Daniel Gollub | fc5c034 | 2014-02-17 14:35:28 +0100 | [diff] [blame] | 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); |
| 134 | failed_shutdown: |
| 135 | rados_shutdown(rbd_data->cluster); |
| 136 | failed_early: |
| 137 | return 1; |
| 138 | } |
| 139 | |
| 140 | static void _fio_rbd_disconnect(struct rbd_data *rbd_data) |
| 141 | { |
| 142 | if (!rbd_data) |
| 143 | return; |
| 144 | |
| 145 | /* shutdown everything */ |
| 146 | if (rbd_data->image) { |
| 147 | rbd_close(rbd_data->image); |
| 148 | rbd_data->image = NULL; |
| 149 | } |
| 150 | |
| 151 | if (rbd_data->io_ctx) { |
| 152 | rados_ioctx_destroy(rbd_data->io_ctx); |
| 153 | rbd_data->io_ctx = NULL; |
| 154 | } |
| 155 | |
| 156 | if (rbd_data->cluster) { |
| 157 | rados_shutdown(rbd_data->cluster); |
| 158 | rbd_data->cluster = NULL; |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | static void _fio_rbd_finish_write_aiocb(rbd_completion_t comp, void *data) |
| 163 | { |
| 164 | struct io_u *io_u = (struct io_u *)data; |
| 165 | struct fio_rbd_iou *fio_rbd_iou = |
| 166 | (struct fio_rbd_iou *)io_u->engine_data; |
| 167 | |
| 168 | fio_rbd_iou->io_complete = 1; |
| 169 | |
| 170 | /* if write needs to be verified - we should not release comp here |
| 171 | without fetching the result */ |
| 172 | |
| 173 | rbd_aio_release(comp); |
| 174 | /* TODO handle error */ |
| 175 | |
| 176 | return; |
| 177 | } |
| 178 | |
| 179 | static void _fio_rbd_finish_read_aiocb(rbd_completion_t comp, void *data) |
| 180 | { |
| 181 | struct io_u *io_u = (struct io_u *)data; |
| 182 | struct fio_rbd_iou *fio_rbd_iou = |
| 183 | (struct fio_rbd_iou *)io_u->engine_data; |
| 184 | |
| 185 | fio_rbd_iou->io_complete = 1; |
| 186 | |
| 187 | /* if read needs to be verified - we should not release comp here |
| 188 | without fetching the result */ |
| 189 | rbd_aio_release(comp); |
| 190 | |
| 191 | /* TODO handle error */ |
| 192 | |
| 193 | return; |
| 194 | } |
| 195 | |
| 196 | static struct io_u *fio_rbd_event(struct thread_data *td, int event) |
| 197 | { |
| 198 | struct rbd_data *rbd_data = td->io_ops->data; |
| 199 | |
| 200 | return rbd_data->aio_events[event]; |
| 201 | } |
| 202 | |
| 203 | static int fio_rbd_getevents(struct thread_data *td, unsigned int min, |
| 204 | unsigned int max, struct timespec *t) |
| 205 | { |
| 206 | struct rbd_data *rbd_data = td->io_ops->data; |
| 207 | unsigned int events = 0; |
| 208 | struct io_u *io_u; |
| 209 | int i; |
| 210 | struct fio_rbd_iou *fov; |
| 211 | |
| 212 | do { |
| 213 | io_u_qiter(&td->io_u_all, io_u, i) { |
| 214 | if (!(io_u->flags & IO_U_F_FLIGHT)) |
| 215 | continue; |
| 216 | |
| 217 | fov = (struct fio_rbd_iou *)io_u->engine_data; |
| 218 | |
| 219 | if (fov->io_complete) { |
| 220 | fov->io_complete = 0; |
| 221 | rbd_data->aio_events[events] = io_u; |
| 222 | events++; |
| 223 | } |
| 224 | |
| 225 | } |
| 226 | if (events < min) |
| 227 | usleep(100); |
| 228 | else |
| 229 | break; |
| 230 | |
| 231 | } while (1); |
| 232 | |
| 233 | return events; |
| 234 | } |
| 235 | |
| 236 | static int fio_rbd_queue(struct thread_data *td, struct io_u *io_u) |
| 237 | { |
| 238 | int r = -1; |
| 239 | struct rbd_data *rbd_data = td->io_ops->data; |
| 240 | rbd_completion_t comp; |
| 241 | |
| 242 | fio_ro_check(td, io_u); |
| 243 | |
