Loading...
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_trans.h"
20#include "xfs_bmap.h"
21#include "xfs_attr.h"
22#include "xfs_attr_remote.h"
23#include "xfs_trace.h"
24#include "xfs_error.h"
25
26#define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
27
28/*
29 * Remote Attribute Values
30 * =======================
31 *
32 * Remote extended attribute values are conceptually simple -- they're written
33 * to data blocks mapped by an inode's attribute fork, and they have an upper
34 * size limit of 64k. Setting a value does not involve the XFS log.
35 *
36 * However, on a v5 filesystem, maximally sized remote attr values require one
37 * block more than 64k worth of space to hold both the remote attribute value
38 * header (64 bytes). On a 4k block filesystem this results in a 68k buffer;
39 * on a 64k block filesystem, this would be a 128k buffer. Note that the log
40 * format can only handle a dirty buffer of XFS_MAX_BLOCKSIZE length (64k).
41 * Therefore, we /must/ ensure that remote attribute value buffers never touch
42 * the logging system and therefore never have a log item.
43 */
44
45/*
46 * Each contiguous block has a header, so it is not just a simple attribute
47 * length to FSB conversion.
48 */
49int
50xfs_attr3_rmt_blocks(
51 struct xfs_mount *mp,
52 int attrlen)
53{
54 if (xfs_has_crc(mp)) {
55 int buflen = XFS_ATTR3_RMT_BUF_SPACE(mp, mp->m_sb.sb_blocksize);
56 return (attrlen + buflen - 1) / buflen;
57 }
58 return XFS_B_TO_FSB(mp, attrlen);
59}
60
61/*
62 * Checking of the remote attribute header is split into two parts. The verifier
63 * does CRC, location and bounds checking, the unpacking function checks the
64 * attribute parameters and owner.
65 */
66static xfs_failaddr_t
67xfs_attr3_rmt_hdr_ok(
68 void *ptr,
69 xfs_ino_t ino,
70 uint32_t offset,
71 uint32_t size,
72 xfs_daddr_t bno)
73{
74 struct xfs_attr3_rmt_hdr *rmt = ptr;
75
76 if (bno != be64_to_cpu(rmt->rm_blkno))
77 return __this_address;
78 if (offset != be32_to_cpu(rmt->rm_offset))
79 return __this_address;
80 if (size != be32_to_cpu(rmt->rm_bytes))
81 return __this_address;
82 if (ino != be64_to_cpu(rmt->rm_owner))
83 return __this_address;
84
85 /* ok */
86 return NULL;
87}
88
89static xfs_failaddr_t
90xfs_attr3_rmt_verify(
91 struct xfs_mount *mp,
92 struct xfs_buf *bp,
93 void *ptr,
94 int fsbsize,
95 xfs_daddr_t bno)
96{
97 struct xfs_attr3_rmt_hdr *rmt = ptr;
98
99 if (!xfs_verify_magic(bp, rmt->rm_magic))
100 return __this_address;
101 if (!uuid_equal(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid))
102 return __this_address;
103 if (be64_to_cpu(rmt->rm_blkno) != bno)
104 return __this_address;
105 if (be32_to_cpu(rmt->rm_bytes) > fsbsize - sizeof(*rmt))
106 return __this_address;
107 if (be32_to_cpu(rmt->rm_offset) +
108 be32_to_cpu(rmt->rm_bytes) > XFS_XATTR_SIZE_MAX)
109 return __this_address;
110 if (rmt->rm_owner == 0)
111 return __this_address;
112
113 return NULL;
114}
115
116static int
117__xfs_attr3_rmt_read_verify(
118 struct xfs_buf *bp,
119 bool check_crc,
120 xfs_failaddr_t *failaddr)
121{
122 struct xfs_mount *mp = bp->b_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_has_crc(mp))
130 return 0;
131
132 ptr = bp->b_addr;
133 bno = xfs_buf_daddr(bp);
134 len = BBTOB(bp->b_length);
135 ASSERT(len >= blksize);
136
137 while (len > 0) {
138 if (check_crc &&
139 !xfs_verify_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF)) {
