virtiofsd: Use --thread-pool-size=0 to mean no thread pool
[qemu/kevin.git] / util / iov.c
blobf3a9e92a378f3eb5255b939b14039fe6680b1f52
1 /*
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.
7 * Author(s):
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.
19 #include "qemu/osdep.h"
20 #include "qemu-common.h"
21 #include "qemu/iov.h"
22 #include "qemu/sockets.h"
23 #include "qemu/cutils.h"
25 size_t iov_from_buf_full(const struct iovec *iov, unsigned int iov_cnt,
26 size_t offset, const void *buf, size_t bytes)
28 size_t done;
29 unsigned int i;
30 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) {
31 if (offset < iov[i].iov_len) {
32 size_t len = MIN(iov[i].iov_len - offset, bytes - done);
33 memcpy(iov[i].iov_base + offset, buf + done, len);
34 done += len;
35 offset = 0;
36 } else {
37 offset -= iov[i].iov_len;
40 assert(offset == 0);
41 return done;
44 size_t iov_to_buf_full(const struct iovec *iov, const unsigned int iov_cnt,
45 size_t offset, void *buf, size_t bytes)
47 size_t done;
48 unsigned int i;
49 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) {
50 if (offset < iov[i].iov_len) {
51 size_t len = MIN(iov[i].iov_len - offset, bytes - done);
52 memcpy(buf + done, iov[i].iov_base + offset, len);
53 done += len;
54 offset = 0;
55 } else {
56 offset -= iov[i].iov_len;
59 assert(offset == 0);
60 return done;
63 size_t iov_memset(const struct iovec *iov, const unsigned int iov_cnt,
64 size_t offset, int fillc, size_t bytes)
66 size_t done;
67 unsigned int i;
68 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) {
69 if (offset < iov[i].iov_len) {
70 size_t len = MIN(iov[i].iov_len - offset, bytes - done);
71 memset(iov[i].iov_base + offset, fillc, len);
72 done += len;
73 offset = 0;
74 } else {
75 offset -= iov[i].iov_len;
78 assert(offset == 0);
79 return done;
82 size_t iov_size(const struct iovec *iov, const unsigned int iov_cnt)
84 size_t len;
85 unsigned int i;
87 len = 0;
88 for (i = 0; i < iov_cnt; i++) {
89 len += iov[i].iov_len;
91 return len;
94 /* helper function for iov_send_recv() */
95 static ssize_t
96 do_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, bool do_send)
98 #ifdef CONFIG_POSIX
99 ssize_t ret;
100 struct msghdr msg;
101 memset(&msg, 0, sizeof(msg));
102 msg.msg_iov = iov;
103 msg.msg_iovlen = iov_cnt;
104 do {
105 ret = do_send
106 ? sendmsg(sockfd, &msg, 0)
107 : recvmsg(sockfd, &msg, 0);
108 } while (ret < 0 && errno == EINTR);
109 return ret;
110 #else
111 /* else send piece-by-piece */
112 /*XXX Note: windows has WSASend() and WSARecv() */
113 unsigned i = 0;
114 ssize_t ret = 0;
115 while (i < iov_cnt) {
116 ssize_t r = do_send
117 ? send(sockfd, iov[i].iov_base, iov[i].iov_len, 0)
118 : recv(sockfd, iov[i].iov_base, iov[i].iov_len, 0);
119 if (r > 0) {
120 ret += r;
121 } else if (!r) {
122 break;
123 } else if (errno == EINTR) {
124 continue;
125 } else {
126 /* else it is some "other" error,
127 * only return if there was no data processed. */
128 if (ret == 0) {
129 ret = -1;
131 break;
133 i++;
135 return ret;
136 #endif
139 ssize_t iov_send_recv(int sockfd, const struct iovec *_iov, unsigned iov_cnt,
140 size_t offset, size_t bytes,
141 bool do_send)
143 ssize_t total = 0;
144 ssize_t ret;
145 size_t orig_len, tail;
146 unsigned niov;
147 struct iovec *local_iov, *iov;
149 if (bytes <= 0) {
150 return 0;
153 local_iov = g_new0(struct iovec, iov_cnt);
154 iov_copy(local_iov, iov_cnt, _iov, iov_cnt, offset, bytes);
155 offset = 0;
156 iov = local_iov;
158 while (bytes > 0) {
