| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 4 | * Copyright (C) 1994, Karl Keyte: Added support for disk statistics |
| 5 | * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE |
| 6 | * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> |
| 7 | * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> |
| 8 | * - July2000 |
| 9 | * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 |
| 10 | */ |
| 11 | |
| 12 | /* |
| 13 | * This handles all read/write requests to block devices |
| 14 | */ |
| 15 | #include <linux/kernel.h> |
| 16 | #include <linux/module.h> |
| 17 | #include <linux/backing-dev.h> |
| 18 | #include <linux/bio.h> |
| 19 | #include <linux/blkdev.h> |
| 20 | #include <linux/blk-mq.h> |
| 21 | #include <linux/highmem.h> |
| 22 | #include <linux/mm.h> |
| 23 | #include <linux/kernel_stat.h> |
| 24 | #include <linux/string.h> |
| 25 | #include <linux/init.h> |
| 26 | #include <linux/completion.h> |
| 27 | #include <linux/slab.h> |
| 28 | #include <linux/swap.h> |
| 29 | #include <linux/writeback.h> |
| 30 | #include <linux/task_io_accounting_ops.h> |
| 31 | #include <linux/fault-inject.h> |
| 32 | #include <linux/list_sort.h> |
| 33 | #include <linux/delay.h> |
| 34 | #include <linux/ratelimit.h> |
| 35 | #include <linux/pm_runtime.h> |
| 36 | #include <linux/blk-cgroup.h> |
| 37 | #include <linux/debugfs.h> |
| 38 | #include <linux/bpf.h> |
| 39 | |
| 40 | #define CREATE_TRACE_POINTS |
| 41 | #include <trace/events/block.h> |
| 42 | |
| 43 | #include "blk.h" |
| 44 | #include "blk-mq.h" |
| 45 | #include "blk-mq-sched.h" |
| 46 | #include "blk-pm.h" |
| 47 | #include "blk-rq-qos.h" |
| 48 | |
| 49 | #ifdef CONFIG_DEBUG_FS |
| 50 | struct dentry *blk_debugfs_root; |
| 51 | #endif |
| 52 | |
| 53 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); |
| 54 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); |
| 55 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); |
| 56 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_split); |
| 57 | EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug); |
| 58 | |
| 59 | DEFINE_IDA(blk_queue_ida); |
| 60 | |
| 61 | /* |
| 62 | * For queue allocation |
| 63 | */ |
| 64 | struct kmem_cache *blk_requestq_cachep; |
| 65 | |
| 66 | /* |
| 67 | * Controlling structure to kblockd |
| 68 | */ |
| 69 | static struct workqueue_struct *kblockd_workqueue; |
| 70 | |
| 71 | /** |
| 72 | * blk_queue_flag_set - atomically set a queue flag |
| 73 | * @flag: flag to be set |
| 74 | * @q: request queue |
| 75 | */ |
| 76 | void blk_queue_flag_set(unsigned int flag, struct request_queue *q) |
| 77 | { |
| 78 | set_bit(flag, &q->queue_flags); |
| 79 | } |
| 80 | EXPORT_SYMBOL(blk_queue_flag_set); |
| 81 | |
| 82 | /** |
| 83 | * blk_queue_flag_clear - atomically clear a queue flag |
| 84 | * @flag: flag to be cleared |
| 85 | * @q: request queue |
| 86 | */ |
| 87 | void blk_queue_flag_clear(unsigned int flag, struct request_queue *q) |
| 88 | { |
| 89 | clear_bit(flag, &q->queue_flags); |
| 90 | } |
| 91 | EXPORT_SYMBOL(blk_queue_flag_clear); |
| 92 | |
| 93 | /** |
| 94 | * blk_queue_flag_test_and_set - atomically test and set a queue flag |
| 95 | * @flag: flag to be set |
| 96 | * @q: request queue |
| 97 | * |
| 98 | * Returns the previous value of @flag - 0 if the flag was not set and 1 if |
| 99 | * the flag was already set. |
| 100 | */ |
| 101 | bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q) |
| 102 | { |
| 103 | return test_and_set_bit(flag, &q->queue_flags); |
| 104 | } |
| 105 | EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set); |
| 106 | |
| 107 | void blk_rq_init(struct request_queue *q, struct request *rq) |
| 108 | { |
| 109 | memset(rq, 0, sizeof(*rq)); |
| 110 | |
| 111 | INIT_LIST_HEAD(&rq->queuelist); |
| 112 | rq->q = q; |
| 113 | rq->__sector = (sector_t) -1; |
| 114 | INIT_HLIST_NODE(&rq->hash); |
| 115 | RB_CLEAR_NODE(&rq->rb_node); |
| 116 | rq->tag = -1; |
| 117 | rq->internal_tag = -1; |
| 118 | rq->start_time_ns = ktime_get_ns(); |
| 119 | rq->part = NULL; |
| 120 | } |
| 121 | EXPORT_SYMBOL(blk_rq_init); |
| 122 | |
| 123 | #define REQ_OP_NAME(name) [REQ_OP_##name] = #name |
| 124 | static const char *const blk_op_name[] = { |
| 125 | REQ_OP_NAME(READ), |
| 126 | REQ_OP_NAME(WRITE), |
| 127 | REQ_OP_NAME(FLUSH), |
| 128 | REQ_OP_NAME(DISCARD), |
| 129 | REQ_OP_NAME(SECURE_ERASE), |
| 130 | REQ_OP_NAME(ZONE_RESET), |
| 131 | REQ_OP_NAME(WRITE_SAME), |
| 132 | REQ_OP_NAME(WRITE_ZEROES), |
| 133 | REQ_OP_NAME(SCSI_IN), |
| 134 | REQ_OP_NAME(SCSI_OUT), |
| 135 | REQ_OP_NAME(DRV_IN), |
| 136 | REQ_OP_NAME(DRV_OUT), |
| 137 | }; |
| 138 | #undef REQ_OP_NAME |
| 139 | |
| 140 | /** |
| 141 | * blk_op_str - Return string XXX in the REQ_OP_XXX. |
| 142 | * @op: REQ_OP_XXX. |
| 143 | * |
| 144 | * Description: Centralize block layer function to convert REQ_OP_XXX into |
| 145 | * string format. Useful in the debugging and tracing bio or request. For |
| 146 | * invalid REQ_OP_XXX it returns string "UNKNOWN". |
| 147 | */ |
| 148 | inline const char *blk_op_str(unsigned int op) |
| 149 | { |
| 150 | const char *op_str = "UNKNOWN"; |
| 151 | |
| 152 | if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op]) |
| 153 | op_str = blk_op_name[op]; |
| 154 | |
| 155 | return op_str; |
| 156 | } |
| 157 | EXPORT_SYMBOL_GPL(blk_op_str); |
| 158 | |
| 159 | static const struct { |
| 160 | int errno; |
| 161 | const char *name; |
| 162 | } blk_errors[] = { |
| 163 | [BLK_STS_OK] = { 0, "" }, |
| 164 | [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" }, |
| 165 | [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" }, |
| 166 | [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" }, |
| 167 | [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" }, |
| 168 | [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" }, |
| 169 | [BLK_STS_NEXUS] = { -EBADE, "critical nexus" }, |
| 170 | [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" }, |
| 171 | [BLK_STS_PROTECTION] = { -EILSEQ, "protection" }, |
| 172 | [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" }, |
| 173 | [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" }, |
| 174 | [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" }, |
| 175 | |
| 176 | /* device mapper special case, should not leak out: */ |
| 177 | [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" }, |
| 178 | |
| 179 | /* everything else not covered above: */ |
| 180 | [BLK_STS_IOERR] = { -EIO, "I/O" }, |
| 181 | }; |
| 182 | |
| 183 | blk_status_t errno_to_blk_status(int errno) |
| 184 | { |
| 185 | int i; |
| 186 | |
| 187 | for (i = 0; i < ARRAY_SIZE(blk_errors); i++) { |
| 188 | if (blk_errors[i].errno == errno) |
| 189 | return (__force blk_status_t)i; |
| 190 | } |
| 191 | |
| 192 | return BLK_STS_IOERR; |
| 193 | } |
| 194 | EXPORT_SYMBOL_GPL(errno_to_blk_status); |
| 195 | |
| 196 | int blk_status_to_errno(blk_status_t status) |
| 197 | { |
| 198 | int idx = (__force int)status; |
| 199 | |
| 200 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
| 201 | return -EIO; |
| 202 | return blk_errors[idx].errno; |
| 203 | } |
| 204 | EXPORT_SYMBOL_GPL(blk_status_to_errno); |
| 205 | |
| 206 | static void print_req_error(struct request *req, blk_status_t status, |
| 207 | const char *caller) |
| 208 | { |
