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" | |
11 | ||
12 | struct fio_rbd_iou { | |
13 | struct io_u *io_u; | |
14 | int io_complete; | |
15 | }; | |
16 | ||
17 | struct rbd_data { | |
18 | rados_t cluster; | |
19 | rados_ioctx_t io_ctx; | |
20 | rbd_image_t image; | |
21 | struct io_u **aio_events; | |
22 | }; | |
23 | ||
24 | struct rbd_options { | |
25 | struct thread_data *td; | |
26 | char *rbd_name; | |
27 | char *pool_name; | |
28 | char *client_name; | |
29 | }; | |
30 | ||
31 | static struct fio_option options[] = { | |
32 | { | |
33 | .name = "rbdname", | |
34 | .lname = "rbd engine rbdname", | |
35 | .type = FIO_OPT_STR_STORE, | |
36 | .help = "RBD name for RBD engine", | |
37 | .off1 = offsetof(struct rbd_options, rbd_name), | |
38 | .category = FIO_OPT_C_ENGINE, | |
39 | .group = FIO_OPT_G_RBD, | |
40 | }, | |
41 | { | |
42 | .name = "pool", | |
43 | .lname = "rbd engine pool", | |
44 | .type = FIO_OPT_STR_STORE, | |
45 | .help = "Name of the pool hosting the RBD for the RBD engine", | |
46 | .off1 = offsetof(struct rbd_options, pool_name), | |
47 | .category = FIO_OPT_C_ENGINE, | |
48 | .group = FIO_OPT_G_RBD, | |
49 | }, | |
50 | { | |
51 | .name = "clientname", | |
52 | .lname = "rbd engine clientname", | |
53 | .type = FIO_OPT_STR_STORE, | |
54 | .help = "Name of the ceph client to access the RBD for the RBD engine", | |
55 | .off1 = offsetof(struct rbd_options, client_name), | |
56 | .category = FIO_OPT_C_ENGINE, | |
57 | .group = FIO_OPT_G_RBD, | |
58 | }, | |
59 | { | |
60 | .name = NULL, | |
61 | }, | |
62 | }; | |
63 | ||
64 | static int _fio_setup_rbd_data(struct thread_data *td, | |
65 | struct rbd_data **rbd_data_ptr) | |
66 | { | |
67 | struct rbd_data *rbd_data; | |
68 | ||
69 | if (td->io_ops->data) | |
70 | return 0; | |
71 | ||
72 | rbd_data = malloc(sizeof(struct rbd_data)); | |
73 | if (!rbd_data) | |
74 | goto failed; | |
75 | ||
76 | memset(rbd_data, 0, sizeof(struct rbd_data)); | |
77 | ||
78 | rbd_data->aio_events = malloc(td->o.iodepth * sizeof(struct io_u *)); | |
79 | if (!rbd_data->aio_events) | |
80 | goto failed; | |
81 | ||
82 | memset(rbd_data->aio_events, 0, td->o.iodepth * sizeof(struct io_u *)); | |
83 | ||
84 | *rbd_data_ptr = rbd_data; | |
85 | ||
86 | return 0; | |
87 | ||
88 | failed: | |
89 | return 1; | |
90 | ||
91 | } | |
92 | ||
93 | static int _fio_rbd_connect(struct thread_data *td) | |
94 | { | |
95 | struct rbd_data *rbd_data = td->io_ops->data; | |
96 | struct rbd_options *o = td->eo; | |
97 | int r; | |
98 | ||
99 | r = rados_create(&(rbd_data->cluster), o->client_name); | |
100 | if (r < 0) { | |
101 | log_err("rados_create failed.\n"); | |
102 | goto failed_early; | |
103 | } | |
104 | ||
105 | r = rados_conf_read_file(rbd_data->cluster, NULL); | |
106 | if (r < 0) { | |
107 | log_err("rados_conf_read_file failed.\n"); | |
108 | goto failed_early; | |
109 | } | |
110 | ||
111 | r = rados_connect(rbd_data->cluster); | |
112 | if (r < 0) { | |
113 | log_err("rados_connect failed.\n"); | |
114 | goto failed_shutdown; | |
115 | } | |
116 | ||
117 | r = rados_ioctx_create(rbd_data->cluster, o->pool_name, | |
118 | &(rbd_data->io_ctx)); | |
119 | if (r < 0) { | |
120 | log_err("rados_ioctx_create failed.\n"); | |
121 | goto failed_shutdown; | |
122 | } | |