159 /* Find the start position, skipping `offset' bytes:
160 * first, skip all full-sized vector elements, */
161 for (niov = 0; niov < iov_cnt && offset >= iov[niov].iov_len; ++niov) {
162 offset -= iov[niov].iov_len;
165 /* niov == iov_cnt would only be valid if bytes == 0, which
166 * we already ruled out in the loop condition. */
167 assert(niov < iov_cnt);
168 iov += niov;
169 iov_cnt -= niov;
171 if (offset) {
172 /* second, skip `offset' bytes from the (now) first element,
173 * undo it on exit */
174 iov[0].iov_base += offset;
175 iov[0].iov_len -= offset;
177 /* Find the end position skipping `bytes' bytes: */
178 /* first, skip all full-sized elements */
179 tail = bytes;
180 for (niov = 0; niov < iov_cnt && iov[niov].iov_len <= tail; ++niov) {
181 tail -= iov[niov].iov_len;
183 if (tail) {
184 /* second, fixup the last element, and remember the original
185 * length */
186 assert(niov < iov_cnt);
187 assert(iov[niov].iov_len > tail);
188 orig_len = iov[niov].iov_len;
189 iov[niov++].iov_len = tail;
190 ret = do_send_recv(sockfd, iov, niov, do_send);
191 /* Undo the changes above before checking for errors */
192 iov[niov-1].iov_len = orig_len;
193 } else {
194 ret = do_send_recv(sockfd, iov, niov, do_send);
196 if (offset) {
197 iov[0].iov_base -= offset;
198 iov[0].iov_len += offset;
201 if (ret < 0) {
202 assert(errno != EINTR);
203 g_free(local_iov);
204 if (errno == EAGAIN && total > 0) {
205 return total;
207 return -1;
210 if (ret == 0 && !do_send) {
211 /* recv returns 0 when the peer has performed an orderly
212 * shutdown. */
213 break;
216 /* Prepare for the next iteration */
217 offset += ret;
218 total += ret;
219 bytes -= ret;
222 g_free(local_iov);
223 return total;
227 void iov_hexdump(const struct iovec *iov, const unsigned int iov_cnt,
228 FILE *fp, const char *prefix, size_t limit)
230 int v;
231 size_t size = 0;
232 char *buf;
234 for (v = 0; v < iov_cnt; v++) {
235 size += iov[v].iov_len;
237 size = size > limit ? limit : size;
238 buf = g_malloc(size);
239 iov_to_buf(iov, iov_cnt, 0, buf, size);
240 qemu_hexdump(fp, prefix, buf, size);
241 g_free(buf);
244 unsigned iov_copy(struct iovec *dst_iov, unsigned int dst_iov_cnt,
245 const struct iovec *iov, unsigned int iov_cnt,
246 size_t offset, size_t bytes)
248 size_t len;
249 unsigned int i, j;
250 for (i = 0, j = 0;
251 i < iov_cnt && j < dst_iov_cnt && (offset || bytes); i++) {
252 if (offset >= iov[i].iov_len) {
253 offset -= iov[i].iov_len;
254 continue;
256 len = MIN(bytes, iov[i].iov_len - offset);
258 dst_iov[j].iov_base = iov[i].iov_base + offset;
259 dst_iov[j].iov_len = len;
260 j++;
261 bytes -= len;
262 offset = 0;
264 assert(offset == 0);
265 return j;
268 /* io vectors */
270 void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint)
272 qiov->iov = g_new(struct iovec, alloc_hint);
273 qiov->niov = 0;
274 qiov->nalloc = alloc_hint;
275 qiov->size = 0;
278 void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int niov)
280 int i;
282 qiov->iov = iov;
283 qiov->niov = niov;
284 qiov->nalloc = -1;
285 qiov->size = 0;
286 for (i = 0; i < niov; i++)
287 qiov->size += iov[i].iov_len;
290 void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len)
292 assert(qiov->nalloc != -1);
294 if (qiov->niov == qiov->nalloc) {
295 qiov->nalloc = 2 * qiov->nalloc + 1;
296 qiov->iov = g_renew(struct iovec, qiov->iov, qiov->nalloc);
298 qiov->iov[qiov->niov].iov_base = base;
299 qiov->iov[qiov->niov].iov_len = len;
300 qiov->size += len;
301 ++qiov->niov;
305 * Concatenates (partial) iovecs from src_iov to the end of dst.
