Commit | Line | Data |
---|---|---|
ed92ac0c | 1 | /* |
da751ca9 JA |
2 | * net engine |
3 | * | |
4 | * IO engine that reads/writes to/from sockets. | |
5 | * | |
ed92ac0c JA |
6 | */ |
7 | #include <stdio.h> | |
8 | #include <stdlib.h> | |
9 | #include <unistd.h> | |
10 | #include <errno.h> | |
11 | #include <assert.h> | |
12 | #include <netinet/in.h> | |
13 | #include <arpa/inet.h> | |
14 | #include <netdb.h> | |
5fdd124a | 15 | #include <sys/poll.h> |
ed92ac0c JA |
16 | |
17 | #include "../fio.h" | |
ed92ac0c | 18 | |
b5af8293 JA |
19 | struct netio_data { |
20 | int listenfd; | |
21 | int send_to_net; | |
9cce02e8 JA |
22 | int use_splice; |
23 | int pipes[2]; | |
b5af8293 JA |
24 | char host[64]; |
25 | struct sockaddr_in addr; | |
26 | }; | |
ed92ac0c JA |
27 | |
28 | static int fio_netio_prep(struct thread_data *td, struct io_u *io_u) | |
29 | { | |
b5af8293 | 30 | struct netio_data *nd = td->io_ops->data; |
ed92ac0c JA |
31 | struct fio_file *f = io_u->file; |
32 | ||
7a6499da JA |
33 | /* |
34 | * Make sure we don't see spurious reads to a receiver, and vice versa | |
35 | */ | |
b5af8293 JA |
36 | if ((nd->send_to_net && io_u->ddir == DDIR_READ) || |
37 | (!nd->send_to_net && io_u->ddir == DDIR_WRITE)) { | |
e1161c32 | 38 | td_verror(td, EINVAL, "bad direction"); |
7a6499da | 39 | return 1; |
ed92ac0c | 40 | } |
7a6499da | 41 | |
ed92ac0c JA |
42 | if (io_u->ddir == DDIR_SYNC) |
43 | return 0; | |
44 | if (io_u->offset == f->last_completed_pos) | |
45 | return 0; | |
46 | ||
e01547d2 JA |
47 | /* |
48 | * If offset is different from last end position, it's a seek. | |
49 | * As network io is purely sequential, we don't allow seeks. | |
50 | */ | |
e1161c32 | 51 | td_verror(td, EINVAL, "cannot seek"); |
ed92ac0c JA |
52 | return 1; |
53 | } | |
54 | ||
cd963e18 | 55 | static int splice_io_u(int fdin, int fdout, unsigned int len) |
ed92ac0c | 56 | { |
9cce02e8 | 57 | int bytes = 0; |
7a6499da | 58 | |
9cce02e8 | 59 | while (len) { |
cd963e18 | 60 | int ret = splice(fdin, NULL, fdout, NULL, len, 0); |
9cce02e8 JA |
61 | |
62 | if (ret < 0) { | |
63 | if (!bytes) | |
64 | bytes = ret; | |
65 | ||
66 | break; | |
67 | } else if (!ret) | |
68 | break; | |
69 | ||
70 | bytes += ret; | |
f657a2fb | 71 | len -= ret; |
9cce02e8 JA |
72 | } |
73 | ||
74 | return bytes; | |
75 | } | |
76 | ||
77 | /* | |
cd963e18 | 78 | * Receive bytes from a socket and fill them into the internal pipe |
9cce02e8 | 79 | */ |
cd963e18 | 80 | static int splice_in(struct thread_data *td, struct io_u *io_u) |
9cce02e8 JA |
81 | { |
82 | struct netio_data *nd = td->io_ops->data; | |
9cce02e8 | 83 | |
cd963e18 | 84 | return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen); |
9cce02e8 JA |
85 | } |
86 | ||
87 | /* | |
cd963e18 | 88 | * Transmit 'len' bytes from the internal pipe |
9cce02e8 | 89 | */ |
cd963e18 JA |
90 | static int splice_out(struct thread_data *td, struct io_u *io_u, |
91 | unsigned int len) | |
9cce02e8 JA |
92 | { |
93 | struct netio_data *nd = td->io_ops->data; | |
cd963e18 JA |
94 | |
95 | return splice_io_u(nd->pipes[0], io_u->file->fd, len); | |
96 | } | |
97 | ||
98 | static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len) | |
