1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.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"
22 #include "xfs_attr_remote.h"
23 #include "xfs_trace.h"
24 #include "xfs_error.h"
25 #include "xfs_health.h"
27 #define ATTR_RMTVALUE_MAPSIZE 1 /* # of map entries at once */
30 * Remote Attribute Values
31 * =======================
33 * Remote extended attribute values are conceptually simple -- they're written
34 * to data blocks mapped by an inode's attribute fork, and they have an upper
35 * size limit of 64k. Setting a value does not involve the XFS log.
37 * However, on a v5 filesystem, maximally sized remote attr values require one
38 * block more than 64k worth of space to hold both the remote attribute value
39 * header (64 bytes). On a 4k block filesystem this results in a 68k buffer;
40 * on a 64k block filesystem, this would be a 128k buffer. Note that the log
41 * format can only handle a dirty buffer of XFS_MAX_BLOCKSIZE length (64k).
42 * Therefore, we /must/ ensure that remote attribute value buffers never touch
43 * the logging system and therefore never have a log item.
46 /* How many bytes can be stored in a remote value buffer? */
48 xfs_attr3_rmt_buf_space(
51 unsigned int blocksize
= mp
->m_attr_geo
->blksize
;
54 return blocksize
- sizeof(struct xfs_attr3_rmt_hdr
);
59 /* Compute number of fsblocks needed to store a remote attr value */
66 * Each contiguous block has a header, so it is not just a simple
67 * attribute length to FSB conversion.
70 return howmany(attrlen
, xfs_attr3_rmt_buf_space(mp
));
72 return XFS_B_TO_FSB(mp
, attrlen
);
76 * Checking of the remote attribute header is split into two parts. The verifier
77 * does CRC, location and bounds checking, the unpacking function checks the
78 * attribute parameters and owner.
88 struct xfs_attr3_rmt_hdr
*rmt
= ptr
;
90 if (bno
!= be64_to_cpu(rmt
->rm_blkno
))
91 return __this_address
;
92 if (offset
!= be32_to_cpu(rmt
->rm_offset
))
93 return __this_address
;
94 if (size
!= be32_to_cpu(rmt
->rm_bytes
))
95 return __this_address
;
96 if (ino
!= be64_to_cpu(rmt
->rm_owner
))
97 return __this_address
;
103 static xfs_failaddr_t
104 xfs_attr3_rmt_verify(
105 struct xfs_mount
*mp
,
110 struct xfs_attr3_rmt_hdr
*rmt
= ptr
;
112 if (!xfs_verify_magic(bp
, rmt
->rm_magic
))
113 return __this_address
;
114 if (!uuid_equal(&rmt
->rm_uuid
, &mp
->m_sb
.sb_meta_uuid
))
115 return __this_address
;
116 if (be64_to_cpu(rmt
->rm_blkno
) != bno
)
117 return __this_address
;
118 if (be32_to_cpu(rmt
->rm_bytes
) > mp
->m_attr_geo
->blksize
- sizeof(*rmt
))
119 return __this_address
;
120 if (be32_to_cpu(rmt
->rm_offset
) +
121 be32_to_cpu(rmt
->rm_bytes
) > XFS_XATTR_SIZE_MAX
)
122 return __this_address
;
123 if (rmt
->rm_owner
== 0)
124 return __this_address
;
130 __xfs_attr3_rmt_read_verify(
133 xfs_failaddr_t
*failaddr
)
135 struct xfs_mount
*mp
= bp
->b_mount
;
139 unsigned int blksize
= mp
->m_attr_geo
->blksize
;
141 /* no verification of non-crc buffers */
142 if (!xfs_has_crc(mp
))
146 bno
= xfs_buf_daddr(bp
);
147 len
= BBTOB(bp
->b_length
);
148 ASSERT(len
>= blksize
);
152 !xfs_verify_cksum(ptr
, blksize
, XFS_ATTR3_RMT_CRC_OFF
)) {
153 *failaddr
= __this_address
;
156 *failaddr
= xfs_attr3_rmt_verify(mp
, bp
, ptr
, bno
);
158 return -EFSCORRUPTED
;
161 bno
+= BTOBB(blksize
);
165 *failaddr
= __this_address
;
166 return -EFSCORRUPTED
;
173 xfs_attr3_rmt_read_verify(
179 error
= __xfs_attr3_rmt_read_verify(bp
, true, &fa
);
181 xfs_verifier_error(bp
, error
, fa
);
184 static xfs_failaddr_t
185 xfs_attr3_rmt_verify_struct(
191 error
= __xfs_attr3_rmt_read_verify(bp
, false, &fa
);
192 return error
? fa
: NULL
;
196 xfs_attr3_rmt_write_verify(
199 struct xfs_mount
*mp
= bp
->b_mount
;
201 unsigned int blksize
= mp
->m_attr_geo
->blksize
;
206 /* no verification of non-crc buffers */
207 if (!xfs_has_crc(mp
))
211 bno
= xfs_buf_daddr(bp
);
212 len
= BBTOB(bp
->b_length
);
213 ASSERT(len
>= blksize
);
216 struct xfs_attr3_rmt_hdr
*rmt
= (struct xfs_attr3_rmt_hdr
*)ptr
;
218 fa
= xfs_attr3_rmt_verify(mp
, bp
, ptr
, bno
);
220 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
225 * Ensure we aren't writing bogus LSNs to disk. See
226 * xfs_attr3_rmt_hdr_set() for the explanation.
