Commit | Line | Data |
---|---|---|
fc5c0345 DG |
1 | /* |
2 | * rbd engine | |
3 | * | |
4 | * IO engine using Ceph's librbd to test RADOS Block Devices. | |
5 | * | |
6 | */ | |
7 | ||
8 | #include <rbd/librbd.h> | |
9 | ||
10 | #include "../fio.h" | |
d220c761 | 11 | #include "../optgroup.h" |
a4c4c346 | 12 | #ifdef CONFIG_RBD_BLKIN |
13 | #include <zipkin_c.h> | |
14 | #endif | |
fc5c0345 | 15 | |
53b6c979 PL |
16 | #ifdef CONFIG_RBD_POLL |
17 | /* add for poll */ | |
18 | #include <poll.h> | |
19 | #include <sys/eventfd.h> | |
20 | #endif | |
21 | ||
fc5c0345 DG |
22 | struct fio_rbd_iou { |
23 | struct io_u *io_u; | |
d8b64af2 | 24 | rbd_completion_t completion; |
d8b64af2 | 25 | int io_seen; |
b8ecbef6 | 26 | int io_complete; |
550beaad | 27 | #ifdef CONFIG_RBD_BLKIN |
a4c4c346 | 28 | struct blkin_trace_info info; |
550beaad | 29 | #endif |
fc5c0345 DG |
30 | }; |
31 | ||
32 | struct rbd_data { | |
33 | rados_t cluster; | |
34 | rados_ioctx_t io_ctx; | |
35 | rbd_image_t image; | |
36 | struct io_u **aio_events; | |
6f9961ac | 37 | struct io_u **sort_events; |
53b6c979 | 38 | int fd; /* add for poll */ |
e0f35b53 | 39 | bool connected; |
fc5c0345 DG |
40 | }; |
41 | ||
42 | struct rbd_options { | |
a1f871c7 | 43 | void *pad; |
6e20c6e7 | 44 | char *cluster_name; |
fc5c0345 DG |
45 | char *rbd_name; |
46 | char *pool_name; | |
47 | char *client_name; | |
d7d702c7 | 48 | int busy_poll; |
fc5c0345 DG |
49 | }; |
50 | ||
51 | static struct fio_option options[] = { | |
6e20c6e7 T |
52 | { |
53 | .name = "clustername", | |
54 | .lname = "ceph cluster name", | |
55 | .type = FIO_OPT_STR_STORE, | |
56 | .help = "Cluster name for ceph", | |
57 | .off1 = offsetof(struct rbd_options, cluster_name), | |
58 | .category = FIO_OPT_C_ENGINE, | |
59 | .group = FIO_OPT_G_RBD, | |
60 | }, | |
fc5c0345 | 61 | { |
d8b64af2 JA |
62 | .name = "rbdname", |
63 | .lname = "rbd engine rbdname", | |
64 | .type = FIO_OPT_STR_STORE, | |
65 | .help = "RBD name for RBD engine", | |
66 | .off1 = offsetof(struct rbd_options, rbd_name), | |
67 | .category = FIO_OPT_C_ENGINE, | |
68 | .group = FIO_OPT_G_RBD, | |
69 | }, | |
fc5c0345 | 70 | { |
d7d702c7 JA |
71 | .name = "pool", |
72 | .lname = "rbd engine pool", | |
73 | .type = FIO_OPT_STR_STORE, | |
74 | .help = "Name of the pool hosting the RBD for the RBD engine", | |
75 | .off1 = offsetof(struct rbd_options, pool_name), | |
76 | .category = FIO_OPT_C_ENGINE, | |
77 | .group = FIO_OPT_G_RBD, | |
d8b64af2 | 78 | }, |
fc5c0345 | 79 | { |
d7d702c7 JA |
80 | .name = "clientname", |
81 | .lname = "rbd engine clientname", | |
82 | .type = FIO_OPT_STR_STORE, | |
83 | .help = "Name of the ceph client to access the RBD for the RBD engine", | |
84 | .off1 = offsetof(struct rbd_options, client_name), | |
85 | .category = FIO_OPT_C_ENGINE, | |
86 | .group = FIO_OPT_G_RBD, | |
87 | }, | |
88 | { | |
89 | .name = "busy_poll", | |
