2 * Helpers for getting linearized buffers from iov / filling buffers into iovs
4 * Copyright IBM, Corp. 2007, 2008
5 * Copyright (C) 2010 Red Hat, Inc.
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Amit Shah <amit.shah@redhat.com>
10 * Michael Tokarev <mjt@tls.msk.ru>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
15 * Contributions after 2012-01-13 are licensed under the terms of the
16 * GNU GPL, version 2 or (at your option) any later version.
20 #include "qemu/sockets.h"
22 size_t iov_from_buf(const struct iovec
*iov
, unsigned int iov_cnt
,
23 size_t offset
, const void *buf
, size_t bytes
)
27 for (i
= 0, done
= 0; (offset
|| done
< bytes
) && i
< iov_cnt
; i
++) {
28 if (offset
< iov
[i
].iov_len
) {
29 size_t len
= MIN(iov
[i
].iov_len
- offset
, bytes
- done
);
30 memcpy(iov
[i
].iov_base
+ offset
, buf
+ done
, len
);
34 offset
-= iov
[i
].iov_len
;
41 size_t iov_to_buf(const struct iovec
*iov
, const unsigned int iov_cnt
,
42 size_t offset
, void *buf
, size_t bytes
)
46 for (i
= 0, done
= 0; (offset
|| done
< bytes
) && i
< iov_cnt
; i
++) {
47 if (offset
< iov
[i
].iov_len
) {
48 size_t len
= MIN(iov
[i
].iov_len
- offset
, bytes
- done
);
49 memcpy(buf
+ done
, iov
[i
].iov_base
+ offset
, len
);
53 offset
-= iov
[i
].iov_len
;
60 size_t iov_memset(const struct iovec
*iov
, const unsigned int iov_cnt
,
61 size_t offset
, int fillc
, size_t bytes
)
65 for (i
= 0, done
= 0; (offset
|| done
< bytes
) && i
< iov_cnt
; i
++) {
66 if (offset
< iov
[i
].iov_len
) {
67 size_t len
= MIN(iov
[i
].iov_len
- offset
, bytes
- done
);
68 memset(iov
[i
].iov_base
+ offset
, fillc
, len
);
72 offset
-= iov
[i
].iov_len
;
79 size_t iov_size(const struct iovec
*iov
, const unsigned int iov_cnt
)
85 for (i
= 0; i
< iov_cnt
; i
++) {
86 len
+= iov
[i
].iov_len
;
91 /* helper function for iov_send_recv() */
93 do_send_recv(int sockfd
, struct iovec
*iov
, unsigned iov_cnt
, bool do_send
)
98 memset(&msg
, 0, sizeof(msg
));
100 msg
.msg_iovlen
= iov_cnt
;
103 ? sendmsg(sockfd
, &msg
, 0)
104 : recvmsg(sockfd
, &msg
, 0);
105 } while (ret
< 0 && errno
== EINTR
);
108 /* else send piece-by-piece */
109 /*XXX Note: windows has WSASend() and WSARecv() */
112 while (i
< iov_cnt
) {
114 ? send(sockfd
, iov
[i
].iov_base
, iov
[i
].iov_len
, 0)
115 : recv(sockfd
, iov
[i
].iov_base
, iov
[i
].iov_len
, 0);
120 } else if (errno
== EINTR
) {
123 /* else it is some "other" error,
124 * only return if there was no data processed. */
136 ssize_t
iov_send_recv(int sockfd
, struct iovec
*iov
, unsigned iov_cnt
,
137 size_t offset
, size_t bytes
,
142 size_t orig_len
, tail
;
146 /* Find the start position, skipping `offset' bytes:
147 * first, skip all full-sized vector elements, */
148 for (niov
= 0; niov
< iov_cnt
&& offset
>= iov
[niov
].iov_len
; ++niov
) {
149 offset
-= iov
[niov
].iov_len
;
152 /* niov == iov_cnt would only be valid if bytes == 0, which
153 * we already ruled out in the loop condition. */
154 assert(niov
< iov_cnt
);
159 /* second, skip `offset' bytes from the (now) first element,
161 iov
[0].iov_base
+= offset
;
162 iov
[0].iov_len
-= offset
;
164 /* Find the end position skipping `bytes' bytes: */
165 /* first, skip all full-sized elements */
167 for (niov
= 0; niov
< iov_cnt
&& iov
[niov
].iov_len
<= tail
; ++niov
) {
168 tail
-= iov
[niov
].iov_len
;
171 /* second, fixup the last element, and remember the original
173 assert(niov
< iov_cnt
);
174 assert(iov
[niov
].iov_len
