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 | 31 | |
7a6499da JA |
32 | /* |
33 | * Make sure we don't see spurious reads to a receiver, and vice versa | |
34 | */ | |
b5af8293 JA |
35 | if ((nd->send_to_net && io_u->ddir == DDIR_READ) || |
36 | (!nd->send_to_net && io_u->ddir == DDIR_WRITE)) { | |
e1161c32 | 37 | td_verror(td, EINVAL, "bad direction"); |
7a6499da | 38 | return 1; |
ed92ac0c | 39 | } |
7a6499da | 40 | |
f85ac25a | 41 | return 0; |
ed92ac0c JA |
42 | } |
43 | ||
cd963e18 | 44 | static int splice_io_u(int fdin, int fdout, unsigned int len) |
ed92ac0c | 45 | { |
9cce02e8 | 46 | int bytes = 0; |
7a6499da | 47 | |
9cce02e8 | 48 | while (len) { |
cd963e18 | 49 | int ret = splice(fdin, NULL, fdout, NULL, len, 0); |
9cce02e8 JA |
50 | |
51 | if (ret < 0) { | |
52 | if (!bytes) | |
53 | bytes = ret; | |
54 | ||
55 | break; | |
56 | } else if (!ret) | |
57 | break; | |
58 | ||
59 | bytes += ret; | |
f657a2fb | 60 | len -= ret; |
9cce02e8 JA |
61 | } |
62 | ||
63 | return bytes; | |
64 | } | |
65 | ||
66 | /* | |
cd963e18 | 67 | * Receive bytes from a socket and fill them into the internal pipe |
9cce02e8 | 68 | */ |
cd963e18 | 69 | static int splice_in(struct thread_data *td, struct io_u *io_u) |
9cce02e8 JA |
70 | { |
71 | struct netio_data *nd = td->io_ops->data; | |
9cce02e8 | 72 | |
cd963e18 | 73 | return splice_io_u(io_u->file->fd, nd->pipes[1], io_u->xfer_buflen); |
9cce02e8 JA |
74 | } |
75 | ||
76 | /* | |
cd963e18 | 77 | * Transmit 'len' bytes from the internal pipe |
9cce02e8 | 78 | */ |
cd963e18 JA |
79 | static int splice_out(struct thread_data *td, struct io_u *io_u, |
80 | unsigned int len) | |
9cce02e8 JA |
81 | { |
82 | struct netio_data *nd = td->io_ops->data; | |
cd963e18 JA |
83 | |
84 | return splice_io_u(nd->pipes[0], io_u->file->fd, len); | |
85 | } | |
86 | ||
87 | static int vmsplice_io_u(struct io_u *io_u, int fd, unsigned int len) | |
88 | { | |
9cce02e8 JA |
89 | struct iovec iov = { |
90 | .iov_base = io_u->xfer_buf, | |
91 | .iov_len = len, | |
92 | }; | |
93 | int bytes = 0; | |
94 | ||
95 | while (iov.iov_len) { | |
cd963e18 | 96 | int ret = vmsplice(fd, &iov, 1, SPLICE_F_MOVE); |
9cce02e8 JA |
97 | |
98 | if (ret < 0) { | |
99 | if (!bytes) | |
100 | bytes = ret; | |
101 | break; | |
102 | } else if (!ret) | |
103 | break; | |
104 | ||
105 | iov.iov_len -= ret; | |
cd963e18 | 106 | iov.iov_base += ret; |
f657a2fb | 107 | bytes += ret; |
9cce02e8 JA |
108 | } |
109 | ||
110 | return bytes; | |
cd963e18 | 111 | |
9cce02e8 JA |
112 | } |
113 | ||
114 | /* | |
cd963e18 | 115 | * vmsplice() pipe to io_u buffer |
9cce02e8 | 116 | */ |
cd963e18 JA |
117 | static int vmsplice_io_u_out(struct thread_data *td, struct io_u *io_u, |
118 | unsigned int len) | |
9cce02e8 JA |
119 | { |
120 | struct netio_data *nd = td->io_ops->data; | |
9cce02e8 | 121 | |
cd963e18 JA |
122 | return vmsplice_io_u(io_u, nd->pipes[0], len); |
123 | } | |
9cce02e8 | 124 | |
cd963e18 JA |
125 | /* |