90 | .lname = "Busy poll", | |
91 | .type = FIO_OPT_BOOL, | |
92 | .help = "Busy poll for completions instead of sleeping", | |
fea585d4 | 93 | .off1 = offsetof(struct rbd_options, busy_poll), |
d7d702c7 JA |
94 | .def = "0", |
95 | .category = FIO_OPT_C_ENGINE, | |
96 | .group = FIO_OPT_G_RBD, | |
d8b64af2 | 97 | }, |
fc5c0345 | 98 | { |
d8b64af2 JA |
99 | .name = NULL, |
100 | }, | |
fc5c0345 DG |
101 | }; |
102 | ||
103 | static int _fio_setup_rbd_data(struct thread_data *td, | |
104 | struct rbd_data **rbd_data_ptr) | |
105 | { | |
6f9961ac | 106 | struct rbd_data *rbd; |
fc5c0345 | 107 | |
565e784d | 108 | if (td->io_ops_data) |
fc5c0345 DG |
109 | return 0; |
110 | ||
6f9961ac JA |
111 | rbd = calloc(1, sizeof(struct rbd_data)); |
112 | if (!rbd) | |
fc5c0345 DG |
113 | goto failed; |
114 | ||
e0f35b53 PL |
115 | rbd->connected = false; |
116 | ||
53b6c979 PL |
117 | /* add for poll, init fd: -1 */ |
118 | rbd->fd = -1; | |
53b6c979 | 119 | |
6f9961ac JA |
120 | rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *)); |
121 | if (!rbd->aio_events) | |
fc5c0345 DG |
122 | goto failed; |
123 | ||
6f9961ac JA |
124 | rbd->sort_events = calloc(td->o.iodepth, sizeof(struct io_u *)); |
125 | if (!rbd->sort_events) | |
126 | goto failed; | |
fc5c0345 | 127 | |
6f9961ac | 128 | *rbd_data_ptr = rbd; |
fc5c0345 DG |
129 | return 0; |
130 | ||
131 | failed: | |
5a4adfd2 JQ |
132 | if (rbd) { |
133 | if (rbd->aio_events) | |
134 | free(rbd->aio_events); | |
135 | if (rbd->sort_events) | |
136 | free(rbd->sort_events); | |
6f9961ac | 137 | free(rbd); |
5a4adfd2 | 138 | } |
fc5c0345 DG |
139 | return 1; |
140 | ||
141 | } | |
142 | ||
73ae55c3 JA |
143 | #ifdef CONFIG_RBD_POLL |
144 | static bool _fio_rbd_setup_poll(struct rbd_data *rbd) | |
145 | { | |
146 | int r; | |
147 | ||
148 | /* add for rbd poll */ | |
149 | rbd->fd = eventfd(0, EFD_NONBLOCK); | |
150 | if (rbd->fd < 0) { | |
151 | log_err("eventfd failed.\n"); | |
152 | return false; | |
153 | } | |
154 | ||
155 | r = rbd_set_image_notification(rbd->image, rbd->fd, EVENT_TYPE_EVENTFD); | |
156 | if (r < 0) { | |
157 | log_err("rbd_set_image_notification failed.\n"); | |
158 | close(rbd->fd); | |
159 | rbd->fd = -1; | |
160 | return false; | |
161 | } | |
162 | ||
163 | return true; | |
164 | } | |
165 | #else | |
166 | static bool _fio_rbd_setup_poll(struct rbd_data *rbd) | |
167 | { | |
168 | return true; | |
169 | } | |
170 | #endif | |
171 | ||
fc5c0345 DG |
172 | static int _fio_rbd_connect(struct thread_data *td) |
173 | { | |
565e784d | 174 | struct rbd_data *rbd = td->io_ops_data; |
fc5c0345 DG |
175 | struct rbd_options *o = td->eo; |
176 | int r; | |
177 | ||
6e20c6e7 T |
178 | if (o->cluster_name) { |
179 | char *client_name = NULL; | |
180 | ||
181 | /* | |
89e5ec96 | 182 | * If we specify cluser name, the rados_create2 |
6e20c6e7 T |
