xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_remote.c (revision 4f6cce39)
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_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_alloc.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_bmap.h"
35 #include "xfs_bmap_util.h"
36 #include "xfs_attr.h"
37 #include "xfs_attr_leaf.h"
38 #include "xfs_attr_remote.h"
39 #include "xfs_trans_space.h"
40 #include "xfs_trace.h"
41 #include "xfs_cksum.h"
42 #include "xfs_buf_item.h"
43 #include "xfs_error.h"
44 
45 #define ATTR_RMTVALUE_MAPSIZE	1	/* # of map entries at once */
46 
47 /*
48  * Each contiguous block has a header, so it is not just a simple attribute
49  * length to FSB conversion.
50  */
51 int
52 xfs_attr3_rmt_blocks(
53 	struct xfs_mount *mp,
54 	int		attrlen)
55 {
56 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
57 		int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
58 		return (attrlen + buflen - 1) / buflen;
59 	}
60 	return XFS_B_TO_FSB(mp, attrlen);
61 }
62 
63 /*
64  * Checking of the remote attribute header is split into two parts. The verifier
65  * does CRC, location and bounds checking, the unpacking function checks the
66  * attribute parameters and owner.
67  */
68 static bool
69 xfs_attr3_rmt_hdr_ok(
70 	void			*ptr,
71 	xfs_ino_t		ino,
72 	uint32_t		offset,
73 	uint32_t		size,
74 	xfs_daddr_t		bno)
75 {
76 	struct xfs_attr3_rmt_hdr *rmt = ptr;
77 
78 	if (bno != be64_to_cpu(rmt->rm_blkno))
79 		return false;
80 	if (offset != be32_to_cpu(rmt->rm_offset))
81 		return false;
82 	if (size != be32_to_cpu(rmt->rm_bytes))
83 		return false;
84 	if (ino != be64_to_cpu(rmt->rm_owner))
85 		return false;
86 
87 	/* ok */
88 	return true;
89 }
90 
91 static bool
92 xfs_attr3_rmt_verify(
93 	struct xfs_mount	*mp,
94 	void			*ptr,
95 	int			fsbsize,
96 	xfs_daddr_t		bno)
97 {
98 	struct xfs_attr3_rmt_hdr *rmt = ptr;
99 
100 	if (!xfs_sb_version_hascrc(&mp->m_sb))
101 		return false;
102 	if (rmt->rm_magic != cpu_to_be32(XFS_ATTR3_RMT_MAGIC))
103 		return false;
104 	if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid))
105 		return false;
106 	if (be64_to_cpu(rmt->rm_blkno) != bno)
107 		return false;
108 	if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
109 		return false;
110 	if (be32_to_cpu(rmt->rm_offset) +
111 				be32_to_cpu(rmt->rm_bytes) > XFS_XATTR_SIZE_MAX)
112 		return false;
113 	if (rmt->rm_owner == 0)
114 		return false;
115 
116 	return true;
117 }
118 
119 static void
120 xfs_attr3_rmt_read_verify(
121 	struct xfs_buf	*bp)
122 {
123 	struct xfs_mount *mp = bp->b_target->bt_mount;
124 	char		*ptr;
125 	int		len;
126 	xfs_daddr_t	bno;
127 	int		blksize = mp->m_attr_geo->blksize;
128 
129 	/* no verification of non-crc buffers */
130 	if (!xfs_sb_version_hascrc(&mp->m_sb))
131 		return;
132 
133 	ptr = bp->b_addr;
134 	bno = bp->b_bn;
135 	len = BBTOB(bp->b_length);
136 	ASSERT(len >= blksize);
137 
138 	while (len > 0) {
139 		if (!xfs_verify_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF)) {