99 | { | |
9cce02e8 JA |
100 | struct iovec iov = { |
101 | .iov_base = io_u->xfer_buf, | |
102 | .iov_len = len, | |
103 | }; | |
104 | int bytes = 0; | |
105 | ||
106 | while (iov.iov_len) { | |
cd963e18 | 107 | int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE); |
9cce02e8 JA |
108 | |
109 | if (ret < 0) { | |
110 | if (!bytes) | |
111 | bytes = ret; | |
112 | break; | |
113 | } else if (!ret) | |
114 | break; | |
115 | ||
116 | iov.iov_len -= ret; | |
cd963e18 | 117 | iov.iov_base += ret; |
f657a2fb | 118 | bytes += ret; |
9cce02e8 JA |
119 | } |
120 | ||
121 | return bytes; | |
cd963e18 | 122 | |
9cce02e8 JA |
123 | } |
124 | ||
125 | /* | |
cd963e18 | 126 | * vmsplice() pipe to io_u buffer |
9cce02e8 | 127 | */ |
cd963e18 JA |
128 | static int vmsplice_io_u_out(struct thread_data *td, struct io_u *io_u, |
129 | unsigned int len) | |
9cce02e8 JA |
130 | { |
131 | struct netio_data *nd = td->io_ops->data; | |
9cce02e8 | 132 | |
cd963e18 JA |
133 | return vmsplice_io_u(io_u, nd->pipes[0], len); |
134 | } | |
9cce02e8 | 135 | |
cd963e18 JA |
136 | /* |
137 | * vmsplice() io_u to pipe | |
138 | */ | |
139 | static int vmsplice_io_u_in(struct thread_data *td, struct io_u *io_u) | |
140 | { | |
141 | struct netio_data *nd = td->io_ops->data; | |
ed92ac0c | 142 | |
cd963e18 | 143 | return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen); |
9cce02e8 JA |
144 | } |
145 | ||
cd963e18 JA |
146 | /* |
147 | * splice receive - transfer socket data into a pipe using splice, then map | |
148 | * that pipe data into the io_u using vmsplice. | |
149 | */ | |
9cce02e8 JA |
150 | static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u) |
151 | { | |
152 | int ret; | |
153 | ||
154 | ret = splice_in(td, io_u); | |
cd963e18 JA |
155 | if (ret > 0) |
156 | return vmsplice_io_u_out(td, io_u, ret); | |
9cce02e8 | 157 | |
cd963e18 | 158 | return ret; |
9cce02e8 JA |
159 | } |
160 | ||
cd963e18 JA |
161 | /* |
162 | * splice transmit - map data from the io_u into a pipe by using vmsplice, | |
163 | * then transfer that pipe to a socket using splice. | |
164 | */ | |
9cce02e8 JA |
165 | static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u) |
166 | { | |
167 | int ret; | |
168 | ||
169 | ret = vmsplice_io_u_in(td, io_u); | |
cd963e18 JA |
170 | if (ret > 0) |
171 | return splice_out(td, io_u, ret); | |
9cce02e8 | 172 | |
cd963e18 | 173 | return ret; |
9cce02e8 JA |
174 | } |
175 | ||
176 | static int fio_netio_send(struct thread_data *td, struct io_u *io_u) | |
177 | { | |
178 | int flags = 0; | |
179 | ||
180 | /* | |
181 | * if we are going to write more, set MSG_MORE | |
182 | */ | |
183 | if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size) | |
184 | flags = MSG_MORE; | |
185 | ||
186 | return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags); | |
187 | } | |
188 | ||
189 | static int fio_netio_recv(struct io_u *io_u) | |
190 | { | |
191 | int flags = MSG_WAITALL; | |
192 | ||
193 | return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags); | |
194 | } | |
195 | ||
196 | static int fio_netio_queue(struct thread_data *td, struct io_u *io_u) | |
197 | { | |
198 | struct netio_data *nd = td->io_ops->data; | |
199 | int ret; | |
