xref: /openbmc/linux/fs/xfs/libxfs/xfs_attr_remote.c (revision be709d48)
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 	struct xfs_buf		*bp,
83 	void			*ptr,
84 	int			fsbsize,
85 	xfs_daddr_t		bno)
86 {
87 	struct xfs_attr3_rmt_hdr *rmt = ptr;
88 
89 	if (!xfs_sb_version_hascrc(&mp->m_sb))
90 		return __this_address;
91 	if (!xfs_verify_magic(bp, rmt->rm_magic))
92 		return __this_address;
93 	if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid))
94 		return __this_address;
95 	if (be64_to_cpu(rmt->rm_blkno) != bno)
96 		return __this_address;
97 	if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
98 		return __this_address;
99 	if (be32_to_cpu(rmt->rm_offset) +
100 				be32_to_cpu(rmt->rm_bytes) > XFS_XATTR_SIZE_MAX)
101 		return __this_address;
102 	if (rmt->rm_owner == 0)
103 		return __this_address;
104 
105 	return NULL;
106 }
107 
108 static int
109 __xfs_attr3_rmt_read_verify(
110 	struct xfs_buf	*bp,
111 	bool		check_crc,
112 	xfs_failaddr_t	*failaddr)
113 {
114 	struct xfs_mount *mp = bp->b_target->bt_mount;
115 	char		*ptr;
116 	int		len;
117 	xfs_daddr_t	bno;
118 	int		blksize = mp->m_attr_geo->blksize;
119 
120 	/* no verification of non-crc buffers */
121 	if (!xfs_sb_version_hascrc(&mp->m_sb))
122 		return 0;
123 
124 	ptr = bp->b_addr;
125 	bno = bp->b_bn;
126 	len = BBTOB(bp->b_length);
127 	ASSERT(len >= blksize);
128 
129 	while (len > 0) {
130 		if (check_crc &&
131 		    !xfs_verify_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF)) {
132 			*failaddr = __this_address;
133 			return -EFSBADCRC;
134 		}
135 		*failaddr = xfs_attr3_rmt_verify(mp, bp, ptr, blksize, bno);
136 		if (*failaddr)
137 			return -EFSCORRUPTED;
138 		len -= blksize;
139 		ptr += blksize;
140 		bno += BTOBB(blksize);
141 	}
142 
143 	if (len != 0) {
144 		*failaddr = __this_address;
145 		return -EFSCORRUPTED;
146 	}
147 
148 	return 0;
149 }
150 
151 static void
152 xfs_attr3_rmt_read_verify(
153 	struct xfs_buf	*bp)
154 {
155 	xfs_failaddr_t	fa;
156 	int		error;
157 
158 	error = __xfs_attr3_rmt_read_verify(bp, true, &fa);
159 	if (error)
160 		xfs_verifier_error(bp, error, fa);
161 }
162 
163 static xfs_failaddr_t
164 xfs_attr3_rmt_verify_struct(
165 	struct xfs_buf	*bp)
166 {
167 	xfs_failaddr_t	fa;
168 	int		error;
169 
170 	error = __xfs_attr3_rmt_read_verify(bp, false, &fa);
171 	return error ? fa : NULL;
172 }
173 
174 static void
175 xfs_attr3_rmt_write_verify(
176 	struct xfs_buf	*bp)
177 {
178 	struct xfs_mount *mp = bp->b_target->bt_mount;
179 	xfs_failaddr_t	fa;
180 	int		blksize = mp->m_attr_geo->blksize;
181 	char		*ptr;
182 	int		len;
183 	xfs_daddr_t	bno;
184 
185 	/* no verification of non-crc buffers */
186 	if (!xfs_sb_version_hascrc(&mp->m_sb))
187 		return;
188 
189 	ptr = bp->b_addr;
190 	bno = bp->b_bn;
191 	len = BBTOB(bp->b_length);
192 	ASSERT(len >= blksize);
193 
194 	while (len > 0) {
195 		struct xfs_attr3_rmt_hdr *rmt = (struct xfs_attr3_rmt_hdr *)ptr;
196 
197 		fa = xfs_attr3_rmt_verify(mp, bp, ptr, blksize, bno);
198 		if (fa) {
199 			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
200 			return;
201 		}
202 
203 		/*
204 		 * Ensure we aren't writing bogus LSNs to disk. See
205 		 * xfs_attr3_rmt_hdr_set() for the explanation.
