xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_remote.c (revision 96de2506)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * Copyright (c) 2013 Red Hat, Inc.
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_bit.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_da_format.h"
17 #include "xfs_da_btree.h"
18 #include "xfs_inode.h"
19 #include "xfs_alloc.h"
20 #include "xfs_trans.h"
21 #include "xfs_inode_item.h"
22 #include "xfs_bmap.h"
23 #include "xfs_bmap_util.h"
24 #include "xfs_attr.h"
25 #include "xfs_attr_leaf.h"
26 #include "xfs_attr_remote.h"
27 #include "xfs_trans_space.h"
28 #include "xfs_trace.h"
29 #include "xfs_cksum.h"
30 #include "xfs_buf_item.h"
31 #include "xfs_error.h"
32 
33 #define ATTR_RMTVALUE_MAPSIZE	1	/* # of map entries at once */
34 
35 /*
36  * Each contiguous block has a header, so it is not just a simple attribute
37  * length to FSB conversion.
38  */
39 int
40 xfs_attr3_rmt_blocks(
41 	struct xfs_mount *mp,
42 	int		attrlen)
43 {
44 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
45 		int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
46 		return (attrlen + buflen - 1) / buflen;
47 	}
48 	return XFS_B_TO_FSB(mp, attrlen);
49 }
50 
51 /*
52  * Checking of the remote attribute header is split into two parts. The verifier
53  * does CRC, location and bounds checking, the unpacking function checks the
54  * attribute parameters and owner.
55  */
56 static xfs_failaddr_t
57 xfs_attr3_rmt_hdr_ok(
58 	void			*ptr,
59 	xfs_ino_t		ino,
60 	uint32_t		offset,
61 	uint32_t		size,
62 	xfs_daddr_t		bno)
63 {
64 	struct xfs_attr3_rmt_hdr *rmt = ptr;
65 
66 	if (bno != be64_to_cpu(rmt->rm_blkno))
67 		return __this_address;
68 	if (offset != be32_to_cpu(rmt->rm_offset))
69 		return __this_address;
70 	if (size != be32_to_cpu(rmt->rm_bytes))
71 		return __this_address;
72 	if (ino != be64_to_cpu(rmt->rm_owner))
73 		return __this_address;
74 
75 	/* ok */
76 	return NULL;
77 }
78 
79 static xfs_failaddr_t
80 xfs_attr3_rmt_verify(
81 	struct xfs_mount	*mp,
82 	void			*ptr,
83 	int			fsbsize,
84 	xfs_daddr_t		bno)
85 {
86 	struct xfs_attr3_rmt_hdr *rmt = ptr;
87 
88 	if (!xfs_sb_version_hascrc(&mp->m_sb))
89 		return __this_address;
90 	if (rmt->rm_magic != cpu_to_be32(XFS_ATTR3_RMT_MAGIC))
91 		return __this_address;
92 	if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid))
93 		return __this_address;
94 	if (be64_to_cpu(rmt->rm_blkno) != bno)
95 		return __this_address;
96 	if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
97 		return __this_address;
98 	if (be32_to_cpu(rmt->rm_offset) +
99 				be32_to_cpu(rmt->rm_bytes) > XFS_XATTR_SIZE_MAX)
100 		return __this_address;
101 	if (rmt->rm_owner == 0)
102 		return __this_address;
103 
104 	return NULL;
105 }
106 
107 static int
108 __xfs_attr3_rmt_read_verify(
109 	struct xfs_buf	*bp,
110 	bool		check_crc,
111 	xfs_failaddr_t	*failaddr)
112 {
113 	struct xfs_mount *mp = bp->b_target->bt_mount;
114 	char		*ptr;
115 	int		len;
116 	xfs_daddr_t	bno;
117 	int		blksize = mp->m_attr_geo->blksize;
118 
119 	/* no verification of non-crc buffers */
120 	if (!xfs_sb_version_hascrc(&mp->m_sb))
121 		return 0;
122 
123 	ptr = bp->b_addr;
124 	bno = bp->b_bn;
125 	len = BBTOB(bp->b_length);
126 	ASSERT(len >= blksize);
127 
128 	while (len > 0) {
129 		if (check_crc &&
130 		    !xfs_verify_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF)) {
131 			*failaddr = __this_address;
132 			return -EFSBADCRC;
133 		}