140 			xfs_buf_ioerror(bp, -EFSBADCRC);
141 			break;
142 		}
143 		if (!xfs_attr3_rmt_verify(mp, ptr, blksize, bno)) {
144 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
145 			break;
146 		}
147 		len -= blksize;
148 		ptr += blksize;
149 		bno += BTOBB(blksize);
150 	}
151 
152 	if (bp->b_error)
153 		xfs_verifier_error(bp);
154 	else
155 		ASSERT(len == 0);
156 }
157 
158 static void
159 xfs_attr3_rmt_write_verify(
160 	struct xfs_buf	*bp)
161 {
162 	struct xfs_mount *mp = bp->b_target->bt_mount;
163 	int		blksize = mp->m_attr_geo->blksize;
164 	char		*ptr;
165 	int		len;
166 	xfs_daddr_t	bno;
167 
168 	/* no verification of non-crc buffers */
169 	if (!xfs_sb_version_hascrc(&mp->m_sb))
170 		return;
171 
172 	ptr = bp->b_addr;
173 	bno = bp->b_bn;
174 	len = BBTOB(bp->b_length);
175 	ASSERT(len >= blksize);
176 
177 	while (len > 0) {
178 		struct xfs_attr3_rmt_hdr *rmt = (struct xfs_attr3_rmt_hdr *)ptr;
179 
180 		if (!xfs_attr3_rmt_verify(mp, ptr, blksize, bno)) {
181 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
182 			xfs_verifier_error(bp);
183 			return;
184 		}
185 
186 		/*
187 		 * Ensure we aren't writing bogus LSNs to disk. See
188 		 * xfs_attr3_rmt_hdr_set() for the explanation.
189 		 */
190 		if (rmt->rm_lsn != cpu_to_be64(NULLCOMMITLSN)) {
191 			xfs_buf_ioerror(bp, -EFSCORRUPTED);
192 			xfs_verifier_error(bp);
193 			return;
194 		}
195 		xfs_update_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF);
196 
197 		len -= blksize;
198 		ptr += blksize;
199 		bno += BTOBB(blksize);
200 	}
201 	ASSERT(len == 0);
202 }
203 
204 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
205 	.name = "xfs_attr3_rmt",
206 	.verify_read = xfs_attr3_rmt_read_verify,
207 	.verify_write = xfs_attr3_rmt_write_verify,
208 };
209 
210 STATIC int
211 xfs_attr3_rmt_hdr_set(
212 	struct xfs_mount	*mp,
213 	void			*ptr,
214 	xfs_ino_t		ino,
215 	uint32_t		offset,
216 	uint32_t		size,
217 	xfs_daddr_t		bno)
218 {
219 	struct xfs_attr3_rmt_hdr *rmt = ptr;
220 
221 	if (!xfs_sb_version_hascrc(&mp->m_sb))
222 		return 0;
223 
224 	rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
225 	rmt->rm_offset = cpu_to_be32(offset);
226 	rmt->rm_bytes = cpu_to_be32(size);
227 	uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid);
228 	rmt->rm_owner = cpu_to_be64(ino);
229 	rmt->rm_blkno = cpu_to_be64(bno);
230 
231 	/*
232 	 * Remote attribute blocks are written synchronously, so we don't
233 	 * have an LSN that we can stamp in them that makes any sense to log
234 	 * recovery. To ensure that log recovery handles overwrites of these
235 	 * blocks sanely (i.e. once they've been freed and reallocated as some
236 	 * other type of metadata) we need to ensure that the LSN has a value
237 	 * that tells log recovery to ignore the LSN and overwrite the buffer
238 	 * with whatever is in it's log. To do this, we use the magic
239 	 * NULLCOMMITLSN to indicate that the LSN is invalid.
