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" |
fc5c0345 | 12 | |
53b6c979 PL |
13 | #ifdef CONFIG_RBD_POLL |
14 | /* add for poll */ | |
15 | #include <poll.h> | |
16 | #include <sys/eventfd.h> | |
17 | #endif | |
18 | ||
fc5c0345 DG |
19 | struct fio_rbd_iou { |
20 | struct io_u *io_u; | |
d8b64af2 | 21 | rbd_completion_t completion; |
d8b64af2 | 22 | int io_seen; |
b8ecbef6 | 23 | int io_complete; |
fc5c0345 DG |
24 | }; |
25 | ||
26 | struct rbd_data { | |
27 | rados_t cluster; | |
28 | rados_ioctx_t io_ctx; | |
29 | rbd_image_t image; | |
30 | struct io_u **aio_events; | |
6f9961ac | 31 | struct io_u **sort_events; |
53b6c979 | 32 | int fd; /* add for poll */ |
e0f35b53 | 33 | bool connected; |
fc5c0345 DG |
34 | }; |
35 | ||
36 | struct rbd_options { | |
a1f871c7 | 37 | void *pad; |
6e20c6e7 | 38 | char *cluster_name; |
fc5c0345 DG |
39 | char *rbd_name; |
40 | char *pool_name; | |
41 | char *client_name; | |
d7d702c7 | 42 | int busy_poll; |
fc5c0345 DG |
43 | }; |
44 | ||
45 | static struct fio_option options[] = { | |
6e20c6e7 T |
46 | { |
47 | .name = "clustername", | |
48 | .lname = "ceph cluster name", | |
49 | .type = FIO_OPT_STR_STORE, | |
50 | .help = "Cluster name for ceph", | |
51 | .off1 = offsetof(struct rbd_options, cluster_name), | |
52 | .category = FIO_OPT_C_ENGINE, | |
53 | .group = FIO_OPT_G_RBD, | |
54 | }, | |
fc5c0345 | 55 | { |
d8b64af2 JA |
56 | .name = "rbdname", |
57 | .lname = "rbd engine rbdname", | |
58 | .type = FIO_OPT_STR_STORE, | |
59 | .help = "RBD name for RBD engine", | |
60 | .off1 = offsetof(struct rbd_options, rbd_name), | |
61 | .category = FIO_OPT_C_ENGINE, | |
62 | .group = FIO_OPT_G_RBD, | |
63 | }, | |
fc5c0345 | 64 | { |
d7d702c7 JA |
65 | .name = "pool", |
66 | .lname = "rbd engine pool", | |
67 | .type = FIO_OPT_STR_STORE, | |
68 | .help = "Name of the pool hosting the RBD for the RBD engine", | |
69 | .off1 = offsetof(struct rbd_options, pool_name), | |
70 | .category = FIO_OPT_C_ENGINE, | |
71 | .group = FIO_OPT_G_RBD, | |
d8b64af2 | 72 | }, |
fc5c0345 | 73 | { |
d7d702c7 JA |
74 | .name = "clientname", |
75 | .lname = "rbd engine clientname", | |
76 | .type = FIO_OPT_STR_STORE, | |
77 | .help = "Name of the ceph client to access the RBD for the RBD engine", | |
78 | .off1 = offsetof(struct rbd_options, client_name), | |
79 | .category = FIO_OPT_C_ENGINE, | |
80 | .group = FIO_OPT_G_RBD, | |
81 | }, | |
82 | { | |
83 | .name = "busy_poll", | |
84 | .lname = "Busy poll", | |
85 | .type = FIO_OPT_BOOL, | |
86 | .help = "Busy poll for completions instead of sleeping", | |
fea585d4 | 87 | .off1 = offsetof(struct rbd_options, busy_poll), |
d7d702c7 JA |
88 | .def = "0", |
89 | .category = FIO_OPT_C_ENGINE, | |
90 | .group = FIO_OPT_G_RBD, | |
d8b64af2 | 91 | }, |
fc5c0345 | 92 | { |
d8b64af2 JA |
93 | .name = NULL, |
94 | }, | |
