blob: 03934da74dc280d3692c21e0db607fbf4ddbe6b7 [file] [log] [blame]
Amit Shahe4d56392010-04-27 18:04:05 +05301/*
2 * Helpers for getting linearized buffers from iov / filling buffers into iovs
3 *
4 * Copyright IBM, Corp. 2007, 2008
5 * Copyright (C) 2010 Red Hat, Inc.
6 *
7 * Author(s):
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Amit Shah <amit.shah@redhat.com>
Michael Tokarev2278a692012-06-07 20:08:19 +040010 * Michael Tokarev <mjt@tls.msk.ru>
Amit Shahe4d56392010-04-27 18:04:05 +053011 *
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
Paolo Bonzini6b620ca2012-01-13 17:44:23 +010014 *
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.
Amit Shahe4d56392010-04-27 18:04:05 +053017 */
18
Paolo Bonzini1de7afc2012-12-17 18:20:00 +010019#include "qemu/iov.h"
Amit Shahe4d56392010-04-27 18:04:05 +053020
Michael Tokarev25e5e4c2012-03-14 11:18:54 +040021#ifdef _WIN32
22# include <windows.h>
23# include <winsock2.h>
24#else
25# include <sys/types.h>
26# include <sys/socket.h>
27#endif
28
Michael S. Tsirkin844b5ce2012-09-24 12:50:32 +020029size_t iov_from_buf(const struct iovec *iov, unsigned int iov_cnt,
Michael Tokarev2278a692012-06-07 20:08:19 +040030 size_t offset, const void *buf, size_t bytes)
Amit Shahe4d56392010-04-27 18:04:05 +053031{
Michael Tokarev2278a692012-06-07 20:08:19 +040032 size_t done;
Amit Shahe4d56392010-04-27 18:04:05 +053033 unsigned int i;
Michael Tokarev2278a692012-06-07 20:08:19 +040034 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) {
35 if (offset < iov[i].iov_len) {
36 size_t len = MIN(iov[i].iov_len - offset, bytes - done);
37 memcpy(iov[i].iov_base + offset, buf + done, len);
38 done += len;
39 offset = 0;
40 } else {
41 offset -= iov[i].iov_len;
Amit Shahfa6111f2010-04-27 18:04:06 +053042 }
Amit Shahfa6111f2010-04-27 18:04:06 +053043 }
Michael Tokarev2278a692012-06-07 20:08:19 +040044 assert(offset == 0);
45 return done;
Amit Shahfa6111f2010-04-27 18:04:06 +053046}
47
Michael Tokarev2278a692012-06-07 20:08:19 +040048size_t iov_to_buf(const struct iovec *iov, const unsigned int iov_cnt,
49 size_t offset, void *buf, size_t bytes)
Hannes Reinecke348e7b82011-07-11 15:02:23 +020050{
Michael Tokarev2278a692012-06-07 20:08:19 +040051 size_t done;
Hannes Reinecke348e7b82011-07-11 15:02:23 +020052 unsigned int i;
Michael Tokarev2278a692012-06-07 20:08:19 +040053 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) {
54 if (offset < iov[i].iov_len) {
55 size_t len = MIN(iov[i].iov_len - offset, bytes - done);
56 memcpy(buf + done, iov[i].iov_base + offset, len);
57 done += len;
58 offset = 0;
59 } else {
60 offset -= iov[i].iov_len;
Hannes Reinecke348e7b82011-07-11 15:02:23 +020061 }
Hannes Reinecke348e7b82011-07-11 15:02:23 +020062 }
Michael Tokarev2278a692012-06-07 20:08:19 +040063 assert(offset == 0);
64 return done;
Hannes Reinecke348e7b82011-07-11 15:02:23 +020065}
66
Michael Tokarevdcf6f5e2012-03-11 18:05:12 +040067size_t iov_memset(const struct iovec *iov, const unsigned int iov_cnt,
Michael Tokarev2278a692012-06-07 20:08:19 +040068 size_t offset, int fillc, size_t bytes)
Gerd Hoffmann8d150282011-07-13 15:16:08 +020069{
Michael Tokarev2278a692012-06-07 20:08:19 +040070 size_t done;
Gerd Hoffmann8d150282011-07-13 15:16:08 +020071 unsigned int i;
Michael Tokarev2278a692012-06-07 20:08:19 +040072 for (i = 0, done = 0; (offset || done < bytes) && i < iov_cnt; i++) {
73 if (offset < iov[i].iov_len) {
74 size_t len = MIN(iov[i].iov_len - offset, bytes - done);
75 memset(iov[i].iov_base + offset, fillc, len);
76 done += len;
77 offset = 0;
78 } else {
79 offset -= iov[i].iov_len;
Gerd Hoffmann8d150282011-07-13 15:16:08 +020080 }
Gerd Hoffmann8d150282011-07-13 15:16:08 +020081 }
Michael Tokarev2278a692012-06-07 20:08:19 +040082 assert(offset == 0);
83 return done;
Gerd Hoffmann8d150282011-07-13 15:16:08 +020084}