183 | * will not assume 'client.'. name is considered |
184 | * as a full type.id namestr | |
185 | */ | |
89e5ec96 JQ |
186 | if (o->client_name) { |
187 | if (!index(o->client_name, '.')) { | |
188 | client_name = calloc(1, strlen("client.") + | |
189 | strlen(o->client_name) + 1); | |
190 | strcat(client_name, "client."); | |
6ec0dcd1 JQ |
191 | strcat(client_name, o->client_name); |
192 | } else { | |
193 | client_name = o->client_name; | |
89e5ec96 | 194 | } |
6e20c6e7 | 195 | } |
6ec0dcd1 | 196 | |
6e20c6e7 | 197 | r = rados_create2(&rbd->cluster, o->cluster_name, |
6ec0dcd1 JQ |
198 | client_name, 0); |
199 | ||
200 | if (client_name && !index(o->client_name, '.')) | |
201 | free(client_name); | |
6e20c6e7 T |
202 | } else |
203 | r = rados_create(&rbd->cluster, o->client_name); | |
204 | ||
fc5c0345 DG |
205 | if (r < 0) { |
206 | log_err("rados_create failed.\n"); | |
207 | goto failed_early; | |
208 | } | |
209 | ||
6f9961ac | 210 | r = rados_conf_read_file(rbd->cluster, NULL); |
fc5c0345 DG |
211 | if (r < 0) { |
212 | log_err("rados_conf_read_file failed.\n"); | |
213 | goto failed_early; | |
214 | } | |
215 | ||
6f9961ac | 216 | r = rados_connect(rbd->cluster); |
fc5c0345 DG |
217 | if (r < 0) { |
218 | log_err("rados_connect failed.\n"); | |
219 | goto failed_shutdown; | |
220 | } | |
221 | ||
6f9961ac | 222 | r = rados_ioctx_create(rbd->cluster, o->pool_name, &rbd->io_ctx); |
fc5c0345 DG |
223 | if (r < 0) { |
224 | log_err("rados_ioctx_create failed.\n"); | |
225 | goto failed_shutdown; | |
226 | } | |
227 | ||
6f9961ac | 228 | r = rbd_open(rbd->io_ctx, o->rbd_name, &rbd->image, NULL /*snap */ ); |
fc5c0345 DG |
229 | if (r < 0) { |
230 | log_err("rbd_open failed.\n"); | |
231 | goto failed_open; | |
232 | } | |
53b6c979 | 233 | |
73ae55c3 JA |
234 | if (!_fio_rbd_setup_poll(rbd)) |
235 | goto failed_poll; | |
53b6c979 | 236 | |
fc5c0345 DG |
237 | return 0; |
238 | ||
73ae55c3 JA |
239 | failed_poll: |
240 | rbd_close(rbd->image); | |
241 | rbd->image = NULL; | |
fc5c0345 | 242 | failed_open: |
6f9961ac JA |
243 | rados_ioctx_destroy(rbd->io_ctx); |
244 | rbd->io_ctx = NULL; | |
fc5c0345 | 245 | failed_shutdown: |
6f9961ac JA |
246 | rados_shutdown(rbd->cluster); |
247 | rbd->cluster = NULL; | |
fc5c0345 DG |
248 | failed_early: |
249 | return 1; | |
250 | } | |
251 | ||
6f9961ac | 252 | static void _fio_rbd_disconnect(struct rbd_data *rbd) |
fc5c0345 | 253 | { |
6f9961ac | 254 | if (!rbd) |
fc5c0345 DG |
255 | return; |
256 | ||
53b6c979 | 257 | /* close eventfd */ |
73ae55c3 | 258 | if (rbd->fd != -1) { |
53b6c979 PL |
259 | close(rbd->fd); |
260 | rbd->fd = -1; | |
261 | } | |
53b6c979 | 262 | |
fc5c0345 | 263 | /* shutdown everything */ |
6f9961ac JA |
264 | if (rbd->image) { |
265 | rbd_close(rbd->image); | |
266 | rbd->image = NULL; | |
fc5c0345 DG |
267 | } |
268 | ||
6f9961ac JA |
269 | if (rbd->io_ctx) { |
270 | rados_ioctx_destroy(rbd->io_ctx); | |
