1 /* 2 * QEMU System Emulator block driver 3 * 4 * Copyright (c) 2011 IBM Corp. 5 * Copyright (c) 2012 Red Hat, Inc. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qemu-common.h" 28 #include "trace.h" 29 #include "block/block.h" 30 #include "block/blockjob.h" 31 #include "block/block_int.h" 32 #include "sysemu/block-backend.h" 33 #include "qapi/qmp/qerror.h" 34 #include "qapi/qmp/qjson.h" 35 #include "qemu/coroutine.h" 36 #include "qmp-commands.h" 37 #include "qemu/timer.h" 38 #include "qapi-event.h" 39 40 /* Transactional group of block jobs */ 41 struct BlockJobTxn { 42 43 /* Is this txn being cancelled? */ 44 bool aborting; 45 46 /* List of jobs */ 47 QLIST_HEAD(, BlockJob) jobs; 48 49 /* Reference count */ 50 int refcnt; 51 }; 52 53 void *block_job_create(const BlockJobDriver *driver, BlockDriverState *bs, 54 int64_t speed, BlockCompletionFunc *cb, 55 void *opaque, Error **errp) 56 { 57 BlockJob *job; 58 59 if (bs->job) { 60 error_setg(errp, QERR_DEVICE_IN_USE, bdrv_get_device_name(bs)); 61 return NULL; 62 } 63 bdrv_ref(bs); 64 job = g_malloc0(driver->instance_size); 65 error_setg(&job->blocker, "block device is in use by block job: %s", 66 BlockJobType_lookup[driver->job_type]); 67 bdrv_op_block_all(bs, job->blocker); 68 bdrv_op_unblock(bs, BLOCK_OP_TYPE_DATAPLANE, job->blocker); 69 70 job->driver = driver; 71 job->id = g_strdup(bdrv_get_device_name(bs)); 72 job->bs = bs; 73 job->cb = cb; 74 job->opaque = opaque; 75 job->busy = true; 76 job->refcnt = 1; 77 bs->job = job; 78 79 /* Only set speed when necessary to avoid NotSupported error */ 80 if (speed != 0) { 81 Error *local_err = NULL; 82 83 block_job_set_speed(job, speed, &local_err); 84 if (local_err) { 85 block_job_unref(job); 86 error_propagate(errp, local_err); 87 return NULL; 88 } 89 } 90 return job; 91 } 92 93 void block_job_ref(BlockJob *job) 94 { 95 ++job->refcnt; 96 } 97 98 void block_job_unref(BlockJob *job) 99 { 100 if (--job->refcnt == 0) { 101 job->bs->job = NULL; 102 bdrv_op_unblock_all(job->bs, job->blocker); 103 bdrv_unref(job->bs); 104 error_free(job->blocker); 105 g_free(job->id); 106 g_free(job); 107 } 108 } 109 110 static void block_job_completed_single(BlockJob *job) 111 { 112 if (!job->ret) { 113 if (job->driver->commit) { 114 job->driver->commit(job); 115 } 116 } else { 117 if (job->driver->abort) { 118 job->driver->abort(job); 119 } 120 } 121 job->cb(job->opaque, job->ret); 122 if (job->txn) { 123 block_job_txn_unref(job->txn); 124 } 125 block_job_unref(job); 126 } 127 128 static void block_job_completed_txn_abort(BlockJob *job) 129 { 130 AioContext *ctx; 131 BlockJobTxn *txn = job->txn; 132 BlockJob *other_job, *next; 133 134 if (txn->aborting) { 135 /* 136 * We are cancelled by another job, which will handle everything. 