85
Hannes Reinecke348e7b82011-07-11 15:02:23 +020086size_t iov_size(const struct iovec *iov, const unsigned int iov_cnt)
Amit Shahfa6111f2010-04-27 18:04:06 +053087{
88 size_t len;
89 unsigned int i;
90
91 len = 0;
Hannes Reinecke348e7b82011-07-11 15:02:23 +020092 for (i = 0; i < iov_cnt; i++) {
Amit Shahfa6111f2010-04-27 18:04:06 +053093 len += iov[i].iov_len;
94 }
95 return len;
96}
Gerd Hoffmann3a1dca92011-07-12 13:35:10 +020097
Michael Tokarev25e5e4c2012-03-14 11:18:54 +040098/* helper function for iov_send_recv() */
99static ssize_t
100do_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, bool do_send)
101{
Paolo Bonzini9adea5f2013-04-21 12:01:06 +0200102#ifdef CONFIG_POSIX
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400103 ssize_t ret;
104 struct msghdr msg;
105 memset(&msg, 0, sizeof(msg));
106 msg.msg_iov = iov;
107 msg.msg_iovlen = iov_cnt;
108 do {
109 ret = do_send
110 ? sendmsg(sockfd, &msg, 0)
111 : recvmsg(sockfd, &msg, 0);
112 } while (ret < 0 && errno == EINTR);
113 return ret;
114#else
115 /* else send piece-by-piece */
116 /*XXX Note: windows has WSASend() and WSARecv() */
Stefan Weilc0958552012-07-11 07:09:05 +0200117 unsigned i = 0;
118 ssize_t ret = 0;
119 while (i < iov_cnt) {
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400120 ssize_t r = do_send
121 ? send(sockfd, iov[i].iov_base, iov[i].iov_len, 0)
122 : recv(sockfd, iov[i].iov_base, iov[i].iov_len, 0);
123 if (r > 0) {
124 ret += r;
125 } else if (!r) {
126 break;
127 } else if (errno == EINTR) {
128 continue;
129 } else {
130 /* else it is some "other" error,
131 * only return if there was no data processed. */
132 if (ret == 0) {
Stefan Weilc0958552012-07-11 07:09:05 +0200133 ret = -1;
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400134 }
135 break;
136 }
Stefan Weilc0958552012-07-11 07:09:05 +0200137 i++;
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400138 }
Stefan Weilc0958552012-07-11 07:09:05 +0200139 return ret;
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400140#endif
141}
142
143ssize_t iov_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt,
144 size_t offset, size_t bytes,
145 bool do_send)
146{
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100147 ssize_t total = 0;
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400148 ssize_t ret;
Paolo Bonzini5209d672013-03-27 17:36:29 +0100149 size_t orig_len, tail;
Paolo Bonzinif48869a2013-03-27 17:36:30 +0100150 unsigned niov;
Paolo Bonzini5209d672013-03-27 17:36:29 +0100151
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100152 while (bytes > 0) {
153 /* Find the start position, skipping `offset' bytes:
154 * first, skip all full-sized vector elements, */
155 for (niov = 0; niov < iov_cnt && offset >= iov[niov].iov_len; ++niov) {
156 offset -= iov[niov].iov_len;
157 }
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400158
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100159 /* niov == iov_cnt would only be valid if bytes == 0, which
160 * we already ruled out in the loop condition. */
Paolo Bonzinif48869a2013-03-27 17:36:30 +0100161 assert(niov < iov_cnt);
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100162 iov += niov;
163 iov_cnt -= niov;
164
165 if (offset) {
166 /* second, skip `offset' bytes from the (now) first element,
167 * undo it on exit */
168 iov[0].iov_base += offset;
169 iov[0].iov_len -= offset;
170 }
171 /* Find the end position skipping `bytes' bytes: */
172 /* first, skip all full-sized elements */
173 tail = bytes;
174 for (niov = 0; niov < iov_cnt && iov[niov].iov_len <= tail; ++niov) {
175 tail -= iov[niov].iov_len;
176 }
177 if (tail) {
178 /* second, fixup the last element, and remember the original
179 * length */
180 assert(niov < iov_cnt);
181 assert(iov[niov].iov_len > tail);
182 orig_len = iov[niov].iov_len;