271 | rbd->io_ctx = NULL; | |
fc5c0345 DG |
272 | } |
273 | ||
6f9961ac JA |
274 | if (rbd->cluster) { |
275 | rados_shutdown(rbd->cluster); | |
276 | rbd->cluster = NULL; | |
fc5c0345 DG |
277 | } |
278 | } | |
279 | ||
d8b64af2 | 280 | static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data) |
fc5c0345 | 281 | { |
dbf388d2 JA |
282 | struct fio_rbd_iou *fri = data; |
283 | struct io_u *io_u = fri->io_u; | |
d8b64af2 | 284 | ssize_t ret; |
fc5c0345 | 285 | |
d8b64af2 JA |
286 | /* |
287 | * Looks like return value is 0 for success, or < 0 for | |
288 | * a specific error. So we have to assume that it can't do | |
289 | * partial completions. | |
290 | */ | |
291 | ret = rbd_aio_get_return_value(fri->completion); | |
292 | if (ret < 0) { | |
d7658ced | 293 | io_u->error = -ret; |
d8b64af2 JA |
294 | io_u->resid = io_u->xfer_buflen; |
295 | } else | |
296 | io_u->error = 0; | |
20cf5aab JD |
297 | |
298 | fri->io_complete = 1; | |
d8b64af2 | 299 | } |
fc5c0345 | 300 | |
d8b64af2 JA |
301 | static struct io_u *fio_rbd_event(struct thread_data *td, int event) |
302 | { | |
565e784d | 303 | struct rbd_data *rbd = td->io_ops_data; |
fc5c0345 | 304 | |
6f9961ac | 305 | return rbd->aio_events[event]; |
fc5c0345 DG |
306 | } |
307 | ||
6f9961ac | 308 | static inline int fri_check_complete(struct rbd_data *rbd, struct io_u *io_u, |
d8b64af2 | 309 | unsigned int *events) |
fc5c0345 | 310 | { |
d8b64af2 | 311 | struct fio_rbd_iou *fri = io_u->engine_data; |
fc5c0345 | 312 | |
b8ecbef6 | 313 | if (fri->io_complete) { |
d8b64af2 | 314 | fri->io_seen = 1; |
6f9961ac | 315 | rbd->aio_events[*events] = io_u; |
d8b64af2 | 316 | (*events)++; |
fc5c0345 | 317 | |
d8b64af2 JA |
318 | rbd_aio_release(fri->completion); |
319 | return 1; | |
320 | } | |
fc5c0345 | 321 | |
d8b64af2 | 322 | return 0; |
fc5c0345 DG |
323 | } |
324 | ||
6f9961ac JA |
325 | static inline int rbd_io_u_seen(struct io_u *io_u) |
326 | { | |
327 | struct fio_rbd_iou *fri = io_u->engine_data; | |
328 | ||
329 | return fri->io_seen; | |
330 | } | |
331 | ||
332 | static void rbd_io_u_wait_complete(struct io_u *io_u) | |
333 | { | |
334 | struct fio_rbd_iou *fri = io_u->engine_data; | |
335 | ||
336 | rbd_aio_wait_for_complete(fri->completion); | |
337 | } | |
338 | ||
339 | static int rbd_io_u_cmp(const void *p1, const void *p2) | |
340 | { | |
341 | const struct io_u **a = (const struct io_u **) p1; | |
342 | const struct io_u **b = (const struct io_u **) p2; | |
343 | uint64_t at, bt; | |
344 | ||
345 | at = utime_since_now(&(*a)->start_time); | |
346 | bt = utime_since_now(&(*b)->start_time); | |
347 | ||
348 | if (at < bt) | |
349 | return -1; | |
350 | else if (at == bt) | |
351 | return 0; | |
352 | else | |
353 | return 1; | |
354 | } | |
355 | ||
d8b64af2 JA |
356 | static int rbd_iter_events(struct thread_data *td, unsigned int *events, |