306 * It starts copying after skipping `soffset' bytes at the
307 * beginning of src and adds individual vectors from src to
308 * dst copies up to `sbytes' bytes total, or up to the end
309 * of src_iov if it comes first. This way, it is okay to specify
310 * very large value for `sbytes' to indicate "up to the end
311 * of src".
312 * Only vector pointers are processed, not the actual data buffers.
314 size_t qemu_iovec_concat_iov(QEMUIOVector *dst,
315 struct iovec *src_iov, unsigned int src_cnt,
316 size_t soffset, size_t sbytes)
318 int i;
319 size_t done;
321 if (!sbytes) {
322 return 0;
324 assert(dst->nalloc != -1);
325 for (i = 0, done = 0; done < sbytes && i < src_cnt; i++) {
326 if (soffset < src_iov[i].iov_len) {
327 size_t len = MIN(src_iov[i].iov_len - soffset, sbytes - done);
328 qemu_iovec_add(dst, src_iov[i].iov_base + soffset, len);
329 done += len;
330 soffset = 0;
331 } else {
332 soffset -= src_iov[i].iov_len;
335 assert(soffset == 0); /* offset beyond end of src */
337 return done;
341 * Concatenates (partial) iovecs from src to the end of dst.
342 * It starts copying after skipping `soffset' bytes at the
343 * beginning of src and adds individual vectors from src to
344 * dst copies up to `sbytes' bytes total, or up to the end
345 * of src if it comes first. This way, it is okay to specify
346 * very large value for `sbytes' to indicate "up to the end
347 * of src".
348 * Only vector pointers are processed, not the actual data buffers.
350 void qemu_iovec_concat(QEMUIOVector *dst,
351 QEMUIOVector *src, size_t soffset, size_t sbytes)
353 qemu_iovec_concat_iov(dst, src->iov, src->niov, soffset, sbytes);
357 * qiov_find_iov
359 * Return pointer to iovec structure, where byte at @offset in original vector
360 * @iov exactly is.
361 * Set @remaining_offset to be offset inside that iovec to the same byte.
363 static struct iovec *iov_skip_offset(struct iovec *iov, size_t offset,
364 size_t *remaining_offset)
366 while (offset > 0 && offset >= iov->iov_len) {
367 offset -= iov->iov_len;
368 iov++;
370 *remaining_offset = offset;
372 return iov;
376 * qiov_slice
378 * Find subarray of iovec's, containing requested range. @head would
379 * be offset in first iov (returned by the function), @tail would be
380 * count of extra bytes in last iovec (returned iov + @niov - 1).
382 static struct iovec *qiov_slice(QEMUIOVector *qiov,
383 size_t offset, size_t len,
384 size_t *head, size_t *tail, int *niov)
386 struct iovec *iov, *end_iov;
388 assert(offset + len <= qiov->size);
390 iov = iov_skip_offset(qiov->iov, offset, head);
391 end_iov = iov_skip_offset(iov, *head + len, tail);
393 if (*tail > 0) {
394 assert(*tail < end_iov->iov_len);
395 *tail = end_iov->iov_len - *tail;
396 end_iov++;
399 *niov = end_iov - iov;
401 return iov;
404 int qemu_iovec_subvec_niov(QEMUIOVector *qiov, size_t offset, size_t len)
406 size_t head, tail;
407 int niov;
409 qiov_slice(qiov, offset, len, &head, &tail, &niov);
411 return niov;
415 * Compile new iovec, combining @head_buf buffer, sub-qiov of @mid_qiov,
416 * and @tail_buf buffer into new qiov.