240 	 */
241 	rmt->rm_lsn = cpu_to_be64(NULLCOMMITLSN);
242 
243 	return sizeof(struct xfs_attr3_rmt_hdr);
244 }
245 
246 /*
247  * Helper functions to copy attribute data in and out of the one disk extents
248  */
249 STATIC int
250 xfs_attr_rmtval_copyout(
251 	struct xfs_mount *mp,
252 	struct xfs_buf	*bp,
253 	xfs_ino_t	ino,
254 	int		*offset,
255 	int		*valuelen,
256 	__uint8_t	**dst)
257 {
258 	char		*src = bp->b_addr;
259 	xfs_daddr_t	bno = bp->b_bn;
260 	int		len = BBTOB(bp->b_length);
261 	int		blksize = mp->m_attr_geo->blksize;
262 
263 	ASSERT(len >= blksize);
264 
265 	while (len > 0 && *valuelen > 0) {
266 		int hdr_size = 0;
267 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
268 
269 		byte_cnt = min(*valuelen, byte_cnt);
270 
271 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
272 			if (!xfs_attr3_rmt_hdr_ok(src, ino, *offset,
273 						  byte_cnt, bno)) {
274 				xfs_alert(mp,
275 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
276 					bno, *offset, byte_cnt, ino);
277 				return -EFSCORRUPTED;
278 			}
279 			hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
280 		}
281 
282 		memcpy(*dst, src + hdr_size, byte_cnt);
283 
284 		/* roll buffer forwards */
285 		len -= blksize;
286 		src += blksize;
287 		bno += BTOBB(blksize);
288 
289 		/* roll attribute data forwards */
290 		*valuelen -= byte_cnt;
291 		*dst += byte_cnt;
292 		*offset += byte_cnt;
293 	}
294 	return 0;
295 }
296 
297 STATIC void
298 xfs_attr_rmtval_copyin(
299 	struct xfs_mount *mp,
300 	struct xfs_buf	*bp,
301 	xfs_ino_t	ino,
302 	int		*offset,
303 	int		*valuelen,
304 	__uint8_t	**src)
305 {
306 	char		*dst = bp->b_addr;
307 	xfs_daddr_t	bno = bp->b_bn;
308 	int		len = BBTOB(bp->b_length);
309 	int		blksize = mp->m_attr_geo->blksize;
310 
311 	ASSERT(len >= blksize);
312 
313 	while (len > 0 && *valuelen > 0) {
314 		int hdr_size;
315 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
316 
317 		byte_cnt = min(*valuelen, byte_cnt);
318 		hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
319 						 byte_cnt, bno);
320 
321 		memcpy(dst + hdr_size, *src, byte_cnt);
322 
323 		/*
324 		 * If this is the last block, zero the remainder of it.
325 		 * Check that we are actually the last block, too.
326 		 */
327 		if (byte_cnt + hdr_size < blksize) {
328 			ASSERT(*valuelen - byte_cnt == 0);
329 			ASSERT(len == blksize);
330 			memset(dst + hdr_size + byte_cnt, 0,
331 					blksize - hdr_size - byte_cnt);
332 		}
333 
334 		/* roll buffer forwards */
335 		len -= blksize;
336 		dst += blksize;
337 		bno += BTOBB(blksize);
338 
339 		/* roll attribute data forwards */
340 		*valuelen -= byte_cnt;
341 		*src += byte_cnt;
342 		*offset += byte_cnt;
343 	}
344 }
345 
346 /*
347  * Read the value associated with an attribute from the out-of-line buffer
348  * that we stored it in.