206 		 */
207 		if (rmt->rm_lsn != cpu_to_be64(NULLCOMMITLSN)) {
208 			xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
209 			return;
210 		}
211 		xfs_update_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF);
212 
213 		len -= blksize;
214 		ptr += blksize;
215 		bno += BTOBB(blksize);
216 	}
217 
218 	if (len != 0)
219 		xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
220 }
221 
222 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
223 	.name = "xfs_attr3_rmt",
224 	.magic = { 0, cpu_to_be32(XFS_ATTR3_RMT_MAGIC) },
225 	.verify_read = xfs_attr3_rmt_read_verify,
226 	.verify_write = xfs_attr3_rmt_write_verify,
227 	.verify_struct = xfs_attr3_rmt_verify_struct,
228 };
229 
230 STATIC int
231 xfs_attr3_rmt_hdr_set(
232 	struct xfs_mount	*mp,
233 	void			*ptr,
234 	xfs_ino_t		ino,
235 	uint32_t		offset,
236 	uint32_t		size,
237 	xfs_daddr_t		bno)
238 {
239 	struct xfs_attr3_rmt_hdr *rmt = ptr;
240 
241 	if (!xfs_sb_version_hascrc(&mp->m_sb))
242 		return 0;
243 
244 	rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
245 	rmt->rm_offset = cpu_to_be32(offset);
246 	rmt->rm_bytes = cpu_to_be32(size);
247 	uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid);
248 	rmt->rm_owner = cpu_to_be64(ino);
249 	rmt->rm_blkno = cpu_to_be64(bno);
250 
251 	/*
252 	 * Remote attribute blocks are written synchronously, so we don't
253 	 * have an LSN that we can stamp in them that makes any sense to log
254 	 * recovery. To ensure that log recovery handles overwrites of these
255 	 * blocks sanely (i.e. once they've been freed and reallocated as some
256 	 * other type of metadata) we need to ensure that the LSN has a value
257 	 * that tells log recovery to ignore the LSN and overwrite the buffer
258 	 * with whatever is in it's log. To do this, we use the magic
259 	 * NULLCOMMITLSN to indicate that the LSN is invalid.
260 	 */
261 	rmt->rm_lsn = cpu_to_be64(NULLCOMMITLSN);
262 
263 	return sizeof(struct xfs_attr3_rmt_hdr);
264 }
265 
266 /*
267  * Helper functions to copy attribute data in and out of the one disk extents
268  */
269 STATIC int
270 xfs_attr_rmtval_copyout(
271 	struct xfs_mount *mp,
272 	struct xfs_buf	*bp,
273 	xfs_ino_t	ino,
274 	int		*offset,
275 	int		*valuelen,
276 	uint8_t		**dst)
277 {
278 	char		*src = bp->b_addr;
279 	xfs_daddr_t	bno = bp->b_bn;
280 	int		len = BBTOB(bp->b_length);
281 	int		blksize = mp->m_attr_geo->blksize;
282 
283 	ASSERT(len >= blksize);
284 
285 	while (len > 0 && *valuelen > 0) {
286 		int hdr_size = 0;
287 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
288 
289 		byte_cnt = min(*valuelen, byte_cnt);
290 
291 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
292 			if (xfs_attr3_rmt_hdr_ok(src, ino, *offset,
293 						  byte_cnt, bno)) {
294 				xfs_alert(mp,
295 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
296 					bno, *offset, byte_cnt, ino);
297 				return -EFSCORRUPTED;
298 			}
299 			hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
300 		}
301 
302 		memcpy(*dst, src + hdr_size, byte_cnt);
303 
304 		/* roll buffer forwards */
305 		len -= blksize;
306 		src += blksize;
307 		bno += BTOBB(blksize);
308 
309 		/* roll attribute data forwards */
310 		*valuelen -= byte_cnt;
311 		*dst += byte_cnt;
312 		*offset += byte_cnt;
313 	}
314 	return 0;
315 }
316 
317 STATIC void
318 xfs_attr_rmtval_copyin(
319 	struct xfs_mount *mp,
320 	struct xfs_buf	*bp,
321 	xfs_ino_t	ino,
322 	int		*offset,
323 	int		*valuelen,
324 	uint8_t		**src)
325 {
326 	char		*dst = bp->b_addr;
327 	xfs_daddr_t	bno = bp->b_bn;
328 	int		len = BBTOB(bp->b_length);
329 	int		blksize = mp->m_attr_geo->blksize;
330 
331 	ASSERT(len >= blksize);
332 
333 	while (len > 0 && *valuelen > 0) {
334 		int hdr_size;
335 		int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
336 
337 		byte_cnt = min(*valuelen, byte_cnt);
338 		hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
339 						 byte_cnt, bno);
340 
341 		memcpy(dst + hdr_size, *src, byte_cnt);
342 
343 		/*
344 		 * If this is the last block, zero the remainder of it.
345 		 * Check that we are actually the last block, too.