357 | unsigned int min_evts, int wait) | |
82340a9f | 358 | { |
565e784d | 359 | struct rbd_data *rbd = td->io_ops_data; |
d8b64af2 JA |
360 | unsigned int this_events = 0; |
361 | struct io_u *io_u; | |
53b6c979 PL |
362 | int i, sidx = 0; |
363 | ||
364 | #ifdef CONFIG_RBD_POLL | |
365 | int ret = 0; | |
366 | int event_num = 0; | |
367 | struct fio_rbd_iou *fri = NULL; | |
368 | rbd_completion_t comps[min_evts]; | |
82340a9f | 369 | |
53b6c979 PL |
370 | struct pollfd pfd; |
371 | pfd.fd = rbd->fd; | |
372 | pfd.events = POLLIN; | |
373 | ||
374 | ret = poll(&pfd, 1, -1); | |
73ae55c3 | 375 | if (ret <= 0) |
53b6c979 | 376 | return 0; |
73ae55c3 | 377 | |
53b6c979 PL |
378 | assert(pfd.revents & POLLIN); |
379 | ||
380 | event_num = rbd_poll_io_events(rbd->image, comps, min_evts); | |
381 | ||
382 | for (i = 0; i < event_num; i++) { | |
383 | fri = rbd_aio_get_arg(comps[i]); | |
384 | io_u = fri->io_u; | |
385 | #else | |
d8b64af2 | 386 | io_u_qiter(&td->io_u_all, io_u, i) { |
53b6c979 | 387 | #endif |
d8b64af2 JA |
388 | if (!(io_u->flags & IO_U_F_FLIGHT)) |
389 | continue; | |
6f9961ac | 390 | if (rbd_io_u_seen(io_u)) |
d8b64af2 | 391 | continue; |
82340a9f | 392 | |
6f9961ac | 393 | if (fri_check_complete(rbd, io_u, events)) |
d8b64af2 | 394 | this_events++; |
6f9961ac JA |
395 | else if (wait) |
396 | rbd->sort_events[sidx++] = io_u; | |
397 | } | |
82340a9f | 398 | |
6f9961ac JA |
399 | if (!wait || !sidx) |
400 | return this_events; | |
401 | ||
402 | /* | |
403 | * Sort events, oldest issue first, then wait on as many as we | |
404 | * need in order of age. If we have enough events, stop waiting, | |
405 | * and just check if any of the older ones are done. | |
406 | */ | |
407 | if (sidx > 1) | |
408 | qsort(rbd->sort_events, sidx, sizeof(struct io_u *), rbd_io_u_cmp); | |
409 | ||
410 | for (i = 0; i < sidx; i++) { | |
411 | io_u = rbd->sort_events[i]; | |
412 | ||
413 | if (fri_check_complete(rbd, io_u, events)) { | |
414 | this_events++; | |
415 | continue; | |
d8b64af2 | 416 | } |
6f9961ac JA |
417 | |
418 | /* | |
419 | * Stop waiting when we have enough, but continue checking | |
420 | * all pending IOs if they are complete. | |
421 | */ | |
d8b64af2 | 422 | if (*events >= min_evts) |
6f9961ac JA |
423 | continue; |
424 | ||
425 | rbd_io_u_wait_complete(io_u); | |
426 | ||
427 | if (fri_check_complete(rbd, io_u, events)) | |
428 | this_events++; | |
d8b64af2 | 429 | } |
fc5c0345 | 430 | |
d8b64af2 | 431 | return this_events; |
fc5c0345 DG |
432 | } |
433 | ||
434 | static int fio_rbd_getevents(struct thread_data *td, unsigned int min, | |
1f440ece | 435 | unsigned int max, const struct timespec *t) |
fc5c0345 | 436 | { |
d8b64af2 | 437 | unsigned int this_events, events = 0; |
d7d702c7 | 438 | struct rbd_options *o = td->eo; |
d8b64af2 | 439 | int wait = 0; |
fc5c0345 DG |
440 | |
441 | do { | |