140 *failaddr = __this_address;
141 return -EFSBADCRC;
142 }
143 *failaddr = xfs_attr3_rmt_verify(mp, bp, ptr, blksize, bno);
144 if (*failaddr)
145 return -EFSCORRUPTED;
146 len -= blksize;
147 ptr += blksize;
148 bno += BTOBB(blksize);
149 }
150
151 if (len != 0) {
152 *failaddr = __this_address;
153 return -EFSCORRUPTED;
154 }
155
156 return 0;
157}
158
159static void
160xfs_attr3_rmt_read_verify(
161 struct xfs_buf *bp)
162{
163 xfs_failaddr_t fa;
164 int error;
165
166 error = __xfs_attr3_rmt_read_verify(bp, true, &fa);
167 if (error)
168 xfs_verifier_error(bp, error, fa);
169}
170
171static xfs_failaddr_t
172xfs_attr3_rmt_verify_struct(
173 struct xfs_buf *bp)
174{
175 xfs_failaddr_t fa;
176 int error;
177
178 error = __xfs_attr3_rmt_read_verify(bp, false, &fa);
179 return error ? fa : NULL;
180}
181
182static void
183xfs_attr3_rmt_write_verify(
184 struct xfs_buf *bp)
185{
186 struct xfs_mount *mp = bp->b_mount;
187 xfs_failaddr_t fa;
188 int blksize = mp->m_attr_geo->blksize;
189 char *ptr;
190 int len;
191 xfs_daddr_t bno;
192
193 /* no verification of non-crc buffers */
194 if (!xfs_has_crc(mp))
195 return;
196
197 ptr = bp->b_addr;
198 bno = xfs_buf_daddr(bp);
199 len = BBTOB(bp->b_length);
200 ASSERT(len >= blksize);
201
202 while (len > 0) {
203 struct xfs_attr3_rmt_hdr *rmt = (struct xfs_attr3_rmt_hdr *)ptr;
204
205 fa = xfs_attr3_rmt_verify(mp, bp, ptr, blksize, bno);
206 if (fa) {
207 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
208 return;
209 }
210
211 /*
212 * Ensure we aren't writing bogus LSNs to disk. See
213 * xfs_attr3_rmt_hdr_set() for the explanation.
214 */
215 if (rmt->rm_lsn != cpu_to_be64(NULLCOMMITLSN)) {
216 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
217 return;
218 }
219 xfs_update_cksum(ptr, blksize, XFS_ATTR3_RMT_CRC_OFF);
220
221 len -= blksize;
222 ptr += blksize;
223 bno += BTOBB(blksize);
224 }
225
226 if (len != 0)
227 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
228}
229
230const struct xfs_buf_ops xfs_attr3_rmt_buf_ops = {
231 .name = "xfs_attr3_rmt",
232 .magic = { 0, cpu_to_be32(XFS_ATTR3_RMT_MAGIC) },
233 .verify_read = xfs_attr3_rmt_read_verify,
234 .verify_write = xfs_attr3_rmt_write_verify,
235 .verify_struct = xfs_attr3_rmt_verify_struct,
236};
237
238STATIC int
239xfs_attr3_rmt_hdr_set(
240 struct xfs_mount *mp,
241 void *ptr,
242 xfs_ino_t ino,
243 uint32_t offset,
244 uint32_t size,
245 xfs_daddr_t bno)
246{
247 struct xfs_attr3_rmt_hdr *rmt = ptr;
248
249 if (!xfs_has_crc(mp))
250 return 0;
251
252 rmt->rm_magic = cpu_to_be32(XFS_ATTR3_RMT_MAGIC);
253 rmt->rm_offset = cpu_to_be32(offset);
254 rmt->rm_bytes = cpu_to_be32(size);
255 uuid_copy(&rmt->rm_uuid, &mp->m_sb.sb_meta_uuid);
256 rmt->rm_owner = cpu_to_be64(ino);
257 rmt->rm_blkno = cpu_to_be64(bno);
258
259 /*
260 * Remote attribute blocks are written synchronously, so we don't
261 * have an LSN that we can stamp in them that makes any sense to log
262 * recovery. To ensure that log recovery handles overwrites of these
263 * blocks sanely (i.e. once they've been freed and reallocated as some
264 * other type of metadata) we need to ensure that the LSN has a value
265 * that tells log recovery to ignore the LSN and overwrite the buffer
266 * with whatever is in it's log. To do this, we use the magic
267 * NULLCOMMITLSN to indicate that the LSN is invalid.