| 209 | int idx = (__force int)status; |
| 210 | |
| 211 | if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors))) |
| 212 | return; |
| 213 | |
| 214 | printk_ratelimited(KERN_ERR |
| 215 | "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x " |
| 216 | "phys_seg %u prio class %u\n", |
| 217 | caller, blk_errors[idx].name, |
| 218 | req->rq_disk ? req->rq_disk->disk_name : "?", |
| 219 | blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)), |
| 220 | req->cmd_flags & ~REQ_OP_MASK, |
| 221 | req->nr_phys_segments, |
| 222 | IOPRIO_PRIO_CLASS(req->ioprio)); |
| 223 | } |
| 224 | |
| 225 | static void req_bio_endio(struct request *rq, struct bio *bio, |
| 226 | unsigned int nbytes, blk_status_t error) |
| 227 | { |
| 228 | if (error) |
| 229 | bio->bi_status = error; |
| 230 | |
| 231 | if (unlikely(rq->rq_flags & RQF_QUIET)) |
| 232 | bio_set_flag(bio, BIO_QUIET); |
| 233 | |
| 234 | bio_advance(bio, nbytes); |
| 235 | |
| 236 | /* don't actually finish bio if it's part of flush sequence */ |
| 237 | if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ)) |
| 238 | bio_endio(bio); |
| 239 | } |
| 240 | |
| 241 | void blk_dump_rq_flags(struct request *rq, char *msg) |
| 242 | { |
| 243 | printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg, |
| 244 | rq->rq_disk ? rq->rq_disk->disk_name : "?", |
| 245 | (unsigned long long) rq->cmd_flags); |
| 246 | |
| 247 | printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", |
| 248 | (unsigned long long)blk_rq_pos(rq), |
| 249 | blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); |
| 250 | printk(KERN_INFO " bio %p, biotail %p, len %u\n", |
| 251 | rq->bio, rq->biotail, blk_rq_bytes(rq)); |
| 252 | } |
| 253 | EXPORT_SYMBOL(blk_dump_rq_flags); |
| 254 | |
| 255 | /** |
| 256 | * blk_sync_queue - cancel any pending callbacks on a queue |
| 257 | * @q: the queue |
| 258 | * |
| 259 | * Description: |
| 260 | * The block layer may perform asynchronous callback activity |
| 261 | * on a queue, such as calling the unplug function after a timeout. |
| 262 | * A block device may call blk_sync_queue to ensure that any |
| 263 | * such activity is cancelled, thus allowing it to release resources |
| 264 | * that the callbacks might use. The caller must already have made sure |
| 265 | * that its ->make_request_fn will not re-add plugging prior to calling |
| 266 | * this function. |
| 267 | * |
| 268 | * This function does not cancel any asynchronous activity arising |
| 269 | * out of elevator or throttling code. That would require elevator_exit() |
| 270 | * and blkcg_exit_queue() to be called with queue lock initialized. |
| 271 | * |
| 272 | */ |
| 273 | void blk_sync_queue(struct request_queue *q) |
| 274 | { |
| 275 | del_timer_sync(&q->timeout); |
| 276 | cancel_work_sync(&q->timeout_work); |
| 277 | } |
| 278 | EXPORT_SYMBOL(blk_sync_queue); |
| 279 | |
| 280 | /** |
| 281 | * blk_set_pm_only - increment pm_only counter |
| 282 | * @q: request queue pointer |
| 283 | */ |
| 284 | void blk_set_pm_only(struct request_queue *q) |
| 285 | { |
| 286 | atomic_inc(&q->pm_only); |
| 287 | } |
| 288 | EXPORT_SYMBOL_GPL(blk_set_pm_only); |
| 289 | |
| 290 | void blk_clear_pm_only(struct request_queue *q) |
| 291 | { |
| 292 | int pm_only; |
| 293 | |
| 294 | pm_only = atomic_dec_return(&q->pm_only); |
| 295 | WARN_ON_ONCE(pm_only < 0); |
| 296 | if (pm_only == 0) |
| 297 | wake_up_all(&q->mq_freeze_wq); |
| 298 | } |
| 299 | EXPORT_SYMBOL_GPL(blk_clear_pm_only); |
| 300 | |
| 301 | void blk_put_queue(struct request_queue *q) |
| 302 | { |
| 303 | kobject_put(&q->kobj); |
| 304 | } |
| 305 | EXPORT_SYMBOL(blk_put_queue); |
| 306 | |
| 307 | void blk_set_queue_dying(struct request_queue *q) |
| 308 | { |
| 309 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
| 310 | |
| 311 | /* |
| 312 | * When queue DYING flag is set, we need to block new req |
| 313 | * entering queue, so we call blk_freeze_queue_start() to |
| 314 | * prevent I/O from crossing blk_queue_enter(). |
| 315 | */ |
| 316 | blk_freeze_queue_start(q); |
| 317 | |
| 318 | if (queue_is_mq(q)) |
| 319 | blk_mq_wake_waiters(q); |
| 320 | |
| 321 | /* Make blk_queue_enter() reexamine the DYING flag. */ |
| 322 | wake_up_all(&q->mq_freeze_wq); |
| 323 | } |
| 324 | EXPORT_SYMBOL_GPL(blk_set_queue_dying); |
| 325 | |
| 326 | /** |
| 327 | * blk_cleanup_queue - shutdown a request queue |
| 328 | * @q: request queue to shutdown |
| 329 | * |
| 330 | * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and |
| 331 | * put it. All future requests will be failed immediately with -ENODEV. |
| 332 | */ |
| 333 | void blk_cleanup_queue(struct request_queue *q) |
| 334 | { |
| 335 | /* mark @q DYING, no new request or merges will be allowed afterwards */ |
| 336 | mutex_lock(&q->sysfs_lock); |
| 337 | blk_set_queue_dying(q); |
| 338 | |
| 339 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q); |
| 340 | blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q); |
| 341 | blk_queue_flag_set(QUEUE_FLAG_DYING, q); |
| 342 | mutex_unlock(&q->sysfs_lock); |
| 343 | |
| 344 | /* |
| 345 | * Drain all requests queued before DYING marking. Set DEAD flag to |
| 346 | * prevent that q->request_fn() gets invoked after draining finished. |
| 347 | */ |
| 348 | blk_freeze_queue(q); |
| 349 | |
| 350 | rq_qos_exit(q); |
| 351 | |
| 352 | blk_queue_flag_set(QUEUE_FLAG_DEAD, q); |
| 353 | |
| 354 | /* for synchronous bio-based driver finish in-flight integrity i/o */ |
| 355 | blk_flush_integrity(); |
| 356 | |
| 357 | /* @q won't process any more request, flush async actions */ |
| 358 | del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer); |
| 359 | blk_sync_queue(q); |
| 360 | |
| 361 | if (queue_is_mq(q)) |
| 362 | blk_mq_exit_queue(q); |
| 363 | |
| 364 | /* |
| 365 | * In theory, request pool of sched_tags belongs to request queue. |
| 366 | * However, the current implementation requires tag_set for freeing |
| 367 | * requests, so free the pool now. |
| 368 | * |
| 369 | * Queue has become frozen, there can't be any in-queue requests, so |
| 370 | * it is safe to free requests now. |
| 371 | */ |
| 372 | mutex_lock(&q->sysfs_lock); |
| 373 | if (q->elevator) |
| 374 | blk_mq_sched_free_requests(q); |
| 375 | mutex_unlock(&q->sysfs_lock); |
| 376 | |
| 377 | percpu_ref_exit(&q->q_usage_counter); |
| 378 | |
| 379 | /* @q is and will stay empty, shutdown and put */ |
| 380 | blk_put_queue(q); |
| 381 | } |
| 382 | EXPORT_SYMBOL(blk_cleanup_queue); |
| 383 | |
| 384 | struct request_queue *blk_alloc_queue(gfp_t gfp_mask) |
| 385 | { |
| 386 | return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); |
| 387 | } |
| 388 | EXPORT_SYMBOL(blk_alloc_queue); |
| 389 | |
| 390 | /** |
| 391 | * blk_queue_enter() - try to increase q->q_usage_counter |
| 392 | * @q: request queue pointer |
| 393 | * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT |
| 394 | */ |
| 395 | int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags) |
| 396 | { |
| 397 | const bool pm = flags & BLK_MQ_REQ_PREEMPT; |
| 398 | |
| 399 | while (true) { |
| 400 | bool success = false; |
| 401 | |
| 402 | rcu_read_lock(); |
| 403 | if (percpu_ref_tryget_live(&q->q_usage_counter)) { |
| 404 | /* |
| 405 | * The code that increments the pm_only counter is |