134 		*failaddr = xfs_attr3_rmt_verify(mp, ptr, blksize, bno);
135 		if (*failaddr)
136 			return -EFSCORRUPTED;
137 		len -= blksize;
138 		ptr += blksize;
139 		bno += BTOBB(blksize);
140 	}
141 
142 	if (len != 0) {
143 		*failaddr = __this_address;
144 		return -EFSCORRUPTED;
145 	}
146 
147 	return 0;
148 }
149 
150 static void
151 xfs_attr3_rmt_read_verify(
152 	struct xfs_buf	*bp)
153 {
154 	xfs_failaddr_t	fa;
155 	int		error;
156 
157 	error = __xfs_attr3_rmt_read_verify(bp, true, &fa);
158 	if (error)
159 		xfs_verifier_error(bp, error, fa);
160 }
161 
162 static xfs_failaddr_t
163 xfs_attr3_rmt_verify_struct(
164 	struct xfs_buf	*bp)
165 {
166 	xfs_failaddr_t	fa;
167 	int		error;
168 
169 	error = __xfs_attr3_rmt_read_verify(bp, false, &fa);
170 	return error ? fa : NULL;
171 }
172 
173 static void
174 xfs_attr3_rmt_write_verify(
175 	struct xfs_buf	*bp)
176 {
177 	struct xfs_mount *mp = bp->b_target->bt_mount;
178 	xfs_failaddr_t	fa;
179 	int		blksize = mp->m_attr_geo->blksize;
180 	char		*ptr;
181 	int		len;
182 	xfs_daddr_t	bno;
183 
184 	/* no verification of non-crc buffers */
185 	if (!xfs_sb_version_hascrc(&mp->m_sb))
186 		return;
187 
188 	ptr = bp->b_addr;
189 	bno = bp->b_bn;
190 	len = BBTOB(bp->b_length);
191 	ASSERT(len >= blksize);
192 
193 	while (len > 0) {
194 		struct xfs_attr3_rmt_hdr *rmt = (struct xfs_attr3_rmt_hdr *)ptr;
195 
196 		fa = xfs_attr3_rmt_verify(mp, ptr, blksize, bno);
197 		if (fa) {
198 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
199 			return;
200 		}
201 
202 		/*
203 		 * Ensure we aren't writing bogus LSNs to disk. See
204 		 * xfs_attr3_rmt_hdr_set() for the explanation.
205 		 */
206 		if (rmt->rm_lsn != cpu_to_be64(NULLCOMMITLSN)) {
207 			xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
208 			return;
209 		}
210 		xfs_update_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF);
211 
212 		len -= blksize;
213 		ptr += blksize;
214 		bno += BTOBB(blksize);
215 	}
216 
217 	if (len != 0)
218 		xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
219 }
220 
221 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
222 	.name = "xfs_attr3_rmt",
223 	.verify_read = xfs_attr3_rmt_read_verify,
224 	.verify_write = xfs_attr3_rmt_write_verify,
225 	.verify_struct = xfs_attr3_rmt_verify_struct,
226 };
227 
228 STATIC int
229 xfs_attr3_rmt_hdr_set(
230 	struct xfs_mount	*mp,
231 	void			*ptr,
232 	xfs_ino_t		ino,
233 	uint32_t		offset,
234 	uint32_t		size,
235 	xfs_daddr_t		bno)
236 {
237 	struct xfs_attr3_rmt_hdr *rmt = ptr;
238 
239 	if (!xfs_sb_version_hascrc(&mp->m_sb))
240 		return 0;
241 
242 	rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
243 	rmt->rm_offset = cpu_to_be32(offset);
244 	rmt->rm_bytes = cpu_to_be32(size);
245 	uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid);
246 	rmt->rm_owner = cpu_to_be64(ino);
247 	rmt->rm_blkno = cpu_to_be64(bno);
248 
249 	/*
250 	 * Remote attribute blocks are written synchronously, so we don't
251 	 * have an LSN that we can stamp in them that makes any sense to log
252 	 * recovery. To ensure that log recovery handles overwrites of these
253 	 * blocks sanely (i.e. once they've been freed and reallocated as some
254 	 * other type of metadata) we need to ensure that the LSN has a value
255 	 * that tells log recovery to ignore the LSN and overwrite the buffer
256 	 * with whatever is in it's log. To do this, we use the magic
257 	 * NULLCOMMITLSN to indicate that the LSN is invalid.