fc5c0345 DG |
95 | }; |
96 | ||
97 | static int _fio_setup_rbd_data(struct thread_data *td, | |
98 | struct rbd_data **rbd_data_ptr) | |
99 | { | |
6f9961ac | 100 | struct rbd_data *rbd; |
fc5c0345 | 101 | |
565e784d | 102 | if (td->io_ops_data) |
fc5c0345 DG |
103 | return 0; |
104 | ||
6f9961ac JA |
105 | rbd = calloc(1, sizeof(struct rbd_data)); |
106 | if (!rbd) | |
fc5c0345 DG |
107 | goto failed; |
108 | ||
e0f35b53 PL |
109 | rbd->connected = false; |
110 | ||
53b6c979 PL |
111 | /* add for poll, init fd: -1 */ |
112 | rbd->fd = -1; | |
53b6c979 | 113 | |
6f9961ac JA |
114 | rbd->aio_events = calloc(td->o.iodepth, sizeof(struct io_u *)); |
115 | if (!rbd->aio_events) | |
fc5c0345 DG |
116 | goto failed; |
117 | ||
6f9961ac JA |
118 | rbd->sort_events = calloc(td->o.iodepth, sizeof(struct io_u *)); |
119 | if (!rbd->sort_events) | |
120 | goto failed; | |
fc5c0345 | 121 | |
6f9961ac | 122 | *rbd_data_ptr = rbd; |
fc5c0345 DG |
123 | return 0; |
124 | ||
125 | failed: | |
5a4adfd2 JQ |
126 | if (rbd) { |
127 | if (rbd->aio_events) | |
128 | free(rbd->aio_events); | |
129 | if (rbd->sort_events) | |
130 | free(rbd->sort_events); | |
6f9961ac | 131 | free(rbd); |
5a4adfd2 | 132 | } |
fc5c0345 DG |
133 | return 1; |
134 | ||
135 | } | |
136 | ||
73ae55c3 JA |
137 | #ifdef CONFIG_RBD_POLL |
138 | static bool _fio_rbd_setup_poll(struct rbd_data *rbd) | |
139 | { | |
140 | int r; | |
141 | ||
142 | /* add for rbd poll */ | |
4001b1ba | 143 | rbd->fd = eventfd(0, EFD_SEMAPHORE); |
73ae55c3 JA |
144 | if (rbd->fd < 0) { |
145 | log_err("eventfd failed.\n"); | |
146 | return false; | |
147 | } | |
148 | ||
149 | r = rbd_set_image_notification(rbd->image, rbd->fd, EVENT_TYPE_EVENTFD); | |
150 | if (r < 0) { | |
151 | log_err("rbd_set_image_notification failed.\n"); | |
152 | close(rbd->fd); | |
153 | rbd->fd = -1; | |
154 | return false; | |
155 | } | |
156 | ||
157 | return true; | |
158 | } | |
159 | #else | |
160 | static bool _fio_rbd_setup_poll(struct rbd_data *rbd) | |
161 | { | |
162 | return true; | |
163 | } | |
164 | #endif | |
165 | ||
fc5c0345 DG |
166 | static int _fio_rbd_connect(struct thread_data *td) |
167 | { | |
565e784d | 168 | struct rbd_data *rbd = td->io_ops_data; |
fc5c0345 DG |
169 | struct rbd_options *o = td->eo; |
170 | int r; | |
171 | ||
6e20c6e7 T |
172 | if (o->cluster_name) { |
173 | char *client_name = NULL; | |
174 | ||
175 | /* | |
89e5ec96 | 176 | * If we specify cluser name, the rados_create2 |
6e20c6e7 T |
177 | * will not assume 'client.'. name is considered |
178 | * as a full type.id namestr | |
179 | */ | |
89e5ec96 JQ |
180 | if (o->client_name) { |
181 | if (!index(o->client_name, '.')) { | |
182 | client_name = calloc(1, strlen("client.") + | |
183 | strlen(o->client_name) + 1); | |
184 | strcat(client_name, "client."); | |
6ec0dcd1 JQ |