137 */ 138 return; 139 } 140 txn->aborting = true; 141 /* We are the first failed job. Cancel other jobs. */ 142 QLIST_FOREACH(other_job, &txn->jobs, txn_list) { 143 ctx = bdrv_get_aio_context(other_job->bs); 144 aio_context_acquire(ctx); 145 } 146 QLIST_FOREACH(other_job, &txn->jobs, txn_list) { 147 if (other_job == job || other_job->completed) { 148 /* Other jobs are "effectively" cancelled by us, set the status for 149 * them; this job, however, may or may not be cancelled, depending 150 * on the caller, so leave it. */ 151 if (other_job != job) { 152 other_job->cancelled = true; 153 } 154 continue; 155 } 156 block_job_cancel_sync(other_job); 157 assert(other_job->completed); 158 } 159 QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) { 160 ctx = bdrv_get_aio_context(other_job->bs); 161 block_job_completed_single(other_job); 162 aio_context_release(ctx); 163 } 164 } 165 166 static void block_job_completed_txn_success(BlockJob *job) 167 { 168 AioContext *ctx; 169 BlockJobTxn *txn = job->txn; 170 BlockJob *other_job, *next; 171 /* 172 * Successful completion, see if there are other running jobs in this 173 * txn. 174 */ 175 QLIST_FOREACH(other_job, &txn->jobs, txn_list) { 176 if (!other_job->completed) { 177 return; 178 } 179 } 180 /* We are the last completed job, commit the transaction. */ 181 QLIST_FOREACH_SAFE(other_job, &txn->jobs, txn_list, next) { 182 ctx = bdrv_get_aio_context(other_job->bs); 183 aio_context_acquire(ctx); 184 assert(other_job->ret == 0); 185 block_job_completed_single(other_job); 186 aio_context_release(ctx); 187 } 188 } 189 190 void block_job_completed(BlockJob *job, int ret) 191 { 192 BlockDriverState *bs = job->bs; 193 194 assert(bs->job == job); 195 assert(!job->completed); 196 job->completed = true; 197 job->ret = ret; 198 if (!job->txn) { 199 block_job_completed_single(job); 200 } else if (ret < 0 || block_job_is_cancelled(job)) { 201 block_job_completed_txn_abort(job); 202 } else { 203 block_job_completed_txn_success(job); 204 } 205 } 206 207 void block_job_set_speed(BlockJob *job, int64_t speed, Error **errp) 208 { 209 Error *local_err = NULL; 210 211 if (!job->driver->set_speed) { 212 error_setg(errp, QERR_UNSUPPORTED); 213 return; 214 } 215 job->driver->set_speed(job, speed, &local_err); 216 if (local_err) { 217 error_propagate(errp, local_err); 218 return; 219 } 220 221 job->speed = speed; 222 } 223 224 void block_job_complete(BlockJob *job, Error **errp) 225 { 226 if (job->pause_count || job->cancelled || !job->driver->complete) { 227 error_setg(errp, QERR_BLOCK_JOB_NOT_READY, job->id); 228 return; 229 } 230 231 job->driver->complete(job, errp); 232 } 233 234 void block_job_pause(BlockJob *job) 235 { 236 job->pause_count++; 237 } 238 239 bool block_job_is_paused(BlockJob *job) 240 { 241 return job->pause_count > 0; 242 } 243 244 void block_job_resume(BlockJob *job) 245 { 246 assert(job->pause_count > 0); 247 job->pause_count--; 248 if (job->pause_count) { 249 return; 250 } 251 block_job_enter(job); 252 } 253 254 void block_job_enter(BlockJob *job) 255 { 256 block_job_iostatus_reset(job); 257 if (job->co && !job->busy) { 258 qemu_coroutine_enter(job->co, NULL); 259 } 260 } 261 262 void block_job_cancel(BlockJob *job) 263 { 264 job->cancelled = true; 265 block_job_enter(job); 266 } 267 268 bool block_job_is_cancelled(BlockJob *job) 269 { 270 return job->cancelled; 271 } 272 273 void block_job_iostatus_reset(BlockJob *job) 274 { 275 job->iostatus = BLOCK_DEVICE_IO_STATUS_OK; 276 if (job->driver->iostatus_reset) { 277 job->driver->iostatus_reset(job); 278 } 279 } 280 281 static int