> tail
);
175 orig_len
= iov
[niov
].iov_len
;
176 iov
[niov
++].iov_len
= tail
;
177 ret
= do_send_recv(sockfd
, iov
, niov
, do_send
);
178 /* Undo the changes above before checking for errors */
179 iov
[niov
-1].iov_len
= orig_len
;
181 ret
= do_send_recv(sockfd
, iov
, niov
, do_send
);
184 iov
[0].iov_base
-= offset
;
185 iov
[0].iov_len
+= offset
;
189 assert(errno
!= EINTR
);
190 if (errno
== EAGAIN
&& total
> 0) {
196 if (ret
== 0 && !do_send
) {
197 /* recv returns 0 when the peer has performed an orderly
202 /* Prepare for the next iteration */
212 void iov_hexdump(const struct iovec
*iov
, const unsigned int iov_cnt
,
213 FILE *fp
, const char *prefix
, size_t limit
)
219 for (v
= 0; v
< iov_cnt
; v
++) {
220 size
+= iov
[v
].iov_len
;
222 size
= size
> limit
? limit
: size
;
223 buf
= g_malloc(size
);
224 iov_to_buf(iov
, iov_cnt
, 0, buf
, size
);
225 qemu_hexdump(buf
, fp
, prefix
, size
);
229 unsigned iov_copy(struct iovec
*dst_iov
, unsigned int dst_iov_cnt
,
230 const struct iovec
*iov
, unsigned int iov_cnt
,
231 size_t offset
, size_t bytes
)
235 for (i
= 0, j
= 0; i
< iov_cnt
&& j
< dst_iov_cnt
&& bytes
; i
++) {
236 if (offset
>= iov
[i
].iov_len
) {
237 offset
-= iov
[i
].iov_len
;
240 len
= MIN(bytes
, iov
[i
].iov_len
- offset
);
242 dst_iov
[j
].iov_base
= iov
[i
].iov_base
+ offset
;
243 dst_iov
[j
].iov_len
= len
;
254 void qemu_iovec_init(QEMUIOVector
*qiov
, int alloc_hint
)
256 qiov
->iov
= g_malloc(alloc_hint
* sizeof(struct iovec
));
258 qiov
->nalloc
= alloc_hint
;
262 void qemu_iovec_init_external(QEMUIOVector
*qiov
, struct iovec
*iov
, int niov
)
270 for (i
= 0; i
< niov
; i
++)
271 qiov
->size
+= iov
[i
].iov_len
;
274 void qemu_iovec_add(QEMUIOVector
*qiov
, void *base
, size_t len
)
276 assert(qiov
->nalloc
!= -1);
278 if (qiov
->niov
== qiov
->nalloc
) {
279 qiov
->nalloc
= 2 * qiov
->nalloc
+ 1;
280 qiov
->iov
= g_realloc(qiov
->iov
, qiov
->nalloc
* sizeof(struct iovec
));
282 qiov
->iov
[qiov
->niov
].iov_base
= base
;
283 qiov
->iov
[qiov
->niov
].iov_len
= len
;
289 * Concatenates (partial) iovecs from src_iov to the end of dst.
290 * It starts copying after skipping `soffset' bytes at the
291 * beginning of src and adds individual vectors from src to
292 * dst copies up to `sbytes' bytes total, or up to the end
293 * of src_iov if it comes first. This way, it is okay to specify
294 * very large value for `sbytes' to indicate "up to the end
296 * Only vector pointers are processed, not the actual data buffers.
298 void qemu_iovec_concat_iov(QEMUIOVector
*dst
,
299 struct iovec
*src_iov
, unsigned int src_cnt
,
300 size_t soffset
, size_t sbytes
)
308 assert(dst
->nalloc
!= -1);
309 for (i
= 0, done
= 0; done
< sbytes
&& i
< src_cnt
; i
++) {
310 if (soffset
< src_iov
[i
].iov_len
) {
311 size_t len
= MIN(src_iov
[i
].iov_len
- soffset
, sbytes
- done
);
312 qemu_iovec_add(dst
, src_iov
[i
].iov_base
+ soffset
, len
);
316 soffset
-= src_iov
[i
].iov_len
;
319 assert(soffset
== 0); /* offset beyond end of src */
323 * Concatenates (partial) iovecs from src to the end of dst.
324 * It starts copying after skipping `soffset' bytes at the
325 * beginning of src and adds individual vectors from src to
326 * dst copies up to `sbytes' bytes total, or up to the end
327 * of src if it comes first. This way, it is okay to specify
328 * very large value for `sbytes' to indicate "up to the end
330 * Only vector pointers are processed, not the actual data buffers.