185 | strcat(client_name, o->client_name); |
186 | } else { | |
187 | client_name = o->client_name; | |
89e5ec96 | 188 | } |
6e20c6e7 | 189 | } |
6ec0dcd1 | 190 | |
6e20c6e7 | 191 | r = rados_create2(&rbd->cluster, o->cluster_name, |
6ec0dcd1 JQ |
192 | client_name, 0); |
193 | ||
194 | if (client_name && !index(o->client_name, '.')) | |
195 | free(client_name); | |
6e20c6e7 T |
196 | } else |
197 | r = rados_create(&rbd->cluster, o->client_name); | |
198 | ||
fc5c0345 DG |
199 | if (r < 0) { |
200 | log_err("rados_create failed.\n"); | |
201 | goto failed_early; | |
202 | } | |
203 | ||
6f9961ac | 204 | r = rados_conf_read_file(rbd->cluster, NULL); |
fc5c0345 DG |
205 | if (r < 0) { |
206 | log_err("rados_conf_read_file failed.\n"); | |
207 | goto failed_early; | |
208 | } | |
209 | ||
6f9961ac | 210 | r = rados_connect(rbd->cluster); |
fc5c0345 DG |
211 | if (r < 0) { |
212 | log_err("rados_connect failed.\n"); | |
213 | goto failed_shutdown; | |
214 | } | |
215 | ||
6f9961ac | 216 | r = rados_ioctx_create(rbd->cluster, o->pool_name, &rbd->io_ctx); |
fc5c0345 DG |
217 | if (r < 0) { |
218 | log_err("rados_ioctx_create failed.\n"); | |
219 | goto failed_shutdown; | |
220 | } | |
221 | ||
6f9961ac | 222 | r = rbd_open(rbd->io_ctx, o->rbd_name, &rbd->image, NULL /*snap */ ); |
fc5c0345 DG |
223 | if (r < 0) { |
224 | log_err("rbd_open failed.\n"); | |
225 | goto failed_open; | |
226 | } | |
53b6c979 | 227 | |
73ae55c3 JA |
228 | if (!_fio_rbd_setup_poll(rbd)) |
229 | goto failed_poll; | |
53b6c979 | 230 | |
fc5c0345 DG |
231 | return 0; |
232 | ||
73ae55c3 JA |
233 | failed_poll: |
234 | rbd_close(rbd->image); | |
235 | rbd->image = NULL; | |
fc5c0345 | 236 | failed_open: |
6f9961ac JA |
237 | rados_ioctx_destroy(rbd->io_ctx); |
238 | rbd->io_ctx = NULL; | |
fc5c0345 | 239 | failed_shutdown: |
6f9961ac JA |
240 | rados_shutdown(rbd->cluster); |
241 | rbd->cluster = NULL; | |
fc5c0345 DG |
242 | failed_early: |
243 | return 1; | |
244 | } | |
245 | ||
6f9961ac | 246 | static void _fio_rbd_disconnect(struct rbd_data *rbd) |
fc5c0345 | 247 | { |
6f9961ac | 248 | if (!rbd) |
fc5c0345 DG |
249 | return; |
250 | ||
53b6c979 | 251 | /* close eventfd */ |
73ae55c3 | 252 | if (rbd->fd != -1) { |
53b6c979 PL |
253 | close(rbd->fd); |
254 | rbd->fd = -1; | |
255 | } | |
53b6c979 | 256 | |
fc5c0345 | 257 | /* shutdown everything */ |
6f9961ac JA |
258 | if (rbd->image) { |
259 | rbd_close(rbd->image); | |
260 | rbd->image = NULL; | |
fc5c0345 DG |
261 | } |
262 | ||
6f9961ac JA |
263 | if (rbd->io_ctx) { |
264 | rados_ioctx_destroy(rbd->io_ctx); | |
265 | rbd->io_ctx = NULL; | |
fc5c0345 DG |
266 | } |
267 | ||
6f9961ac JA |
268 | if (rbd->cluster) { |
269 | rados_shutdown(rbd->cluster); | |
270 | rbd->cluster = NULL; | |
fc5c0345 DG |
271 | } |
272 | } | |
273 | ||
d8b64af2 | 274 | static void _fio_rbd_finish_aiocb(rbd_completion_t comp, void *data) |
