]> git.proxmox.com Git - qemu.git/blob - util/iov.c
iov: handle partial writes from sendmsg and recvmsg
[qemu.git] / util / iov.c
1 /*
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>
10 * Michael Tokarev <mjt@tls.msk.ru>
11 *
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
14 *
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.
17 */
18
19 #include "qemu/iov.h"
20
21 #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
29 size_t iov_from_buf(const struct iovec *iov, unsigned int iov_cnt,
30 size_t offset, const void *buf, size_t bytes)
31 {
32 size_t done;
33 unsigned int i;
34 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;
42 }
43 }
44 assert(offset == 0);
45 return done;
46 }
47
48 size_t iov_to_buf(const struct iovec *iov, const unsigned int iov_cnt,
49 size_t offset, void *buf, size_t bytes)
50 {
51 size_t done;
52 unsigned int i;
53 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;
61 }
62 }
63 assert(offset == 0);
64 return done;
65 }
66
67 size_t iov_memset(const struct iovec *iov, const unsigned int iov_cnt,
68 size_t offset, int fillc, size_t bytes)
69 {
70 size_t done;
71 unsigned int i;
72 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;
80 }
81 }
82 assert(offset == 0);
83 return done;
84 }
85
86 size_t iov_size(const struct iovec *iov, const unsigned int iov_cnt)
87 {
88 size_t len;
89 unsigned int i;
90
91 len = 0;
92 for (i = 0; i < iov_cnt; i++) {
93 len += iov[i].iov_len;
94 }
95 return len;
96 }
97
98 /* helper function for iov_send_recv() */
99 static ssize_t
100 do_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt, bool do_send)
101 {
102 #if defined CONFIG_IOVEC && defined CONFIG_POSIX
103 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() */
117 unsigned i = 0;
118 ssize_t ret = 0;
119 while (i < iov_cnt) {
120 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) {
133 ret = -1;
134 }
135 break;
136 }
137 i++;
138 }
139 return ret;
140 #endif
141 }
142
143 ssize_t iov_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt,
144 size_t offset, size_t bytes,
145 bool do_send)
146 {
147 ssize_t total = 0;
148 ssize_t ret;
149 size_t orig_len, tail;
150 unsigned niov;
151
152 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 }
158
159 /* niov == iov_cnt would only be valid if bytes == 0, which
160 * we already ruled out in the loop condition. */
161 assert(niov < iov_cnt);
162 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;
184 }
185
186 ret = do_send_recv(sockfd, iov, niov, do_send);
187
188 /* Undo the changes above before checking for errors */
189 if (tail) {
190 iov[niov-1].iov_len = orig_len;
191 }
192 if (offset) {
193 iov[0].iov_base -= offset;
194 iov[0].iov_len += offset;
195 }
196
197 if (ret < 0) {
198 assert(errno != EINTR);
199 if (errno == EAGAIN && total > 0) {
200 return total;
201 }
202 return -1;
203 }
204
205 /* Prepare for the next iteration */
206 offset += ret;
207 total += ret;
208 bytes -= ret;
209 }
210
211 return total;
212 }
213
214
215 void iov_hexdump(const struct iovec *iov, const unsigned int iov_cnt,
216 FILE *fp, const char *prefix, size_t limit)
217 {
218 int v;
219 size_t size = 0;
220 char *buf;
221
222 for (v = 0; v < iov_cnt; v++) {
223 size += iov[v].iov_len;
224 }
225 size = size > limit ? limit : size;
226 buf = g_malloc(size);
227 iov_to_buf(iov, iov_cnt, 0, buf, size);
228 hexdump(buf, fp, prefix, size);
229 g_free(buf);
230 }
231
232 unsigned iov_copy(struct iovec *dst_iov, unsigned int dst_iov_cnt,
233 const struct iovec *iov, unsigned int iov_cnt,
234 size_t offset, size_t bytes)
235 {
236 size_t len;
237 unsigned int i, j;
238 for (i = 0, j = 0; i < iov_cnt && j < dst_iov_cnt && bytes; i++) {
239 if (offset >= iov[i].iov_len) {
240 offset -= iov[i].iov_len;
241 continue;
242 }
243 len = MIN(bytes, iov[i].iov_len - offset);
244
245 dst_iov[j].iov_base = iov[i].iov_base + offset;
246 dst_iov[j].iov_len = len;
247 j++;
248 bytes -= len;
249 offset = 0;
250 }
251 assert(offset == 0);
252 return j;
253 }
254
255 /* io vectors */
256
257 void qemu_iovec_init(QEMUIOVector *qiov, int alloc_hint)
258 {
259 qiov->iov = g_malloc(alloc_hint * sizeof(struct iovec));
260 qiov->niov = 0;
261 qiov->nalloc = alloc_hint;
262 qiov->size = 0;
263 }
264
265 void qemu_iovec_init_external(QEMUIOVector *qiov, struct iovec *iov, int niov)
266 {
267 int i;
268
269 qiov->iov = iov;
270 qiov->niov = niov;
271 qiov->nalloc = -1;
272 qiov->size = 0;
273 for (i = 0; i < niov; i++)
274 qiov->size += iov[i].iov_len;
275 }
276
277 void qemu_iovec_add(QEMUIOVector *qiov, void *base, size_t len)
278 {
279 assert(qiov->nalloc != -1);
280
281 if (qiov->niov == qiov->nalloc) {
282 qiov->nalloc = 2 * qiov->nalloc + 1;
283 qiov->iov = g_realloc(qiov->iov, qiov->nalloc * sizeof(struct iovec));
284 }
285 qiov->iov[qiov->niov].iov_base = base;
286 qiov->iov[qiov->niov].iov_len = len;
287 qiov->size += len;
288 ++qiov->niov;
289 }
290
291 /*
292 * Concatenates (partial) iovecs from src_iov to the end of dst.