349  */
350 int
351 xfs_attr_rmtval_get(
352 	struct xfs_da_args	*args)
353 {
354 	struct xfs_bmbt_irec	map[ATTR_RMTVALUE_MAPSIZE];
355 	struct xfs_mount	*mp = args->dp->i_mount;
356 	struct xfs_buf		*bp;
357 	xfs_dablk_t		lblkno = args->rmtblkno;
358 	__uint8_t		*dst = args->value;
359 	int			valuelen;
360 	int			nmap;
361 	int			error;
362 	int			blkcnt = args->rmtblkcnt;
363 	int			i;
364 	int			offset = 0;
365 
366 	trace_xfs_attr_rmtval_get(args);
367 
368 	ASSERT(!(args->flags & ATTR_KERNOVAL));
369 	ASSERT(args->rmtvaluelen == args->valuelen);
370 
371 	valuelen = args->rmtvaluelen;
372 	while (valuelen > 0) {
373 		nmap = ATTR_RMTVALUE_MAPSIZE;
374 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
375 				       blkcnt, map, &nmap,
376 				       XFS_BMAPI_ATTRFORK);
377 		if (error)
378 			return error;
379 		ASSERT(nmap >= 1);
380 
381 		for (i = 0; (i < nmap) && (valuelen > 0); i++) {
382 			xfs_daddr_t	dblkno;
383 			int		dblkcnt;
384 
385 			ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
386 			       (map[i].br_startblock != HOLESTARTBLOCK));
387 			dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
388 			dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
389 			error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
390 						   dblkno, dblkcnt, 0, &bp,
391 						   &xfs_attr3_rmt_buf_ops);
392 			if (error)
393 				return error;
394 
395 			error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
396 							&offset, &valuelen,
397 							&dst);
398 			xfs_buf_relse(bp);
399 			if (error)
400 				return error;
401 
402 			/* roll attribute extent map forwards */
403 			lblkno += map[i].br_blockcount;
404 			blkcnt -= map[i].br_blockcount;
405 		}
406 	}
407 	ASSERT(valuelen == 0);
408 	return 0;
409 }
410 
411 /*
412  * Write the value associated with an attribute into the out-of-line buffer
413  * that we have defined for it.
414  */
415 int
416 xfs_attr_rmtval_set(
417 	struct xfs_da_args	*args)
418 {
419 	struct xfs_inode	*dp = args->dp;
420 	struct xfs_mount	*mp = dp->i_mount;
421 	struct xfs_bmbt_irec	map;
422 	xfs_dablk_t		lblkno;
423 	xfs_fileoff_t		lfileoff = 0;
424 	__uint8_t		*src = args->value;
425 	int			blkcnt;
426 	int			valuelen;
427 	int			nmap;
428 	int			error;
429 	int			offset = 0;
430 
431 	trace_xfs_attr_rmtval_set(args);
432 
433 	/*
434 	 * Find a "hole" in the attribute address space large enough for
435 	 * us to drop the new attribute's value into. Because CRC enable
436 	 * attributes have headers, we can't just do a straight byte to FSB
437 	 * conversion and have to take the header space into account.
438 	 */
439 	blkcnt = xfs_attr3_rmt_blocks(mp, args->rmtvaluelen);
440 	error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
441 						   XFS_ATTR_FORK);
442 	if (error)
443 		return error;
444 
445 	args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
446 	args->rmtblkcnt = blkcnt;
447 
448 	/*
449 	 * Roll through the "value", allocating blocks on disk as required.
450 	 */
451 	while (blkcnt > 0) {
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_defer_init(args->dfops, 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->dfops);
469 		if (!error)
470 			error = xfs_defer_finish(&args->trans, args->dfops, dp);
471 		if (error) {
472 			args->trans = NULL;
473 			xfs_defer_cancel(args->dfops);
474 			return error;
475 		}
476 
477 		ASSERT(nmap == 1);
478 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
479 		       (map.br_startblock != HOLESTARTBLOCK));
480 		lblkno += map.br_blockcount;
481 		blkcnt -= map.br_blockcount;
482 
483 		/*
484 		 * Start the next trans in the chain.
485 		 */
486 		error = xfs_trans_roll(&args->trans, dp);
487 		if (error)
488 			return error;
489 	}
490 
491 	/*
492 	 * Roll through the "value", copying the attribute value to the
493 	 * already-allocated blocks.  Blocks are written synchronously
494 	 * so that we can know they are all on disk before we turn off
495 	 * the INCOMPLETE flag.