| 406 | * responsible for ensuring that that counter is |
| 407 | * globally visible before the queue is unfrozen. |
| 408 | */ |
| 409 | if (pm || !blk_queue_pm_only(q)) { |
| 410 | success = true; |
| 411 | } else { |
| 412 | percpu_ref_put(&q->q_usage_counter); |
| 413 | } |
| 414 | } |
| 415 | rcu_read_unlock(); |
| 416 | |
| 417 | if (success) |
| 418 | return 0; |
| 419 | |
| 420 | if (flags & BLK_MQ_REQ_NOWAIT) |
| 421 | return -EBUSY; |
| 422 | |
| 423 | /* |
| 424 | * read pair of barrier in blk_freeze_queue_start(), |
| 425 | * we need to order reading __PERCPU_REF_DEAD flag of |
| 426 | * .q_usage_counter and reading .mq_freeze_depth or |
| 427 | * queue dying flag, otherwise the following wait may |
| 428 | * never return if the two reads are reordered. |
| 429 | */ |
| 430 | smp_rmb(); |
| 431 | |
| 432 | wait_event(q->mq_freeze_wq, |
| 433 | (!q->mq_freeze_depth && |
| 434 | (pm || (blk_pm_request_resume(q), |
| 435 | !blk_queue_pm_only(q)))) || |
| 436 | blk_queue_dying(q)); |
| 437 | if (blk_queue_dying(q)) |
| 438 | return -ENODEV; |
| 439 | } |
| 440 | } |
| 441 | |
| 442 | void blk_queue_exit(struct request_queue *q) |
| 443 | { |
| 444 | percpu_ref_put(&q->q_usage_counter); |
| 445 | } |
| 446 | |
| 447 | static void blk_queue_usage_counter_release(struct percpu_ref *ref) |
| 448 | { |
| 449 | struct request_queue *q = |
| 450 | container_of(ref, struct request_queue, q_usage_counter); |
| 451 | |
| 452 | wake_up_all(&q->mq_freeze_wq); |
| 453 | } |
| 454 | |
| 455 | static void blk_rq_timed_out_timer(struct timer_list *t) |
| 456 | { |
| 457 | struct request_queue *q = from_timer(q, t, timeout); |
| 458 | |
| 459 | kblockd_schedule_work(&q->timeout_work); |
| 460 | } |
| 461 | |
| 462 | static void blk_timeout_work(struct work_struct *work) |
| 463 | { |
| 464 | } |
| 465 | |
| 466 | /** |
| 467 | * blk_alloc_queue_node - allocate a request queue |
| 468 | * @gfp_mask: memory allocation flags |
| 469 | * @node_id: NUMA node to allocate memory from |
| 470 | */ |
| 471 | struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) |
| 472 | { |
| 473 | struct request_queue *q; |
| 474 | int ret; |
| 475 | |
| 476 | q = kmem_cache_alloc_node(blk_requestq_cachep, |
| 477 | gfp_mask | __GFP_ZERO, node_id); |
| 478 | if (!q) |
| 479 | return NULL; |
| 480 | |
| 481 | INIT_LIST_HEAD(&q->queue_head); |
| 482 | q->last_merge = NULL; |
| 483 | |
| 484 | q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); |
| 485 | if (q->id < 0) |
| 486 | goto fail_q; |
| 487 | |
| 488 | ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); |
| 489 | if (ret) |
| 490 | goto fail_id; |
| 491 | |
| 492 | q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id); |
| 493 | if (!q->backing_dev_info) |
| 494 | goto fail_split; |
| 495 | |
| 496 | q->stats = blk_alloc_queue_stats(); |
| 497 | if (!q->stats) |
| 498 | goto fail_stats; |
| 499 | |
| 500 | q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES; |
| 501 | q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK; |
| 502 | q->backing_dev_info->name = "block"; |
| 503 | q->node = node_id; |
| 504 | |
| 505 | timer_setup(&q->backing_dev_info->laptop_mode_wb_timer, |
| 506 | laptop_mode_timer_fn, 0); |
| 507 | timer_setup(&q->timeout, blk_rq_timed_out_timer, 0); |
| 508 | INIT_WORK(&q->timeout_work, blk_timeout_work); |
| 509 | INIT_LIST_HEAD(&q->icq_list); |
| 510 | #ifdef CONFIG_BLK_CGROUP |
| 511 | INIT_LIST_HEAD(&q->blkg_list); |
| 512 | #endif |
| 513 | |
| 514 | kobject_init(&q->kobj, &blk_queue_ktype); |
| 515 | |
| 516 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
| 517 | mutex_init(&q->blk_trace_mutex); |
| 518 | #endif |
| 519 | mutex_init(&q->sysfs_lock); |
| 520 | spin_lock_init(&q->queue_lock); |
| 521 | |
| 522 | init_waitqueue_head(&q->mq_freeze_wq); |
| 523 | mutex_init(&q->mq_freeze_lock); |
| 524 | |
| 525 | /* |
| 526 | * Init percpu_ref in atomic mode so that it's faster to shutdown. |
| 527 | * See blk_register_queue() for details. |
| 528 | */ |
| 529 | if (percpu_ref_init(&q->q_usage_counter, |
| 530 | blk_queue_usage_counter_release, |
| 531 | PERCPU_REF_INIT_ATOMIC, GFP_KERNEL)) |
| 532 | goto fail_bdi; |
| 533 | |
| 534 | if (blkcg_init_queue(q)) |
| 535 | goto fail_ref; |
| 536 | |
| 537 | return q; |
| 538 | |
| 539 | fail_ref: |
| 540 | percpu_ref_exit(&q->q_usage_counter); |
| 541 | fail_bdi: |
| 542 | blk_free_queue_stats(q->stats); |
| 543 | fail_stats: |
| 544 | bdi_put(q->backing_dev_info); |
| 545 | fail_split: |
| 546 | bioset_exit(&q->bio_split); |
| 547 | fail_id: |
| 548 | ida_simple_remove(&blk_queue_ida, q->id); |
| 549 | fail_q: |
| 550 | kmem_cache_free(blk_requestq_cachep, q); |
| 551 | return NULL; |
| 552 | } |
| 553 | EXPORT_SYMBOL(blk_alloc_queue_node); |
| 554 | |
| 555 | bool blk_get_queue(struct request_queue *q) |
| 556 | { |
| 557 | if (likely(!blk_queue_dying(q))) { |
| 558 | __blk_get_queue(q); |
| 559 | return true; |
| 560 | } |
| 561 | |
| 562 | return false; |
| 563 | } |
| 564 | EXPORT_SYMBOL(blk_get_queue); |
| 565 | |
| 566 | /** |
| 567 | * blk_get_request - allocate a request |
| 568 | * @q: request queue to allocate a request for |
| 569 | * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC. |
| 570 | * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT. |
| 571 | */ |
| 572 | struct request *blk_get_request(struct request_queue *q, unsigned int op, |
| 573 | blk_mq_req_flags_t flags) |
| 574 | { |
| 575 | struct request *req; |
| 576 | |
| 577 | WARN_ON_ONCE(op & REQ_NOWAIT); |
| 578 | WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT)); |
| 579 | |
| 580 | req = blk_mq_alloc_request(q, op, flags); |
| 581 | if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn) |
| 582 | q->mq_ops->initialize_rq_fn(req); |
| 583 | |
| 584 | return req; |
| 585 | } |
| 586 | EXPORT_SYMBOL(blk_get_request); |
| 587 | |
| 588 | void blk_put_request(struct request *req) |
| 589 | { |
| 590 | blk_mq_free_request(req); |
| 591 | } |
| 592 | EXPORT_SYMBOL(blk_put_request); |
| 593 | |
| 594 | bool bio_attempt_back_merge(struct request *req, struct bio *bio, |
| 595 | unsigned int nr_segs) |
| 596 | { |
| 597 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
| 598 | |
| 599 | if (!ll_back_merge_fn(req, bio, nr_segs)) |
| 600 | return false; |
| 601 | |
| 602 | trace_block_bio_backmerge(req->q, req, bio); |
| 603 | |
| 604 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
| 605 | blk_rq_set_mixed_merge(req); |
| 606 | |
| 607 | req->biotail->bi_next = bio; |
| 608 | req->biotail = bio; |
| 609 | req->__data_len += bio->bi_iter.bi_size; |
| 610 | |
| 611 | blk_account_io_start(req, false); |
| 612 | return true; |
| 613 | } |
| 614 | |
| 615 | bool bio_attempt_front_merge(struct request *req, struct bio *bio, |
| 616 | unsigned int nr_segs) |
| 617 | { |
| 618 | const int ff = bio->bi_opf & REQ_FAILFAST_MASK; |
| 619 | |
| 620 | if (!ll_front_merge_fn(req, bio, nr_segs)) |
| 621 | return false; |
| 622 | |
| 623 | trace_block_bio_frontmerge(req->q, req, bio); |
| 624 | |
| 625 | if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) |
| 626 | blk_rq_set_mixed_merge(req); |
| 627 | |
| 628 | bio->bi_next = req->bio; |
| 629 | req->bio = bio; |
| 630 | |
| 631 | req->__sector = bio->bi_iter.bi_sector; |
| 632 | req->__data_len += bio->bi_iter.bi_size; |
