2 * This file and its contents are supplied under the terms of the
3 * Common Development and Distribution License ("CDDL"), version 1.0.
4 * You may only use this file in accordance with the terms of version
7 * A full copy of the text of the CDDL should have accompanied this
8 * source. A copy of the CDDL is also available via the Internet at
9 * http://www.illumos.org/license/CDDL.
13 * Copyright (c) 2014 by Chunwei Chen. All rights reserved.
14 * Copyright (c) 2016 by Delphix. All rights reserved.
18 * ARC buffer data (ABD).
20 * ABDs are an abstract data structure for the ARC which can use two
21 * different ways of storing the underlying data:
23 * (a) Linear buffer. In this case, all the data in the ABD is stored in one
24 * contiguous buffer in memory (from a zio_[data_]buf_* kmem cache).
26 * +-------------------+
29 * | abd_size = ... | +--------------------------------+
30 * | abd_buf ------------->| raw buffer of size abd_size |
31 * +-------------------+ +--------------------------------+
34 * (b) Scattered buffer. In this case, the data in the ABD is split into
35 * equal-sized chunks (from the abd_chunk_cache kmem_cache), with pointers
36 * to the chunks recorded in an array at the end of the ABD structure.
38 * +-------------------+
42 * | abd_offset = 0 | +-----------+
43 * | abd_chunks[0] ----------------------------->| chunk 0 |
44 * | abd_chunks[1] ---------------------+ +-----------+
45 * | ... | | +-----------+
46 * | abd_chunks[N-1] ---------+ +------->| chunk 1 |
47 * +-------------------+ | +-----------+
50 * +----------------->| chunk N-1 |
53 * Using a large proportion of scattered ABDs decreases ARC fragmentation since
54 * when we are at the limit of allocatable space, using equal-size chunks will
55 * allow us to quickly reclaim enough space for a new large allocation (assuming
56 * it is also scattered).
58 * In addition to directly allocating a linear or scattered ABD, it is also
59 * possible to create an ABD by requesting the "sub-ABD" starting at an offset
60 * within an existing ABD. In linear buffers this is simple (set abd_buf of
61 * the new ABD to the starting point within the original raw buffer), but
62 * scattered ABDs are a little more complex. The new ABD makes a copy of the
63 * relevant abd_chunks pointers (but not the underlying data). However, to
64 * provide arbitrary rather than only chunk-aligned starting offsets, it also
65 * tracks an abd_offset field which represents the starting point of the data
66 * within the first chunk in abd_chunks. For both linear and scattered ABDs,
67 * creating an offset ABD marks the original ABD as the offset's parent, and the
68 * original ABD's abd_children refcount is incremented. This data allows us to
69 * ensure the root ABD isn't deleted before its children.
71 * Most consumers should never need to know what type of ABD they're using --
72 * the ABD public API ensures that it's possible to transparently switch from
73 * using a linear ABD to a scattered one when doing so would be beneficial.
75 * If you need to use the data within an ABD directly, if you know it's linear
76 * (because you allocated it) you can use abd_to_buf() to access the underlying
77 * raw buffer. Otherwise, you should use one of the abd_borrow_buf* functions
78 * which will allocate a raw buffer if necessary. Use the abd_return_buf*
79 * functions to return any raw buffers that are no longer necessary when you're
82 * There are a variety of ABD APIs that implement basic buffer operations:
83 * compare, copy, read, write, and fill with zeroes. If you need a custom
84 * function which progressively accesses the whole ABD, use the abd_iterate_*
89 #include <sys/param.h>
91 #include <sys/zfs_context.h>
92 #include <sys/zfs_znode.h>
94 typedef struct abd_stats
{
95 kstat_named_t abdstat_struct_size
;
96 kstat_named_t abdstat_scatter_cnt
;
97 kstat_named_t abdstat_scatter_data_size
;
98 kstat_named_t abdstat_scatter_chunk_waste
;
99 kstat_named_t abdstat_linear_cnt
;
100 kstat_named_t abdstat_linear_data_size
;
103 static abd_stats_t abd_stats
= {
104 /* Amount of memory occupied by all of the abd_t struct allocations */
105 { "struct_size", KSTAT_DATA_UINT64
},
107 * The number of scatter ABDs which are currently allocated, excluding
108 * ABDs which don't own their data (for instance the ones which were
109 * allocated through abd_get_offset()).