228 if (rmt
->rm_lsn
!= cpu_to_be64(NULLCOMMITLSN
)) {
229 xfs_verifier_error(bp
, -EFSCORRUPTED
, __this_address
);
232 xfs_update_cksum(ptr
, blksize
, XFS_ATTR3_RMT_CRC_OFF
);
236 bno
+= BTOBB(blksize
);
240 xfs_verifier_error(bp
, -EFSCORRUPTED
, __this_address
);
243 const struct xfs_buf_ops xfs_attr3_rmt_buf_ops
= {
244 .name
= "xfs_attr3_rmt",
245 .magic
= { 0, cpu_to_be32(XFS_ATTR3_RMT_MAGIC
) },
246 .verify_read
= xfs_attr3_rmt_read_verify
,
247 .verify_write
= xfs_attr3_rmt_write_verify
,
248 .verify_struct
= xfs_attr3_rmt_verify_struct
,
252 xfs_attr3_rmt_hdr_set(
253 struct xfs_mount
*mp
,
260 struct xfs_attr3_rmt_hdr
*rmt
= ptr
;
262 if (!xfs_has_crc(mp
))
265 rmt
->rm_magic
= cpu_to_be32(XFS_ATTR3_RMT_MAGIC
);
266 rmt
->rm_offset
= cpu_to_be32(offset
);
267 rmt
->rm_bytes
= cpu_to_be32(size
);
268 uuid_copy(&rmt
->rm_uuid
, &mp
->m_sb
.sb_meta_uuid
);
269 rmt
->rm_owner
= cpu_to_be64(ino
);
270 rmt
->rm_blkno
= cpu_to_be64(bno
);
273 * Remote attribute blocks are written synchronously, so we don't
274 * have an LSN that we can stamp in them that makes any sense to log
275 * recovery. To ensure that log recovery handles overwrites of these
276 * blocks sanely (i.e. once they've been freed and reallocated as some
277 * other type of metadata) we need to ensure that the LSN has a value
278 * that tells log recovery to ignore the LSN and overwrite the buffer
279 * with whatever is in it's log. To do this, we use the magic
280 * NULLCOMMITLSN to indicate that the LSN is invalid.