200 | ||
7101d9c2 JA |
201 | fio_ro_check(td, io_u); |
202 | ||
9cce02e8 JA |
203 | if (io_u->ddir == DDIR_WRITE) { |
204 | if (nd->use_splice) | |
205 | ret = fio_netio_splice_out(td, io_u); | |
206 | else | |
207 | ret = fio_netio_send(td, io_u); | |
d4f12dd0 | 208 | } else if (io_u->ddir == DDIR_READ) { |
9cce02e8 JA |
209 | if (nd->use_splice) |
210 | ret = fio_netio_splice_in(td, io_u); | |
211 | else | |
212 | ret = fio_netio_recv(io_u); | |
d4f12dd0 | 213 | } else |
7a6499da | 214 | ret = 0; /* must be a SYNC */ |
ed92ac0c | 215 | |
cec6b55d | 216 | if (ret != (int) io_u->xfer_buflen) { |
22819ec2 | 217 | if (ret >= 0) { |
cec6b55d JA |
218 | io_u->resid = io_u->xfer_buflen - ret; |
219 | io_u->error = 0; | |
36167d82 | 220 | return FIO_Q_COMPLETED; |
ed92ac0c JA |
221 | } else |
222 | io_u->error = errno; | |
223 | } | |
224 | ||
36167d82 | 225 | if (io_u->error) |
e1161c32 | 226 | td_verror(td, io_u->error, "xfer"); |
ed92ac0c | 227 | |
36167d82 | 228 | return FIO_Q_COMPLETED; |
ed92ac0c JA |
229 | } |
230 | ||
b5af8293 | 231 | static int fio_netio_connect(struct thread_data *td, struct fio_file *f) |
ed92ac0c | 232 | { |
b5af8293 | 233 | struct netio_data *nd = td->io_ops->data; |
ed92ac0c | 234 | |
b5af8293 JA |
235 | f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); |
236 | if (f->fd < 0) { | |
237 | td_verror(td, errno, "socket"); | |
238 | return 1; | |
ed92ac0c JA |
239 | } |
240 | ||
b5af8293 JA |
241 | if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) { |
242 | td_verror(td, errno, "connect"); | |
243 | return 1; | |
ed92ac0c JA |
244 | } |
245 | ||
246 | return 0; | |
ed92ac0c JA |
247 | } |
248 | ||
b5af8293 | 249 | static int fio_netio_accept(struct thread_data *td, struct fio_file *f) |
5fdd124a | 250 | { |
b5af8293 JA |
251 | struct netio_data *nd = td->io_ops->data; |
252 | socklen_t socklen = sizeof(nd->addr); | |
5fdd124a | 253 | struct pollfd pfd; |
b5af8293 | 254 | int ret; |
5fdd124a | 255 | |
6d86144d | 256 | log_info("fio: waiting for connection\n"); |
5fdd124a JA |
257 | |
258 | /* | |
259 | * Accept loop. poll for incoming events, accept them. Repeat until we | |
260 | * have all connections. | |
261 | */ | |
b5af8293 JA |
262 | while (!td->terminate) { |
263 | pfd.fd = nd->listenfd; | |
5fdd124a JA |
264 | pfd.events = POLLIN; |
265 | ||
266 | ret = poll(&pfd, 1, -1); | |
267 | if (ret < 0) { | |
268 | if (errno == EINTR) | |
269 | continue; | |
270 | ||
e1161c32 | 271 | td_verror(td, errno, "poll"); |
5fdd124a JA |
272 | break; |
273 | } else if (!ret) | |
274 | continue; | |
275 | ||
0c09442b JA |
276 | /* |
277 | * should be impossible | |
278 | */ | |
279 | if (!(pfd.revents & POLLIN)) | |
280 | continue; | |
281 | ||
b5af8293 JA |
282 | f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen); |
283 | if (f->fd < 0) { | |
284 | td_verror(td, errno, "accept"); | |
285 | return 1; | |
286 | } | |
287 | break; | |
288 | } | |
5fdd124a | 289 | |
b5af8293 JA |
290 | return 0; |
291 | } | |
292 | ||
b5af8293 JA |
293 | static int fio_netio_open_file(struct thread_data *td, struct fio_file *f) |
294 | { | |
295 | if (td_read(td)) | |
296 | return fio_netio_accept(td, f); | |
297 | else | |