111 { "scatter_cnt", KSTAT_DATA_UINT64
},
112 /* Amount of data stored in all scatter ABDs tracked by scatter_cnt */
113 { "scatter_data_size", KSTAT_DATA_UINT64
},
115 * The amount of space wasted at the end of the last chunk across all
116 * scatter ABDs tracked by scatter_cnt.
118 { "scatter_chunk_waste", KSTAT_DATA_UINT64
},
120 * The number of linear ABDs which are currently allocated, excluding
121 * ABDs which don't own their data (for instance the ones which were
122 * allocated through abd_get_offset() and abd_get_from_buf()). If an
123 * ABD takes ownership of its buf then it will become tracked.
125 { "linear_cnt", KSTAT_DATA_UINT64
},
126 /* Amount of data stored in all linear ABDs tracked by linear_cnt */
127 { "linear_data_size", KSTAT_DATA_UINT64
},
130 #define ABDSTAT(stat) (abd_stats.stat.value.ui64)
131 #define ABDSTAT_INCR(stat, val) \
132 atomic_add_64(&abd_stats.stat.value.ui64, (val))
133 #define ABDSTAT_BUMP(stat) ABDSTAT_INCR(stat, 1)
134 #define ABDSTAT_BUMPDOWN(stat) ABDSTAT_INCR(stat, -1)
137 * It is possible to make all future ABDs be linear by setting this to B_FALSE.
138 * Otherwise, ABDs are allocated scattered by default unless the caller uses
139 * abd_alloc_linear().
141 boolean_t zfs_abd_scatter_enabled
= B_TRUE
;
144 * The size of the chunks ABD allocates. Because the sizes allocated from the
145 * kmem_cache can't change, this tunable can only be modified at boot. Changing
146 * it at runtime would cause ABD iteration to work incorrectly for ABDs which
147 * were allocated with the old size, so a safeguard has been put in place which
148 * will cause the machine to panic if you change it and try to access the data
149 * within a scattered ABD.
151 size_t zfs_abd_chunk_size
= 4096;
154 extern vmem_t
*zio_alloc_arena
;
157 kmem_cache_t
*abd_chunk_cache
;
158 static kstat_t
*abd_ksp
;
160 extern inline boolean_t
abd_is_linear(abd_t
*abd
);
161 extern inline void abd_copy(abd_t
*dabd
, abd_t
*sabd
, size_t size
);
162 extern inline void abd_copy_from_buf(abd_t
*abd
, const void *buf
, size_t size
);
163 extern inline void abd_copy_to_buf(void* buf
, abd_t
*abd
, size_t size
);
164 extern inline int abd_cmp_buf(abd_t
*abd
, const void *buf
, size_t size
);
165 extern inline void abd_zero(abd_t
*abd
, size_t size
);
170 void *c
= kmem_cache_alloc(abd_chunk_cache
, KM_PUSHPAGE
);
171 ASSERT3P(c
, !=, NULL
);
176 abd_free_chunk(void *c
)
178 kmem_cache_free(abd_chunk_cache
, c
);
184 vmem_t
*data_alloc_arena
= NULL
;
187 data_alloc_arena
= zio_alloc_arena
;
191 * Since ABD chunks do not appear in crash dumps, we pass KMC_NOTOUCH
192 * so that no allocator metadata is stored with the buffers.