d8b64af2 | 442 | this_events = rbd_iter_events(td, &events, min, wait); |
fc5c0345 | 443 | |
d8b64af2 | 444 | if (events >= min) |
fc5c0345 | 445 | break; |
d8b64af2 JA |
446 | if (this_events) |
447 | continue; | |
fc5c0345 | 448 | |
d7d702c7 JA |
449 | if (!o->busy_poll) |
450 | wait = 1; | |
451 | else | |
452 | nop; | |
fc5c0345 DG |
453 | } while (1); |
454 | ||
455 | return events; | |
456 | } | |
457 | ||
458 | static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u) | |
459 | { | |
565e784d | 460 | struct rbd_data *rbd = td->io_ops_data; |
d8b64af2 JA |
461 | struct fio_rbd_iou *fri = io_u->engine_data; |
462 | int r = -1; | |
fc5c0345 DG |
463 | |
464 | fio_ro_check(td, io_u); | |
465 | ||
d8b64af2 | 466 | fri->io_seen = 0; |
b8ecbef6 | 467 | fri->io_complete = 0; |
d8b64af2 | 468 | |
dbf388d2 | 469 | r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb, |
d8b64af2 | 470 | &fri->completion); |
dbf388d2 JA |
471 | if (r < 0) { |
472 | log_err("rbd_aio_create_completion failed.\n"); | |
473 | goto failed; | |
474 | } | |
fc5c0345 | 475 | |
dbf388d2 | 476 | if (io_u->ddir == DDIR_WRITE) { |
a4c4c346 | 477 | #ifdef CONFIG_RBD_BLKIN |
478 | blkin_init_trace_info(&fri->info); | |
479 | r = rbd_aio_write_traced(rbd->image, io_u->offset, io_u->xfer_buflen, | |
480 | io_u->xfer_buf, fri->completion, &fri->info); | |
481 | #else | |
6f9961ac JA |
482 | r = rbd_aio_write(rbd->image, io_u->offset, io_u->xfer_buflen, |
483 | io_u->xfer_buf, fri->completion); | |
a4c4c346 | 484 | #endif |
fc5c0345 DG |
485 | if (r < 0) { |
486 | log_err("rbd_aio_write failed.\n"); | |
dbf388d2 | 487 | goto failed_comp; |
fc5c0345 DG |
488 | } |
489 | ||
490 | } else if (io_u->ddir == DDIR_READ) { | |
a4c4c346 | 491 | #ifdef CONFIG_RBD_BLKIN |
492 | blkin_init_trace_info(&fri->info); | |
493 | r = rbd_aio_read_traced(rbd->image, io_u->offset, io_u->xfer_buflen, | |
494 | io_u->xfer_buf, fri->completion, &fri->info); | |
495 | #else | |
6f9961ac JA |
496 | r = rbd_aio_read(rbd->image, io_u->offset, io_u->xfer_buflen, |
497 | io_u->xfer_buf, fri->completion); | |
a4c4c346 | 498 | #endif |
fc5c0345 DG |
499 | |
500 | if (r < 0) { | |
501 | log_err("rbd_aio_read failed.\n"); | |
dbf388d2 | 502 | goto failed_comp; |
fc5c0345 | 503 | } |
dbf388d2 | 504 | } else if (io_u->ddir == DDIR_TRIM) { |
6f9961ac JA |
505 | r = rbd_aio_discard(rbd->image, io_u->offset, |
506 | io_u->xfer_buflen, fri->completion); | |
82340a9f | 507 | if (r < 0) { |
dbf388d2 JA |
508 | log_err("rbd_aio_discard failed.\n"); |
509 | goto failed_comp; | |
82340a9f | 510 | } |
dbf388d2 | 511 | } else if (io_u->ddir == DDIR_SYNC) { |
6f9961ac | 512 | r = rbd_aio_flush(rbd->image, fri->completion); |
fc5c0345 DG |
513 | if (r < 0) { |
514 | log_err("rbd_flush failed.\n"); | |
dbf388d2 | 515 | goto failed_comp; |
fc5c0345 | 516 | } |
fc5c0345 DG |
517 | } else { |
518 | dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__, | |