fc5c0345 | 275 | { |
dbf388d2 JA |
276 | struct fio_rbd_iou *fri = data; |
277 | struct io_u *io_u = fri->io_u; | |
d8b64af2 | 278 | ssize_t ret; |
fc5c0345 | 279 | |
d8b64af2 JA |
280 | /* |
281 | * Looks like return value is 0 for success, or < 0 for | |
282 | * a specific error. So we have to assume that it can't do | |
283 | * partial completions. | |
284 | */ | |
285 | ret = rbd_aio_get_return_value(fri->completion); | |
286 | if (ret < 0) { | |
d7658ced | 287 | io_u->error = -ret; |
d8b64af2 JA |
288 | io_u->resid = io_u->xfer_buflen; |
289 | } else | |
290 | io_u->error = 0; | |
20cf5aab JD |
291 | |
292 | fri->io_complete = 1; | |
d8b64af2 | 293 | } |
fc5c0345 | 294 | |
d8b64af2 JA |
295 | static struct io_u *fio_rbd_event(struct thread_data *td, int event) |
296 | { | |
565e784d | 297 | struct rbd_data *rbd = td->io_ops_data; |
fc5c0345 | 298 | |
6f9961ac | 299 | return rbd->aio_events[event]; |
fc5c0345 DG |
300 | } |
301 | ||
6f9961ac | 302 | static inline int fri_check_complete(struct rbd_data *rbd, struct io_u *io_u, |
d8b64af2 | 303 | unsigned int *events) |
fc5c0345 | 304 | { |
d8b64af2 | 305 | struct fio_rbd_iou *fri = io_u->engine_data; |
fc5c0345 | 306 | |
b8ecbef6 | 307 | if (fri->io_complete) { |
d8b64af2 | 308 | fri->io_seen = 1; |
6f9961ac | 309 | rbd->aio_events[*events] = io_u; |
d8b64af2 | 310 | (*events)++; |
fc5c0345 | 311 | |
d8b64af2 JA |
312 | rbd_aio_release(fri->completion); |
313 | return 1; | |
314 | } | |
fc5c0345 | 315 | |
d8b64af2 | 316 | return 0; |
fc5c0345 DG |
317 | } |
318 | ||
c7cc1c75 | 319 | #ifndef CONFIG_RBD_POLL |
6f9961ac JA |
320 | static inline int rbd_io_u_seen(struct io_u *io_u) |
321 | { | |
322 | struct fio_rbd_iou *fri = io_u->engine_data; | |
323 | ||
324 | return fri->io_seen; | |
325 | } | |
c7cc1c75 | 326 | #endif |
6f9961ac JA |
327 | |
328 | static void rbd_io_u_wait_complete(struct io_u *io_u) | |
329 | { | |
330 | struct fio_rbd_iou *fri = io_u->engine_data; | |
331 | ||
332 | rbd_aio_wait_for_complete(fri->completion); | |
333 | } | |
334 | ||
335 | static int rbd_io_u_cmp(const void *p1, const void *p2) | |
336 | { | |
337 | const struct io_u **a = (const struct io_u **) p1; | |
338 | const struct io_u **b = (const struct io_u **) p2; | |
339 | uint64_t at, bt; | |
340 | ||
341 | at = utime_since_now(&(*a)->start_time); | |
342 | bt = utime_since_now(&(*b)->start_time); | |
343 | ||
344 | if (at < bt) | |
345 | return -1; | |
346 | else if (at == bt) | |
347 | return 0; | |
348 | else | |
349 | return 1; | |
350 | } | |
351 | ||
d8b64af2 JA |
352 | static int rbd_iter_events(struct thread_data *td, unsigned int *events, |
353 | unsigned int min_evts, int wait) | |
82340a9f | 354 | { |
565e784d | 355 | struct rbd_data *rbd = td->io_ops_data; |
d8b64af2 JA |
356 | unsigned int this_events = 0; |