123 | ||
124 | r = rbd_open(rbd_data->io_ctx, o->rbd_name, &(rbd_data->image), | |
125 | NULL /*snap */ ); | |
126 | if (r < 0) { | |
127 | log_err("rbd_open failed.\n"); | |
128 | goto failed_open; | |
129 | } | |
130 | return 0; | |
131 | ||
132 | failed_open: | |
133 | rados_ioctx_destroy(rbd_data->io_ctx); | |
086f9935 | 134 | rbd_data->io_ctx = NULL; |
fc5c0345 DG |
135 | failed_shutdown: |
136 | rados_shutdown(rbd_data->cluster); | |
086f9935 | 137 | rbd_data->cluster = NULL; |
fc5c0345 DG |
138 | failed_early: |
139 | return 1; | |
140 | } | |
141 | ||
142 | static void _fio_rbd_disconnect(struct rbd_data *rbd_data) | |
143 | { | |
144 | if (!rbd_data) | |
145 | return; | |
146 | ||
147 | /* shutdown everything */ | |
148 | if (rbd_data->image) { | |
149 | rbd_close(rbd_data->image); | |
150 | rbd_data->image = NULL; | |
151 | } | |
152 | ||
153 | if (rbd_data->io_ctx) { | |
154 | rados_ioctx_destroy(rbd_data->io_ctx); | |
155 | rbd_data->io_ctx = NULL; | |
156 | } | |
157 | ||
158 | if (rbd_data->cluster) { | |
159 | rados_shutdown(rbd_data->cluster); | |
160 | rbd_data->cluster = NULL; | |
161 | } | |
162 | } | |
163 | ||
164 | static void _fio_rbd_finish_write_aiocb(rbd_completion_t comp, void *data) | |
165 | { | |
166 | struct io_u *io_u = (struct io_u *)data; | |
167 | struct fio_rbd_iou *fio_rbd_iou = | |
168 | (struct fio_rbd_iou *)io_u->engine_data; | |
169 | ||
170 | fio_rbd_iou->io_complete = 1; | |
171 | ||
172 | /* if write needs to be verified - we should not release comp here | |
173 | without fetching the result */ | |
174 | ||
175 | rbd_aio_release(comp); | |
176 | /* TODO handle error */ | |
177 | ||
178 | return; | |
179 | } | |
180 | ||
181 | static void _fio_rbd_finish_read_aiocb(rbd_completion_t comp, void *data) | |
182 | { | |
183 | struct io_u *io_u = (struct io_u *)data; | |
184 | struct fio_rbd_iou *fio_rbd_iou = | |
185 | (struct fio_rbd_iou *)io_u->engine_data; | |
186 | ||
187 | fio_rbd_iou->io_complete = 1; | |
188 | ||
189 | /* if read needs to be verified - we should not release comp here | |
190 | without fetching the result */ | |
191 | rbd_aio_release(comp); | |
192 | ||
193 | /* TODO handle error */ | |
194 | ||
195 | return; | |
196 | } | |
197 | ||
198 | static struct io_u *fio_rbd_event(struct thread_data *td, int event) | |
199 | { | |
200 | struct rbd_data *rbd_data = td->io_ops->data; | |
201 | ||
202 | return rbd_data->aio_events[event]; | |
203 | } | |
204 | ||
205 | static int fio_rbd_getevents(struct thread_data *td, unsigned int min, | |
206 | unsigned int max, struct timespec *t) | |
207 | { | |
208 | struct rbd_data *rbd_data = td->io_ops->data; | |
209 | unsigned int events = 0; | |
210 | struct io_u *io_u; | |
211 | int i; | |
212 | struct fio_rbd_iou *fov; | |
213 | ||
214 | do { | |
215 | io_u_qiter(&td->io_u_all, io_u, i) { | |
216 | if (!(io_u->flags & IO_U_F_FLIGHT)) | |
217 | continue; | |
218 | ||
219 | fov = (struct fio_rbd_iou *)io_u->engine_data; | |
220 | ||
221 | if (fov->io_complete) { | |
222 | fov->io_complete = 0; | |
223 | rbd_data->aio_events[events] = io_u; | |
224 | events++; | |
225 | } | |
226 | ||
227 | } | |
228 | if (events < min) | |
229 | usleep(100); | |
230 | else | |
231 | break; | |
232 | ||
233 | } while (1); | |
234 | ||