519 | io_u->ddir); | |
ef2f4a50 | 520 | r = -EINVAL; |
dbf388d2 | 521 | goto failed_comp; |
fc5c0345 DG |
522 | } |
523 | ||
524 | return FIO_Q_QUEUED; | |
dbf388d2 JA |
525 | failed_comp: |
526 | rbd_aio_release(fri->completion); | |
fc5c0345 | 527 | failed: |
d7658ced | 528 | io_u->error = -r; |
fc5c0345 DG |
529 | td_verror(td, io_u->error, "xfer"); |
530 | return FIO_Q_COMPLETED; | |
531 | } | |
532 | ||
533 | static int fio_rbd_init(struct thread_data *td) | |
534 | { | |
535 | int r; | |
e0f35b53 | 536 | struct rbd_data *rbd = td->io_ops_data; |
35448275 JA |
537 | |
538 | if (rbd->connected) | |
e0f35b53 | 539 | return 0; |
fc5c0345 DG |
540 | |
541 | r = _fio_rbd_connect(td); | |
542 | if (r) { | |
543 | log_err("fio_rbd_connect failed, return code: %d .\n", r); | |
544 | goto failed; | |
545 | } | |
546 | ||
547 | return 0; | |
548 | ||
549 | failed: | |
550 | return 1; | |
fc5c0345 DG |
551 | } |
552 | ||
553 | static void fio_rbd_cleanup(struct thread_data *td) | |
554 | { | |
565e784d | 555 | struct rbd_data *rbd = td->io_ops_data; |
fc5c0345 | 556 | |
6f9961ac JA |
557 | if (rbd) { |
558 | _fio_rbd_disconnect(rbd); | |
559 | free(rbd->aio_events); | |
560 | free(rbd->sort_events); | |
561 | free(rbd); | |
fc5c0345 | 562 | } |
fc5c0345 DG |
563 | } |
564 | ||
565 | static int fio_rbd_setup(struct thread_data *td) | |
566 | { | |
fc5c0345 DG |
567 | rbd_image_info_t info; |
568 | struct fio_file *f; | |
6f9961ac | 569 | struct rbd_data *rbd = NULL; |
6f9961ac | 570 | int r; |
fc5c0345 | 571 | |
fc5c0345 | 572 | /* allocate engine specific structure to deal with librbd. */ |
6f9961ac | 573 | r = _fio_setup_rbd_data(td, &rbd); |
fc5c0345 DG |
574 | if (r) { |
575 | log_err("fio_setup_rbd_data failed.\n"); | |
576 | goto cleanup; | |
577 | } | |
565e784d | 578 | td->io_ops_data = rbd; |
fc5c0345 | 579 | |
d8b64af2 JA |
580 | /* librbd does not allow us to run first in the main thread and later |
581 | * in a fork child. It needs to be the same process context all the | |
582 | * time. | |
fc5c0345 DG |
583 | */ |
584 | td->o.use_thread = 1; | |
585 | ||
586 | /* connect in the main thread to determine to determine | |
587 | * the size of the given RADOS block device. And disconnect | |
588 | * later on. | |
589 | */ | |
590 | r = _fio_rbd_connect(td); | |
591 | if (r) { | |
592 | log_err("fio_rbd_connect failed.\n"); | |
593 | goto cleanup; | |
594 | } | |
e0f35b53 | 595 | rbd->connected = true; |
fc5c0345 DG |
596 | |
597 | /* get size of the RADOS block device */ | |
6f9961ac | 598 | r = rbd_stat(rbd->image, &info, sizeof(info)); |
fc5c0345 DG |
599 | if (r < 0) { |
600 | log_err("rbd_status failed.\n"); | |
6747f469 | 601 | goto cleanup; |
206c546d PL |
602 | } else if (info.size == 0) { |
603 | log_err("image size should be larger than zero.\n"); | |
604 | r = -EINVAL; | |
6747f469 | 605 | goto cleanup; |