258 	 */
259 	rmt->rm_lsn = cpu_to_be64(NULLCOMMITLSN);
260 
261 	return sizeof(struct xfs_attr3_rmt_hdr);
262 }
263 
264 /*
265  * Helper functions to copy attribute data in and out of the one disk extents
266  */
267 STATIC int
268 xfs_attr_rmtval_copyout(
269 	struct xfs_mount *mp,
270 	struct xfs_buf	*bp,
271 	xfs_ino_t	ino,
272 	int		*offset,
273 	int		*valuelen,
274 	uint8_t		**dst)
275 {
276 	char		*src = bp->b_addr;
277 	xfs_daddr_t	bno = bp->b_bn;
278 	int		len = BBTOB(bp->b_length);
279 	int		blksize = mp->m_attr_geo->blksize;
280 
281 	ASSERT(len >= blksize);
282 
283 	while (len > 0 && *valuelen > 0) {
284 		int hdr_size = 0;
285 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
286 
287 		byte_cnt = min(*valuelen, byte_cnt);
288 
289 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
290 			if (xfs_attr3_rmt_hdr_ok(src, ino, *offset,
291 						  byte_cnt, bno)) {
292 				xfs_alert(mp,
293 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
294 					bno, *offset, byte_cnt, ino);
295 				return -EFSCORRUPTED;
296 			}
297 			hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
298 		}
299 
300 		memcpy(*dst, src + hdr_size, byte_cnt);
301 
302 		/* roll buffer forwards */
303 		len -= blksize;
304 		src += blksize;
305 		bno += BTOBB(blksize);
306 
307 		/* roll attribute data forwards */
308 		*valuelen -= byte_cnt;
309 		*dst += byte_cnt;
310 		*offset += byte_cnt;
311 	}
312 	return 0;
313 }
314 
315 STATIC void
316 xfs_attr_rmtval_copyin(
317 	struct xfs_mount *mp,
318 	struct xfs_buf	*bp,
319 	xfs_ino_t	ino,
320 	int		*offset,
321 	int		*valuelen,
322 	uint8_t		**src)
323 {
324 	char		*dst = bp->b_addr;
325 	xfs_daddr_t	bno = bp->b_bn;
326 	int		len = BBTOB(bp->b_length);
327 	int		blksize = mp->m_attr_geo->blksize;
328 
329 	ASSERT(len >= blksize);
330 
331 	while (len > 0 && *valuelen > 0) {
332 		int hdr_size;
333 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
334 
335 		byte_cnt = min(*valuelen, byte_cnt);
336 		hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
337 						 byte_cnt, bno);
338 
339 		memcpy(dst + hdr_size, *src, byte_cnt);
340 
341 		/*
342 		 * If this is the last block, zero the remainder of it.
343 		 * Check that we are actually the last block, too.
344 		 */
345 		if (byte_cnt + hdr_size < blksize) {
346 			ASSERT(*valuelen - byte_cnt == 0);
347 			ASSERT(len == blksize);
348 			memset(dst + hdr_size + byte_cnt, 0,
349 					blksize - hdr_size - byte_cnt);
350 		}
351 
352 		/* roll buffer forwards */
353 		len -= blksize;
354 		dst += blksize;
355 		bno += BTOBB(blksize);
356 
357 		/* roll attribute data forwards */
358 		*valuelen -= byte_cnt;
359 		*src += byte_cnt;
360 		*offset += byte_cnt;
361 	}
362 }
363 
364 /*
365  * Read the value associated with an attribute from the out-of-line buffer
366  * that we stored it in.