418 void qemu_iovec_init_extended(
419 QEMUIOVector *qiov,
420 void *head_buf, size_t head_len,
421 QEMUIOVector *mid_qiov, size_t mid_offset, size_t mid_len,
422 void *tail_buf, size_t tail_len)
424 size_t mid_head, mid_tail;
425 int total_niov, mid_niov = 0;
426 struct iovec *p, *mid_iov = NULL;
428 if (mid_len) {
429 mid_iov = qiov_slice(mid_qiov, mid_offset, mid_len,
430 &mid_head, &mid_tail, &mid_niov);
433 total_niov = !!head_len + mid_niov + !!tail_len;
434 if (total_niov == 1) {
435 qemu_iovec_init_buf(qiov, NULL, 0);
436 p = &qiov->local_iov;
437 } else {
438 qiov->niov = qiov->nalloc = total_niov;
439 qiov->size = head_len + mid_len + tail_len;
440 p = qiov->iov = g_new(struct iovec, qiov->niov);
443 if (head_len) {
444 p->iov_base = head_buf;
445 p->iov_len = head_len;
446 p++;
449 assert(!mid_niov == !mid_len);
450 if (mid_niov) {
451 memcpy(p, mid_iov, mid_niov * sizeof(*p));
452 p[0].iov_base = (uint8_t *)p[0].iov_base + mid_head;
453 p[0].iov_len -= mid_head;
454 p[mid_niov - 1].iov_len -= mid_tail;
455 p += mid_niov;
458 if (tail_len) {
459 p->iov_base = tail_buf;
460 p->iov_len = tail_len;
465 * Check if the contents of subrange of qiov data is all zeroes.
467 bool qemu_iovec_is_zero(QEMUIOVector *qiov, size_t offset, size_t bytes)
469 struct iovec *iov;
470 size_t current_offset;
472 assert(offset + bytes <= qiov->size);
474 iov = iov_skip_offset(qiov->iov, offset, &current_offset);
476 while (bytes) {
477 uint8_t *base = (uint8_t *)iov->iov_base + current_offset;
478 size_t len = MIN(iov->iov_len - current_offset, bytes);
480 if (!buffer_is_zero(base, len)) {
481 return false;
484 current_offset = 0;
485 bytes -= len;
486 iov++;
489 return true;
492 void qemu_iovec_init_slice(QEMUIOVector *qiov, QEMUIOVector *source,
493 size_t offset, size_t len)
495 qemu_iovec_init_extended(qiov, NULL, 0, source, offset, len, NULL, 0);
498 void qemu_iovec_destroy(QEMUIOVector *qiov)
500 if (qiov->nalloc != -1) {
501 g_free(qiov->iov);
504 memset(qiov, 0, sizeof(*qiov));
507 void qemu_iovec_reset(QEMUIOVector *qiov)
509 assert(qiov->nalloc != -1);
511 qiov->niov = 0;
512 qiov->size = 0;
515 size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset,
516 void *buf, size_t bytes)
518 return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes);
521 size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset,
522 const void *buf, size_t bytes)
524 return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes);
527 size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset,
528 int fillc, size_t bytes)
530 return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes);
534 * Check that I/O vector contents are identical
536 * The IO vectors must have the same structure (same length of all parts).
537 * A typical usage is to compare vectors created with qemu_iovec_clone().
539 * @a: I/O vector
540 * @b: I/O vector
541 * @ret: Offset to first mismatching byte or -1 if match
543 ssize_t qemu_iovec_compare(QEMUIOVector *a, QEMUIOVector *b)
545 int i;
546 ssize_t offset = 0;
548 assert(a->niov == b->niov);
549 for (i = 0; i < a->niov; i++) {
550 size_t len = 0;
551 uint8_t *p = (uint8_t *)a->iov[i].iov_base;
552 uint8_t *q = (uint8_t *)b->iov[i].iov_base;
554 assert(a->iov[i].iov_len == b->iov[i].iov_len);
555 while (len < a->iov[i].iov_len && *p++ == *q++) {
556 len++;
559 offset += len;
561 if (len != a->iov[i].iov_len) {
562 return offset;
565 return -1;
568 typedef struct {
569 int src_index;
570 struct iovec *src_iov;
571 void *dest_base;
572 } IOVectorSortElem;
574 static int sortelem_cmp_src_base(const void *a, const void *b)
576 const IOVectorSortElem *elem_a = a;
577 const IOVectorSortElem *elem_b = b;
579 /* Don't overflow */
580 if (elem_a->src_iov->iov_base < elem_b->src_iov->iov_base) {
581 return -1;
582 } else if (elem_a->src_iov->iov_base > elem_b->src_iov->iov_base) {
583 return 1;
584 } else {
585 return 0;
589 static int sortelem_cmp_src_index(const void *a, const void *b)
591 const IOVectorSortElem *elem_a = a;
592 const IOVectorSortElem *elem_b = b;
594 return elem_a->src_index - elem_b->src_index;
598 * Copy contents of I/O vector
600 * The relative relationships of overlapping iovecs are preserved. This is
601 * necessary to ensure identical semantics in the cloned I/O vector.