194 abd_chunk_cache
= kmem_cache_create("abd_chunk", zfs_abd_chunk_size
, 0,
195 NULL
, NULL
, NULL
, NULL
, data_alloc_arena
, KMC_NOTOUCH
);
197 abd_ksp
= kstat_create("zfs", 0, "abdstats", "misc", KSTAT_TYPE_NAMED
,
198 sizeof (abd_stats
) / sizeof (kstat_named_t
), KSTAT_FLAG_VIRTUAL
);
199 if (abd_ksp
!= NULL
) {
200 abd_ksp
->ks_data
= &abd_stats
;
201 kstat_install(abd_ksp
);
208 if (abd_ksp
!= NULL
) {
209 kstat_delete(abd_ksp
);
213 kmem_cache_destroy(abd_chunk_cache
);
214 abd_chunk_cache
= NULL
;
218 abd_chunkcnt_for_bytes(size_t size
)
220 return (P2ROUNDUP(size
, zfs_abd_chunk_size
) / zfs_abd_chunk_size
);
224 abd_scatter_chunkcnt(abd_t
*abd
)
226 ASSERT(!abd_is_linear(abd
));
227 return (abd_chunkcnt_for_bytes(
228 abd
->abd_u
.abd_scatter
.abd_offset
+ abd
->abd_size
));
232 abd_verify(abd_t
*abd
)
234 ASSERT3U(abd
->abd_size
, >, 0);
235 ASSERT3U(abd
->abd_size
, <=, SPA_MAXBLOCKSIZE
);
236 ASSERT3U(abd
->abd_flags
, ==, abd
->abd_flags
& (ABD_FLAG_LINEAR
|
237 ABD_FLAG_OWNER
| ABD_FLAG_META
));
238 IMPLY(abd
->abd_parent
!= NULL
, !(abd
->abd_flags
& ABD_FLAG_OWNER
));
239 IMPLY(abd
->abd_flags
& ABD_FLAG_META
, abd
->abd_flags
& ABD_FLAG_OWNER
);
240 if (abd_is_linear(abd
)) {
241 ASSERT3P(abd
->abd_u
.abd_linear
.abd_buf
, !=, NULL
);
243 ASSERT3U(abd
->abd_u
.abd_scatter
.abd_offset
, <,
245 size_t n
= abd_scatter_chunkcnt(abd
);
246 for (int i
= 0; i
< n
; i
++) {
248 abd
->abd_u
.abd_scatter
.abd_chunks
[i
], !=, NULL
);
253 static inline abd_t
*
254 abd_alloc_struct(size_t chunkcnt
)
256 size_t size
= offsetof(abd_t
, abd_u
.abd_scatter
.abd_chunks
[chunkcnt
]);
257 abd_t
*abd
= kmem_alloc(size
, KM_PUSHPAGE
);
258 ASSERT3P(abd
, !=, NULL
);
259 ABDSTAT_INCR(abdstat_struct_size
, size
);
265 abd_free_struct(abd_t
*abd
)
267 size_t chunkcnt
= abd_is_linear(abd
) ? 0 : abd_scatter_chunkcnt(abd
);
268 int size
= offsetof(abd_t
, abd_u
.abd_scatter
.abd_chunks
[chunkcnt
]);
269 kmem_free(abd
, size
);
270 ABDSTAT_INCR(abdstat_struct_size
, -size
);
274 * Allocate an ABD, along with its own underlying data buffers. Use this if you
275 * don't care whether the ABD is linear or not.
278 abd_alloc(size_t size
, boolean_t is_metadata
)
280 if (!zfs_abd_scatter_enabled
)
281 return (abd_alloc_linear(size
, is_metadata
));
283 VERIFY3U(size
, <=, SPA_MAXBLOCKSIZE
);
285 size_t n
= abd_chunkcnt_for_bytes(size
);
286 abd_t
*abd
= abd_alloc_struct(n
);
288 abd
->abd_flags
= ABD_FLAG_OWNER
;
290 abd
->abd_flags
|= ABD_FLAG_META
;
292 abd
->abd_size
= size
;
293 abd
->abd_parent
= NULL
;
294 refcount_create(&abd
->abd_children
);
296 abd
->abd_u
.abd_scatter
.abd_offset
= 0;
297 abd
->abd_u
.abd_scatter
.abd_chunk_size
= zfs_abd_chunk_size
;
299 for (int i
= 0; i
< n
; i
++) {
300 void *c
= abd_alloc_chunk();
301 ASSERT3P(c
, !=, NULL
);
302 abd
->abd_u
.abd_scatter
.abd_chunks
[i
] = c
;
305 ABDSTAT_BUMP(abdstat_scatter_cnt
);
306 ABDSTAT_INCR(abdstat_scatter_data_size
, size
);
307 ABDSTAT_INCR(abdstat_scatter_chunk_waste
,
308 n
* zfs_abd_chunk_size
- size
);
314 abd_free_scatter(abd_t
*abd
)
316 size_t n
= abd_scatter_chunkcnt(abd
);
317 for (int i
= 0; i
< n
; i
++) {
318 abd_free_chunk(abd
->abd_u
.abd_scatter
.abd_chunks
[i
]);
321 refcount_destroy(&abd
->abd_children
);
322 ABDSTAT_BUMPDOWN(abdstat_scatter_cnt
);
323 ABDSTAT_INCR(abdstat_scatter_data_size
, -(int)abd
->abd_size
);
324 ABDSTAT_INCR(abdstat_scatter_chunk_waste
,
325 abd
->abd_size
- n
* zfs_abd_chunk_size
);
327 abd_free_struct(abd
);
331 * Allocate an ABD that must be linear, along with its own underlying data
332 * buffer. Only use this when it would be very annoying to write your ABD
333 * consumer with a scattered ABD.