367  */
368 int
369 xfs_attr_rmtval_get(
370 	struct xfs_da_args	*args)
371 {
372 	struct xfs_bmbt_irec	map[ATTR_RMTVALUE_MAPSIZE];
373 	struct xfs_mount	*mp = args->dp->i_mount;
374 	struct xfs_buf		*bp;
375 	xfs_dablk_t		lblkno = args->rmtblkno;
376 	uint8_t			*dst = args->value;
377 	int			valuelen;
378 	int			nmap;
379 	int			error;
380 	int			blkcnt = args->rmtblkcnt;
381 	int			i;
382 	int			offset = 0;
383 
384 	trace_xfs_attr_rmtval_get(args);
385 
386 	ASSERT(!(args->flags & ATTR_KERNOVAL));
387 	ASSERT(args->rmtvaluelen == args->valuelen);
388 
389 	valuelen = args->rmtvaluelen;
390 	while (valuelen > 0) {
391 		nmap = ATTR_RMTVALUE_MAPSIZE;
392 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
393 				       blkcnt, map, &nmap,
394 				       XFS_BMAPI_ATTRFORK);
395 		if (error)
396 			return error;
397 		ASSERT(nmap >= 1);
398 
399 		for (i = 0; (i < nmap) && (valuelen > 0); i++) {
400 			xfs_daddr_t	dblkno;
401 			int		dblkcnt;
402 
403 			ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
404 			       (map[i].br_startblock != HOLESTARTBLOCK));
405 			dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
406 			dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
407 			error = xfs_trans_read_buf(mp, args->trans,
408 						   mp->m_ddev_targp,
409 						   dblkno, dblkcnt, 0, &bp,
410 						   &xfs_attr3_rmt_buf_ops);
411 			if (error)
412 				return error;
413 
414 			error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
415 							&offset, &valuelen,
416 							&dst);
417 			xfs_trans_brelse(args->trans, bp);
418 			if (error)
419 				return error;
420 
421 			/* roll attribute extent map forwards */
422 			lblkno += map[i].br_blockcount;
423 			blkcnt -= map[i].br_blockcount;
424 		}
425 	}
426 	ASSERT(valuelen == 0);
427 	return 0;
428 }
429 
430 /*
431  * Write the value associated with an attribute into the out-of-line buffer
432  * that we have defined for it.
433  */
434 int
435 xfs_attr_rmtval_set(
436 	struct xfs_da_args	*args)
437 {
438 	struct xfs_inode	*dp = args->dp;
439 	struct xfs_mount	*mp = dp->i_mount;
440 	struct xfs_bmbt_irec	map;
441 	xfs_dablk_t		lblkno;
442 	xfs_fileoff_t		lfileoff = 0;
443 	uint8_t			*src = args->value;
444 	int			blkcnt;
445 	int			valuelen;
446 	int			nmap;
447 	int			error;
448 	int			offset = 0;
449 
450 	trace_xfs_attr_rmtval_set(args);
451 
452 	/*
453 	 * Find a "hole" in the attribute address space large enough for
454 	 * us to drop the new attribute's value into. Because CRC enable
455 	 * attributes have headers, we can't just do a straight byte to FSB
456 	 * conversion and have to take the header space into account.
457 	 */
458 	blkcnt = xfs_attr3_rmt_blocks(mp, args->rmtvaluelen);
459 	error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
460 						   XFS_ATTR_FORK);
461 	if (error)
462 		return error;
463 
464 	args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
465 	args->rmtblkcnt = blkcnt;
466 
467 	/*
468 	 * Roll through the "value", allocating blocks on disk as required.
469 	 */
470 	while (blkcnt > 0) {
471 		/*
472 		 * Allocate a single extent, up to the size of the value.
473 		 *
474 		 * Note that we have to consider this a data allocation as we
475 		 * write the remote attribute without logging the contents.
476 		 * Hence we must ensure that we aren't using blocks that are on
477 		 * the busy list so that we don't overwrite blocks which have
478 		 * recently been freed but their transactions are not yet
479 		 * committed to disk. If we overwrite the contents of a busy
480 		 * extent and then crash then the block may not contain the
481 		 * correct metadata after log recovery occurs.
482 		 */
483 		nmap = 1;
484 		error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
485 				  blkcnt, XFS_BMAPI_ATTRFORK, args->total, &map,
486 				  &nmap);
487 		if (error)
488 			return error;
489 		error = xfs_defer_finish(&args->trans);
490 		if (error)
491 			return error;
492 
493 		ASSERT(nmap == 1);
494 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
495 		       (map.br_startblock != HOLESTARTBLOCK));
496 		lblkno += map.br_blockcount;
497 		blkcnt -= map.br_blockcount;
498 
499 		/*
500 		 * Start the next trans in the chain.