block_job_finish_sync(BlockJob *job, 282 void (*finish)(BlockJob *, Error **errp), 283 Error **errp) 284 { 285 BlockDriverState *bs = job->bs; 286 Error *local_err = NULL; 287 int ret; 288 289 assert(bs->job == job); 290 291 block_job_ref(job); 292 finish(job, &local_err); 293 if (local_err) { 294 error_propagate(errp, local_err); 295 block_job_unref(job); 296 return -EBUSY; 297 } 298 while (!job->completed) { 299 aio_poll(job->deferred_to_main_loop ? qemu_get_aio_context() : 300 bdrv_get_aio_context(bs), 301 true); 302 } 303 ret = (job->cancelled && job->ret == 0) ? -ECANCELED : job->ret; 304 block_job_unref(job); 305 return ret; 306 } 307 308 /* A wrapper around block_job_cancel() taking an Error ** parameter so it may be 309 * used with block_job_finish_sync() without the need for (rather nasty) 310 * function pointer casts there. */ 311 static void block_job_cancel_err(BlockJob *job, Error **errp) 312 { 313 block_job_cancel(job); 314 } 315 316 int block_job_cancel_sync(BlockJob *job) 317 { 318 return block_job_finish_sync(job, &block_job_cancel_err, NULL); 319 } 320 321 int block_job_complete_sync(BlockJob *job, Error **errp) 322 { 323 return block_job_finish_sync(job, &block_job_complete, errp); 324 } 325 326 void block_job_sleep_ns(BlockJob *job, QEMUClockType type, int64_t ns) 327 { 328 assert(job->busy); 329 330 /* Check cancellation *before* setting busy = false, too! */ 331 if (block_job_is_cancelled(job)) { 332 return; 333 } 334 335 job->busy = false; 336 if (block_job_is_paused(job)) { 337 qemu_coroutine_yield(); 338 } else { 339 co_aio_sleep_ns(bdrv_get_aio_context(job->bs), type, ns); 340 } 341 job->busy = true; 342 } 343 344 void block_job_yield(BlockJob *job) 345 { 346 assert(job->busy); 347 348 /* Check cancellation *before* setting busy = false, too! */ 349 if (block_job_is_cancelled(job)) { 350 return; 351 } 352 353 job->busy = false; 354 qemu_coroutine_yield(); 355 job->busy = true; 356 } 357 358 BlockJobInfo *block_job_query(BlockJob *job) 359 { 360 BlockJobInfo *info = g_new0(BlockJobInfo, 1); 361 info->type = g_strdup(BlockJobType_lookup[job->driver->job_type]); 362 info->device = g_strdup(job->id); 363 info->len = job->len; 364 info->busy = job->busy; 365 info->paused = job->pause_count > 0; 366 info->offset = job->offset; 367 info->speed = job->speed; 368 info->io_status = job->iostatus; 369 info->ready = job->ready; 370 return info; 371 } 372 373 static void block_job_iostatus_set_err(BlockJob *job, int error) 374 { 375 if (job->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { 376 job->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : 377 BLOCK_DEVICE_IO_STATUS_FAILED; 378 } 379 } 380 381 void block_job_event_cancelled(BlockJob *job) 382 { 383 qapi_event_send_block_job_cancelled(job->driver->job_type, 384 job->id, 385 job->len, 386 job->offset, 387 job->speed, 388 &error_abort); 389 } 390 391 void block_job_event_completed(BlockJob *job, const char *msg) 392 { 393 qapi_event_send_block_job_completed(job->driver->job_type, 394 job->id, 395 job->len, 396 job->offset, 397 job->speed, 398 !!msg, 399 msg, 400 &error_abort); 401 } 402 403 void block_job_event_ready(BlockJob *job) 404 { 405 job->ready = true; 406 407 qapi_event_send_block_job_ready(job->driver->job_type, 408 job->id, 409 job->len, 410 job->offset, 411 job->speed, &error_abort); 412 } 413 414 BlockErrorAction block_job_error_action(BlockJob *job, BlockDriverState *bs, 415 BlockdevOnError on_err, 416 int is_read, int error) 417 { 418 BlockErrorAction action; 419 420 switch (on_err) { 421 case BLOCKDEV_ON_ERROR_ENOSPC: 422 action = (error == ENOSPC) ? 