| 633 | |
| 634 | blk_account_io_start(req, false); |
| 635 | return true; |
| 636 | } |
| 637 | |
| 638 | bool bio_attempt_discard_merge(struct request_queue *q, struct request *req, |
| 639 | struct bio *bio) |
| 640 | { |
| 641 | unsigned short segments = blk_rq_nr_discard_segments(req); |
| 642 | |
| 643 | if (segments >= queue_max_discard_segments(q)) |
| 644 | goto no_merge; |
| 645 | if (blk_rq_sectors(req) + bio_sectors(bio) > |
| 646 | blk_rq_get_max_sectors(req, blk_rq_pos(req))) |
| 647 | goto no_merge; |
| 648 | |
| 649 | req->biotail->bi_next = bio; |
| 650 | req->biotail = bio; |
| 651 | req->__data_len += bio->bi_iter.bi_size; |
| 652 | req->nr_phys_segments = segments + 1; |
| 653 | |
| 654 | blk_account_io_start(req, false); |
| 655 | return true; |
| 656 | no_merge: |
| 657 | req_set_nomerge(q, req); |
| 658 | return false; |
| 659 | } |
| 660 | |
| 661 | /** |
| 662 | * blk_attempt_plug_merge - try to merge with %current's plugged list |
| 663 | * @q: request_queue new bio is being queued at |
| 664 | * @bio: new bio being queued |
| 665 | * @nr_segs: number of segments in @bio |
| 666 | * @same_queue_rq: pointer to &struct request that gets filled in when |
| 667 | * another request associated with @q is found on the plug list |
| 668 | * (optional, may be %NULL) |
| 669 | * |
| 670 | * Determine whether @bio being queued on @q can be merged with a request |
| 671 | * on %current's plugged list. Returns %true if merge was successful, |
| 672 | * otherwise %false. |
| 673 | * |
| 674 | * Plugging coalesces IOs from the same issuer for the same purpose without |
| 675 | * going through @q->queue_lock. As such it's more of an issuing mechanism |
| 676 | * than scheduling, and the request, while may have elvpriv data, is not |
| 677 | * added on the elevator at this point. In addition, we don't have |
| 678 | * reliable access to the elevator outside queue lock. Only check basic |
| 679 | * merging parameters without querying the elevator. |
| 680 | * |
| 681 | * Caller must ensure !blk_queue_nomerges(q) beforehand. |
| 682 | */ |
| 683 | bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, |
| 684 | unsigned int nr_segs, struct request **same_queue_rq) |
| 685 | { |
| 686 | struct blk_plug *plug; |
| 687 | struct request *rq; |
| 688 | struct list_head *plug_list; |
| 689 | |
| 690 | plug = current->plug; |
| 691 | if (!plug) |
| 692 | return false; |
| 693 | |
| 694 | plug_list = &plug->mq_list; |
| 695 | |
| 696 | list_for_each_entry_reverse(rq, plug_list, queuelist) { |
| 697 | bool merged = false; |
| 698 | |
| 699 | if (rq->q == q && same_queue_rq) { |
| 700 | /* |
| 701 | * Only blk-mq multiple hardware queues case checks the |
| 702 | * rq in the same queue, there should be only one such |
| 703 | * rq in a queue |
| 704 | **/ |
| 705 | *same_queue_rq = rq; |
| 706 | } |
| 707 | |
| 708 | if (rq->q != q || !blk_rq_merge_ok(rq, bio)) |
| 709 | continue; |
| 710 | |
| 711 | switch (blk_try_merge(rq, bio)) { |
| 712 | case ELEVATOR_BACK_MERGE: |
| 713 | merged = bio_attempt_back_merge(rq, bio, nr_segs); |
| 714 | break; |
| 715 | case ELEVATOR_FRONT_MERGE: |
| 716 | merged = bio_attempt_front_merge(rq, bio, nr_segs); |
| 717 | break; |
| 718 | case ELEVATOR_DISCARD_MERGE: |
| 719 | merged = bio_attempt_discard_merge(q, rq, bio); |
| 720 | break; |
| 721 | default: |
| 722 | break; |
| 723 | } |
| 724 | |
| 725 | if (merged) |
| 726 | return true; |
| 727 | } |
| 728 | |
| 729 | return false; |
| 730 | } |
| 731 | |
| 732 | static void handle_bad_sector(struct bio *bio, sector_t maxsector) |
| 733 | { |
| 734 | char b[BDEVNAME_SIZE]; |
| 735 | |
| 736 | printk(KERN_INFO "attempt to access beyond end of device\n"); |
| 737 | printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n", |
| 738 | bio_devname(bio, b), bio->bi_opf, |
| 739 | (unsigned long long)bio_end_sector(bio), |
| 740 | (long long)maxsector); |
| 741 | } |
| 742 | |
| 743 | #ifdef CONFIG_FAIL_MAKE_REQUEST |
| 744 | |
| 745 | static DECLARE_FAULT_ATTR(fail_make_request); |
| 746 | |
| 747 | static int __init setup_fail_make_request(char *str) |
| 748 | { |
| 749 | return setup_fault_attr(&fail_make_request, str); |
| 750 | } |
| 751 | __setup("fail_make_request=", setup_fail_make_request); |
| 752 | |
| 753 | static bool should_fail_request(struct hd_struct *part, unsigned int bytes) |
| 754 | { |
| 755 | return part->make_it_fail && should_fail(&fail_make_request, bytes); |
| 756 | } |
| 757 | |
| 758 | static int __init fail_make_request_debugfs(void) |
| 759 | { |
| 760 | struct dentry *dir = fault_create_debugfs_attr("fail_make_request", |
| 761 | NULL, &fail_make_request); |
| 762 | |
| 763 | return PTR_ERR_OR_ZERO(dir); |
| 764 | } |
| 765 | |
| 766 | late_initcall(fail_make_request_debugfs); |
| 767 | |
| 768 | #else /* CONFIG_FAIL_MAKE_REQUEST */ |
| 769 | |
| 770 | static inline bool should_fail_request(struct hd_struct *part, |
| 771 | unsigned int bytes) |
| 772 | { |
| 773 | return false; |
| 774 | } |
| 775 | |
| 776 | #endif /* CONFIG_FAIL_MAKE_REQUEST */ |
| 777 | |
| 778 | static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part) |
| 779 | { |
| 780 | const int op = bio_op(bio); |
| 781 | |
| 782 | if (part->policy && op_is_write(op)) { |
| 783 | char b[BDEVNAME_SIZE]; |
| 784 | |
| 785 | if (op_is_flush(bio->bi_opf) && !bio_sectors(bio)) |
| 786 | return false; |
| 787 | |
| 788 | WARN_ONCE(1, |
| 789 | "generic_make_request: Trying to write " |
| 790 | "to read-only block-device %s (partno %d)\n", |
| 791 | bio_devname(bio, b), part->partno); |
| 792 | /* Older lvm-tools actually trigger this */ |
| 793 | return false; |
| 794 | } |
| 795 | |
| 796 | return false; |
| 797 | } |
| 798 | |
| 799 | static noinline int should_fail_bio(struct bio *bio) |
| 800 | { |
| 801 | if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size)) |
| 802 | return -EIO; |
| 803 | return 0; |
| 804 | } |
| 805 | ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO); |
| 806 | |
| 807 | /* |
| 808 | * Check whether this bio extends beyond the end of the device or partition. |
| 809 | * This may well happen - the kernel calls bread() without checking the size of |
| 810 | * the device, e.g., when mounting a file system. |
| 811 | */ |
| 812 | static inline int bio_check_eod(struct bio *bio, sector_t maxsector) |
| 813 | { |
| 814 | unsigned int nr_sectors = bio_sectors(bio); |
| 815 | |
| 816 | if (nr_sectors && maxsector && |
| 817 | (nr_sectors > maxsector || |
| 818 | bio->bi_iter.bi_sector > maxsector - nr_sectors)) { |
| 819 | handle_bad_sector(bio, maxsector); |
| 820 | return -EIO; |
| 821 | } |
| 822 | return 0; |
| 823 | } |
| 824 | |
| 825 | /* |
| 826 | * Remap block n of partition p to block n+start(p) of the disk. |
| 827 | */ |
| 828 | static inline int blk_partition_remap(struct bio *bio) |
| 829 | { |
| 830 | struct hd_struct *p; |
| 831 | int ret = -EIO; |
| 832 | |
| 833 | rcu_read_lock(); |
| 834 | p = __disk_get_part(bio->bi_disk, bio->bi_partno); |
| 835 | if (unlikely(!p)) |
| 836 | goto out; |
| 837 | if (unlikely(should_fail_request(p, bio->bi_iter.bi_size))) |
| 838 | goto out; |
| 839 | if (unlikely(bio_check_ro(bio, p))) |
| 840 | goto out; |
| 841 | |
| 842 | /* |
| 843 | * Zone reset does not include bi_size so bio_sectors() is always 0. |