282 rmt
->rm_lsn
= cpu_to_be64(NULLCOMMITLSN
);
284 return sizeof(struct xfs_attr3_rmt_hdr
);
288 * Helper functions to copy attribute data in and out of the one disk extents
291 xfs_attr_rmtval_copyout(
292 struct xfs_mount
*mp
,
294 struct xfs_inode
*dp
,
296 unsigned int *offset
,
297 unsigned int *valuelen
,
300 char *src
= bp
->b_addr
;
301 xfs_daddr_t bno
= xfs_buf_daddr(bp
);
302 unsigned int len
= BBTOB(bp
->b_length
);
303 unsigned int blksize
= mp
->m_attr_geo
->blksize
;
305 ASSERT(len
>= blksize
);
307 while (len
> 0 && *valuelen
> 0) {
308 unsigned int hdr_size
= 0;
309 unsigned int byte_cnt
= xfs_attr3_rmt_buf_space(mp
);
311 byte_cnt
= min(*valuelen
, byte_cnt
);
313 if (xfs_has_crc(mp
)) {
314 if (xfs_attr3_rmt_hdr_ok(src
, owner
, *offset
,
317 "remote attribute header mismatch bno/off/len/owner (0x%llx/0x%x/Ox%x/0x%llx)",
318 bno
, *offset
, byte_cnt
, owner
);
319 xfs_dirattr_mark_sick(dp
, XFS_ATTR_FORK
);
320 return -EFSCORRUPTED
;
322 hdr_size
= sizeof(struct xfs_attr3_rmt_hdr
);
325 memcpy(*dst
, src
+ hdr_size
, byte_cnt
);
327 /* roll buffer forwards */
330 bno
+= BTOBB(blksize
);
332 /* roll attribute data forwards */
333 *valuelen
-= byte_cnt
;
341 xfs_attr_rmtval_copyin(
342 struct xfs_mount
*mp
,
345 unsigned int *offset
,
346 unsigned int *valuelen
,
349 char *dst
= bp
->b_addr
;
350 xfs_daddr_t bno
= xfs_buf_daddr(bp
);
351 unsigned int len
= BBTOB(bp
->b_length
);
352 unsigned int blksize
= mp
->m_attr_geo
->blksize
;
354 ASSERT(len
>= blksize
);
356 while (len
> 0 && *valuelen
> 0) {
357 unsigned int hdr_size
;
358 unsigned int byte_cnt
= xfs_attr3_rmt_buf_space(mp
);
360 byte_cnt
= min(*valuelen
, byte_cnt
);
361 hdr_size
= xfs_attr3_rmt_hdr_set(mp
, dst
, ino
, *offset
,
364 memcpy(dst
+ hdr_size
, *src
, byte_cnt
);
367 * If this is the last block, zero the remainder of it.
368 * Check that we are actually the last block, too.
370 if (byte_cnt
+ hdr_size
< blksize
) {
371 ASSERT(*valuelen
- byte_cnt
== 0);
372 ASSERT(len
== blksize
);
373 memset(dst
+ hdr_size
+ byte_cnt
, 0,
374 blksize
- hdr_size
- byte_cnt
);
377 /* roll buffer forwards */
380 bno
+= BTOBB(blksize
);
382 /* roll attribute data forwards */
383 *valuelen
-= byte_cnt
;
390 * Read the value associated with an attribute from the out-of-line buffer
391 * that we stored it in.
393 * Returns 0 on successful retrieval, otherwise an error.
397 struct xfs_da_args
*args
)
399 struct xfs_bmbt_irec map
[ATTR_RMTVALUE_MAPSIZE
];
400 struct xfs_mount
*mp
= args
->dp
->i_mount
;
402 xfs_dablk_t lblkno
= args
->rmtblkno
;
403 uint8_t *dst
= args
->value
;
404 unsigned int valuelen
;
407 unsigned int blkcnt
= args
->rmtblkcnt
;
409 unsigned int offset
= 0;
411 trace_xfs_attr_rmtval_get(args
);
413 ASSERT(args
->valuelen
!= 0);
414 ASSERT(args
->rmtvaluelen
== args
->valuelen
);
416 valuelen
= args
->rmtvaluelen
;
417 while (valuelen
> 0) {
418 nmap
= ATTR_RMTVALUE_MAPSIZE
;
419 error
= xfs_bmapi_read(args
->dp
, (xfs_fileoff_t
)lblkno
,
426 for (i
= 0; (i
< nmap
) && (valuelen
> 0); i
++) {
430 ASSERT((map
[i
].br_startblock
!= DELAYSTARTBLOCK
) &&
431 (map
[i
].br_startblock
!= HOLESTARTBLOCK
));
432 dblkno
= XFS_FSB_TO_DADDR(mp
, map
[i
].br_startblock
);
433 dblkcnt
= XFS_FSB_TO_BB(mp
, map
[i
].br_blockcount
);
434 error
= xfs_buf_read(mp
->m_ddev_targp
, dblkno
, dblkcnt
,
435 0, &bp
, &xfs_attr3_rmt_buf_ops
);
436 if (xfs_metadata_is_sick(error
))
437 xfs_dirattr_mark_sick(args
->dp
, XFS_ATTR_FORK
);
441 error
= xfs_attr_rmtval_copyout(mp
, bp
, args
->dp
,
442 args
->owner
, &offset
, &valuelen
, &dst
);
447 /* roll attribute extent map forwards */