423 BLOCK_ERROR_ACTION_STOP : BLOCK_ERROR_ACTION_REPORT; 424 break; 425 case BLOCKDEV_ON_ERROR_STOP: 426 action = BLOCK_ERROR_ACTION_STOP; 427 break; 428 case BLOCKDEV_ON_ERROR_REPORT: 429 action = BLOCK_ERROR_ACTION_REPORT; 430 break; 431 case BLOCKDEV_ON_ERROR_IGNORE: 432 action = BLOCK_ERROR_ACTION_IGNORE; 433 break; 434 default: 435 abort(); 436 } 437 qapi_event_send_block_job_error(job->id, 438 is_read ? IO_OPERATION_TYPE_READ : 439 IO_OPERATION_TYPE_WRITE, 440 action, &error_abort); 441 if (action == BLOCK_ERROR_ACTION_STOP) { 442 /* make the pause user visible, which will be resumed from QMP. */ 443 job->user_paused = true; 444 block_job_pause(job); 445 block_job_iostatus_set_err(job, error); 446 if (bs->blk && bs != job->bs) { 447 blk_iostatus_set_err(bs->blk, error); 448 } 449 } 450 return action; 451 } 452 453 typedef struct { 454 BlockJob *job; 455 QEMUBH *bh; 456 AioContext *aio_context; 457 BlockJobDeferToMainLoopFn *fn; 458 void *opaque; 459 } BlockJobDeferToMainLoopData; 460 461 static void block_job_defer_to_main_loop_bh(void *opaque) 462 { 463 BlockJobDeferToMainLoopData *data = opaque; 464 AioContext *aio_context; 465 466 qemu_bh_delete(data->bh); 467 468 /* Prevent race with block_job_defer_to_main_loop() */ 469 aio_context_acquire(data->aio_context); 470 471 /* Fetch BDS AioContext again, in case it has changed */ 472 aio_context = bdrv_get_aio_context(data->job->bs); 473 aio_context_acquire(aio_context); 474 475 data->job->deferred_to_main_loop = false; 476 data->fn(data->job, data->opaque); 477 478 aio_context_release(aio_context); 479 480 aio_context_release(data->aio_context); 481 482 g_free(data); 483 } 484 485 void block_job_defer_to_main_loop(BlockJob *job, 486 BlockJobDeferToMainLoopFn *fn, 487 void *opaque) 488 { 489 BlockJobDeferToMainLoopData *data = g_malloc(sizeof(*data)); 490 data->job = job; 491 data->bh = qemu_bh_new(block_job_defer_to_main_loop_bh, data); 492 data->aio_context = bdrv_get_aio_context(job->bs); 493 data->fn = fn; 494 data->opaque = opaque; 495 job->deferred_to_main_loop = true; 496 497 qemu_bh_schedule(data->bh); 498 } 499 500 BlockJobTxn *block_job_txn_new(void) 501 { 502 BlockJobTxn *txn = g_new0(BlockJobTxn, 1); 503 QLIST_INIT(&txn->jobs); 504 txn->refcnt = 1; 505 return txn; 506 } 507 508 static void block_job_txn_ref(BlockJobTxn *txn) 509 { 510 txn->refcnt++; 511 } 512 513 void block_job_txn_unref(BlockJobTxn *txn) 514 { 515 if (txn && --txn->refcnt == 0) { 516 g_free(txn); 517 } 518 } 519 520 void block_job_txn_add_job(BlockJobTxn *txn, BlockJob *job) 521 { 522 if (!txn) { 523 return; 524 } 525 526 assert(!job->txn); 527 job->txn = txn; 528 529 QLIST_INSERT_HEAD(&txn->jobs, job, txn_list); 530 block_job_txn_ref(txn); 531 } 532