| 844 | * Include a test for the reset op code and perform the remap if needed. |
| 845 | */ |
| 846 | if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) { |
| 847 | if (bio_check_eod(bio, part_nr_sects_read(p))) |
| 848 | goto out; |
| 849 | bio->bi_iter.bi_sector += p->start_sect; |
| 850 | trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p), |
| 851 | bio->bi_iter.bi_sector - p->start_sect); |
| 852 | } |
| 853 | bio->bi_partno = 0; |
| 854 | ret = 0; |
| 855 | out: |
| 856 | rcu_read_unlock(); |
| 857 | return ret; |
| 858 | } |
| 859 | |
| 860 | static noinline_for_stack bool |
| 861 | generic_make_request_checks(struct bio *bio) |
| 862 | { |
| 863 | struct request_queue *q; |
| 864 | int nr_sectors = bio_sectors(bio); |
| 865 | blk_status_t status = BLK_STS_IOERR; |
| 866 | char b[BDEVNAME_SIZE]; |
| 867 | |
| 868 | might_sleep(); |
| 869 | |
| 870 | q = bio->bi_disk->queue; |
| 871 | if (unlikely(!q)) { |
| 872 | printk(KERN_ERR |
| 873 | "generic_make_request: Trying to access " |
| 874 | "nonexistent block-device %s (%Lu)\n", |
| 875 | bio_devname(bio, b), (long long)bio->bi_iter.bi_sector); |
| 876 | goto end_io; |
| 877 | } |
| 878 | |
| 879 | /* |
| 880 | * For a REQ_NOWAIT based request, return -EOPNOTSUPP |
| 881 | * if queue is not a request based queue. |
| 882 | */ |
| 883 | if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q)) |
| 884 | goto not_supported; |
| 885 | |
| 886 | if (should_fail_bio(bio)) |
| 887 | goto end_io; |
| 888 | |
| 889 | if (bio->bi_partno) { |
| 890 | if (unlikely(blk_partition_remap(bio))) |
| 891 | goto end_io; |
| 892 | } else { |
| 893 | if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0))) |
| 894 | goto end_io; |
| 895 | if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk)))) |
| 896 | goto end_io; |
| 897 | } |
| 898 | |
| 899 | /* |
| 900 | * Filter flush bio's early so that make_request based |
| 901 | * drivers without flush support don't have to worry |
| 902 | * about them. |
| 903 | */ |
| 904 | if (op_is_flush(bio->bi_opf) && |
| 905 | !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) { |
| 906 | bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA); |
| 907 | if (!nr_sectors) { |
| 908 | status = BLK_STS_OK; |
| 909 | goto end_io; |
| 910 | } |
| 911 | } |
| 912 | |
| 913 | if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags)) |
| 914 | bio->bi_opf &= ~REQ_HIPRI; |
| 915 | |
| 916 | switch (bio_op(bio)) { |
| 917 | case REQ_OP_DISCARD: |
| 918 | if (!blk_queue_discard(q)) |
| 919 | goto not_supported; |
| 920 | break; |
| 921 | case REQ_OP_SECURE_ERASE: |
| 922 | if (!blk_queue_secure_erase(q)) |
| 923 | goto not_supported; |
| 924 | break; |
| 925 | case REQ_OP_WRITE_SAME: |
| 926 | if (!q->limits.max_write_same_sectors) |
| 927 | goto not_supported; |
| 928 | break; |
| 929 | case REQ_OP_ZONE_RESET: |
| 930 | if (!blk_queue_is_zoned(q)) |
| 931 | goto not_supported; |
| 932 | break; |
| 933 | case REQ_OP_WRITE_ZEROES: |
| 934 | if (!q->limits.max_write_zeroes_sectors) |
| 935 | goto not_supported; |
| 936 | break; |
| 937 | default: |
| 938 | break; |
| 939 | } |
| 940 | |
| 941 | /* |
| 942 | * Various block parts want %current->io_context and lazy ioc |
| 943 | * allocation ends up trading a lot of pain for a small amount of |
| 944 | * memory. Just allocate it upfront. This may fail and block |
| 945 | * layer knows how to live with it. |
| 946 | */ |
| 947 | create_io_context(GFP_ATOMIC, q->node); |
| 948 | |
| 949 | if (!blkcg_bio_issue_check(q, bio)) |
| 950 | return false; |
| 951 | |
| 952 | if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) { |
| 953 | trace_block_bio_queue(q, bio); |
| 954 | /* Now that enqueuing has been traced, we need to trace |
| 955 | * completion as well. |
| 956 | */ |
| 957 | bio_set_flag(bio, BIO_TRACE_COMPLETION); |
| 958 | } |
| 959 | return true; |
| 960 | |
| 961 | not_supported: |
| 962 | status = BLK_STS_NOTSUPP; |
| 963 | end_io: |
| 964 | bio->bi_status = status; |
| 965 | bio_endio(bio); |
| 966 | return false; |
| 967 | } |
| 968 | |
| 969 | /** |
| 970 | * generic_make_request - hand a buffer to its device driver for I/O |
| 971 | * @bio: The bio describing the location in memory and on the device. |
| 972 | * |
| 973 | * generic_make_request() is used to make I/O requests of block |
| 974 | * devices. It is passed a &struct bio, which describes the I/O that needs |
| 975 | * to be done. |
| 976 | * |
| 977 | * generic_make_request() does not return any status. The |
| 978 | * success/failure status of the request, along with notification of |
| 979 | * completion, is delivered asynchronously through the bio->bi_end_io |
| 980 | * function described (one day) else where. |
| 981 | * |
| 982 | * The caller of generic_make_request must make sure that bi_io_vec |
| 983 | * are set to describe the memory buffer, and that bi_dev and bi_sector are |
| 984 | * set to describe the device address, and the |
| 985 | * bi_end_io and optionally bi_private are set to describe how |
| 986 | * completion notification should be signaled. |
| 987 | * |
| 988 | * generic_make_request and the drivers it calls may use bi_next if this |
| 989 | * bio happens to be merged with someone else, and may resubmit the bio to |
| 990 | * a lower device by calling into generic_make_request recursively, which |
| 991 | * means the bio should NOT be touched after the call to ->make_request_fn. |
| 992 | */ |
| 993 | blk_qc_t generic_make_request(struct bio *bio) |
| 994 | { |
| 995 | /* |
| 996 | * bio_list_on_stack[0] contains bios submitted by the current |
| 997 | * make_request_fn. |
| 998 | * bio_list_on_stack[1] contains bios that were submitted before |
| 999 | * the current make_request_fn, but that haven't been processed |
| 1000 | * yet. |
| 1001 | */ |
| 1002 | struct bio_list bio_list_on_stack[2]; |
| 1003 | blk_qc_t ret = BLK_QC_T_NONE; |
| 1004 | |
| 1005 | if (!generic_make_request_checks(bio)) |
| 1006 | goto out; |
| 1007 | |
| 1008 | /* |
| 1009 | * We only want one ->make_request_fn to be active at a time, else |
| 1010 | * stack usage with stacked devices could be a problem. So use |
| 1011 | * current->bio_list to keep a list of requests submited by a |
| 1012 | * make_request_fn function. current->bio_list is also used as a |
| 1013 | * flag to say if generic_make_request is currently active in this |
| 1014 | * task or not. If it is NULL, then no make_request is active. If |
| 1015 | * it is non-NULL, then a make_request is active, and new requests |
| 1016 | * should be added at the tail |
| 1017 | */ |
| 1018 | if (current->bio_list) { |
| 1019 | bio_list_add(¤t->bio_list[0], bio); |
| 1020 | goto out; |
| 1021 | } |
| 1022 | |
| 1023 | /* following loop may be a bit non-obvious, and so deserves some |
| 1024 | * explanation. |
| 1025 | * Before entering the loop, bio->bi_next is NULL (as all callers |
| 1026 | * ensure that) so we have a list with a single bio. |
| 1027 | * We pretend that we have just taken it off a longer list, so |
| 1028 | * we assign bio_list to a pointer to the bio_list_on_stack, |
| 1029 | * thus initialising the bio_list of new bios to be |
| 1030 | * added. ->make_request() may indeed add some more bios |
| 1031 | * through a recursive call to generic_make_request. If it |