357 | struct io_u *io_u; | |
53b6c979 PL |
358 | int i, sidx = 0; |
359 | ||
360 | #ifdef CONFIG_RBD_POLL | |
361 | int ret = 0; | |
362 | int event_num = 0; | |
363 | struct fio_rbd_iou *fri = NULL; | |
364 | rbd_completion_t comps[min_evts]; | |
4001b1ba JD |
365 | uint64_t counter; |
366 | bool completed; | |
82340a9f | 367 | |
53b6c979 PL |
368 | struct pollfd pfd; |
369 | pfd.fd = rbd->fd; | |
370 | pfd.events = POLLIN; | |
371 | ||
4001b1ba | 372 | ret = poll(&pfd, 1, wait ? -1 : 0); |
73ae55c3 | 373 | if (ret <= 0) |
53b6c979 | 374 | return 0; |
4001b1ba JD |
375 | if (!(pfd.revents & POLLIN)) |
376 | return 0; | |
53b6c979 PL |
377 | |
378 | event_num = rbd_poll_io_events(rbd->image, comps, min_evts); | |
379 | ||
380 | for (i = 0; i < event_num; i++) { | |
381 | fri = rbd_aio_get_arg(comps[i]); | |
382 | io_u = fri->io_u; | |
4001b1ba JD |
383 | |
384 | /* best effort to decrement the semaphore */ | |
385 | ret = read(rbd->fd, &counter, sizeof(counter)); | |
386 | if (ret <= 0) | |
387 | log_err("rbd_iter_events failed to decrement semaphore.\n"); | |
388 | ||
389 | completed = fri_check_complete(rbd, io_u, events); | |
390 | assert(completed); | |
391 | ||
392 | this_events++; | |
393 | } | |
53b6c979 | 394 | #else |
d8b64af2 | 395 | io_u_qiter(&td->io_u_all, io_u, i) { |
d8b64af2 JA |
396 | if (!(io_u->flags & IO_U_F_FLIGHT)) |
397 | continue; | |
6f9961ac | 398 | if (rbd_io_u_seen(io_u)) |
d8b64af2 | 399 | continue; |
82340a9f | 400 | |
6f9961ac | 401 | if (fri_check_complete(rbd, io_u, events)) |
d8b64af2 | 402 | this_events++; |
6f9961ac JA |
403 | else if (wait) |
404 | rbd->sort_events[sidx++] = io_u; | |
405 | } | |
4001b1ba | 406 | #endif |
82340a9f | 407 | |
6f9961ac JA |
408 | if (!wait || !sidx) |
409 | return this_events; | |
410 | ||
411 | /* | |
412 | * Sort events, oldest issue first, then wait on as many as we | |
413 | * need in order of age. If we have enough events, stop waiting, | |
414 | * and just check if any of the older ones are done. | |
415 | */ | |
416 | if (sidx > 1) | |
417 | qsort(rbd->sort_events, sidx, sizeof(struct io_u *), rbd_io_u_cmp); | |
418 | ||
419 | for (i = 0; i < sidx; i++) { | |
420 | io_u = rbd->sort_events[i]; | |
421 | ||
422 | if (fri_check_complete(rbd, io_u, events)) { | |
423 | this_events++; | |
424 | continue; | |
d8b64af2 | 425 | } |
6f9961ac JA |
426 | |
427 | /* | |
428 | * Stop waiting when we have enough, but continue checking | |
429 | * all pending IOs if they are complete. | |
430 | */ | |
d8b64af2 | 431 | if (*events >= min_evts) |
6f9961ac JA |
432 | continue; |
433 | ||
434 | rbd_io_u_wait_complete(io_u); | |
435 | ||
436 | if (fri_check_complete(rbd, io_u, events)) | |
437 | this_events++; | |
d8b64af2 | 438 | } |
fc5c0345 | 439 | |
d8b64af2 | 440 | return this_events; |
fc5c0345 DG |
441 | } |
442 | ||