126 | * vmsplice() io_u to pipe | |
127 | */ | |
128 | static int vmsplice_io_u_in(struct thread_data *td, struct io_u *io_u) | |
129 | { | |
130 | struct netio_data *nd = td->io_ops->data; | |
ed92ac0c | 131 | |
cd963e18 | 132 | return vmsplice_io_u(io_u, nd->pipes[1], io_u->xfer_buflen); |
9cce02e8 JA |
133 | } |
134 | ||
cd963e18 JA |
135 | /* |
136 | * splice receive - transfer socket data into a pipe using splice, then map | |
137 | * that pipe data into the io_u using vmsplice. | |
138 | */ | |
9cce02e8 JA |
139 | static int fio_netio_splice_in(struct thread_data *td, struct io_u *io_u) |
140 | { | |
141 | int ret; | |
142 | ||
143 | ret = splice_in(td, io_u); | |
cd963e18 JA |
144 | if (ret > 0) |
145 | return vmsplice_io_u_out(td, io_u, ret); | |
9cce02e8 | 146 | |
cd963e18 | 147 | return ret; |
9cce02e8 JA |
148 | } |
149 | ||
cd963e18 JA |
150 | /* |
151 | * splice transmit - map data from the io_u into a pipe by using vmsplice, | |
152 | * then transfer that pipe to a socket using splice. | |
153 | */ | |
9cce02e8 JA |
154 | static int fio_netio_splice_out(struct thread_data *td, struct io_u *io_u) |
155 | { | |
156 | int ret; | |
157 | ||
158 | ret = vmsplice_io_u_in(td, io_u); | |
cd963e18 JA |
159 | if (ret > 0) |
160 | return splice_out(td, io_u, ret); | |
9cce02e8 | 161 | |
cd963e18 | 162 | return ret; |
9cce02e8 JA |
163 | } |
164 | ||
165 | static int fio_netio_send(struct thread_data *td, struct io_u *io_u) | |
166 | { | |
167 | int flags = 0; | |
168 | ||
169 | /* | |
170 | * if we are going to write more, set MSG_MORE | |
171 | */ | |
172 | if (td->this_io_bytes[DDIR_WRITE] + io_u->xfer_buflen < td->o.size) | |
173 | flags = MSG_MORE; | |
174 | ||
175 | return send(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags); | |
176 | } | |
177 | ||
178 | static int fio_netio_recv(struct io_u *io_u) | |
179 | { | |
180 | int flags = MSG_WAITALL; | |
181 | ||
182 | return recv(io_u->file->fd, io_u->xfer_buf, io_u->xfer_buflen, flags); | |
183 | } | |
184 | ||
185 | static int fio_netio_queue(struct thread_data *td, struct io_u *io_u) | |
186 | { | |
187 | struct netio_data *nd = td->io_ops->data; | |
188 | int ret; | |
189 | ||
7101d9c2 JA |
190 | fio_ro_check(td, io_u); |
191 | ||
9cce02e8 JA |
192 | if (io_u->ddir == DDIR_WRITE) { |
193 | if (nd->use_splice) | |
194 | ret = fio_netio_splice_out(td, io_u); | |
195 | else | |
196 | ret = fio_netio_send(td, io_u); | |
d4f12dd0 | 197 | } else if (io_u->ddir == DDIR_READ) { |
9cce02e8 JA |
198 | if (nd->use_splice) |
199 | ret = fio_netio_splice_in(td, io_u); | |
200 | else | |
201 | ret = fio_netio_recv(io_u); | |
d4f12dd0 | 202 | } else |
7a6499da | 203 | ret = 0; /* must be a SYNC */ |
ed92ac0c | 204 | |
cec6b55d | 205 | if (ret != (int) io_u->xfer_buflen) { |
22819ec2 | 206 | if (ret >= 0) { |
cec6b55d JA |
207 | io_u->resid = io_u->xfer_buflen - ret; |
208 | io_u->error = 0; | |
36167d82 | 209 | return FIO_Q_COMPLETED; |
ed92ac0c JA |
210 | } else |