235 | return events; | |
236 | } | |
237 | ||
238 | static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u) | |
239 | { | |
240 | int r = -1; | |
241 | struct rbd_data *rbd_data = td->io_ops->data; | |
242 | rbd_completion_t comp; | |
243 | ||
244 | fio_ro_check(td, io_u); | |
245 | ||
246 | if (io_u->ddir == DDIR_WRITE) { | |
247 | r = rbd_aio_create_completion(io_u, | |
248 | (rbd_callback_t) | |
249 | _fio_rbd_finish_write_aiocb, | |
250 | &comp); | |
251 | if (r < 0) { | |
252 | log_err | |
253 | ("rbd_aio_create_completion for DDIR_WRITE failed.\n"); | |
254 | goto failed; | |
255 | } | |
256 | ||
257 | r = rbd_aio_write(rbd_data->image, io_u->offset, | |
258 | io_u->xfer_buflen, io_u->xfer_buf, comp); | |
259 | if (r < 0) { | |
260 | log_err("rbd_aio_write failed.\n"); | |
261 | goto failed; | |
262 | } | |
263 | ||
264 | } else if (io_u->ddir == DDIR_READ) { | |
265 | r = rbd_aio_create_completion(io_u, | |
266 | (rbd_callback_t) | |
267 | _fio_rbd_finish_read_aiocb, | |
268 | &comp); | |
269 | if (r < 0) { | |
270 | log_err | |
271 | ("rbd_aio_create_completion for DDIR_READ failed.\n"); | |
272 | goto failed; | |
273 | } | |
274 | ||
275 | r = rbd_aio_read(rbd_data->image, io_u->offset, | |
276 | io_u->xfer_buflen, io_u->xfer_buf, comp); | |
277 | ||
278 | if (r < 0) { | |
279 | log_err("rbd_aio_read failed.\n"); | |
280 | goto failed; | |
281 | } | |
282 | ||
283 | } else if (io_u->ddir == DDIR_SYNC) { | |
284 | r = rbd_flush(rbd_data->image); | |
285 | if (r < 0) { | |
286 | log_err("rbd_flush failed.\n"); | |
287 | goto failed; | |
288 | } | |
289 | ||
290 | return FIO_Q_COMPLETED; | |
291 | } else { | |
292 | dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__, | |
293 | io_u->ddir); | |
294 | return FIO_Q_COMPLETED; | |
295 | } | |
296 | ||
297 | return FIO_Q_QUEUED; | |
298 | ||
299 | failed: | |
300 | io_u->error = r; | |
301 | td_verror(td, io_u->error, "xfer"); | |
302 | return FIO_Q_COMPLETED; | |
303 | } | |
304 | ||
305 | static int fio_rbd_init(struct thread_data *td) | |
306 | { | |
307 | int r; | |
308 | ||
309 | r = _fio_rbd_connect(td); | |
310 | if (r) { | |
311 | log_err("fio_rbd_connect failed, return code: %d .\n", r); | |
312 | goto failed; | |
313 | } | |
314 | ||
315 | return 0; | |
316 | ||
317 | failed: | |
318 | return 1; | |
319 | ||
320 | } | |
321 | ||
322 | static void fio_rbd_cleanup(struct thread_data *td) | |
323 | { | |
324 | struct rbd_data *rbd_data = td->io_ops->data; | |
325 | ||
326 | if (rbd_data) { | |
327 | _fio_rbd_disconnect(rbd_data); | |
328 | free(rbd_data->aio_events); | |
329 | free(rbd_data); | |
330 | } | |
331 | ||
332 | } | |
333 | ||
334 | static int fio_rbd_setup(struct thread_data *td) | |
335 | { | |
336 | int r = 0; | |
337 | rbd_image_info_t info; | |
338 | struct fio_file *f; | |
339 | struct rbd_data *rbd_data = NULL; | |
340 | int major, minor, extra; | |
341 | ||
342 | /* log version of librbd. No cluster connection required. */ | |
343 | rbd_version(&major, &minor, &extra); | |
344 | log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra); | |
345 | ||
346 | /* allocate engine specific structure to deal with librbd. */ | |
347 | r = _fio_setup_rbd_data(td, &rbd_data); | |
348 | if (r) { | |
349 | log_err("fio_setup_rbd_data failed.\n"); | |