| 244 | if (io_u->ddir == DDIR_WRITE) { |
| 245 | r = rbd_aio_create_completion(io_u, |
| 246 | (rbd_callback_t) |
| 247 | _fio_rbd_finish_write_aiocb, |
| 248 | &comp); |
| 249 | if (r < 0) { |
| 250 | log_err |
| 251 | ("rbd_aio_create_completion for DDIR_WRITE failed.\n"); |
| 252 | goto failed; |
| 253 | } |
| 254 | |
| 255 | r = rbd_aio_write(rbd_data->image, io_u->offset, |
| 256 | io_u->xfer_buflen, io_u->xfer_buf, comp); |
| 257 | if (r < 0) { |
| 258 | log_err("rbd_aio_write failed.\n"); |
| 259 | goto failed; |
| 260 | } |
| 261 | |
| 262 | } else if (io_u->ddir == DDIR_READ) { |
| 263 | r = rbd_aio_create_completion(io_u, |
| 264 | (rbd_callback_t) |
| 265 | _fio_rbd_finish_read_aiocb, |
| 266 | &comp); |
| 267 | if (r < 0) { |
| 268 | log_err |
| 269 | ("rbd_aio_create_completion for DDIR_READ failed.\n"); |
| 270 | goto failed; |
| 271 | } |
| 272 | |
| 273 | r = rbd_aio_read(rbd_data->image, io_u->offset, |
| 274 | io_u->xfer_buflen, io_u->xfer_buf, comp); |
| 275 | |
| 276 | if (r < 0) { |
| 277 | log_err("rbd_aio_read failed.\n"); |
| 278 | goto failed; |
| 279 | } |
| 280 | |
| 281 | } else if (io_u->ddir == DDIR_SYNC) { |
| 282 | r = rbd_flush(rbd_data->image); |
| 283 | if (r < 0) { |
| 284 | log_err("rbd_flush failed.\n"); |
| 285 | goto failed; |
| 286 | } |
| 287 | |
| 288 | return FIO_Q_COMPLETED; |
| 289 | } else { |
| 290 | dprint(FD_IO, "%s: Warning: unhandled ddir: %d\n", __func__, |
| 291 | io_u->ddir); |
| 292 | return FIO_Q_COMPLETED; |
| 293 | } |
| 294 | |
| 295 | return FIO_Q_QUEUED; |
| 296 | |
| 297 | failed: |
| 298 | io_u->error = r; |
| 299 | td_verror(td, io_u->error, "xfer"); |
| 300 | return FIO_Q_COMPLETED; |
| 301 | } |
| 302 | |
| 303 | static int fio_rbd_init(struct thread_data *td) |
| 304 | { |
| 305 | int r; |
| 306 | |
| 307 | r = _fio_rbd_connect(td); |
| 308 | if (r) { |
| 309 | log_err("fio_rbd_connect failed, return code: %d .\n", r); |
| 310 | goto failed; |
| 311 | } |
| 312 | |
| 313 | return 0; |
| 314 | |
| 315 | failed: |
| 316 | return 1; |
| 317 | |
| 318 | } |
| 319 | |
| 320 | static void fio_rbd_cleanup(struct thread_data *td) |
| 321 | { |
| 322 | struct rbd_data *rbd_data = td->io_ops->data; |
| 323 | |
| 324 | if (rbd_data) { |
| 325 | _fio_rbd_disconnect(rbd_data); |
| 326 | free(rbd_data->aio_events); |
| 327 | free(rbd_data); |
| 328 | } |
| 329 | |
| 330 | } |
| 331 | |
| 332 | static int fio_rbd_setup(struct thread_data *td) |
| 333 | { |
| 334 | int r = 0; |
| 335 | rbd_image_info_t info; |
| 336 | struct fio_file *f; |
| 337 | struct rbd_data *rbd_data = NULL; |
| 338 | int major, minor, extra; |
| 339 | |
| 340 | /* log version of librbd. No cluster connection required. */ |
| 341 | rbd_version(&major, &minor, &extra); |
| 342 | log_info("rbd engine: RBD version: %d.%d.%d\n", major, minor, extra); |
| 343 | |
| 344 | /* allocate engine specific structure to deal with librbd. */ |
| 345 | r = _fio_setup_rbd_data(td, &rbd_data); |
| 346 | if (r) { |
| 347 | log_err("fio_setup_rbd_data failed.\n"); |
| 348 | goto cleanup; |
| 349 | } |
| 350 | td->io_ops->data = rbd_data; |
| 351 | |
| 352 | /* librbd does not allow us to run first in the main thread and later in a |
| 353 | * fork child. It needs to be the same process context all the time. |
| 354 | */ |
| 355 | td->o.use_thread = 1; |
| 356 | |