| 1032 | * did, we find a non-NULL value in bio_list and re-enter the loop |
| 1033 | * from the top. In this case we really did just take the bio |
| 1034 | * of the top of the list (no pretending) and so remove it from |
| 1035 | * bio_list, and call into ->make_request() again. |
| 1036 | */ |
| 1037 | BUG_ON(bio->bi_next); |
| 1038 | bio_list_init(&bio_list_on_stack[0]); |
| 1039 | current->bio_list = bio_list_on_stack; |
| 1040 | do { |
| 1041 | struct request_queue *q = bio->bi_disk->queue; |
| 1042 | blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ? |
| 1043 | BLK_MQ_REQ_NOWAIT : 0; |
| 1044 | |
| 1045 | if (likely(blk_queue_enter(q, flags) == 0)) { |
| 1046 | struct bio_list lower, same; |
| 1047 | |
| 1048 | /* Create a fresh bio_list for all subordinate requests */ |
| 1049 | bio_list_on_stack[1] = bio_list_on_stack[0]; |
| 1050 | bio_list_init(&bio_list_on_stack[0]); |
| 1051 | ret = q->make_request_fn(q, bio); |
| 1052 | |
| 1053 | blk_queue_exit(q); |
| 1054 | |
| 1055 | /* sort new bios into those for a lower level |
| 1056 | * and those for the same level |
| 1057 | */ |
| 1058 | bio_list_init(&lower); |
| 1059 | bio_list_init(&same); |
| 1060 | while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL) |
| 1061 | if (q == bio->bi_disk->queue) |
| 1062 | bio_list_add(&same, bio); |
| 1063 | else |
| 1064 | bio_list_add(&lower, bio); |
| 1065 | /* now assemble so we handle the lowest level first */ |
| 1066 | bio_list_merge(&bio_list_on_stack[0], &lower); |
| 1067 | bio_list_merge(&bio_list_on_stack[0], &same); |
| 1068 | bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]); |
| 1069 | } else { |
| 1070 | if (unlikely(!blk_queue_dying(q) && |
| 1071 | (bio->bi_opf & REQ_NOWAIT))) |
| 1072 | bio_wouldblock_error(bio); |
| 1073 | else |
| 1074 | bio_io_error(bio); |
| 1075 | } |
| 1076 | bio = bio_list_pop(&bio_list_on_stack[0]); |
| 1077 | } while (bio); |
| 1078 | current->bio_list = NULL; /* deactivate */ |
| 1079 | |
| 1080 | out: |
| 1081 | return ret; |
| 1082 | } |
| 1083 | EXPORT_SYMBOL(generic_make_request); |
| 1084 | |
| 1085 | /** |
| 1086 | * direct_make_request - hand a buffer directly to its device driver for I/O |
| 1087 | * @bio: The bio describing the location in memory and on the device. |
| 1088 | * |
| 1089 | * This function behaves like generic_make_request(), but does not protect |
| 1090 | * against recursion. Must only be used if the called driver is known |
| 1091 | * to not call generic_make_request (or direct_make_request) again from |
| 1092 | * its make_request function. (Calling direct_make_request again from |
| 1093 | * a workqueue is perfectly fine as that doesn't recurse). |
| 1094 | */ |
| 1095 | blk_qc_t direct_make_request(struct bio *bio) |
| 1096 | { |
| 1097 | struct request_queue *q = bio->bi_disk->queue; |
| 1098 | bool nowait = bio->bi_opf & REQ_NOWAIT; |
| 1099 | blk_qc_t ret; |
| 1100 | |
| 1101 | if (!generic_make_request_checks(bio)) |
| 1102 | return BLK_QC_T_NONE; |
| 1103 | |
| 1104 | if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) { |
| 1105 | if (nowait && !blk_queue_dying(q)) |
| 1106 | bio->bi_status = BLK_STS_AGAIN; |
| 1107 | else |
| 1108 | bio->bi_status = BLK_STS_IOERR; |
| 1109 | bio_endio(bio); |
| 1110 | return BLK_QC_T_NONE; |
| 1111 | } |
| 1112 | |
| 1113 | ret = q->make_request_fn(q, bio); |
| 1114 | blk_queue_exit(q); |
| 1115 | return ret; |
| 1116 | } |
| 1117 | EXPORT_SYMBOL_GPL(direct_make_request); |
| 1118 | |
| 1119 | /** |
| 1120 | * submit_bio - submit a bio to the block device layer for I/O |
| 1121 | * @bio: The &struct bio which describes the I/O |
| 1122 | * |
| 1123 | * submit_bio() is very similar in purpose to generic_make_request(), and |
| 1124 | * uses that function to do most of the work. Both are fairly rough |
| 1125 | * interfaces; @bio must be presetup and ready for I/O. |
| 1126 | * |
| 1127 | */ |
| 1128 | blk_qc_t submit_bio(struct bio *bio) |
| 1129 | { |
| 1130 | /* |
| 1131 | * If it's a regular read/write or a barrier with data attached, |
| 1132 | * go through the normal accounting stuff before submission. |
| 1133 | */ |
| 1134 | if (bio_has_data(bio)) { |
| 1135 | unsigned int count; |
| 1136 | |
| 1137 | if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME)) |
| 1138 | count = queue_logical_block_size(bio->bi_disk->queue) >> 9; |
| 1139 | else |
| 1140 | count = bio_sectors(bio); |
| 1141 | |
| 1142 | if (op_is_write(bio_op(bio))) { |
| 1143 | count_vm_events(PGPGOUT, count); |
| 1144 | } else { |
| 1145 | task_io_account_read(bio->bi_iter.bi_size); |
| 1146 | count_vm_events(PGPGIN, count); |
| 1147 | } |
| 1148 | |
| 1149 | if (unlikely(block_dump)) { |
| 1150 | char b[BDEVNAME_SIZE]; |
| 1151 | printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", |
| 1152 | current->comm, task_pid_nr(current), |
| 1153 | op_is_write(bio_op(bio)) ? "WRITE" : "READ", |
| 1154 | (unsigned long long)bio->bi_iter.bi_sector, |
| 1155 | bio_devname(bio, b), count); |
| 1156 | } |
| 1157 | } |
| 1158 | |
| 1159 | return generic_make_request(bio); |
| 1160 | } |
| 1161 | EXPORT_SYMBOL(submit_bio); |
| 1162 | |
| 1163 | /** |
| 1164 | * blk_cloned_rq_check_limits - Helper function to check a cloned request |
| 1165 | * for new the queue limits |
| 1166 | * @q: the queue |
| 1167 | * @rq: the request being checked |
| 1168 | * |
| 1169 | * Description: |
| 1170 | * @rq may have been made based on weaker limitations of upper-level queues |
| 1171 | * in request stacking drivers, and it may violate the limitation of @q. |
| 1172 | * Since the block layer and the underlying device driver trust @rq |
| 1173 | * after it is inserted to @q, it should be checked against @q before |
| 1174 | * the insertion using this generic function. |
| 1175 | * |
| 1176 | * Request stacking drivers like request-based dm may change the queue |
| 1177 | * limits when retrying requests on other queues. Those requests need |
| 1178 | * to be checked against the new queue limits again during dispatch. |
| 1179 | */ |
| 1180 | static int blk_cloned_rq_check_limits(struct request_queue *q, |
| 1181 | struct request *rq) |
| 1182 | { |
| 1183 | if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) { |
| 1184 | printk(KERN_ERR "%s: over max size limit. (%u > %u)\n", |
| 1185 | __func__, blk_rq_sectors(rq), |
| 1186 | blk_queue_get_max_sectors(q, req_op(rq))); |
| 1187 | return -EIO; |
| 1188 | } |
| 1189 | |
| 1190 | /* |
| 1191 | * queue's settings related to segment counting like q->bounce_pfn |
| 1192 | * may differ from that of other stacking queues. |
| 1193 | * Recalculate it to check the request correctly on this queue's |
| 1194 | * limitation. |
| 1195 | */ |
| 1196 | rq->nr_phys_segments = blk_recalc_rq_segments(rq); |
| 1197 | if (rq->nr_phys_segments > queue_max_segments(q)) { |
| 1198 | printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n", |
| 1199 | __func__, rq->nr_phys_segments, queue_max_segments(q)); |
| 1200 | return -EIO; |
| 1201 | } |
| 1202 | |
| 1203 | return 0; |
| 1204 | } |
| 1205 | |
| 1206 | /** |
| 1207 | * blk_insert_cloned_request - Helper for stacking drivers to submit a request |
| 1208 | * @q: the queue to submit the request |
| 1209 | * @rq: the request being queued |
| 1210 | */ |
| 1211 | blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq) |