211 | io_u->error = errno; | |
212 | } | |
213 | ||
36167d82 | 214 | if (io_u->error) |
e1161c32 | 215 | td_verror(td, io_u->error, "xfer"); |
ed92ac0c | 216 | |
36167d82 | 217 | return FIO_Q_COMPLETED; |
ed92ac0c JA |
218 | } |
219 | ||
b5af8293 | 220 | static int fio_netio_connect(struct thread_data *td, struct fio_file *f) |
ed92ac0c | 221 | { |
b5af8293 | 222 | struct netio_data *nd = td->io_ops->data; |
ed92ac0c | 223 | |
b5af8293 JA |
224 | f->fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); |
225 | if (f->fd < 0) { | |
226 | td_verror(td, errno, "socket"); | |
227 | return 1; | |
ed92ac0c JA |
228 | } |
229 | ||
b5af8293 JA |
230 | if (connect(f->fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) { |
231 | td_verror(td, errno, "connect"); | |
232 | return 1; | |
ed92ac0c JA |
233 | } |
234 | ||
235 | return 0; | |
ed92ac0c JA |
236 | } |
237 | ||
b5af8293 | 238 | static int fio_netio_accept(struct thread_data *td, struct fio_file *f) |
5fdd124a | 239 | { |
b5af8293 JA |
240 | struct netio_data *nd = td->io_ops->data; |
241 | socklen_t socklen = sizeof(nd->addr); | |
5fdd124a | 242 | struct pollfd pfd; |
b5af8293 | 243 | int ret; |
5fdd124a | 244 | |
6d86144d | 245 | log_info("fio: waiting for connection\n"); |
5fdd124a JA |
246 | |
247 | /* | |
248 | * Accept loop. poll for incoming events, accept them. Repeat until we | |
249 | * have all connections. | |
250 | */ | |
b5af8293 JA |
251 | while (!td->terminate) { |
252 | pfd.fd = nd->listenfd; | |
5fdd124a JA |
253 | pfd.events = POLLIN; |
254 | ||
255 | ret = poll(&pfd, 1, -1); | |
256 | if (ret < 0) { | |
257 | if (errno == EINTR) | |
258 | continue; | |
259 | ||
e1161c32 | 260 | td_verror(td, errno, "poll"); |
5fdd124a JA |
261 | break; |
262 | } else if (!ret) | |
263 | continue; | |
264 | ||
0c09442b JA |
265 | /* |
266 | * should be impossible | |
267 | */ | |
268 | if (!(pfd.revents & POLLIN)) | |
269 | continue; | |
270 | ||
b5af8293 JA |
271 | f->fd = accept(nd->listenfd, (struct sockaddr *) &nd->addr, &socklen); |
272 | if (f->fd < 0) { | |
273 | td_verror(td, errno, "accept"); | |
274 | return 1; | |
275 | } | |
276 | break; | |
277 | } | |
5fdd124a | 278 | |
b5af8293 JA |
279 | return 0; |
280 | } | |
281 | ||
b5af8293 JA |
282 | static int fio_netio_open_file(struct thread_data *td, struct fio_file *f) |
283 | { | |
284 | if (td_read(td)) | |
285 | return fio_netio_accept(td, f); | |
286 | else | |
287 | return fio_netio_connect(td, f); | |
288 | } | |
289 | ||
290 | static int fio_netio_setup_connect(struct thread_data *td, const char *host, | |
291 | unsigned short port) | |
292 | { | |
293 | struct netio_data *nd = td->io_ops->data; | |
294 | ||
295 | nd->addr.sin_family = AF_INET; | |
296 | nd->addr.sin_port = htons(port); | |
297 | ||
298 | if (inet_aton(host, &nd->addr.sin_addr) != 1) { | |
299 | struct hostent *hent; | |
300 | ||
301 | hent = gethostbyname(host); | |
302 | if (!hent) { | |
303 | td_verror(td, errno, "gethostbyname"); | |