268 */
269 rmt->rm_lsn = cpu_to_be64(NULLCOMMITLSN);
270
271 return sizeof(struct xfs_attr3_rmt_hdr);
272}
273
274/*
275 * Helper functions to copy attribute data in and out of the one disk extents
276 */
277STATIC int
278xfs_attr_rmtval_copyout(
279 struct xfs_mount *mp,
280 struct xfs_buf *bp,
281 xfs_ino_t ino,
282 int *offset,
283 int *valuelen,
284 uint8_t **dst)
285{
286 char *src = bp->b_addr;
287 xfs_daddr_t bno = xfs_buf_daddr(bp);
288 int len = BBTOB(bp->b_length);
289 int blksize = mp->m_attr_geo->blksize;
290
291 ASSERT(len >= blksize);
292
293 while (len > 0 && *valuelen > 0) {
294 int hdr_size = 0;
295 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
296
297 byte_cnt = min(*valuelen, byte_cnt);
298
299 if (xfs_has_crc(mp)) {
300 if (xfs_attr3_rmt_hdr_ok(src, ino, *offset,
301 byte_cnt, bno)) {
302 xfs_alert(mp,
303"remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
304 bno, *offset, byte_cnt, ino);
305 return -EFSCORRUPTED;
306 }
307 hdr_size = sizeof(struct xfs_attr3_rmt_hdr);
308 }
309
310 memcpy(*dst, src + hdr_size, byte_cnt);
311
312 /* roll buffer forwards */
313 len -= blksize;
314 src += blksize;
315 bno += BTOBB(blksize);
316
317 /* roll attribute data forwards */
318 *valuelen -= byte_cnt;
319 *dst += byte_cnt;
320 *offset += byte_cnt;
321 }
322 return 0;
323}
324
325STATIC void
326xfs_attr_rmtval_copyin(
327 struct xfs_mount *mp,
328 struct xfs_buf *bp,
329 xfs_ino_t ino,
330 int *offset,
331 int *valuelen,
332 uint8_t **src)
333{
334 char *dst = bp->b_addr;
335 xfs_daddr_t bno = xfs_buf_daddr(bp);
336 int len = BBTOB(bp->b_length);
337 int blksize = mp->m_attr_geo->blksize;
338
339 ASSERT(len >= blksize);
340
341 while (len > 0 && *valuelen > 0) {
342 int hdr_size;
343 int byte_cnt = XFS_ATTR3_RMT_BUF_SPACE(mp, blksize);
344
345 byte_cnt = min(*valuelen, byte_cnt);
346 hdr_size = xfs_attr3_rmt_hdr_set(mp, dst, ino, *offset,
347 byte_cnt, bno);
348
349 memcpy(dst + hdr_size, *src, byte_cnt);
350
351 /*
352 * If this is the last block, zero the remainder of it.
353 * Check that we are actually the last block, too.
354 */
355 if (byte_cnt + hdr_size < blksize) {
356 ASSERT(*valuelen - byte_cnt == 0);
357 ASSERT(len == blksize);
358 memset(dst + hdr_size + byte_cnt, 0,
359 blksize - hdr_size - byte_cnt);
360 }
361
362 /* roll buffer forwards */
363 len -= blksize;
364 dst += blksize;
365 bno += BTOBB(blksize);
366
367 /* roll attribute data forwards */
368 *valuelen -= byte_cnt;
369 *src += byte_cnt;
370 *offset += byte_cnt;
371 }
372}
373
374/*
375 * Read the value associated with an attribute from the out-of-line buffer
376 * that we stored it in.
