions.
637 		 */
638 		if (committed)
639 			xfs_trans_ijoin(args->trans, args->dp, 0);
640 
641 		/*
642 		 * Close out trans and start the next one in the chain.
643 		 */
644 		error = xfs_trans_roll(&args->trans, args->dp);
645 		if (error)
646 			return error;
647 	}
648 	return 0;
649 }
650