346 		 */
347 		if (byte_cnt + hdr_size < blksize) {
348 			ASSERT(*valuelen - byte_cnt == 0);
349 			ASSERT(len == blksize);
350 			memset(dst + hdr_size + byte_cnt, 0,
351 					blksize - hdr_size - byte_cnt);
352 		}
353 
354 		/* roll buffer forwards */
355 		len -= blksize;
356 		dst += blksize;
357 		bno += BTOBB(blksize);
358 
359 		/* roll attribute data forwards */
360 		*valuelen -= byte_cnt;
361 		*src += byte_cnt;
362 		*offset += byte_cnt;
363 	}
364 }
365 
366 /*
367  * Read the value associated with an attribute from the out-of-line buffer
368  * that we stored it in.
369  */
370 int
371 xfs_attr_rmtval_get(
372 	struct xfs_da_args	*args)
373 {
374 	struct xfs_bmbt_irec	map[ATTR_RMTVALUE_MAPSIZE];
375 	struct xfs_mount	*mp = args->dp->i_mount;
376 	struct xfs_buf		*bp;
377 	xfs_dablk_t		lblkno = args->rmtblkno;
378 	uint8_t			*dst = args->value;
379 	int			valuelen;
380 	int			nmap;
381 	int			error;
382 	int			blkcnt = args->rmtblkcnt;
383 	int			i;
384 	int			offset = 0;
385 
386 	trace_xfs_attr_rmtval_get(args);
387 
388 	ASSERT(!(args->flags & ATTR_KERNOVAL));
389 	ASSERT(args->rmtvaluelen == args->valuelen);
390 
391 	valuelen = args->rmtvaluelen;
392 	while (valuelen > 0) {
393 		nmap = ATTR_RMTVALUE_MAPSIZE;
394 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
395 				       blkcnt, map, &nmap,
396 				       XFS_BMAPI_ATTRFORK);
397 		if (error)
398 			return error;
399 		ASSERT(nmap >= 1);
400 
401 		for (i = 0; (i < nmap) && (valuelen > 0); i++) {
402 			xfs_daddr_t	dblkno;
403 			int		dblkcnt;
404 
405 			ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
406 			       (map[i].br_startblock != HOLESTARTBLOCK));
407 			dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
408 			dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
409 			error = xfs_trans_read_buf(mp, args->trans,
410 						   mp->m_ddev_targp,
411 						   dblkno, dblkcnt, 0, &bp,
412 						   &xfs_attr3_rmt_buf_ops);
413 			if (error)
414 				return error;
415 
416 			error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
417 							&offset, &valuelen,
418 							&dst);
419 			xfs_trans_brelse(args->trans, bp);
420 			if (error)
421 				return error;
422 
423 			/* roll attribute extent map forwards */
424 			lblkno += map[i].br_blockcount;
425 			blkcnt -= map[i].br_blockcount;
426 		}
427 	}
428 	ASSERT(valuelen == 0);
429 	return 0;
430 }
431 
432 /*
433  * Write the value associated with an attribute into the out-of-line buffer
434  * that we have defined for it.
435  */
436 int
437 xfs_attr_rmtval_set(
438 	struct xfs_da_args	*args)
439 {
440 	struct xfs_inode	*dp = args->dp;
441 	struct xfs_mount	*mp = dp->i_mount;
442 	struct xfs_bmbt_irec	map;
443 	xfs_dablk_t		lblkno;
444 	xfs_fileoff_t		lfileoff = 0;
445 	uint8_t			*src = args->value;
446 	int			blkcnt;
447 	int			valuelen;
448 	int			nmap;
449 	int			error;
450 	int			offset = 0;
451 
452 	trace_xfs_attr_rmtval_set(args);
453 
454 	/*
455 	 * Find a "hole" in the attribute address space large enough for
456 	 * us to drop the new attribute's value into. Because CRC enable
457 	 * attributes have headers, we can't just do a straight byte to FSB
458 	 * conversion and have to take the header space into account.
459 	 */
460 	blkcnt = xfs_attr3_rmt_blocks(mp, args->rmtvaluelen);
461 	error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
462 						   XFS_ATTR_FORK);
463 	if (error)
464 		return error;
465 
466 	args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
467 	args->rmtblkcnt = blkcnt;
468 
469 	/*
470 	 * Roll through the "value", allocating blocks on disk as required.
471 	 */
472 	while (blkcnt > 0) {
473 		/*
474 		 * Allocate a single extent, up to the size of the value.
475 		 *
476 		 * Note that we have to consider this a data allocation as we
477 		 * write the remote attribute without logging the contents.
478 		 * Hence we must ensure that we aren't using blocks that are on
479 		 * the busy list so that we don't overwrite blocks which have
480 		 * recently been freed but their transactions are not yet
481 		 * committed to disk. If we overwrite the contents of a busy
482 		 * extent and then crash then the block may not contain the
483 		 * correct metadata after log recovery occurs.