298 | return fio_netio_connect(td, f); | |
299 | } | |
300 | ||
301 | static int fio_netio_setup_connect(struct thread_data *td, const char *host, | |
302 | unsigned short port) | |
303 | { | |
304 | struct netio_data *nd = td->io_ops->data; | |
305 | ||
306 | nd->addr.sin_family = AF_INET; | |
307 | nd->addr.sin_port = htons(port); | |
308 | ||
309 | if (inet_aton(host, &nd->addr.sin_addr) != 1) { | |
310 | struct hostent *hent; | |
311 | ||
312 | hent = gethostbyname(host); | |
313 | if (!hent) { | |
314 | td_verror(td, errno, "gethostbyname"); | |
315 | return 1; | |
5fdd124a | 316 | } |
b5af8293 JA |
317 | |
318 | memcpy(&nd->addr.sin_addr, hent->h_addr, 4); | |
5fdd124a JA |
319 | } |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
b5af8293 | 324 | static int fio_netio_setup_listen(struct thread_data *td, short port) |
ed92ac0c | 325 | { |
b5af8293 | 326 | struct netio_data *nd = td->io_ops->data; |
5fdd124a | 327 | int fd, opt; |
ed92ac0c | 328 | |
6bedbfaf | 329 | fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); |
ed92ac0c | 330 | if (fd < 0) { |
e1161c32 | 331 | td_verror(td, errno, "socket"); |
ed92ac0c JA |
332 | return 1; |
333 | } | |
334 | ||
335 | opt = 1; | |
336 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) { | |
e1161c32 | 337 | td_verror(td, errno, "setsockopt"); |
ed92ac0c JA |
338 | return 1; |
339 | } | |
6bedbfaf JA |
340 | #ifdef SO_REUSEPORT |
341 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) { | |
e1161c32 | 342 | td_verror(td, errno, "setsockopt"); |
6bedbfaf JA |
343 | return 1; |
344 | } | |
345 | #endif | |
ed92ac0c | 346 | |
b5af8293 JA |
347 | nd->addr.sin_family = AF_INET; |
348 | nd->addr.sin_addr.s_addr = htonl(INADDR_ANY); | |
349 | nd->addr.sin_port = htons(port); | |
ed92ac0c | 350 | |
b5af8293 | 351 | if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) { |
e1161c32 | 352 | td_verror(td, errno, "bind"); |
ed92ac0c JA |
353 | return 1; |
354 | } | |
355 | if (listen(fd, 1) < 0) { | |
e1161c32 | 356 | td_verror(td, errno, "listen"); |
ed92ac0c JA |
357 | return 1; |
358 | } | |
359 | ||
b5af8293 JA |
360 | nd->listenfd = fd; |
361 | return 0; | |
ed92ac0c JA |
362 | } |
363 | ||
9bec88e1 | 364 | static int fio_netio_init(struct thread_data *td) |
ed92ac0c | 365 | { |
b5af8293 | 366 | struct netio_data *nd = td->io_ops->data; |
e01547d2 | 367 | unsigned short port; |
b5af8293 | 368 | char host[64], buf[128]; |
ed92ac0c | 369 | char *sep; |
af52b345 | 370 | int ret; |
ed92ac0c | 371 | |
413dd459 | 372 | if (td_rw(td)) { |
ed92ac0c JA |
373 | log_err("fio: network connections must be read OR write\n"); |
374 | return 1; | |
375 | } | |
16d55aae JA |
376 | if (td_random(td)) { |
377 | log_err("fio: network IO can't be random\n"); | |
378 | return 1; | |
379 | } | |
ed92ac0c | 380 | |
2dc1bbeb | 381 | strcpy(buf, td->o.filename); |
ed92ac0c | 382 | |
9f9214f2 | 383 | sep = strchr(buf, '/'); |
ed92ac0c | 384 | if (!sep) { |
2dc1bbeb | 385 | log_err("fio: bad network host/port <<%s>>\n", td->o.filename); |
ed92ac0c JA |
386 | return 1; |
387 | } | |
388 | ||
389 | *sep = '\0'; | |
390 | sep++; | |
391 | strcpy(host, buf); | |
e01547d2 | 392 | port = atoi(sep); |
ed92ac0c | 393 | |