183 iov[niov++].iov_len = tail;
Michael Tokarev2be178a2013-09-14 13:11:36 +0400184 ret = do_send_recv(sockfd, iov, niov, do_send);
185 /* Undo the changes above before checking for errors */
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100186 iov[niov-1].iov_len = orig_len;
Michael Tokarev2be178a2013-09-14 13:11:36 +0400187 } else {
188 ret = do_send_recv(sockfd, iov, niov, do_send);
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100189 }
190 if (offset) {
191 iov[0].iov_base -= offset;
192 iov[0].iov_len += offset;
193 }
194
195 if (ret < 0) {
196 assert(errno != EINTR);
197 if (errno == EAGAIN && total > 0) {
198 return total;
199 }
200 return -1;
201 }
202
MORITA Kazutaka84004292013-07-23 17:30:12 +0900203 if (ret == 0 && !do_send) {
204 /* recv returns 0 when the peer has performed an orderly
205 * shutdown. */
206 break;
207 }
208
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100209 /* Prepare for the next iteration */
210 offset += ret;
211 total += ret;
212 bytes -= ret;
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400213 }
214
Paolo Bonzini83f75c22013-03-27 17:36:31 +0100215 return total;
Michael Tokarev25e5e4c2012-03-14 11:18:54 +0400216}
217
218
Gerd Hoffmann3a1dca92011-07-12 13:35:10 +0200219void iov_hexdump(const struct iovec *iov, const unsigned int iov_cnt,
220 FILE *fp, const char *prefix, size_t limit)
221{
Peter Crosthwaite6ff66f52013-03-15 16:41:58 +0000222 int v;
223 size_t size = 0;
224 char *buf;
Gerd Hoffmann3a1dca92011-07-12 13:35:10 +0200225
Peter Crosthwaite6ff66f52013-03-15 16:41:58 +0000226 for (v = 0; v < iov_cnt; v++) {
227 size += iov[v].iov_len;
Gerd Hoffmann3a1dca92011-07-12 13:35:10 +0200228 }
Peter Crosthwaite6ff66f52013-03-15 16:41:58 +0000229 size = size > limit ? limit : size;
230 buf = g_malloc(size);
231 iov_to_buf(iov, iov_cnt, 0, buf, size);
Ed Maste3568ac22013-05-16 11:32:28 -0400232 qemu_hexdump(buf, fp, prefix, size);
Peter Crosthwaite6ff66f52013-03-15 16:41:58 +0000233 g_free(buf);
Gerd Hoffmann3a1dca92011-07-12 13:35:10 +0200234}
Paolo Bonzini01912532012-10-24 14:58:39 +0200235
Michael S. Tsirkind3363362012-09-24 13:02:52 +0200236unsigned iov_copy(struct iovec *dst_iov, unsigned int dst_iov_cnt,
237 const struct iovec *iov, unsigned int iov_cnt,
238 size_t offset, size_t bytes)
239{
240 size_t len;
241 unsigned int i, j;
242 for (i = 0, j = 0; i < iov_cnt && j < dst_iov_cnt && bytes; i++) {
243 if (offset >= iov[i].iov_len) {
244 offset -= iov[i].iov_len;
245 continue;
246 }
247 len = MIN(bytes, iov[i].iov_len - offset);
248
249 dst_iov[j].iov_base = iov[i].iov_base + offset;
250 dst_iov[j].iov_len = len;
251 j++;
252 bytes -= len;
253 offset = 0;
254 }
255 assert(offset == 0);
256 return j;
257}
Paolo Bonzinif563a5d2012-10-31 10:42:51 +0100258
Paolo Bonzini01912532012-10-24 14:58:39 +0200259/* io vectors */
260
261void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint)
262{
263 qiov->iov = g_malloc(alloc_hint * sizeof(struct iovec));
264 qiov->niov = 0;
265 qiov->nalloc = alloc_hint;
266 qiov->size = 0;
267}
268
269void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int niov)
270{
271 int i;
272
273 qiov->iov = iov;
274 qiov->niov = niov;
275 qiov->nalloc = -1;
276 qiov->size = 0;
277 for (i = 0; i < niov; i++)
278 qiov->size += iov[i].iov_len;
279}
280
281void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len)
282{
283 assert(qiov->nalloc != -1);
284
285 if (qiov->niov == qiov->nalloc) {
286 qiov->nalloc = 2 * qiov->nalloc + 1;
287 qiov->iov = g_realloc(qiov->iov, qiov->nalloc * sizeof(struct iovec));
288 }
289 qiov->iov[qiov->niov].iov_base = base;
290 qiov->iov[qiov->niov].iov_len = len;
291 qiov->size += len;
292 ++qiov->niov;
293}
294
295/*
Stefan Hajnoczi530c0bb2012-11-22 16:06:06 +0100296 * Concatenates (partial) iovecs from src_iov to the end of dst.