| 1212 | { |
| 1213 | if (blk_cloned_rq_check_limits(q, rq)) |
| 1214 | return BLK_STS_IOERR; |
| 1215 | |
| 1216 | if (rq->rq_disk && |
| 1217 | should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) |
| 1218 | return BLK_STS_IOERR; |
| 1219 | |
| 1220 | if (blk_queue_io_stat(q)) |
| 1221 | blk_account_io_start(rq, true); |
| 1222 | |
| 1223 | /* |
| 1224 | * Since we have a scheduler attached on the top device, |
| 1225 | * bypass a potential scheduler on the bottom device for |
| 1226 | * insert. |
| 1227 | */ |
| 1228 | return blk_mq_request_issue_directly(rq, true); |
| 1229 | } |
| 1230 | EXPORT_SYMBOL_GPL(blk_insert_cloned_request); |
| 1231 | |
| 1232 | /** |
| 1233 | * blk_rq_err_bytes - determine number of bytes till the next failure boundary |
| 1234 | * @rq: request to examine |
| 1235 | * |
| 1236 | * Description: |
| 1237 | * A request could be merge of IOs which require different failure |
| 1238 | * handling. This function determines the number of bytes which |
| 1239 | * can be failed from the beginning of the request without |
| 1240 | * crossing into area which need to be retried further. |
| 1241 | * |
| 1242 | * Return: |
| 1243 | * The number of bytes to fail. |
| 1244 | */ |
| 1245 | unsigned int blk_rq_err_bytes(const struct request *rq) |
| 1246 | { |
| 1247 | unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; |
| 1248 | unsigned int bytes = 0; |
| 1249 | struct bio *bio; |
| 1250 | |
| 1251 | if (!(rq->rq_flags & RQF_MIXED_MERGE)) |
| 1252 | return blk_rq_bytes(rq); |
| 1253 | |
| 1254 | /* |
| 1255 | * Currently the only 'mixing' which can happen is between |
| 1256 | * different fastfail types. We can safely fail portions |
| 1257 | * which have all the failfast bits that the first one has - |
| 1258 | * the ones which are at least as eager to fail as the first |
| 1259 | * one. |
| 1260 | */ |
| 1261 | for (bio = rq->bio; bio; bio = bio->bi_next) { |
| 1262 | if ((bio->bi_opf & ff) != ff) |
| 1263 | break; |
| 1264 | bytes += bio->bi_iter.bi_size; |
| 1265 | } |
| 1266 | |
| 1267 | /* this could lead to infinite loop */ |
| 1268 | BUG_ON(blk_rq_bytes(rq) && !bytes); |
| 1269 | return bytes; |
| 1270 | } |
| 1271 | EXPORT_SYMBOL_GPL(blk_rq_err_bytes); |
| 1272 | |
| 1273 | void blk_account_io_completion(struct request *req, unsigned int bytes) |
| 1274 | { |
| 1275 | if (blk_do_io_stat(req)) { |
| 1276 | const int sgrp = op_stat_group(req_op(req)); |
| 1277 | struct hd_struct *part; |
| 1278 | |
| 1279 | part_stat_lock(); |
| 1280 | part = req->part; |
| 1281 | part_stat_add(part, sectors[sgrp], bytes >> 9); |
| 1282 | part_stat_unlock(); |
| 1283 | } |
| 1284 | } |
| 1285 | |
| 1286 | void blk_account_io_done(struct request *req, u64 now) |
| 1287 | { |
| 1288 | /* |
| 1289 | * Account IO completion. flush_rq isn't accounted as a |
| 1290 | * normal IO on queueing nor completion. Accounting the |
| 1291 | * containing request is enough. |
| 1292 | */ |
| 1293 | if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) { |
| 1294 | const int sgrp = op_stat_group(req_op(req)); |
| 1295 | struct hd_struct *part; |
| 1296 | |
| 1297 | part_stat_lock(); |
| 1298 | part = req->part; |
| 1299 | |
| 1300 | update_io_ticks(part, jiffies); |
| 1301 | part_stat_inc(part, ios[sgrp]); |
| 1302 | part_stat_add(part, nsecs[sgrp], now - req->start_time_ns); |
| 1303 | part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns)); |
| 1304 | part_dec_in_flight(req->q, part, rq_data_dir(req)); |
| 1305 | |
| 1306 | hd_struct_put(part); |
| 1307 | part_stat_unlock(); |
| 1308 | } |
| 1309 | } |
| 1310 | |
| 1311 | void blk_account_io_start(struct request *rq, bool new_io) |
| 1312 | { |
| 1313 | struct hd_struct *part; |
| 1314 | int rw = rq_data_dir(rq); |
| 1315 | |
| 1316 | if (!blk_do_io_stat(rq)) |
| 1317 | return; |
| 1318 | |
| 1319 | part_stat_lock(); |
| 1320 | |
| 1321 | if (!new_io) { |
| 1322 | part = rq->part; |
| 1323 | part_stat_inc(part, merges[rw]); |
| 1324 | } else { |
| 1325 | part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); |
| 1326 | if (!hd_struct_try_get(part)) { |
| 1327 | /* |
| 1328 | * The partition is already being removed, |
| 1329 | * the request will be accounted on the disk only |
| 1330 | * |
| 1331 | * We take a reference on disk->part0 although that |
| 1332 | * partition will never be deleted, so we can treat |
| 1333 | * it as any other partition. |
| 1334 | */ |
| 1335 | part = &rq->rq_disk->part0; |
| 1336 | hd_struct_get(part); |
| 1337 | } |
| 1338 | part_inc_in_flight(rq->q, part, rw); |
| 1339 | rq->part = part; |
| 1340 | } |
| 1341 | |
| 1342 | update_io_ticks(part, jiffies); |
| 1343 | |
| 1344 | part_stat_unlock(); |
| 1345 | } |
| 1346 | |
| 1347 | /* |
| 1348 | * Steal bios from a request and add them to a bio list. |
| 1349 | * The request must not have been partially completed before. |
| 1350 | */ |
| 1351 | void blk_steal_bios(struct bio_list *list, struct request *rq) |
| 1352 | { |
| 1353 | if (rq->bio) { |
| 1354 | if (list->tail) |
| 1355 | list->tail->bi_next = rq->bio; |
| 1356 | else |
| 1357 | list->head = rq->bio; |
| 1358 | list->tail = rq->biotail; |
| 1359 | |
| 1360 | rq->bio = NULL; |
| 1361 | rq->biotail = NULL; |
| 1362 | } |
| 1363 | |
| 1364 | rq->__data_len = 0; |
| 1365 | } |
| 1366 | EXPORT_SYMBOL_GPL(blk_steal_bios); |
| 1367 | |
| 1368 | /** |
| 1369 | * blk_update_request - Special helper function for request stacking drivers |
| 1370 | * @req: the request being processed |
| 1371 | * @error: block status code |
| 1372 | * @nr_bytes: number of bytes to complete @req |
| 1373 | * |
| 1374 | * Description: |
| 1375 | * Ends I/O on a number of bytes attached to @req, but doesn't complete |
| 1376 | * the request structure even if @req doesn't have leftover. |
| 1377 | * If @req has leftover, sets it up for the next range of segments. |
| 1378 | * |
| 1379 | * This special helper function is only for request stacking drivers |
| 1380 | * (e.g. request-based dm) so that they can handle partial completion. |
| 1381 | * Actual device drivers should use blk_mq_end_request instead. |
| 1382 | * |
| 1383 | * Passing the result of blk_rq_bytes() as @nr_bytes guarantees |
| 1384 | * %false return from this function. |
| 1385 | * |
| 1386 | * Note: |
| 1387 | * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both |
| 1388 | * blk_rq_bytes() and in blk_update_request(). |
| 1389 | * |
| 1390 | * Return: |
| 1391 | * %false - this request doesn't have any more data |
| 1392 | * %true - this request has more data |
| 1393 | **/ |
| 1394 | bool blk_update_request(struct request *req, blk_status_t error, |
| 1395 | unsigned int nr_bytes) |
| 1396 | { |
| 1397 | int total_bytes; |
| 1398 | |
| 1399 | trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes); |
| 1400 | |
| 1401 | if (!req->bio) |
| 1402 | return false; |
| 1403 | |
| 1404 | if (unlikely(error && !blk_rq_is_passthrough(req) && |
| 1405 | !(req->rq_flags & RQF_QUIET))) |
| 1406 | print_req_error(req, error, __func__); |
| 1407 | |
| 1408 | blk_account_io_completion(req, nr_bytes); |
| 1409 | |
| 1410 | total_bytes = 0; |
| 1411 | while (req->bio) { |
| 1412 | struct bio *bio = req->bio; |
| 1413 | unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes); |
| 1414 | |