443 | static int fio_rbd_getevents(struct thread_data *td, unsigned int min, | |
1f440ece | 444 | unsigned int max, const struct timespec *t) |
fc5c0345 | 445 | { |
d8b64af2 | 446 | unsigned int this_events, events = 0; |
d7d702c7 | 447 | struct rbd_options *o = td->eo; |
d8b64af2 | 448 | int wait = 0; |
fc5c0345 DG |
449 | |
450 | do { | |
d8b64af2 | 451 | this_events = rbd_iter_events(td, &events, min, wait); |
fc5c0345 | 452 | |
d8b64af2 | 453 | if (events >= min) |
fc5c0345 | 454 | break; |
d8b64af2 JA |
455 | if (this_events) |
456 | continue; | |
fc5c0345 | 457 | |
d7d702c7 JA |
458 | if (!o->busy_poll) |
459 | wait = 1; | |
460 | else | |
461 | nop; | |
fc5c0345 DG |
462 | } while (1); |
463 | ||
464 | return events; | |
465 | } | |
466 | ||
2e4ef4fb JA |
467 | static enum fio_q_status fio_rbd_queue(struct thread_data *td, |
468 | struct io_u *io_u) | |
fc5c0345 | 469 | { |
565e784d | 470 | struct rbd_data *rbd = td->io_ops_data; |
d8b64af2 JA |
471 | struct fio_rbd_iou *fri = io_u->engine_data; |
472 | int r = -1; | |
fc5c0345 DG |
473 | |
474 | fio_ro_check(td, io_u); | |
475 | ||
d8b64af2 | 476 | fri->io_seen = 0; |
b8ecbef6 | 477 | fri->io_complete = 0; |
d8b64af2 | 478 | |
dbf388d2 | 479 | r = rbd_aio_create_completion(fri, _fio_rbd_finish_aiocb, |
d8b64af2 | 480 | &fri->completion); |
dbf388d2 JA |
481 | if (r < 0) { |
482 | log_err("rbd_aio_create_completion failed.\n"); | |
483 | goto failed; | |
484 | } | |
fc5c0345 | 485 | |
dbf388d2 | 486 | if (io_u->ddir == DDIR_WRITE) { |
6f9961ac JA |
487 | r = rbd_aio_write(rbd->image, io_u->offset, io_u->xfer_buflen, |
488 | io_u->xfer_buf, fri->completion); | |
fc5c0345 DG |
489 | if (r < 0) { |
490 | log_err("rbd_aio_write failed.\n"); | |
dbf388d2 | 491 | goto failed_comp; |
fc5c0345 DG |
492 | } |
493 | ||
494 | } else if (io_u->ddir == DDIR_READ) { | |
6f9961ac JA |
495 | r = rbd_aio_read(rbd->image, io_u->offset, io_u->xfer_buflen, |
496 | io_u->xfer_buf, fri->completion); | |
fc5c0345 DG |
497 | |
498 | if (r < 0) { | |
499 | log_err("rbd_aio_read failed.\n"); | |
dbf388d2 | 500 | goto failed_comp; |
fc5c0345 | 501 | } |
dbf388d2 | 502 | } else if (io_u->ddir == DDIR_TRIM) { |
6f9961ac JA |
503 | r = rbd_aio_discard(rbd->image, io_u->offset, |
504 | io_u->xfer_buflen, fri->completion); | |
82340a9f | 505 | if (r < 0) { |
dbf388d2 JA |
506 | log_err("rbd_aio_discard failed.\n"); |
507 | goto failed_comp; | |
82340a9f | 508 | } |
dbf388d2 | 509 | } else if (io_u->ddir == DDIR_SYNC) { |
6f9961ac | 510 | r = rbd_aio_flush(rbd->image, fri->completion); |
fc5c0345 DG |
511 | if (r < 0) { |
512 | log_err("rbd_flush failed.\n"); | |
dbf388d2 | 513 | goto failed_comp; |
fc5c0345 | 514 | } |
fc5c0345 DG |
515 | } else { |
516 | dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__, | |
517 | io_u->ddir); | |
ef2f4a50 | 518 | r = -EINVAL; |