297 * It starts copying after skipping `soffset' bytes at the
298 * beginning of src and adds individual vectors from src to
299 * dst copies up to `sbytes' bytes total, or up to the end
300 * of src_iov if it comes first. This way, it is okay to specify
301 * very large value for `sbytes' to indicate "up to the end
302 * of src".
303 * Only vector pointers are processed, not the actual data buffers.
304 */
305void qemu_iovec_concat_iov(QEMUIOVector *dst,
306 struct iovec *src_iov, unsigned int src_cnt,
307 size_t soffset, size_t sbytes)
308{
309 int i;
310 size_t done;
Aneesh Kumar K.Vfacf98a2013-02-05 11:27:45 +0530311
312 if (!sbytes) {
313 return;
314 }
Stefan Hajnoczi530c0bb2012-11-22 16:06:06 +0100315 assert(dst->nalloc != -1);
316 for (i = 0, done = 0; done < sbytes && i < src_cnt; i++) {
317 if (soffset < src_iov[i].iov_len) {
318 size_t len = MIN(src_iov[i].iov_len - soffset, sbytes - done);
319 qemu_iovec_add(dst, src_iov[i].iov_base + soffset, len);
320 done += len;
321 soffset = 0;
322 } else {
323 soffset -= src_iov[i].iov_len;
324 }
325 }
326 assert(soffset == 0); /* offset beyond end of src */
327}
328
329/*
Paolo Bonzini01912532012-10-24 14:58:39 +0200330 * Concatenates (partial) iovecs from src to the end of dst.
331 * It starts copying after skipping `soffset' bytes at the
332 * beginning of src and adds individual vectors from src to
333 * dst copies up to `sbytes' bytes total, or up to the end
334 * of src if it comes first. This way, it is okay to specify
335 * very large value for `sbytes' to indicate "up to the end
336 * of src".
337 * Only vector pointers are processed, not the actual data buffers.
338 */
339void qemu_iovec_concat(QEMUIOVector *dst,
340 QEMUIOVector *src, size_t soffset, size_t sbytes)
341{
Stefan Hajnoczi530c0bb2012-11-22 16:06:06 +0100342 qemu_iovec_concat_iov(dst, src->iov, src->niov, soffset, sbytes);
Paolo Bonzini01912532012-10-24 14:58:39 +0200343}
344
345void qemu_iovec_destroy(QEMUIOVector *qiov)
346{
347 assert(qiov->nalloc != -1);
348
349 qemu_iovec_reset(qiov);
350 g_free(qiov->iov);
351 qiov->nalloc = 0;
352 qiov->iov = NULL;
353}
354
355void qemu_iovec_reset(QEMUIOVector *qiov)
356{
357 assert(qiov->nalloc != -1);
358
359 qiov->niov = 0;
360 qiov->size = 0;
361}
362
363size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset,
364 void *buf, size_t bytes)
365{
366 return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes);
367}
368
369size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset,
370 const void *buf, size_t bytes)
371{
372 return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes);
373}
374
375size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset,
376 int fillc, size_t bytes)
377{
378 return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes);
379}
Stefan Hajnoczid0277632012-11-21 17:41:10 +0100380
BenoƮt Canetf70d7f72014-02-21 22:21:13 +0100381/**
382 * Check that I/O vector contents are identical
383 *
384 * The IO vectors must have the same structure (same length of all parts).
385 * A typical usage is to compare vectors created with qemu_iovec_clone().