336 abd_alloc_linear(size_t size
, boolean_t is_metadata
)
338 abd_t
*abd
= abd_alloc_struct(0);
340 VERIFY3U(size
, <=, SPA_MAXBLOCKSIZE
);
342 abd
->abd_flags
= ABD_FLAG_LINEAR
| ABD_FLAG_OWNER
;
344 abd
->abd_flags
|= ABD_FLAG_META
;
346 abd
->abd_size
= size
;
347 abd
->abd_parent
= NULL
;
348 refcount_create(&abd
->abd_children
);
351 abd
->abd_u
.abd_linear
.abd_buf
= zio_buf_alloc(size
);
353 abd
->abd_u
.abd_linear
.abd_buf
= zio_data_buf_alloc(size
);
356 ABDSTAT_BUMP(abdstat_linear_cnt
);
357 ABDSTAT_INCR(abdstat_linear_data_size
, size
);
363 abd_free_linear(abd_t
*abd
)
365 if (abd
->abd_flags
& ABD_FLAG_META
) {
366 zio_buf_free(abd
->abd_u
.abd_linear
.abd_buf
, abd
->abd_size
);
368 zio_data_buf_free(abd
->abd_u
.abd_linear
.abd_buf
, abd
->abd_size
);
371 refcount_destroy(&abd
->abd_children
);
372 ABDSTAT_BUMPDOWN(abdstat_linear_cnt
);
373 ABDSTAT_INCR(abdstat_linear_data_size
, -(int)abd
->abd_size
);
375 abd_free_struct(abd
);
379 * Free an ABD. Only use this on ABDs allocated with abd_alloc() or
380 * abd_alloc_linear().
386 ASSERT3P(abd
->abd_parent
, ==, NULL
);
387 ASSERT(abd
->abd_flags
& ABD_FLAG_OWNER
);
388 if (abd_is_linear(abd
))
389 abd_free_linear(abd
);
391 abd_free_scatter(abd
);
395 * Allocate an ABD of the same format (same metadata flag, same scatterize
396 * setting) as another ABD.
399 abd_alloc_sametype(abd_t
*sabd
, size_t size
)
401 boolean_t is_metadata
= (sabd
->abd_flags
& ABD_FLAG_META
) != 0;
402 if (abd_is_linear(sabd
)) {
403 return (abd_alloc_linear(size
, is_metadata
));
405 return (abd_alloc(size
, is_metadata
));
410 * If we're going to use this ABD for doing I/O using the block layer, the
411 * consumer of the ABD data doesn't care if it's scattered or not, and we don't
412 * plan to store this ABD in memory for a long period of time, we should
413 * allocate the ABD type that requires the least data copying to do the I/O.
415 * Currently this is linear ABDs, however if ldi_strategy() can ever issue I/Os
416 * using a scatter/gather list we should switch to that and replace this call
417 * with vanilla abd_alloc().