fc5c0345 | 606 | } |
206c546d | 607 | |
50ea012c | 608 | dprint(FD_IO, "rbd-engine: image size: %" PRIu64 "\n", info.size); |
fc5c0345 DG |
609 | |
610 | /* taken from "net" engine. Pretend we deal with files, | |
611 | * even if we do not have any ideas about files. | |
612 | * The size of the RBD is set instead of a artificial file. | |
613 | */ | |
614 | if (!td->files_index) { | |
5903e7b7 | 615 | add_file(td, td->o.filename ? : "rbd", 0, 0); |
fc5c0345 | 616 | td->o.nr_files = td->o.nr_files ? : 1; |
b53f2c54 | 617 | td->o.open_files++; |
fc5c0345 DG |
618 | } |
619 | f = td->files[0]; | |
620 | f->real_file_size = info.size; | |
621 | ||
fc5c0345 DG |
622 | return 0; |
623 | ||
fc5c0345 DG |
624 | cleanup: |
625 | fio_rbd_cleanup(td); | |
626 | return r; | |
627 | } | |
628 | ||
629 | static int fio_rbd_open(struct thread_data *td, struct fio_file *f) | |
630 | { | |
631 | return 0; | |
632 | } | |
633 | ||
d9b100fc JA |
634 | static int fio_rbd_invalidate(struct thread_data *td, struct fio_file *f) |
635 | { | |
903b2812 | 636 | #if defined(CONFIG_RBD_INVAL) |
565e784d | 637 | struct rbd_data *rbd = td->io_ops_data; |
903b2812 | 638 | |
6f9961ac | 639 | return rbd_invalidate_cache(rbd->image); |
903b2812 | 640 | #else |
d9b100fc | 641 | return 0; |
903b2812 | 642 | #endif |
d9b100fc JA |
643 | } |
644 | ||
fc5c0345 DG |
645 | static void fio_rbd_io_u_free(struct thread_data *td, struct io_u *io_u) |
646 | { | |
d8b64af2 | 647 | struct fio_rbd_iou *fri = io_u->engine_data; |
fc5c0345 | 648 | |
d8b64af2 | 649 | if (fri) { |
fc5c0345 | 650 | io_u->engine_data = NULL; |
d8b64af2 | 651 | free(fri); |
fc5c0345 DG |
652 | } |
653 | } | |
654 | ||
655 | static int fio_rbd_io_u_init(struct thread_data *td, struct io_u *io_u) | |
656 | { | |
d8b64af2 | 657 | struct fio_rbd_iou *fri; |
fc5c0345 | 658 | |
d8b64af2 JA |
659 | fri = calloc(1, sizeof(*fri)); |
660 | fri->io_u = io_u; | |
661 | io_u->engine_data = fri; | |
fc5c0345 DG |
662 | return 0; |
663 | } | |
664 | ||
10aa136b | 665 | static struct ioengine_ops ioengine = { |
d9b100fc JA |
666 | .name = "rbd", |
667 | .version = FIO_IOOPS_VERSION, | |
668 | .setup = fio_rbd_setup, | |
669 | .init = fio_rbd_init, | |
670 | .queue = fio_rbd_queue, | |
671 | .getevents = fio_rbd_getevents, | |
672 | .event = fio_rbd_event, | |
673 | .cleanup = fio_rbd_cleanup, | |
674 | .open_file = fio_rbd_open, | |
675 | .invalidate = fio_rbd_invalidate, | |
676 | .options = options, | |
677 | .io_u_init = fio_rbd_io_u_init, | |
678 | .io_u_free = fio_rbd_io_u_free, | |
679 | .option_struct_size = sizeof(struct rbd_options), | |
fc5c0345 DG |
680 | }; |
681 | ||
682 | static void fio_init fio_rbd_register(void) | |
683 | { | |
684 | register_ioengine(&ioengine); | |
685 | } | |
686 | ||
687 | static void fio_exit fio_rbd_unregister(void) | |
688 | { | |
689 | unregister_ioengine(&ioengine); | |
690 | } |