386 *
387 * @a: I/O vector
388 * @b: I/O vector
389 * @ret: Offset to first mismatching byte or -1 if match
390 */
391ssize_t qemu_iovec_compare(QEMUIOVector *a, QEMUIOVector *b)
392{
393 int i;
394 ssize_t offset = 0;
395
396 assert(a->niov == b->niov);
397 for (i = 0; i < a->niov; i++) {
398 size_t len = 0;
399 uint8_t *p = (uint8_t *)a->iov[i].iov_base;
400 uint8_t *q = (uint8_t *)b->iov[i].iov_base;
401
402 assert(a->iov[i].iov_len == b->iov[i].iov_len);
403 while (len < a->iov[i].iov_len && *p++ == *q++) {
404 len++;
405 }
406
407 offset += len;
408
409 if (len != a->iov[i].iov_len) {
410 return offset;
411 }
412 }
413 return -1;
414}
415
416typedef struct {
417 int src_index;
418 struct iovec *src_iov;
419 void *dest_base;
420} IOVectorSortElem;
421
422static int sortelem_cmp_src_base(const void *a, const void *b)
423{
424 const IOVectorSortElem *elem_a = a;
425 const IOVectorSortElem *elem_b = b;
426
427 /* Don't overflow */
428 if (elem_a->src_iov->iov_base < elem_b->src_iov->iov_base) {
429 return -1;
430 } else if (elem_a->src_iov->iov_base > elem_b->src_iov->iov_base) {
431 return 1;
432 } else {
433 return 0;
434 }
435}
436
437static int sortelem_cmp_src_index(const void *a, const void *b)
438{
439 const IOVectorSortElem *elem_a = a;
440 const IOVectorSortElem *elem_b = b;
441
442 return elem_a->src_index - elem_b->src_index;
443}
444
445/**
446 * Copy contents of I/O vector
447 *
448 * The relative relationships of overlapping iovecs are preserved. This is
449 * necessary to ensure identical semantics in the cloned I/O vector.
450 */
451void qemu_iovec_clone(QEMUIOVector *dest, const QEMUIOVector *src, void *buf)
452{
453 IOVectorSortElem sortelems[src->niov];
454 void *last_end;
455 int i;
456
457 /* Sort by source iovecs by base address */
458 for (i = 0; i < src->niov; i++) {
459 sortelems[i].src_index = i;
460 sortelems[i].src_iov = &src->iov[i];
461 }
462 qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_base);
463
464 /* Allocate buffer space taking into account overlapping iovecs */
465 last_end = NULL;
466 for (i = 0; i < src->niov; i++) {
467 struct iovec *cur = sortelems[i].src_iov;
468 ptrdiff_t rewind = 0;
469
470 /* Detect overlap */
471 if (last_end && last_end > cur->iov_base) {
472 rewind = last_end - cur->iov_base;
473 }
474
475 sortelems[i].dest_base = buf - rewind;
476 buf += cur->iov_len - MIN(rewind, cur->iov_len);
477 last_end = MAX(cur->iov_base + cur->iov_len, last_end);
478 }
479
480 /* Sort by source iovec index and build destination iovec */
481 qsort(sortelems, src->niov, sizeof(sortelems[0]), sortelem_cmp_src_index);
482 for (i = 0; i < src->niov; i++) {
483 qemu_iovec_add(dest, sortelems[i].dest_base, src->iov[i].iov_len);
484 }
485}
486
Stefan Hajnoczid0277632012-11-21 17:41:10 +0100487size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt,
488 size_t bytes)
489{
490 size_t total = 0;
491 struct iovec *cur;
492
493 for (cur = *iov; *iov_cnt > 0; cur++) {
494 if (cur->iov_len > bytes) {
495 cur->iov_base += bytes;
496 cur->iov_len -= bytes;
497 total += bytes;
498 break;
499 }
500
501 bytes -= cur->iov_len;
502 total += cur->iov_len;
503 *iov_cnt -= 1;
504 }
505
506 *iov = cur;
507 return total;
508}
509
510size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt,
511 size_t bytes)
512{
513 size_t total = 0;
514 struct iovec *cur;
515
516 if (*iov_cnt == 0) {
517 return 0;
518 }
519
520 cur = iov + (*iov_cnt - 1);
521
522 while (*iov_cnt > 0) {
523 if (cur->iov_len > bytes) {
524 cur->iov_len -= bytes;
525 total += bytes;
526 break;
527 }
528
529 bytes -= cur->iov_len;
530 total += cur->iov_len;
531 cur--;
532 *iov_cnt -= 1;
533 }
534
535 return total;
536}