420 abd_alloc_for_io(size_t size
, boolean_t is_metadata
)
422 return (abd_alloc_linear(size
, is_metadata
));
426 * Allocate a new ABD to point to offset off of sabd. It shares the underlying
427 * buffer data with sabd. Use abd_put() to free. sabd must not be freed while
428 * any derived ABDs exist.
431 abd_get_offset(abd_t
*sabd
, size_t off
)
436 ASSERT3U(off
, <=, sabd
->abd_size
);
438 if (abd_is_linear(sabd
)) {
439 abd
= abd_alloc_struct(0);
442 * Even if this buf is filesystem metadata, we only track that
443 * if we own the underlying data buffer, which is not true in
444 * this case. Therefore, we don't ever use ABD_FLAG_META here.
446 abd
->abd_flags
= ABD_FLAG_LINEAR
;
448 abd
->abd_u
.abd_linear
.abd_buf
=
449 (char *)sabd
->abd_u
.abd_linear
.abd_buf
+ off
;
451 size_t new_offset
= sabd
->abd_u
.abd_scatter
.abd_offset
+ off
;
452 size_t chunkcnt
= abd_scatter_chunkcnt(sabd
) -
453 (new_offset
/ zfs_abd_chunk_size
);
455 abd
= abd_alloc_struct(chunkcnt
);
458 * Even if this buf is filesystem metadata, we only track that
459 * if we own the underlying data buffer, which is not true in
460 * this case. Therefore, we don't ever use ABD_FLAG_META here.
464 abd
->abd_u
.abd_scatter
.abd_offset
=
465 new_offset
% zfs_abd_chunk_size
;
466 abd
->abd_u
.abd_scatter
.abd_chunk_size
= zfs_abd_chunk_size
;
468 /* Copy the scatterlist starting at the correct offset */
469 (void) memcpy(&abd
->abd_u
.abd_scatter
.abd_chunks
,
470 &sabd
->abd_u
.abd_scatter
.abd_chunks
[new_offset
/
472 chunkcnt
* sizeof (void *));
475 abd
->abd_size
= sabd
->abd_size
- off
;
476 abd
->abd_parent
= sabd
;
477 refcount_create(&abd
->abd_children
);
478 (void) refcount_add_many(&sabd
->abd_children
, abd
->abd_size
, abd
);
484 * Allocate a linear ABD structure for buf. You must free this with abd_put()
485 * since the resulting ABD doesn't own its own buffer.
488 abd_get_from_buf(void *buf
, size_t size
)
490 abd_t
*abd
= abd_alloc_struct(0);
492 VERIFY3U(size
, <=, SPA_MAXBLOCKSIZE
);
495 * Even if this buf is filesystem metadata, we only track that if we
496 * own the underlying data buffer, which is not true in this case.
497 * Therefore, we don't ever use ABD_FLAG_META here.
499 abd
->abd_flags
= ABD_FLAG_LINEAR
;
500 abd
->abd_size
= size
;
501 abd
->abd_parent
= NULL
;
502 refcount_create(&abd
->abd_children
);
504 abd
->abd_u
.abd_linear
.abd_buf
= buf
;
510 * Free an ABD allocated from abd_get_offset() or abd_get_from_buf(). Will not
511 * free the underlying scatterlist or buffer.
517 ASSERT(!(abd
->abd_flags
& ABD_FLAG_OWNER
));
519 if (abd
->abd_parent
!= NULL
) {
520 (void) refcount_remove_many(&abd
->abd_parent
->abd_children
,
524 refcount_destroy(&abd
->abd_children
);
525 abd_free_struct(abd
);
529 * Get the raw buffer associated with a linear ABD.
532 abd_to_buf(abd_t
*abd
)
534 ASSERT(abd_is_linear(abd
));
536 return (abd
->abd_u
.abd_linear
.abd_buf
);
540 * Borrow a raw buffer from an ABD without copying the contents of the ABD
541 * into the buffer. If the ABD is scattered, this will allocate a raw buffer
542 * whose contents are undefined. To copy over the existing data in the ABD, use
543 * abd_borrow_buf_copy() instead.