448 lblkno
+= map
[i
].br_blockcount
;
449 blkcnt
-= map
[i
].br_blockcount
;
452 ASSERT(valuelen
== 0);
457 * Find a "hole" in the attribute address space large enough for us to drop the
458 * new attributes value into
461 xfs_attr_rmt_find_hole(
462 struct xfs_da_args
*args
)
464 struct xfs_inode
*dp
= args
->dp
;
465 struct xfs_mount
*mp
= dp
->i_mount
;
468 xfs_fileoff_t lfileoff
= 0;
471 * Because CRC enable attributes have headers, we can't just do a
472 * straight byte to FSB conversion and have to take the header space
475 blkcnt
= xfs_attr3_rmt_blocks(mp
, args
->rmtvaluelen
);
476 error
= xfs_bmap_first_unused(args
->trans
, args
->dp
, blkcnt
, &lfileoff
,
481 args
->rmtblkno
= (xfs_dablk_t
)lfileoff
;
482 args
->rmtblkcnt
= blkcnt
;
488 xfs_attr_rmtval_set_value(
489 struct xfs_da_args
*args
)
491 struct xfs_inode
*dp
= args
->dp
;
492 struct xfs_mount
*mp
= dp
->i_mount
;
493 struct xfs_bmbt_irec map
;
495 uint8_t *src
= args
->value
;
497 unsigned int valuelen
;
500 unsigned int offset
= 0;
503 * Roll through the "value", copying the attribute value to the
504 * already-allocated blocks. Blocks are written synchronously
505 * so that we can know they are all on disk before we turn off
506 * the INCOMPLETE flag.
508 lblkno
= args
->rmtblkno
;
509 blkcnt
= args
->rmtblkcnt
;
510 valuelen
= args
->rmtvaluelen
;
511 while (valuelen
> 0) {
519 error
= xfs_bmapi_read(dp
, (xfs_fileoff_t
)lblkno
,
525 ASSERT((map
.br_startblock
!= DELAYSTARTBLOCK
) &&
526 (map
.br_startblock
!= HOLESTARTBLOCK
));
528 dblkno
= XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
529 dblkcnt
= XFS_FSB_TO_BB(mp
, map
.br_blockcount
);
531 error
= xfs_buf_get(mp
->m_ddev_targp
, dblkno
, dblkcnt
, &bp
);
534 bp
->b_ops
= &xfs_attr3_rmt_buf_ops
;
536 xfs_attr_rmtval_copyin(mp
, bp
, args
->owner
, &offset
, &valuelen
,
539 error
= xfs_bwrite(bp
); /* GROT: NOTE: synchronous write */
545 /* roll attribute extent map forwards */
546 lblkno
+= map
.br_blockcount
;
547 blkcnt
-= map
.br_blockcount
;
549 ASSERT(valuelen
== 0);
553 /* Mark stale any incore buffers for the remote value. */
555 xfs_attr_rmtval_stale(
556 struct xfs_inode
*ip
,
557 struct xfs_bmbt_irec
*map
,
558 xfs_buf_flags_t incore_flags
)
560 struct xfs_mount
*mp
= ip
->i_mount
;
564 xfs_assert_ilocked(ip
, XFS_ILOCK_EXCL
);
566 if (XFS_IS_CORRUPT(mp
, map
->br_startblock
== DELAYSTARTBLOCK
) ||
567 XFS_IS_CORRUPT(mp
, map
->br_startblock
== HOLESTARTBLOCK
)) {
568 xfs_bmap_mark_sick(ip
, XFS_ATTR_FORK
);
569 return -EFSCORRUPTED
;
572 error
= xfs_buf_incore(mp
->m_ddev_targp
,
573 XFS_FSB_TO_DADDR(mp
, map
->br_startblock
),
574 XFS_FSB_TO_BB(mp
, map
->br_blockcount
),
577 if (error
== -ENOENT
)
588 * Find a hole for the attr and store it in the delayed attr context. This
589 * initializes the context to roll through allocating an attr extent for a
590 * delayed attr operation
593 xfs_attr_rmtval_find_space(
594 struct xfs_attr_intent
*attr
)
596 struct xfs_da_args
*args
= attr
->xattri_da_args
;
597 struct xfs_bmbt_irec
*map
= &attr
->xattri_map
;
600 attr
->xattri_lblkno
= 0;
601 attr
->xattri_blkcnt
= 0;
604 memset(map
, 0, sizeof(struct xfs_bmbt_irec
));
606 error
= xfs_attr_rmt_find_hole(args
);
610 attr
->xattri_blkcnt
= args
->rmtblkcnt
;
611 attr
->xattri_lblkno
= args
->rmtblkno
;
617 * Write one block of the value associated with an attribute into the
618 * out-of-line buffer that we have defined for it. This is similar to a subset
619 * of xfs_attr_rmtval_set, but records the current block to the delayed attr
620 * context, and leaves transaction handling to the caller.