332 void qemu_iovec_concat(QEMUIOVector
*dst
,
333 QEMUIOVector
*src
, size_t soffset
, size_t sbytes
)
335 qemu_iovec_concat_iov(dst
, src
->iov
, src
->niov
, soffset
, sbytes
);
339 * Check if the contents of the iovecs are all zero
341 bool qemu_iovec_is_zero(QEMUIOVector
*qiov
)
344 for (i
= 0; i
< qiov
->niov
; i
++) {
345 size_t offs
= QEMU_ALIGN_DOWN(qiov
->iov
[i
].iov_len
, 4 * sizeof(long));
346 uint8_t *ptr
= qiov
->iov
[i
].iov_base
;
347 if (offs
&& !buffer_is_zero(qiov
->iov
[i
].iov_base
, offs
)) {
350 for (; offs
< qiov
->iov
[i
].iov_len
; offs
++) {
359 void qemu_iovec_destroy(QEMUIOVector
*qiov
)
361 assert(qiov
->nalloc
!= -1);
363 qemu_iovec_reset(qiov
);
369 void qemu_iovec_reset(QEMUIOVector
*qiov
)
371 assert(qiov
->nalloc
!= -1);
377 size_t qemu_iovec_to_buf(QEMUIOVector
*qiov
, size_t offset
,
378 void *buf
, size_t bytes
)
380 return iov_to_buf(qiov
->iov
, qiov
->niov
, offset
, buf
, bytes
);
383 size_t qemu_iovec_from_buf(QEMUIOVector
*qiov
, size_t offset
,
384 const void *buf
, size_t bytes
)
386 return iov_from_buf(qiov
->iov
, qiov
->niov
, offset
, buf
, bytes
);
389 size_t qemu_iovec_memset(QEMUIOVector
*qiov
, size_t offset
,
390 int fillc
, size_t bytes
)
392 return iov_memset(qiov
->iov
, qiov
->niov
, offset
, fillc
, bytes
);
396 * Check that I/O vector contents are identical
398 * The IO vectors must have the same structure (same length of all parts).
399 * A typical usage is to compare vectors created with qemu_iovec_clone().
403 * @ret: Offset to first mismatching byte or -1 if match
405 ssize_t
qemu_iovec_compare(QEMUIOVector
*a
, QEMUIOVector
*b
)
410 assert(a
->niov
== b
->niov
);
411 for (i
= 0; i
< a
->niov
; i
++) {
413 uint8_t *p
= (uint8_t *)a
->iov
[i
].iov_base
;
414 uint8_t *q
= (uint8_t *)b
->iov
[i
].iov_base
;
416 assert(a
->iov
[i
].iov_len
== b
->iov
[i
].iov_len
);
417 while (len
< a
->iov
[i
].iov_len
&& *p
++ == *q
++) {
423 if (len
!= a
->iov
[i
].iov_len
) {
432 struct iovec
*src_iov
;
436 static int sortelem_cmp_src_base(const void *a
, const void *b
)
438 const IOVectorSortElem
*elem_a
= a
;
439 const IOVectorSortElem
*elem_b
= b
;
442 if (elem_a
->src_iov
->iov_base
< elem_b
->src_iov
->iov_base
) {
444 } else if (elem_a
->src_iov
->iov_base
> elem_b
->src_iov
->iov_base
) {
451 static int sortelem_cmp_src_index(const void *a
, const void *b
)
453 const IOVectorSortElem
*elem_a
= a
;
454 const IOVectorSortElem
*elem_b
= b
;
456 return elem_a
->src_index
- elem_b
->src_index
;
460 * Copy contents of I/O vector
462 * The relative relationships of overlapping iovecs are preserved. This is
463 * necessary to ensure identical semantics in the cloned I/O vector.
465 void qemu_iovec_clone(QEMUIOVector
*dest
, const QEMUIOVector
*src
, void *buf
)
467 IOVectorSortElem sortelems
[src
->niov
];
471 /* Sort by source iovecs by base address */
472 for (i
= 0; i
< src
->niov
; i
++) {
473 sortelems
[i
].src_index
= i
;
474 sortelems
[i
].src_iov
= &src
->iov
[i
];
476 qsort(sortelems
, src
->niov
, sizeof(sortelems
[0]), sortelem_cmp_src_base
);
478 /* Allocate buffer space taking into account overlapping iovecs */
480 for (i
= 0; i
< src
->niov
; i
++) {
481 struct iovec
*cur
= sortelems
[i
].src_iov
;
482 ptrdiff_t rewind
= 0;
485 if (last_end
&& last_end
> cur
->iov_base
) {
486 rewind
= last_end
- cur
->iov_base
;
489 sortelems
[i
].dest_base
= buf
- rewind
;
490 buf
+= cur
->iov_len
- MIN(rewind
, cur
->iov_len
);
491 last_end
= MAX(cur
->iov_base
+ cur
->iov_len
, last_end
);
494 /* Sort by source iovec index and build destination iovec */
495 qsort(sortelems
, src
->niov
, sizeof(sortelems
[0]), sortelem_cmp_src_index
);
496 for (i
= 0; i
< src
->niov
; i
++) {
497 qemu_iovec_add(dest
, sortelems
[i
].dest_base
, src
->iov
[i
].iov_len
);
501 size_t iov_discard_front(struct iovec
**iov
, unsigned int *iov_cnt
,
507 for (cur
= *iov
; *iov_cnt
> 0; cur
++) {
508 if (cur
->iov_len
> bytes
) {
509 cur
->iov_base
+= bytes
;
510 cur
->iov_len
-= bytes
;
515 bytes
-= cur
->iov_len
;
516 total
+= cur
->iov_len
;
524 size_t iov_discard_back(struct iovec
*iov
, unsigned int *iov_cnt
,
534 cur
= iov
+ (*iov_cnt
- 1);
536 while (*iov_cnt
> 0) {
537 if (cur
->iov_len
> bytes
) {
538 cur
->iov_len
-= bytes
;
543 bytes
-= cur
->iov_len
;
544 total
+= cur
->iov_len
;