546 abd_borrow_buf(abd_t
*abd
, size_t n
)
550 ASSERT3U(abd
->abd_size
, >=, n
);
551 if (abd_is_linear(abd
)) {
552 buf
= abd_to_buf(abd
);
554 buf
= zio_buf_alloc(n
);
556 (void) refcount_add_many(&abd
->abd_children
, n
, buf
);
562 abd_borrow_buf_copy(abd_t
*abd
, size_t n
)
564 void *buf
= abd_borrow_buf(abd
, n
);
565 if (!abd_is_linear(abd
)) {
566 abd_copy_to_buf(buf
, abd
, n
);
572 * Return a borrowed raw buffer to an ABD. If the ABD is scattered, this will
573 * not change the contents of the ABD and will ASSERT that you didn't modify
574 * the buffer since it was borrowed. If you want any changes you made to buf to
575 * be copied back to abd, use abd_return_buf_copy() instead.
578 abd_return_buf(abd_t
*abd
, void *buf
, size_t n
)
581 ASSERT3U(abd
->abd_size
, >=, n
);
582 if (abd_is_linear(abd
)) {
583 ASSERT3P(buf
, ==, abd_to_buf(abd
));
585 ASSERT0(abd_cmp_buf(abd
, buf
, n
));
586 zio_buf_free(buf
, n
);
588 (void) refcount_remove_many(&abd
->abd_children
, n
, buf
);
592 abd_return_buf_copy(abd_t
*abd
, void *buf
, size_t n
)
594 if (!abd_is_linear(abd
)) {
595 abd_copy_from_buf(abd
, buf
, n
);
597 abd_return_buf(abd
, buf
, n
);
601 * Give this ABD ownership of the buffer that it's storing. Can only be used on
602 * linear ABDs which were allocated via abd_get_from_buf(), or ones allocated
603 * with abd_alloc_linear() which subsequently released ownership of their buf
604 * with abd_release_ownership_of_buf().
607 abd_take_ownership_of_buf(abd_t
*abd
, boolean_t is_metadata
)
609 ASSERT(abd_is_linear(abd
));
610 ASSERT(!(abd
->abd_flags
& ABD_FLAG_OWNER
));
613 abd
->abd_flags
|= ABD_FLAG_OWNER
;
615 abd
->abd_flags
|= ABD_FLAG_META
;
618 ABDSTAT_BUMP(abdstat_linear_cnt
);
619 ABDSTAT_INCR(abdstat_linear_data_size
, abd
->abd_size
);
623 abd_release_ownership_of_buf(abd_t
*abd
)
625 ASSERT(abd_is_linear(abd
));
626 ASSERT(abd
->abd_flags
& ABD_FLAG_OWNER
);
629 abd
->abd_flags
&= ~ABD_FLAG_OWNER
;
630 /* Disable this flag since we no longer own the data buffer */
631 abd
->abd_flags
&= ~ABD_FLAG_META
;
633 ABDSTAT_BUMPDOWN(abdstat_linear_cnt
);
634 ABDSTAT_INCR(abdstat_linear_data_size
, -(int)abd
->abd_size
);
638 abd_t
*iter_abd
; /* ABD being iterated through */
639 size_t iter_pos
; /* position (relative to abd_offset) */
640 void *iter_mapaddr
; /* addr corresponding to iter_pos */
641 size_t iter_mapsize
; /* length of data valid at mapaddr */
645 abd_iter_scatter_chunk_offset(struct abd_iter
*aiter
)
647 ASSERT(!abd_is_linear(aiter
->iter_abd
));
648 return ((aiter
->iter_abd
->abd_u
.abd_scatter
.abd_offset
+
649 aiter
->iter_pos
) % zfs_abd_chunk_size
);
653 abd_iter_scatter_chunk_index(struct abd_iter
*aiter
)
655 ASSERT(!abd_is_linear(aiter
->iter_abd
));
656 return ((aiter
->iter_abd
->abd_u
.abd_scatter
.abd_offset
+
657 aiter
->iter_pos
) / zfs_abd_chunk_size
);
661 * Initialize the abd_iter.