603 void qemu_iovec_clone(QEMUIOVector *dest, const QEMUIOVector *src, void *buf)
605 IOVectorSortElem sortelems[src->niov];
606 void *last_end;
607 int i;
609 /* Sort by source iovecs by base address */
610 for (i = 0; i < src->niov; i++) {
611 sortelems[i].src_index = i;
612 sortelems[i].src_iov = &src->iov[i];
614 qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_base);
616 /* Allocate buffer space taking into account overlapping iovecs */
617 last_end = NULL;
618 for (i = 0; i < src->niov; i++) {
619 struct iovec *cur = sortelems[i].src_iov;
620 ptrdiff_t rewind = 0;
622 /* Detect overlap */
623 if (last_end && last_end > cur->iov_base) {
624 rewind = last_end - cur->iov_base;
627 sortelems[i].dest_base = buf - rewind;
628 buf += cur->iov_len - MIN(rewind, cur->iov_len);
629 last_end = MAX(cur->iov_base + cur->iov_len, last_end);
632 /* Sort by source iovec index and build destination iovec */
633 qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_index);
634 for (i = 0; i < src->niov; i++) {
635 qemu_iovec_add(dest, sortelems[i].dest_base, src->iov[i].iov_len);
639 void iov_discard_undo(IOVDiscardUndo *undo)
641 /* Restore original iovec if it was modified */
642 if (undo->modified_iov) {
643 *undo->modified_iov = undo->orig;
647 size_t iov_discard_front_undoable(struct iovec **iov,
648 unsigned int *iov_cnt,
649 size_t bytes,
650 IOVDiscardUndo *undo)
652 size_t total = 0;
653 struct iovec *cur;
655 if (undo) {
656 undo->modified_iov = NULL;
659 for (cur = *iov; *iov_cnt > 0; cur++) {
660 if (cur->iov_len > bytes) {
661 if (undo) {
662 undo->modified_iov = cur;
663 undo->orig = *cur;
666 cur->iov_base += bytes;
667 cur->iov_len -= bytes;
668 total += bytes;
669 break;
672 bytes -= cur->iov_len;
673 total += cur->iov_len;
674 *iov_cnt -= 1;
677 *iov = cur;
678 return total;
681 size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt,
682 size_t bytes)
684 return iov_discard_front_undoable(iov, iov_cnt, bytes, NULL);
687 size_t iov_discard_back_undoable(struct iovec *iov,
688 unsigned int *iov_cnt,
689 size_t bytes,
690 IOVDiscardUndo *undo)
692 size_t total = 0;
693 struct iovec *cur;
695 if (undo) {
696 undo->modified_iov = NULL;
699 if (*iov_cnt == 0) {
700 return 0;
703 cur = iov + (*iov_cnt - 1);
705 while (*iov_cnt > 0) {
706 if (cur->iov_len > bytes) {
707 if (undo) {
708 undo->modified_iov = cur;
709 undo->orig = *cur;
712 cur->iov_len -= bytes;
713 total += bytes;
714 break;
717 bytes -= cur->iov_len;
718 total += cur->iov_len;
719 cur--;
720 *iov_cnt -= 1;
723 return total;
726 size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt,
727 size_t bytes)
729 return iov_discard_back_undoable(iov, iov_cnt, bytes, NULL);
732 void qemu_iovec_discard_back(QEMUIOVector *qiov, size_t bytes)
734 size_t total;
735 unsigned int niov = qiov->niov;
737 assert(qiov->size >= bytes);
738 total = iov_discard_back(qiov->iov, &niov, bytes);
739 assert(total == bytes);
741 qiov->niov = niov;
742 qiov->size -= bytes;