377 *
378 * Returns 0 on successful retrieval, otherwise an error.
379 */
380int
381xfs_attr_rmtval_get(
382 struct xfs_da_args *args)
383{
384 struct xfs_bmbt_irec map[ATTR_RMTVALUE_MAPSIZE];
385 struct xfs_mount *mp = args->dp->i_mount;
386 struct xfs_buf *bp;
387 xfs_dablk_t lblkno = args->rmtblkno;
388 uint8_t *dst = args->value;
389 int valuelen;
390 int nmap;
391 int error;
392 int blkcnt = args->rmtblkcnt;
393 int i;
394 int offset = 0;
395
396 trace_xfs_attr_rmtval_get(args);
397
398 ASSERT(args->valuelen != 0);
399 ASSERT(args->rmtvaluelen == args->valuelen);
400
401 valuelen = args->rmtvaluelen;
402 while (valuelen > 0) {
403 nmap = ATTR_RMTVALUE_MAPSIZE;
404 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
405 blkcnt, map, &nmap,
406 XFS_BMAPI_ATTRFORK);
407 if (error)
408 return error;
409 ASSERT(nmap >= 1);
410
411 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
412 xfs_daddr_t dblkno;
413 int dblkcnt;
414
415 ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
416 (map[i].br_startblock != HOLESTARTBLOCK));
417 dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
418 dblkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
419 error = xfs_buf_read(mp->m_ddev_targp, dblkno, dblkcnt,
420 0, &bp, &xfs_attr3_rmt_buf_ops);
421 if (error)
422 return error;
423
424 error = xfs_attr_rmtval_copyout(mp, bp, args->dp->i_ino,
425 &offset, &valuelen,
426 &dst);
427 xfs_buf_relse(bp);
428 if (error)
429 return error;
430
431 /* roll attribute extent map forwards */
432 lblkno += map[i].br_blockcount;
433 blkcnt -= map[i].br_blockcount;
434 }
435 }
436 ASSERT(valuelen == 0);
437 return 0;
438}
439
440/*
441 * Find a "hole" in the attribute address space large enough for us to drop the
442 * new attributes value into
443 */
444int
445xfs_attr_rmt_find_hole(
446 struct xfs_da_args *args)
447{
448 struct xfs_inode *dp = args->dp;
449 struct xfs_mount *mp = dp->i_mount;
450 int error;
451 int blkcnt;
452 xfs_fileoff_t lfileoff = 0;
453
454 /*
455 * Because CRC enable attributes have headers, we can't just do a
456 * straight byte to FSB conversion and have to take the header space
457 * into account.
458 */
459 blkcnt = xfs_attr3_rmt_blocks(mp, args->rmtvaluelen);
460 error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
461 XFS_ATTR_FORK);
462 if (error)
463 return error;
464
465 args->rmtblkno = (xfs_dablk_t)lfileoff;
466 args->rmtblkcnt = blkcnt;
467
468 return 0;
469}
470
471int
472xfs_attr_rmtval_set_value(
473 struct xfs_da_args *args)
474{
475 struct xfs_inode *dp = args->dp;
476 struct xfs_mount *mp = dp->i_mount;
477 struct xfs_bmbt_irec map;
478 xfs_dablk_t lblkno;
479 uint8_t *src = args->value;
480 int blkcnt;
481 int valuelen;
482 int nmap;
483 int error;
484 int offset = 0;
485
486 /*
487 * Roll through the "value", copying the attribute value to the
488 * already-allocated blocks. Blocks are written synchronously
489 * so that we can know they are all on disk before we turn off
490 * the INCOMPLETE flag.