350 | goto cleanup; | |
351 | } | |
352 | td->io_ops->data = rbd_data; | |
353 | ||
354 | /* librbd does not allow us to run first in the main thread and later in a | |
355 | * fork child. It needs to be the same process context all the time. | |
356 | */ | |
357 | td->o.use_thread = 1; | |
358 | ||
359 | /* connect in the main thread to determine to determine | |
360 | * the size of the given RADOS block device. And disconnect | |
361 | * later on. | |
362 | */ | |
363 | r = _fio_rbd_connect(td); | |
364 | if (r) { | |
365 | log_err("fio_rbd_connect failed.\n"); | |
366 | goto cleanup; | |
367 | } | |
368 | ||
369 | /* get size of the RADOS block device */ | |
370 | r = rbd_stat(rbd_data->image, &info, sizeof(info)); | |
371 | if (r < 0) { | |
372 | log_err("rbd_status failed.\n"); | |
373 | goto disconnect; | |
374 | } | |
375 | dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size); | |
376 | ||
377 | /* taken from "net" engine. Pretend we deal with files, | |
378 | * even if we do not have any ideas about files. | |
379 | * The size of the RBD is set instead of a artificial file. | |
380 | */ | |
381 | if (!td->files_index) { | |
5903e7b7 | 382 | add_file(td, td->o.filename ? : "rbd", 0, 0); |
fc5c0345 | 383 | td->o.nr_files = td->o.nr_files ? : 1; |
b53f2c54 | 384 | td->o.open_files++; |
fc5c0345 DG |
385 | } |
386 | f = td->files[0]; | |
387 | f->real_file_size = info.size; | |
388 | ||
389 | /* disconnect, then we were only connected to determine | |
390 | * the size of the RBD. | |
391 | */ | |
392 | _fio_rbd_disconnect(rbd_data); | |
393 | return 0; | |
394 | ||
395 | disconnect: | |
396 | _fio_rbd_disconnect(rbd_data); | |
397 | cleanup: | |
398 | fio_rbd_cleanup(td); | |
399 | return r; | |
400 | } | |
401 | ||
402 | static int fio_rbd_open(struct thread_data *td, struct fio_file *f) | |
403 | { | |
404 | return 0; | |
405 | } | |
406 | ||
407 | static void fio_rbd_io_u_free(struct thread_data *td, struct io_u *io_u) | |
408 | { | |
409 | struct fio_rbd_iou *o = io_u->engine_data; | |
410 | ||
411 | if (o) { | |
412 | io_u->engine_data = NULL; | |
413 | free(o); | |
414 | } | |
415 | } | |
416 | ||
417 | static int fio_rbd_io_u_init(struct thread_data *td, struct io_u *io_u) | |
418 | { | |
419 | struct fio_rbd_iou *o; | |
420 | ||
421 | o = malloc(sizeof(*o)); | |
422 | o->io_complete = 0; | |
423 | o->io_u = io_u; | |
424 | io_u->engine_data = o; | |
425 | return 0; | |
426 | } | |
427 | ||
10aa136b | 428 | static struct ioengine_ops ioengine = { |
fc5c0345 DG |
429 | .name = "rbd", |
430 | .version = FIO_IOOPS_VERSION, | |
431 | .setup = fio_rbd_setup, | |
432 | .init = fio_rbd_init, | |
433 | .queue = fio_rbd_queue, | |
434 | .getevents = fio_rbd_getevents, | |
435 | .event = fio_rbd_event, | |
436 | .cleanup = fio_rbd_cleanup, | |
437 | .open_file = fio_rbd_open, | |
438 | .options = options, | |
439 | .io_u_init = fio_rbd_io_u_init, | |
440 | .io_u_free = fio_rbd_io_u_free, | |
441 | .option_struct_size = sizeof(struct rbd_options), | |
442 | }; | |
443 | ||
444 | static void fio_init fio_rbd_register(void) | |
445 | { | |
446 | register_ioengine(&ioengine); | |
447 | } | |
448 | ||
449 | static void fio_exit fio_rbd_unregister(void) | |
450 | { | |
451 | unregister_ioengine(&ioengine); | |
452 | } |