| 357 | /* connect in the main thread to determine to determine |
| 358 | * the size of the given RADOS block device. And disconnect |
| 359 | * later on. |
| 360 | */ |
| 361 | r = _fio_rbd_connect(td); |
| 362 | if (r) { |
| 363 | log_err("fio_rbd_connect failed.\n"); |
| 364 | goto cleanup; |
| 365 | } |
| 366 | |
| 367 | /* get size of the RADOS block device */ |
| 368 | r = rbd_stat(rbd_data->image, &info, sizeof(info)); |
| 369 | if (r < 0) { |
| 370 | log_err("rbd_status failed.\n"); |
| 371 | goto disconnect; |
| 372 | } |
| 373 | dprint(FD_IO, "rbd-engine: image size: %lu\n", info.size); |
| 374 | |
| 375 | /* taken from "net" engine. Pretend we deal with files, |
| 376 | * even if we do not have any ideas about files. |
| 377 | * The size of the RBD is set instead of a artificial file. |
| 378 | */ |
| 379 | if (!td->files_index) { |
Jens Axboe | 5903e7b | 2014-02-26 13:42:13 -0800 | [diff] [blame] | 380 | add_file(td, td->o.filename ? : "rbd", 0, 0); |
Daniel Gollub | fc5c034 | 2014-02-17 14:35:28 +0100 | [diff] [blame] | 381 | td->o.nr_files = td->o.nr_files ? : 1; |
Jens Axboe | b53f2c5 | 2014-04-08 21:07:12 -0600 | [diff] [blame] | 382 | td->o.open_files++; |
Daniel Gollub | fc5c034 | 2014-02-17 14:35:28 +0100 | [diff] [blame] | 383 | } |
| 384 | f = td->files[0]; |
| 385 | f->real_file_size = info.size; |
| 386 | |
| 387 | /* disconnect, then we were only connected to determine |
| 388 | * the size of the RBD. |
| 389 | */ |
| 390 | _fio_rbd_disconnect(rbd_data); |
| 391 | return 0; |
| 392 | |
| 393 | disconnect: |
| 394 | _fio_rbd_disconnect(rbd_data); |
| 395 | cleanup: |
| 396 | fio_rbd_cleanup(td); |
| 397 | return r; |
| 398 | } |
| 399 | |
| 400 | static int fio_rbd_open(struct thread_data *td, struct fio_file *f) |
| 401 | { |
| 402 | return 0; |
| 403 | } |
| 404 | |
| 405 | static void fio_rbd_io_u_free(struct thread_data *td, struct io_u *io_u) |
| 406 | { |
| 407 | struct fio_rbd_iou *o = io_u->engine_data; |
| 408 | |
| 409 | if (o) { |
| 410 | io_u->engine_data = NULL; |
| 411 | free(o); |
| 412 | } |
| 413 | } |
| 414 | |
| 415 | static int fio_rbd_io_u_init(struct thread_data *td, struct io_u *io_u) |
| 416 | { |
| 417 | struct fio_rbd_iou *o; |
| 418 | |
| 419 | o = malloc(sizeof(*o)); |
| 420 | o->io_complete = 0; |
| 421 | o->io_u = io_u; |
| 422 | io_u->engine_data = o; |
| 423 | return 0; |
| 424 | } |
| 425 | |
Jens Axboe | 10aa136 | 2014-04-01 21:10:36 -0600 | [diff] [blame] | 426 | static struct ioengine_ops ioengine = { |
Daniel Gollub | fc5c034 | 2014-02-17 14:35:28 +0100 | [diff] [blame] | 427 | .name = "rbd", |
| 428 | .version = FIO_IOOPS_VERSION, |
| 429 | .setup = fio_rbd_setup, |
| 430 | .init = fio_rbd_init, |
| 431 | .queue = fio_rbd_queue, |
| 432 | .getevents = fio_rbd_getevents, |
| 433 | .event = fio_rbd_event, |
| 434 | .cleanup = fio_rbd_cleanup, |
| 435 | .open_file = fio_rbd_open, |
| 436 | .options = options, |
| 437 | .io_u_init = fio_rbd_io_u_init, |
| 438 | .io_u_free = fio_rbd_io_u_free, |
| 439 | .option_struct_size = sizeof(struct rbd_options), |
| 440 | }; |
| 441 | |
| 442 | static void fio_init fio_rbd_register(void) |
| 443 | { |
| 444 | register_ioengine(&ioengine); |
| 445 | } |
| 446 | |
| 447 | static void fio_exit fio_rbd_unregister(void) |
| 448 | { |
| 449 | unregister_ioengine(&ioengine); |
| 450 | } |