304 | return 1; | |
5fdd124a | 305 | } |
b5af8293 JA |
306 | |
307 | memcpy(&nd->addr.sin_addr, hent->h_addr, 4); | |
5fdd124a JA |
308 | } |
309 | ||
310 | return 0; | |
311 | } | |
312 | ||
b5af8293 | 313 | static int fio_netio_setup_listen(struct thread_data *td, short port) |
ed92ac0c | 314 | { |
b5af8293 | 315 | struct netio_data *nd = td->io_ops->data; |
5fdd124a | 316 | int fd, opt; |
ed92ac0c | 317 | |
6bedbfaf | 318 | fd = socket(AF_INET, SOCK_STREAM, IPPROTO_TCP); |
ed92ac0c | 319 | if (fd < 0) { |
e1161c32 | 320 | td_verror(td, errno, "socket"); |
ed92ac0c JA |
321 | return 1; |
322 | } | |
323 | ||
324 | opt = 1; | |
325 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) { | |
e1161c32 | 326 | td_verror(td, errno, "setsockopt"); |
ed92ac0c JA |
327 | return 1; |
328 | } | |
6bedbfaf JA |
329 | #ifdef SO_REUSEPORT |
330 | if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &opt, sizeof(opt)) < 0) { | |
e1161c32 | 331 | td_verror(td, errno, "setsockopt"); |
6bedbfaf JA |
332 | return 1; |
333 | } | |
334 | #endif | |
ed92ac0c | 335 | |
b5af8293 JA |
336 | nd->addr.sin_family = AF_INET; |
337 | nd->addr.sin_addr.s_addr = htonl(INADDR_ANY); | |
338 | nd->addr.sin_port = htons(port); | |
ed92ac0c | 339 | |
b5af8293 | 340 | if (bind(fd, (struct sockaddr *) &nd->addr, sizeof(nd->addr)) < 0) { |
e1161c32 | 341 | td_verror(td, errno, "bind"); |
ed92ac0c JA |
342 | return 1; |
343 | } | |
344 | if (listen(fd, 1) < 0) { | |
e1161c32 | 345 | td_verror(td, errno, "listen"); |
ed92ac0c JA |
346 | return 1; |
347 | } | |
348 | ||
b5af8293 JA |
349 | nd->listenfd = fd; |
350 | return 0; | |
ed92ac0c JA |
351 | } |
352 | ||
9bec88e1 | 353 | static int fio_netio_init(struct thread_data *td) |
ed92ac0c | 354 | { |
b5af8293 | 355 | struct netio_data *nd = td->io_ops->data; |
e01547d2 | 356 | unsigned short port; |
b5af8293 | 357 | char host[64], buf[128]; |
ed92ac0c | 358 | char *sep; |
af52b345 | 359 | int ret; |
ed92ac0c | 360 | |
413dd459 | 361 | if (td_rw(td)) { |
ed92ac0c JA |
362 | log_err("fio: network connections must be read OR write\n"); |
363 | return 1; | |
364 | } | |
16d55aae JA |
365 | if (td_random(td)) { |
366 | log_err("fio: network IO can't be random\n"); | |
367 | return 1; | |
368 | } | |
ed92ac0c | 369 | |
2dc1bbeb | 370 | strcpy(buf, td->o.filename); |
ed92ac0c | 371 | |
9f9214f2 | 372 | sep = strchr(buf, '/'); |
ed92ac0c | 373 | if (!sep) { |
2dc1bbeb | 374 | log_err("fio: bad network host/port <<%s>>\n", td->o.filename); |
ed92ac0c JA |
375 | return 1; |
376 | } | |
377 | ||
378 | *sep = '\0'; | |
379 | sep++; | |
380 | strcpy(host, buf); | |
e01547d2 | 381 | port = atoi(sep); |
ed92ac0c | 382 | |
413dd459 | 383 | if (td_read(td)) { |
b5af8293 | 384 | nd->send_to_net = 0; |
ed92ac0c JA |
385 | ret = fio_netio_setup_listen(td, port); |
386 | } else { | |
b5af8293 | 387 | nd->send_to_net = 1; |
ed92ac0c JA |
388 | ret = fio_netio_setup_connect(td, host, port); |
389 | } | |
390 | ||