484 		 */
485 		nmap = 1;
486 		error = xfs_bmapi_write(args->trans, dp, (xfs_fileoff_t)lblkno,
487 				  blkcnt, XFS_BMAPI_ATTRFORK, args->total, &map,
488 				  &nmap);
489 		if (error)
490 			return error;
491 		error = xfs_defer_finish(&args->trans);
492 		if (error)
493 			return error;
494 
495 		ASSERT(nmap == 1);
496 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
497 		       (map.br_startblock != HOLESTARTBLOCK));
498 		lblkno += map.br_blockcount;
499 		blkcnt -= map.br_blockcount;
500 
501 		/*
502 		 * Start the next trans in the chain.
503 		 */
504 		error = xfs_trans_roll_inode(&args->trans, dp);
505 		if (error)
506 			return error;
507 	}
508 
509 	/*
510 	 * Roll through the "value", copying the attribute value to the
511 	 * already-allocated blocks.  Blocks are written synchronously
512 	 * so that we can know they are all on disk before we turn off
513 	 * the INCOMPLETE flag.
514 	 */
515 	lblkno = args->rmtblkno;
516 	blkcnt = args->rmtblkcnt;
517 	valuelen = args->rmtvaluelen;
518 	while (valuelen > 0) {
519 		struct xfs_buf	*bp;
520 		xfs_daddr_t	dblkno;
521 		int		dblkcnt;
522 
523 		ASSERT(blkcnt > 0);
524 
525 		nmap = 1;
526 		error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
527 				       blkcnt, &map, &nmap,
528 				       XFS_BMAPI_ATTRFORK);
529 		if (error)
530 			return error;
531 		ASSERT(nmap == 1);
532 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
533 		       (map.br_startblock != HOLESTARTBLOCK));
534 
535 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
536 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
537 
538 		bp = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, 0);
539 		if (!bp)
540 			return -ENOMEM;
541 		bp->b_ops = &xfs_attr3_rmt_buf_ops;
542 
543 		xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
544 				       &valuelen, &src);
545 
546 		error = xfs_bwrite(bp);	/* GROT: NOTE: synchronous write */
547 		xfs_buf_relse(bp);
548 		if (error)
549 			return error;
550 
551 
552 		/* roll attribute extent map forwards */
553 		lblkno += map.br_blockcount;
554 		blkcnt -= map.br_blockcount;
555 	}
556 	ASSERT(valuelen == 0);
557 	return 0;
558 }
559 
560 /*
561  * Remove the value associated with an attribute by deleting the
562  * out-of-line buffer that it is stored on.
563  */
564 int
565 xfs_attr_rmtval_remove(
566 	struct xfs_da_args	*args)
567 {
568 	struct xfs_mount	*mp = args->dp->i_mount;
569 	xfs_dablk_t		lblkno;
570 	int			blkcnt;
571 	int			error;
572 	int			done;
573 
574 	trace_xfs_attr_rmtval_remove(args);
575 
576 	/*
577 	 * Roll through the "value", invalidating the attribute value's blocks.
578 	 */
579 	lblkno = args->rmtblkno;
580 	blkcnt = args->rmtblkcnt;
581 	while (blkcnt > 0) {
582 		struct xfs_bmbt_irec	map;
583 		struct xfs_buf		*bp;
584 		xfs_daddr_t		dblkno;
585 		int			dblkcnt;
586 		int			nmap;
587 
588 		/*
589 		 * Try to remember where we decided to put the value.
590 		 */
591 		nmap = 1;
592 		error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
593 				       blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
594 		if (error)
595 			return error;
596 		ASSERT(nmap == 1);
597 		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
598 		       (map.br_startblock != HOLESTARTBLOCK));
599 
600 		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
601 		dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
602 
603 		/*
604 		 * If the "remote" value is in the cache, remove it.
605 		 */
606 		bp = xfs_buf_incore(mp->m_ddev_targp, dblkno, dblkcnt, XBF_TRYLOCK);
607 		if (bp) {
608 			xfs_buf_stale(bp);
609 			xfs_buf_relse(bp);
610 			bp = NULL;
611 		}
612 
613 		lblkno += map.br_blockcount;
614 		blkcnt -= map.br_blockcount;
615 	}
616 
617 	/*
618 	 * Keep de-allocating extents until the remote-value region is gone.
619 	 */
620 	lblkno = args->rmtblkno;
621 	blkcnt = args->rmtblkcnt;
622 	done = 0;
623 	while (!done) {
624 		error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
625 				    XFS_BMAPI_ATTRFORK, 1, &done);
626 		if (error)
627 			return error;
628 		error = xfs_defer_finish(&args->trans);
629 		if (error)
630 			return error;
631 
632 		/*
633 		 * Close out trans and start the next one in the chain.
634 		 */
635 		error = xfs_trans_roll_inode(&args->trans, args->dp);
636 		if (error)
637 			return error;
638 	}
639 	return 0;
640 }
641