501 		 */
502 		error = xfs_trans_roll_inode(&args->trans, dp);
503 		if (error)
504 			return error;
505 	}
506 
507 	/*
508 	 * Roll through the "value", copying the attribute value to the
509 	 * already-allocated blocks.  Blocks are written synchronously
510 	 * so that we can know they are all on disk before we turn off
511 	 * the INCOMPLETE flag.
512 	 */
513 	lblkno = args->rmtblkno;
514 	blkcnt = args->rmtblkcnt;
515 	valuelen = args->rmtvaluelen;
516 	while (valuelen > 0) {
517 		struct xfs_buf	*bp;
518 		xfs_daddr_t	dblkno;
519 		int		dblkcnt;
520 
521 		ASSERT(blkcnt > 0);
522 
523 		nmap = 1;
524 		error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
525 				       blkcnt, &map, &nmap,
526 				       XFS_BMAPI_ATTRFORK);
527 		if (error)
528 			return error;
529 		ASSERT(nmap == 1);
530 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
531 		       (map.br_startblock != HOLESTARTBLOCK));
532 
533 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
534 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
535 
536 		bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
537 		if (!bp)
538 			return -ENOMEM;
539 		bp->b_ops = &xfs_attr3_rmt_buf_ops;
540 
541 		xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
542 				       &valuelen, &src);
543 
544 		error = xfs_bwrite(bp);	/* GROT: NOTE: synchronous write */
545 		xfs_buf_relse(bp);
546 		if (error)
547 			return error;
548 
549 
550 		/* roll attribute extent map forwards */
551 		lblkno += map.br_blockcount;
552 		blkcnt -= map.br_blockcount;
553 	}
554 	ASSERT(valuelen == 0);
555 	return 0;
556 }
557 
558 /*
559  * Remove the value associated with an attribute by deleting the
560  * out-of-line buffer that it is stored on.
561  */
562 int
563 xfs_attr_rmtval_remove(
564 	struct xfs_da_args	*args)
565 {
566 	struct xfs_mount	*mp = args->dp->i_mount;
567 	xfs_dablk_t		lblkno;
568 	int			blkcnt;
569 	int			error;
570 	int			done;
571 
572 	trace_xfs_attr_rmtval_remove(args);
573 
574 	/*
575 	 * Roll through the "value", invalidating the attribute value's blocks.
576 	 */
577 	lblkno = args->rmtblkno;
578 	blkcnt = args->rmtblkcnt;
579 	while (blkcnt > 0) {
580 		struct xfs_bmbt_irec	map;
581 		struct xfs_buf		*bp;
582 		xfs_daddr_t		dblkno;
583 		int			dblkcnt;
584 		int			nmap;
585 
586 		/*
587 		 * Try to remember where we decided to put the value.
588 		 */
589 		nmap = 1;
590 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
591 				       blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
592 		if (error)
593 			return error;
594 		ASSERT(nmap == 1);
595 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
596 		       (map.br_startblock != HOLESTARTBLOCK));
597 
598 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
599 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
600 
601 		/*
602 		 * If the "remote" value is in the cache, remove it.
603 		 */
604 		bp = xfs_buf_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
605 		if (bp) {
606 			xfs_buf_stale(bp);
607 			xfs_buf_relse(bp);
608 			bp = NULL;
609 		}
610 
611 		lblkno += map.br_blockcount;
612 		blkcnt -= map.br_blockcount;
613 	}
614 
615 	/*
616 	 * Keep de-allocating extents until the remote-value region is gone.
617 	 */
618 	lblkno = args->rmtblkno;
619 	blkcnt = args->rmtblkcnt;
620 	done = 0;
621 	while (!done) {
622 		error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
623 				    XFS_BMAPI_ATTRFORK, 1, &done);
624 		if (error)
625 			return error;
626 		error = xfs_defer_finish(&args->trans);
627 		if (error)
628 			return error;
629 
630 		/*
631 		 * Close out trans and start the next one in the chain.
632 		 */
633 		error = xfs_trans_roll_inode(&args->trans, args->dp);
634 		if (error)
635 			return error;
636 	}
637 	return 0;
638 }
639