7bb48f84 | 391 | return ret; |
ed92ac0c JA |
392 | } |
393 | ||
b5af8293 | 394 | static void fio_netio_cleanup(struct thread_data *td) |
9bec88e1 | 395 | { |
b5af8293 JA |
396 | struct netio_data *nd = td->io_ops->data; |
397 | ||
398 | if (nd) { | |
64b24cd8 JA |
399 | if (nd->listenfd != -1) |
400 | close(nd->listenfd); | |
401 | if (nd->pipes[0] != -1) | |
402 | close(nd->pipes[0]); | |
403 | if (nd->pipes[1] != -1) | |
404 | close(nd->pipes[1]); | |
405 | ||
b5af8293 JA |
406 | free(nd); |
407 | td->io_ops->data = NULL; | |
408 | } | |
409 | } | |
410 | ||
411 | static int fio_netio_setup(struct thread_data *td) | |
412 | { | |
7bb48f84 | 413 | struct netio_data *nd; |
7bb48f84 JA |
414 | |
415 | if (!td->io_ops->data) { | |
416 | nd = malloc(sizeof(*nd));; | |
417 | ||
418 | memset(nd, 0, sizeof(*nd)); | |
419 | nd->listenfd = -1; | |
64b24cd8 | 420 | nd->pipes[0] = nd->pipes[1] = -1; |
7bb48f84 | 421 | td->io_ops->data = nd; |
7bb48f84 | 422 | } |
b5af8293 | 423 | |
9bec88e1 JA |
424 | return 0; |
425 | } | |
426 | ||
9cce02e8 JA |
427 | static int fio_netio_setup_splice(struct thread_data *td) |
428 | { | |
429 | struct netio_data *nd; | |
430 | ||
431 | fio_netio_setup(td); | |
432 | ||
433 | nd = td->io_ops->data; | |
434 | if (nd) { | |
435 | if (pipe(nd->pipes) < 0) | |
436 | return 1; | |
437 | ||
438 | nd->use_splice = 1; | |
439 | return 0; | |
440 | } | |
441 | ||
442 | return 1; | |
443 | } | |
444 | ||
445 | static struct ioengine_ops ioengine_rw = { | |
ed92ac0c JA |
446 | .name = "net", |
447 | .version = FIO_IOOPS_VERSION, | |
ed92ac0c JA |
448 | .prep = fio_netio_prep, |
449 | .queue = fio_netio_queue, | |
ed92ac0c | 450 | .setup = fio_netio_setup, |
9bec88e1 | 451 | .init = fio_netio_init, |
b5af8293 JA |
452 | .cleanup = fio_netio_cleanup, |
453 | .open_file = fio_netio_open_file, | |
454 | .close_file = generic_close_file, | |
ad830ed7 JA |
455 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR | |
456 | FIO_SIGQUIT, | |
ed92ac0c JA |
457 | }; |
458 | ||
9cce02e8 JA |
459 | static struct ioengine_ops ioengine_splice = { |
460 | .name = "netsplice", | |
461 | .version = FIO_IOOPS_VERSION, | |
462 | .prep = fio_netio_prep, | |
463 | .queue = fio_netio_queue, | |
464 | .setup = fio_netio_setup_splice, | |
465 | .init = fio_netio_init, | |
466 | .cleanup = fio_netio_cleanup, | |
467 | .open_file = fio_netio_open_file, | |
468 | .close_file = generic_close_file, | |
ad830ed7 JA |
469 | .flags = FIO_SYNCIO | FIO_DISKLESSIO | FIO_UNIDIR | |
470 | FIO_SIGQUIT, | |
9cce02e8 JA |
471 | }; |
472 | ||
ed92ac0c JA |
473 | static void fio_init fio_netio_register(void) |
474 | { | |
9cce02e8 JA |
475 | register_ioengine(&ioengine_rw); |
476 | register_ioengine(&ioengine_splice); | |
ed92ac0c JA |
477 | } |
478 | ||
479 | static void fio_exit fio_netio_unregister(void) | |
480 | { | |
9cce02e8 JA |
481 | unregister_ioengine(&ioengine_rw); |
482 | unregister_ioengine(&ioengine_splice); | |
ed92ac0c | 483 | } |