496 	 */
497 	lblkno = args->rmtblkno;
498 	blkcnt = args->rmtblkcnt;
499 	valuelen = args->rmtvaluelen;
500 	while (valuelen > 0) {
501 		struct xfs_buf	*bp;
502 		xfs_daddr_t	dblkno;
503 		int		dblkcnt;
504 
505 		ASSERT(blkcnt > 0);
506 
507 		xfs_defer_init(args->dfops, args->firstblock);
508 		nmap = 1;
509 		error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
510 				       blkcnt, &map, &nmap,
511 				       XFS_BMAPI_ATTRFORK);
512 		if (error)
513 			return error;
514 		ASSERT(nmap == 1);
515 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
516 		       (map.br_startblock != HOLESTARTBLOCK));
517 
518 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
519 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
520 
521 		bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
522 		if (!bp)
523 			return -ENOMEM;
524 		bp->b_ops = &xfs_attr3_rmt_buf_ops;
525 
526 		xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
527 				       &valuelen, &src);
528 
529 		error = xfs_bwrite(bp);	/* GROT: NOTE: synchronous write */
530 		xfs_buf_relse(bp);
531 		if (error)
532 			return error;
533 
534 
535 		/* roll attribute extent map forwards */
536 		lblkno += map.br_blockcount;
537 		blkcnt -= map.br_blockcount;
538 	}
539 	ASSERT(valuelen == 0);
540 	return 0;
541 }
542 
543 /*
544  * Remove the value associated with an attribute by deleting the
545  * out-of-line buffer that it is stored on.
546  */
547 int
548 xfs_attr_rmtval_remove(
549 	struct xfs_da_args	*args)
550 {
551 	struct xfs_mount	*mp = args->dp->i_mount;
552 	xfs_dablk_t		lblkno;
553 	int			blkcnt;
554 	int			error;
555 	int			done;
556 
557 	trace_xfs_attr_rmtval_remove(args);
558 
559 	/*
560 	 * Roll through the "value", invalidating the attribute value's blocks.
561 	 */
562 	lblkno = args->rmtblkno;
563 	blkcnt = args->rmtblkcnt;
564 	while (blkcnt > 0) {
565 		struct xfs_bmbt_irec	map;
566 		struct xfs_buf		*bp;
567 		xfs_daddr_t		dblkno;
568 		int			dblkcnt;
569 		int			nmap;
570 
571 		/*
572 		 * Try to remember where we decided to put the value.
573 		 */
574 		nmap = 1;
575 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
576 				       blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
577 		if (error)
578 			return error;
579 		ASSERT(nmap == 1);
580 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
581 		       (map.br_startblock != HOLESTARTBLOCK));
582 
583 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
584 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
585 
586 		/*
587 		 * If the "remote" value is in the cache, remove it.
588 		 */
589 		bp = xfs_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
590 		if (bp) {
591 			xfs_buf_stale(bp);
592 			xfs_buf_relse(bp);
593 			bp = NULL;
594 		}
595 
596 		lblkno += map.br_blockcount;
597 		blkcnt -= map.br_blockcount;
598 	}
599 
600 	/*
601 	 * Keep de-allocating extents until the remote-value region is gone.
602 	 */
603 	lblkno = args->rmtblkno;
604 	blkcnt = args->rmtblkcnt;
605 	done = 0;
606 	while (!done) {
607 		xfs_defer_init(args->dfops, args->firstblock);
608 		error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
609 				    XFS_BMAPI_ATTRFORK, 1, args->firstblock,
610 				    args->dfops, &done);
611 		if (!error)
612 			error = xfs_defer_finish(&args->trans, args->dfops,
613 						args->dp);
614 		if (error) {
615 			args->trans = NULL;
616 			xfs_defer_cancel(args->dfops);
617 			return error;
618 		}
619 
620 		/*
621 		 * Close out trans and start the next one in the chain.
622 		 */
623 		error = xfs_trans_roll(&args->trans, args->dp);
624 		if (error)
625 			return error;
626 	}
627 	return 0;
628 }
629