491 */
492 lblkno = args->rmtblkno;
493 blkcnt = args->rmtblkcnt;
494 valuelen = args->rmtvaluelen;
495 while (valuelen > 0) {
496 struct xfs_buf *bp;
497 xfs_daddr_t dblkno;
498 int dblkcnt;
499
500 ASSERT(blkcnt > 0);
501
502 nmap = 1;
503 error = xfs_bmapi_read(dp, (xfs_fileoff_t)lblkno,
504 blkcnt, &map, &nmap,
505 XFS_BMAPI_ATTRFORK);
506 if (error)
507 return error;
508 ASSERT(nmap == 1);
509 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
510 (map.br_startblock != HOLESTARTBLOCK));
511
512 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
513 dblkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
514
515 error = xfs_buf_get(mp->m_ddev_targp, dblkno, dblkcnt, &bp);
516 if (error)
517 return error;
518 bp->b_ops = &xfs_attr3_rmt_buf_ops;
519
520 xfs_attr_rmtval_copyin(mp, bp, args->dp->i_ino, &offset,
521 &valuelen, &src);
522
523 error = xfs_bwrite(bp); /* GROT: NOTE: synchronous write */
524 xfs_buf_relse(bp);
525 if (error)
526 return error;
527
528
529 /* roll attribute extent map forwards */
530 lblkno += map.br_blockcount;
531 blkcnt -= map.br_blockcount;
532 }
533 ASSERT(valuelen == 0);
534 return 0;
535}
536
537/* Mark stale any incore buffers for the remote value. */
538int
539xfs_attr_rmtval_stale(
540 struct xfs_inode *ip,
541 struct xfs_bmbt_irec *map,
542 xfs_buf_flags_t incore_flags)
543{
544 struct xfs_mount *mp = ip->i_mount;
545 struct xfs_buf *bp;
546 int error;
547
548 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
549
550 if (XFS_IS_CORRUPT(mp, map->br_startblock == DELAYSTARTBLOCK) ||
551 XFS_IS_CORRUPT(mp, map->br_startblock == HOLESTARTBLOCK))
552 return -EFSCORRUPTED;
553
554 error = xfs_buf_incore(mp->m_ddev_targp,
555 XFS_FSB_TO_DADDR(mp, map->br_startblock),
556 XFS_FSB_TO_BB(mp, map->br_blockcount),
557 incore_flags, &bp);
558 if (error) {
559 if (error == -ENOENT)
560 return 0;
561 return error;
562 }
563
564 xfs_buf_stale(bp);
565 xfs_buf_relse(bp);
566 return 0;
567}
568
569/*
570 * Find a hole for the attr and store it in the delayed attr context. This
571 * initializes the context to roll through allocating an attr extent for a
572 * delayed attr operation
573 */
574int
575xfs_attr_rmtval_find_space(
576 struct xfs_attr_intent *attr)
577{
578 struct xfs_da_args *args = attr->xattri_da_args;
579 struct xfs_bmbt_irec *map = &attr->xattri_map;
580 int error;
581
582 attr->xattri_lblkno = 0;
583 attr->xattri_blkcnt = 0;
584 args->rmtblkcnt = 0;
585 args->rmtblkno = 0;
586 memset(map, 0, sizeof(struct xfs_bmbt_irec));
587
588 error = xfs_attr_rmt_find_hole(args);
589 if (error)
590 return error;
591
592 attr->xattri_blkcnt = args->rmtblkcnt;
593 attr->xattri_lblkno = args->rmtblkno;
594
595 return 0;
596}
597
598/*
599 * Write one block of the value associated with an attribute into the
600 * out-of-line buffer that we have defined for it. This is similar to a subset
601 * of xfs_attr_rmtval_set, but records the current block to the delayed attr
602 * context, and leaves transaction handling to the caller.