293 * It starts copying after skipping `soffset' bytes at the
294 * beginning of src and adds individual vectors from src to
295 * dst copies up to `sbytes' bytes total, or up to the end
296 * of src_iov if it comes first. This way, it is okay to specify
297 * very large value for `sbytes' to indicate "up to the end
298 * of src".
299 * Only vector pointers are processed, not the actual data buffers.
300 */
301 void qemu_iovec_concat_iov(QEMUIOVector *dst,
302 struct iovec *src_iov, unsigned int src_cnt,
303 size_t soffset, size_t sbytes)
304 {
305 int i;
306 size_t done;
307
308 if (!sbytes) {
309 return;
310 }
311 assert(dst->nalloc != -1);
312 for (i = 0, done = 0; done < sbytes && i < src_cnt; i++) {
313 if (soffset < src_iov[i].iov_len) {
314 size_t len = MIN(src_iov[i].iov_len - soffset, sbytes - done);
315 qemu_iovec_add(dst, src_iov[i].iov_base + soffset, len);
316 done += len;
317 soffset = 0;
318 } else {
319 soffset -= src_iov[i].iov_len;
320 }
321 }
322 assert(soffset == 0); /* offset beyond end of src */
323 }
324
325 /*
326 * Concatenates (partial) iovecs from src to the end of dst.
327 * It starts copying after skipping `soffset' bytes at the
328 * beginning of src and adds individual vectors from src to
329 * dst copies up to `sbytes' bytes total, or up to the end
330 * of src if it comes first. This way, it is okay to specify
331 * very large value for `sbytes' to indicate "up to the end
332 * of src".
333 * Only vector pointers are processed, not the actual data buffers.
334 */
335 void qemu_iovec_concat(QEMUIOVector *dst,
336 QEMUIOVector *src, size_t soffset, size_t sbytes)
337 {
338 qemu_iovec_concat_iov(dst, src->iov, src->niov, soffset, sbytes);
339 }
340
341 void qemu_iovec_destroy(QEMUIOVector *qiov)
342 {
343 assert(qiov->nalloc != -1);
344
345 qemu_iovec_reset(qiov);
346 g_free(qiov->iov);
347 qiov->nalloc = 0;
348 qiov->iov = NULL;
349 }
350
351 void qemu_iovec_reset(QEMUIOVector *qiov)
352 {
353 assert(qiov->nalloc != -1);
354
355 qiov->niov = 0;
356 qiov->size = 0;
357 }
358
359 size_t qemu_iovec_to_buf(QEMUIOVector *qiov, size_t offset,
360 void *buf, size_t bytes)
361 {
362 return iov_to_buf(qiov->iov, qiov->niov, offset, buf, bytes);
363 }
364
365 size_t qemu_iovec_from_buf(QEMUIOVector *qiov, size_t offset,
366 const void *buf, size_t bytes)
367 {
368 return iov_from_buf(qiov->iov, qiov->niov, offset, buf, bytes);
369 }
370
371 size_t qemu_iovec_memset(QEMUIOVector *qiov, size_t offset,
372 int fillc, size_t bytes)
373 {
374 return iov_memset(qiov->iov, qiov->niov, offset, fillc, bytes);
375 }
376
377 size_t iov_discard_front(struct iovec **iov, unsigned int *iov_cnt,
378 size_t bytes)
379 {
380 size_t total = 0;
381 struct iovec *cur;
382
383 for (cur = *iov; *iov_cnt > 0; cur++) {
384 if (cur->iov_len > bytes) {
385 cur->iov_base += bytes;
386 cur->iov_len -= bytes;
387 total += bytes;
388 break;
389 }
390
391 bytes -= cur->iov_len;
392 total += cur->iov_len;
393 *iov_cnt -= 1;
394 }
395
396 *iov = cur;
397 return total;
398 }
399
400 size_t iov_discard_back(struct iovec *iov, unsigned int *iov_cnt,
401 size_t bytes)
402 {
403 size_t total = 0;
404 struct iovec *cur;
405
406 if (*iov_cnt == 0) {
407 return 0;
408 }
409
410 cur = iov + (*iov_cnt - 1);
411
412 while (*iov_cnt > 0) {
413 if (cur->iov_len > bytes) {
414 cur->iov_len -= bytes;
415 total += bytes;
416 break;
417 }
418
419 bytes -= cur->iov_len;
420 total += cur->iov_len;
421 cur--;
422 *iov_cnt -= 1;
423 }
424
425 return total;
426 }