623 xfs_attr_rmtval_set_blk(
624 struct xfs_attr_intent
*attr
)
626 struct xfs_da_args
*args
= attr
->xattri_da_args
;
627 struct xfs_inode
*dp
= args
->dp
;
628 struct xfs_bmbt_irec
*map
= &attr
->xattri_map
;
633 error
= xfs_bmapi_write(args
->trans
, dp
,
634 (xfs_fileoff_t
)attr
->xattri_lblkno
,
635 attr
->xattri_blkcnt
, XFS_BMAPI_ATTRFORK
, args
->total
,
640 ASSERT((map
->br_startblock
!= DELAYSTARTBLOCK
) &&
641 (map
->br_startblock
!= HOLESTARTBLOCK
));
643 /* roll attribute extent map forwards */
644 attr
->xattri_lblkno
+= map
->br_blockcount
;
645 attr
->xattri_blkcnt
-= map
->br_blockcount
;
651 * Remove the value associated with an attribute by deleting the
652 * out-of-line buffer that it is stored on.
655 xfs_attr_rmtval_invalidate(
656 struct xfs_da_args
*args
)
663 * Roll through the "value", invalidating the attribute value's blocks.
665 lblkno
= args
->rmtblkno
;
666 blkcnt
= args
->rmtblkcnt
;
668 struct xfs_bmbt_irec map
;
672 * Try to remember where we decided to put the value.
675 error
= xfs_bmapi_read(args
->dp
, (xfs_fileoff_t
)lblkno
,
676 blkcnt
, &map
, &nmap
, XFS_BMAPI_ATTRFORK
);
679 if (XFS_IS_CORRUPT(args
->dp
->i_mount
, nmap
!= 1)) {
680 xfs_bmap_mark_sick(args
->dp
, XFS_ATTR_FORK
);
681 return -EFSCORRUPTED
;
683 error
= xfs_attr_rmtval_stale(args
->dp
, &map
, XBF_TRYLOCK
);
687 lblkno
+= map
.br_blockcount
;
688 blkcnt
-= map
.br_blockcount
;
694 * Remove the value associated with an attribute by deleting the out-of-line
695 * buffer that it is stored on. Returns -EAGAIN for the caller to refresh the
696 * transaction and re-call the function. Callers should keep calling this
697 * routine until it returns something other than -EAGAIN.
700 xfs_attr_rmtval_remove(
701 struct xfs_attr_intent
*attr
)
703 struct xfs_da_args
*args
= attr
->xattri_da_args
;
707 * Unmap value blocks for this attr.
709 error
= xfs_bunmapi(args
->trans
, args
->dp
, args
->rmtblkno
,
710 args
->rmtblkcnt
, XFS_BMAPI_ATTRFORK
, 1, &done
);
715 * We don't need an explicit state here to pick up where we left off. We
716 * can figure it out using the !done return code. The actual value of
717 * attr->xattri_dela_state may be some value reminiscent of the calling
718 * function, but it's value is irrelevant with in the context of this
719 * function. Once we are done here, the next state is set as needed by
723 trace_xfs_attr_rmtval_remove_return(attr
->xattri_dela_state
,