664 abd_iter_init(struct abd_iter
*aiter
, abd_t
*abd
)
667 aiter
->iter_abd
= abd
;
669 aiter
->iter_mapaddr
= NULL
;
670 aiter
->iter_mapsize
= 0;
674 * Advance the iterator by a certain amount. Cannot be called when a chunk is
675 * in use. This can be safely called when the aiter has already exhausted, in
676 * which case this does nothing.
679 abd_iter_advance(struct abd_iter
*aiter
, size_t amount
)
681 ASSERT3P(aiter
->iter_mapaddr
, ==, NULL
);
682 ASSERT0(aiter
->iter_mapsize
);
684 /* There's nothing left to advance to, so do nothing */
685 if (aiter
->iter_pos
== aiter
->iter_abd
->abd_size
)
688 aiter
->iter_pos
+= amount
;
692 * Map the current chunk into aiter. This can be safely called when the aiter
693 * has already exhausted, in which case this does nothing.
696 abd_iter_map(struct abd_iter
*aiter
)
701 ASSERT3P(aiter
->iter_mapaddr
, ==, NULL
);
702 ASSERT0(aiter
->iter_mapsize
);
704 /* Panic if someone has changed zfs_abd_chunk_size */
705 IMPLY(!abd_is_linear(aiter
->iter_abd
), zfs_abd_chunk_size
==
706 aiter
->iter_abd
->abd_u
.abd_scatter
.abd_chunk_size
);
708 /* There's nothing left to iterate over, so do nothing */
709 if (aiter
->iter_pos
== aiter
->iter_abd
->abd_size
)
712 if (abd_is_linear(aiter
->iter_abd
)) {
713 offset
= aiter
->iter_pos
;
714 aiter
->iter_mapsize
= aiter
->iter_abd
->abd_size
- offset
;
715 paddr
= aiter
->iter_abd
->abd_u
.abd_linear
.abd_buf
;
717 size_t index
= abd_iter_scatter_chunk_index(aiter
);
718 offset
= abd_iter_scatter_chunk_offset(aiter
);
719 aiter
->iter_mapsize
= zfs_abd_chunk_size
- offset
;
720 paddr
= aiter
->iter_abd
->abd_u
.abd_scatter
.abd_chunks
[index
];
722 aiter
->iter_mapaddr
= (char *)paddr
+ offset
;
726 * Unmap the current chunk from aiter. This can be safely called when the aiter
727 * has already exhausted, in which case this does nothing.
730 abd_iter_unmap(struct abd_iter
*aiter
)
732 /* There's nothing left to unmap, so do nothing */
733 if (aiter
->iter_pos
== aiter
->iter_abd
->abd_size
)
736 ASSERT3P(aiter
->iter_mapaddr
, !=, NULL
);
737 ASSERT3U(aiter
->iter_mapsize
, >, 0);
739 aiter
->iter_mapaddr
= NULL
;
740 aiter
->iter_mapsize
= 0;
744 abd_iterate_func(abd_t
*abd
, size_t off
, size_t size
,
745 abd_iter_func_t
*func
, void *private)
748 struct abd_iter aiter
;
751 ASSERT3U(off
+ size
, <=, abd
->abd_size
);
753 abd_iter_init(&aiter
, abd
);
754 abd_iter_advance(&aiter
, off
);
757 abd_iter_map(&aiter
);
759 size_t len
= MIN(aiter
.iter_mapsize
, size
);
762 ret
= func(aiter
.iter_mapaddr
, len
, private);
764 abd_iter_unmap(&aiter
);
770 abd_iter_advance(&aiter
, len
);
781 abd_copy_to_buf_off_cb(void *buf
, size_t size
, void *private)
783 struct buf_arg
*ba_ptr
= private;
785 (void) memcpy(ba_ptr
->arg_buf
, buf
, size
);
786 ba_ptr
->arg_buf
= (char *)ba_ptr
->arg_buf
+ size
;
792 * Copy abd to buf. (off is the offset in abd.)