dbf388d2 | 519 | goto failed_comp; |
fc5c0345 DG |
520 | } |
521 | ||
522 | return FIO_Q_QUEUED; | |
dbf388d2 JA |
523 | failed_comp: |
524 | rbd_aio_release(fri->completion); | |
fc5c0345 | 525 | failed: |
d7658ced | 526 | io_u->error = -r; |
fc5c0345 DG |
527 | td_verror(td, io_u->error, "xfer"); |
528 | return FIO_Q_COMPLETED; | |
529 | } | |
530 | ||
531 | static int fio_rbd_init(struct thread_data *td) | |
532 | { | |
533 | int r; | |
e0f35b53 | 534 | struct rbd_data *rbd = td->io_ops_data; |
35448275 JA |
535 | |
536 | if (rbd->connected) | |
e0f35b53 | 537 | return 0; |
fc5c0345 DG |
538 | |
539 | r = _fio_rbd_connect(td); | |
540 | if (r) { | |
541 | log_err("fio_rbd_connect failed, return code: %d .\n", r); | |
542 | goto failed; | |
543 | } | |
544 | ||
545 | return 0; | |
546 | ||
547 | failed: | |
548 | return 1; | |
fc5c0345 DG |
549 | } |
550 | ||
551 | static void fio_rbd_cleanup(struct thread_data *td) | |
552 | { | |
565e784d | 553 | struct rbd_data *rbd = td->io_ops_data; |
fc5c0345 | 554 | |
6f9961ac JA |
555 | if (rbd) { |
556 | _fio_rbd_disconnect(rbd); | |
557 | free(rbd->aio_events); | |
558 | free(rbd->sort_events); | |
559 | free(rbd); | |
fc5c0345 | 560 | } |
fc5c0345 DG |
561 | } |
562 | ||
563 | static int fio_rbd_setup(struct thread_data *td) | |
564 | { | |
fc5c0345 DG |
565 | rbd_image_info_t info; |
566 | struct fio_file *f; | |
6f9961ac | 567 | struct rbd_data *rbd = NULL; |
6f9961ac | 568 | int r; |
fc5c0345 | 569 | |
fc5c0345 | 570 | /* allocate engine specific structure to deal with librbd. */ |
6f9961ac | 571 | r = _fio_setup_rbd_data(td, &rbd); |
fc5c0345 DG |
572 | if (r) { |
573 | log_err("fio_setup_rbd_data failed.\n"); | |
574 | goto cleanup; | |
575 | } | |
565e784d | 576 | td->io_ops_data = rbd; |
fc5c0345 | 577 | |
d8b64af2 JA |
578 | /* librbd does not allow us to run first in the main thread and later |
579 | * in a fork child. It needs to be the same process context all the | |
580 | * time. | |
fc5c0345 DG |
581 | */ |
582 | td->o.use_thread = 1; | |
583 | ||
584 | /* connect in the main thread to determine to determine | |
585 | * the size of the given RADOS block device. And disconnect | |
586 | * later on. | |
587 | */ | |
588 | r = _fio_rbd_connect(td); | |
589 | if (r) { | |
590 | log_err("fio_rbd_connect failed.\n"); | |
591 | goto cleanup; | |
592 | } | |
e0f35b53 | 593 | rbd->connected = true; |
fc5c0345 DG |
594 | |
595 | /* get size of the RADOS block device */ | |
6f9961ac | 596 | r = rbd_stat(rbd->image, &info, sizeof(info)); |
fc5c0345 DG |
597 | if (r < 0) { |
598 | log_err("rbd_status failed.\n"); | |
6747f469 | 599 | goto cleanup; |
206c546d PL |
600 | } else if (info.size == 0) { |
601 | log_err("image size should be larger than zero.\n"); | |
602 | r = -EINVAL; | |
6747f469 | 603 | goto cleanup; |
fc5c0345 | 604 | } |
206c546d | 605 | |