413dd459 | 394 | if (td_read(td)) { |
b5af8293 | 395 | nd->send_to_net = 0; |
ed92ac0c JA |
396 | ret = fio_netio_setup_listen(td, port); |
397 | } else { | |
b5af8293 | 398 | nd->send_to_net = 1; |
ed92ac0c JA |
399 | ret = fio_netio_setup_connect(td, host, port); |
400 | } | |
401 | ||
7bb48f84 | 402 | return ret; |
ed92ac0c JA |
403 | } |
404 | ||
b5af8293 | 405 | static void fio_netio_cleanup(struct thread_data *td) |
9bec88e1 | 406 | { |
b5af8293 JA |
407 | struct netio_data *nd = td->io_ops->data; |
408 | ||
409 | if (nd) { | |
64b24cd8 JA |
410 | if (nd->listenfd != -1) |
411 | close(nd->listenfd); | |
412 | if (nd->pipes[0] != -1) | |
413 | close(nd->pipes[0]); | |
414 | if (nd->pipes[1] != -1) | |
415 | close(nd->pipes[1]); | |
416 | ||
b5af8293 JA |
417 | free(nd); |
418 | td->io_ops->data = NULL; | |
419 | } | |
420 | } | |
421 | ||
422 | static int fio_netio_setup(struct thread_data *td) | |
423 | { | |
7bb48f84 | 424 | struct netio_data *nd; |
7bb48f84 JA |
425 | |
426 | if (!td->io_ops->data) { | |
427 | nd = malloc(sizeof(*nd));; | |
428 | ||
429 | memset(nd, 0, sizeof(*nd)); | |
430 | nd->listenfd = -1; | |
64b24cd8 | 431 | nd->pipes[0] = nd->pipes[1] = -1; |
7bb48f84 | 432 | td->io_ops->data = nd; |
7bb48f84 | 433 | } |
b5af8293 | 434 | |
9bec88e1 JA |
435 | return 0; |
436 | } | |
437 | ||
9cce02e8 JA |
438 | static int fio_netio_setup_splice(struct thread_data *td) |
439 | { | |
440 | struct netio_data *nd; | |
441 | ||
442 | fio_netio_setup(td); | |
443 | ||
444 | nd = td->io_ops->data; | |
445 | if (nd) { | |
446 | if (pipe(nd->pipes) < 0) | |
447 | return 1; | |
448 | ||
449 | nd->use_splice = 1; | |
450 | return 0; | |
451 | } | |
452 | ||
453 | return 1; | |
454 | } | |
455 | ||
456 | static struct ioengine_ops ioengine_rw = { | |
ed92ac0c JA |
457 | .name = "net", |
458 | .version = FIO_IOOPS_VERSION, | |
ed92ac0c JA |
459 | .prep = fio_netio_prep, |
460 | .queue = fio_netio_queue, | |
ed92ac0c | 461 | .setup = fio_netio_setup, |
9bec88e1 | 462 | .init = fio_netio_init, |
b5af8293 JA |
463 | .cleanup = fio_netio_cleanup, |
464 | .open_file = fio_netio_open_file, | |
465 | .close_file = generic_close_file, | |
dc372f07 | 466 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR, |
ed92ac0c JA |
467 | }; |
468 | ||
9cce02e8 JA |
469 | static struct ioengine_ops ioengine_splice = { |
470 | .name = "netsplice", | |
471 | .version = FIO_IOOPS_VERSION, | |
472 | .prep = fio_netio_prep, | |
473 | .queue = fio_netio_queue, | |
474 | .setup = fio_netio_setup_splice, | |
475 | .init = fio_netio_init, | |
476 | .cleanup = fio_netio_cleanup, | |
477 | .open_file = fio_netio_open_file, | |
478 | .close_file = generic_close_file, | |
b67740d3 | 479 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR, |
9cce02e8 JA |
480 | }; |
481 | ||
ed92ac0c JA |
482 | static void fio_init fio_netio_register(void) |
483 | { | |
9cce02e8 JA |
484 | register_ioengine(&ioengine_rw); |
485 | register_ioengine(&ioengine_splice); | |
ed92ac0c JA |
486 | } |
487 | ||
488 | static void fio_exit fio_netio_unregister(void) | |
489 | { | |
9cce02e8 JA |
490 | unregister_ioengine(&ioengine_rw); |
491 | unregister_ioengine(&ioengine_splice); | |
ed92ac0c | 492 | } |