603 */
604int
605xfs_attr_rmtval_set_blk(
606 struct xfs_attr_intent *attr)
607{
608 struct xfs_da_args *args = attr->xattri_da_args;
609 struct xfs_inode *dp = args->dp;
610 struct xfs_bmbt_irec *map = &attr->xattri_map;
611 int nmap;
612 int error;
613
614 nmap = 1;
615 error = xfs_bmapi_write(args->trans, dp,
616 (xfs_fileoff_t)attr->xattri_lblkno,
617 attr->xattri_blkcnt, XFS_BMAPI_ATTRFORK, args->total,
618 map, &nmap);
619 if (error)
620 return error;
621
622 ASSERT(nmap == 1);
623 ASSERT((map->br_startblock != DELAYSTARTBLOCK) &&
624 (map->br_startblock != HOLESTARTBLOCK));
625
626 /* roll attribute extent map forwards */
627 attr->xattri_lblkno += map->br_blockcount;
628 attr->xattri_blkcnt -= map->br_blockcount;
629
630 return 0;
631}
632
633/*
634 * Remove the value associated with an attribute by deleting the
635 * out-of-line buffer that it is stored on.
636 */
637int
638xfs_attr_rmtval_invalidate(
639 struct xfs_da_args *args)
640{
641 xfs_dablk_t lblkno;
642 int blkcnt;
643 int error;
644
645 /*
646 * Roll through the "value", invalidating the attribute value's blocks.
647 */
648 lblkno = args->rmtblkno;
649 blkcnt = args->rmtblkcnt;
650 while (blkcnt > 0) {
651 struct xfs_bmbt_irec map;
652 int nmap;
653
654 /*
655 * Try to remember where we decided to put the value.
656 */
657 nmap = 1;
658 error = xfs_bmapi_read(args->dp, (xfs_fileoff_t)lblkno,
659 blkcnt, &map, &nmap, XFS_BMAPI_ATTRFORK);
660 if (error)
661 return error;
662 if (XFS_IS_CORRUPT(args->dp->i_mount, nmap != 1))
663 return -EFSCORRUPTED;
664 error = xfs_attr_rmtval_stale(args->dp, &map, XBF_TRYLOCK);
665 if (error)
666 return error;
667
668 lblkno += map.br_blockcount;
669 blkcnt -= map.br_blockcount;
670 }
671 return 0;
672}
673
674/*
675 * Remove the value associated with an attribute by deleting the out-of-line
676 * buffer that it is stored on. Returns -EAGAIN for the caller to refresh the
677 * transaction and re-call the function. Callers should keep calling this
678 * routine until it returns something other than -EAGAIN.
679 */
680int
681xfs_attr_rmtval_remove(
682 struct xfs_attr_intent *attr)
683{
684 struct xfs_da_args *args = attr->xattri_da_args;
685 int error, done;
686
687 /*
688 * Unmap value blocks for this attr.
689 */
690 error = xfs_bunmapi(args->trans, args->dp, args->rmtblkno,
691 args->rmtblkcnt, XFS_BMAPI_ATTRFORK, 1, &done);
692 if (error)
693 return error;
694
695 /*
696 * We don't need an explicit state here to pick up where we left off. We
697 * can figure it out using the !done return code. The actual value of
698 * attr->xattri_dela_state may be some value reminiscent of the calling
699 * function, but it's value is irrelevant with in the context of this
700 * function. Once we are done here, the next state is set as needed by
701 * the parent
702 */
703 if (!done) {
704 trace_xfs_attr_rmtval_remove_return(attr->xattri_dela_state,
705 args->dp);
706 return -EAGAIN;
707 }
708
709 args->rmtblkno = 0;
710 args->rmtblkcnt = 0;
711 return 0;
712}
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 */
51int
52xfs_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 */
68static bool
69xfs_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
91static bool
92xfs_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
119static void
120xfs_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
158static void
159xfs_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
204const 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
210STATIC int
211xfs_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 */
249STATIC int
250xfs_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
297STATIC void
298xfs_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 */
350int
351xfs_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 */
415int
416xfs_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 */
547int
548xfs_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}