795 abd_copy_to_buf_off(void *buf
, abd_t
*abd
, size_t off
, size_t size
)
797 struct buf_arg ba_ptr
= { buf
};
799 (void) abd_iterate_func(abd
, off
, size
, abd_copy_to_buf_off_cb
,
804 abd_cmp_buf_off_cb(void *buf
, size_t size
, void *private)
807 struct buf_arg
*ba_ptr
= private;
809 ret
= memcmp(buf
, ba_ptr
->arg_buf
, size
);
810 ba_ptr
->arg_buf
= (char *)ba_ptr
->arg_buf
+ size
;
816 * Compare the contents of abd to buf. (off is the offset in abd.)
819 abd_cmp_buf_off(abd_t
*abd
, const void *buf
, size_t off
, size_t size
)
821 struct buf_arg ba_ptr
= { (void *) buf
};
823 return (abd_iterate_func(abd
, off
, size
, abd_cmp_buf_off_cb
, &ba_ptr
));
827 abd_copy_from_buf_off_cb(void *buf
, size_t size
, void *private)
829 struct buf_arg
*ba_ptr
= private;
831 (void) memcpy(buf
, ba_ptr
->arg_buf
, size
);
832 ba_ptr
->arg_buf
= (char *)ba_ptr
->arg_buf
+ size
;
838 * Copy from buf to abd. (off is the offset in abd.)
841 abd_copy_from_buf_off(abd_t
*abd
, const void *buf
, size_t off
, size_t size
)
843 struct buf_arg ba_ptr
= { (void *) buf
};
845 (void) abd_iterate_func(abd
, off
, size
, abd_copy_from_buf_off_cb
,
851 abd_zero_off_cb(void *buf
, size_t size
, void *private)
853 (void) memset(buf
, 0, size
);
858 * Zero out the abd from a particular offset to the end.
861 abd_zero_off(abd_t
*abd
, size_t off
, size_t size
)
863 (void) abd_iterate_func(abd
, off
, size
, abd_zero_off_cb
, NULL
);
867 * Iterate over two ABDs and call func incrementally on the two ABDs' data in
868 * equal-sized chunks (passed to func as raw buffers). func could be called many
869 * times during this iteration.
872 abd_iterate_func2(abd_t
*dabd
, abd_t
*sabd
, size_t doff
, size_t soff
,
873 size_t size
, abd_iter_func2_t
*func
, void *private)
876 struct abd_iter daiter
, saiter
;
881 ASSERT3U(doff
+ size
, <=, dabd
->abd_size
);
882 ASSERT3U(soff
+ size
, <=, sabd
->abd_size
);
884 abd_iter_init(&daiter
, dabd
);
885 abd_iter_init(&saiter
, sabd
);
886 abd_iter_advance(&daiter
, doff
);
887 abd_iter_advance(&saiter
, soff
);
890 abd_iter_map(&daiter
);
891 abd_iter_map(&saiter
);
893 size_t dlen
= MIN(daiter
.iter_mapsize
, size
);
894 size_t slen
= MIN(saiter
.iter_mapsize
, size
);
895 size_t len
= MIN(dlen
, slen
);
896 ASSERT(dlen
> 0 || slen
> 0);
898 ret
= func(daiter
.iter_mapaddr
, saiter
.iter_mapaddr
, len
,
901 abd_iter_unmap(&saiter
);
902 abd_iter_unmap(&daiter
);
908 abd_iter_advance(&daiter
, len
);
909 abd_iter_advance(&saiter
, len
);
917 abd_copy_off_cb(void *dbuf
, void *sbuf
, size_t size
, void *private)
919 (void) memcpy(dbuf
, sbuf
, size
);
924 * Copy from sabd to dabd starting from soff and doff.
927 abd_copy_off(abd_t
*dabd
, abd_t
*sabd
, size_t doff
, size_t soff
, size_t size
)
929 (void) abd_iterate_func2(dabd
, sabd
, doff
, soff
, size
,
930 abd_copy_off_cb
, NULL
);
935 abd_cmp_cb(void *bufa
, void *bufb
, size_t size
, void *private)
937 return (memcmp(bufa
, bufb
, size
));
941 * Compares the first size bytes of two ABDs.
944 abd_cmp(abd_t
*dabd
, abd_t
*sabd
, size_t size
)
946 return (abd_iterate_func2(dabd
, sabd
, 0, 0, size
, abd_cmp_cb
, NULL
));