50ea012c | 606 | dprint(FD_IO, "rbd-engine: image size: %" PRIu64 "\n", info.size); |
fc5c0345 DG |
607 | |
608 | /* taken from "net" engine. Pretend we deal with files, | |
609 | * even if we do not have any ideas about files. | |
610 | * The size of the RBD is set instead of a artificial file. | |
611 | */ | |
612 | if (!td->files_index) { | |
5903e7b7 | 613 | add_file(td, td->o.filename ? : "rbd", 0, 0); |
fc5c0345 | 614 | td->o.nr_files = td->o.nr_files ? : 1; |
b53f2c54 | 615 | td->o.open_files++; |
fc5c0345 DG |
616 | } |
617 | f = td->files[0]; | |
618 | f->real_file_size = info.size; | |
619 | ||
fc5c0345 DG |
620 | return 0; |
621 | ||
fc5c0345 DG |
622 | cleanup: |
623 | fio_rbd_cleanup(td); | |
624 | return r; | |
625 | } | |
626 | ||
627 | static int fio_rbd_open(struct thread_data *td, struct fio_file *f) | |
628 | { | |
629 | return 0; | |
630 | } | |
631 | ||
d9b100fc JA |
632 | static int fio_rbd_invalidate(struct thread_data *td, struct fio_file *f) |
633 | { | |
903b2812 | 634 | #if defined(CONFIG_RBD_INVAL) |
565e784d | 635 | struct rbd_data *rbd = td->io_ops_data; |
903b2812 | 636 | |
6f9961ac | 637 | return rbd_invalidate_cache(rbd->image); |
903b2812 | 638 | #else |
d9b100fc | 639 | return 0; |
903b2812 | 640 | #endif |
d9b100fc JA |
641 | } |
642 | ||
fc5c0345 DG |
643 | static void fio_rbd_io_u_free(struct thread_data *td, struct io_u *io_u) |
644 | { | |
d8b64af2 | 645 | struct fio_rbd_iou *fri = io_u->engine_data; |
fc5c0345 | 646 | |
d8b64af2 | 647 | if (fri) { |
fc5c0345 | 648 | io_u->engine_data = NULL; |
d8b64af2 | 649 | free(fri); |
fc5c0345 DG |
650 | } |
651 | } | |
652 | ||
653 | static int fio_rbd_io_u_init(struct thread_data *td, struct io_u *io_u) | |
654 | { | |
d8b64af2 | 655 | struct fio_rbd_iou *fri; |
fc5c0345 | 656 | |
d8b64af2 JA |
657 | fri = calloc(1, sizeof(*fri)); |
658 | fri->io_u = io_u; | |
659 | io_u->engine_data = fri; | |
fc5c0345 DG |
660 | return 0; |
661 | } | |
662 | ||
10aa136b | 663 | static struct ioengine_ops ioengine = { |
d9b100fc JA |
664 | .name = "rbd", |
665 | .version = FIO_IOOPS_VERSION, | |
666 | .setup = fio_rbd_setup, | |
667 | .init = fio_rbd_init, | |
668 | .queue = fio_rbd_queue, | |
669 | .getevents = fio_rbd_getevents, | |
670 | .event = fio_rbd_event, | |
671 | .cleanup = fio_rbd_cleanup, | |
672 | .open_file = fio_rbd_open, | |
673 | .invalidate = fio_rbd_invalidate, | |
674 | .options = options, | |
675 | .io_u_init = fio_rbd_io_u_init, | |
676 | .io_u_free = fio_rbd_io_u_free, | |
677 | .option_struct_size = sizeof(struct rbd_options), | |
fc5c0345 DG |
678 | }; |
679 | ||
680 | static void fio_init fio_rbd_register(void) | |
681 | { | |
682 | register_ioengine(&ioengine); | |
683 | } | |
684 | ||
685 | static void fio_exit fio_rbd_unregister(void) | |
686 | { | |
687 | unregister_ioengine(&ioengine); | |
688 | } |