| 1415 | if (bio_bytes == bio->bi_iter.bi_size) |
| 1416 | req->bio = bio->bi_next; |
| 1417 | |
| 1418 | /* Completion has already been traced */ |
| 1419 | bio_clear_flag(bio, BIO_TRACE_COMPLETION); |
| 1420 | req_bio_endio(req, bio, bio_bytes, error); |
| 1421 | |
| 1422 | total_bytes += bio_bytes; |
| 1423 | nr_bytes -= bio_bytes; |
| 1424 | |
| 1425 | if (!nr_bytes) |
| 1426 | break; |
| 1427 | } |
| 1428 | |
| 1429 | /* |
| 1430 | * completely done |
| 1431 | */ |
| 1432 | if (!req->bio) { |
| 1433 | /* |
| 1434 | * Reset counters so that the request stacking driver |
| 1435 | * can find how many bytes remain in the request |
| 1436 | * later. |
| 1437 | */ |
| 1438 | req->__data_len = 0; |
| 1439 | return false; |
| 1440 | } |
| 1441 | |
| 1442 | req->__data_len -= total_bytes; |
| 1443 | |
| 1444 | /* update sector only for requests with clear definition of sector */ |
| 1445 | if (!blk_rq_is_passthrough(req)) |
| 1446 | req->__sector += total_bytes >> 9; |
| 1447 | |
| 1448 | /* mixed attributes always follow the first bio */ |
| 1449 | if (req->rq_flags & RQF_MIXED_MERGE) { |
| 1450 | req->cmd_flags &= ~REQ_FAILFAST_MASK; |
| 1451 | req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK; |
| 1452 | } |
| 1453 | |
| 1454 | if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) { |
| 1455 | /* |
| 1456 | * If total number of sectors is less than the first segment |
| 1457 | * size, something has gone terribly wrong. |
| 1458 | */ |
| 1459 | if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { |
| 1460 | blk_dump_rq_flags(req, "request botched"); |
| 1461 | req->__data_len = blk_rq_cur_bytes(req); |
| 1462 | } |
| 1463 | |
| 1464 | /* recalculate the number of segments */ |
| 1465 | req->nr_phys_segments = blk_recalc_rq_segments(req); |
| 1466 | } |
| 1467 | |
| 1468 | return true; |
| 1469 | } |
| 1470 | EXPORT_SYMBOL_GPL(blk_update_request); |
| 1471 | |
| 1472 | #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE |
| 1473 | /** |
| 1474 | * rq_flush_dcache_pages - Helper function to flush all pages in a request |
| 1475 | * @rq: the request to be flushed |
| 1476 | * |
| 1477 | * Description: |
| 1478 | * Flush all pages in @rq. |
| 1479 | */ |
| 1480 | void rq_flush_dcache_pages(struct request *rq) |
| 1481 | { |
| 1482 | struct req_iterator iter; |
| 1483 | struct bio_vec bvec; |
| 1484 | |
| 1485 | rq_for_each_segment(bvec, rq, iter) |
| 1486 | flush_dcache_page(bvec.bv_page); |
| 1487 | } |
| 1488 | EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); |
| 1489 | #endif |
| 1490 | |
| 1491 | /** |
| 1492 | * blk_lld_busy - Check if underlying low-level drivers of a device are busy |
| 1493 | * @q : the queue of the device being checked |
| 1494 | * |
| 1495 | * Description: |
| 1496 | * Check if underlying low-level drivers of a device are busy. |
| 1497 | * If the drivers want to export their busy state, they must set own |
| 1498 | * exporting function using blk_queue_lld_busy() first. |
| 1499 | * |
| 1500 | * Basically, this function is used only by request stacking drivers |
| 1501 | * to stop dispatching requests to underlying devices when underlying |
| 1502 | * devices are busy. This behavior helps more I/O merging on the queue |
| 1503 | * of the request stacking driver and prevents I/O throughput regression |
| 1504 | * on burst I/O load. |
| 1505 | * |
| 1506 | * Return: |
| 1507 | * 0 - Not busy (The request stacking driver should dispatch request) |
| 1508 | * 1 - Busy (The request stacking driver should stop dispatching request) |
| 1509 | */ |
| 1510 | int blk_lld_busy(struct request_queue *q) |
| 1511 | { |
| 1512 | if (queue_is_mq(q) && q->mq_ops->busy) |
| 1513 | return q->mq_ops->busy(q); |
| 1514 | |
| 1515 | return 0; |
| 1516 | } |
| 1517 | EXPORT_SYMBOL_GPL(blk_lld_busy); |
| 1518 | |
| 1519 | /** |
| 1520 | * blk_rq_unprep_clone - Helper function to free all bios in a cloned request |
| 1521 | * @rq: the clone request to be cleaned up |
| 1522 | * |
| 1523 | * Description: |
| 1524 | * Free all bios in @rq for a cloned request. |
| 1525 | */ |
| 1526 | void blk_rq_unprep_clone(struct request *rq) |
| 1527 | { |
| 1528 | struct bio *bio; |
| 1529 | |
| 1530 | while ((bio = rq->bio) != NULL) { |
| 1531 | rq->bio = bio->bi_next; |
| 1532 | |
| 1533 | bio_put(bio); |
| 1534 | } |
| 1535 | } |
| 1536 | EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); |
| 1537 | |
| 1538 | /* |
| 1539 | * Copy attributes of the original request to the clone request. |
| 1540 | * The actual data parts (e.g. ->cmd, ->sense) are not copied. |
| 1541 | */ |
| 1542 | static void __blk_rq_prep_clone(struct request *dst, struct request *src) |
| 1543 | { |
| 1544 | dst->__sector = blk_rq_pos(src); |
| 1545 | dst->__data_len = blk_rq_bytes(src); |
| 1546 | if (src->rq_flags & RQF_SPECIAL_PAYLOAD) { |
| 1547 | dst->rq_flags |= RQF_SPECIAL_PAYLOAD; |
| 1548 | dst->special_vec = src->special_vec; |
| 1549 | } |
| 1550 | dst->nr_phys_segments = src->nr_phys_segments; |
| 1551 | dst->ioprio = src->ioprio; |
| 1552 | dst->extra_len = src->extra_len; |
| 1553 | } |
| 1554 | |
| 1555 | /** |
| 1556 | * blk_rq_prep_clone - Helper function to setup clone request |
| 1557 | * @rq: the request to be setup |
| 1558 | * @rq_src: original request to be cloned |
| 1559 | * @bs: bio_set that bios for clone are allocated from |
| 1560 | * @gfp_mask: memory allocation mask for bio |
| 1561 | * @bio_ctr: setup function to be called for each clone bio. |
| 1562 | * Returns %0 for success, non %0 for failure. |
| 1563 | * @data: private data to be passed to @bio_ctr |
| 1564 | * |
| 1565 | * Description: |
| 1566 | * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. |
| 1567 | * The actual data parts of @rq_src (e.g. ->cmd, ->sense) |
| 1568 | * are not copied, and copying such parts is the caller's responsibility. |
| 1569 | * Also, pages which the original bios are pointing to are not copied |
| 1570 | * and the cloned bios just point same pages. |
| 1571 | * So cloned bios must be completed before original bios, which means |
| 1572 | * the caller must complete @rq before @rq_src. |
| 1573 | */ |
| 1574 | int blk_rq_prep_clone(struct request *rq, struct request *rq_src, |
| 1575 | struct bio_set *bs, gfp_t gfp_mask, |
| 1576 | int (*bio_ctr)(struct bio *, struct bio *, void *), |
| 1577 | void *data) |
| 1578 | { |
| 1579 | struct bio *bio, *bio_src; |
| 1580 | |
| 1581 | if (!bs) |
| 1582 | bs = &fs_bio_set; |
| 1583 | |
| 1584 | __rq_for_each_bio(bio_src, rq_src) { |
| 1585 | bio = bio_clone_fast(bio_src, gfp_mask, bs); |
| 1586 | if (!bio) |
| 1587 | goto free_and_out; |
| 1588 | |
| 1589 | if (bio_ctr && bio_ctr(bio, bio_src, data)) |
| 1590 | goto free_and_out; |
| 1591 | |
| 1592 | if (rq->bio) { |
| 1593 | rq->biotail->bi_next = bio; |
| 1594 | rq->biotail = bio; |
| 1595 | } else |
| 1596 | rq->bio = rq->biotail = bio; |
| 1597 | } |
| 1598 | |
| 1599 | __blk_rq_prep_clone(rq, rq_src); |
| 1600 | |
| 1601 | return 0; |
| 1602 | |
| 1603 | free_and_out: |
| 1604 | if (bio) |
| 1605 | bio_put(bio); |
| 1606 | blk_rq_unprep_clone(rq); |
| 1607 | |
| 1608 | return -ENOMEM; |
| 1609 | } |
| 1610 | EXPORT_SYMBOL_GPL(blk_rq_prep_clone); |
| 1611 | |
| 1612 | int kblockd_schedule_work(struct work_struct *work) |
| 1613 | { |
| 1614 | return queue_work(kblockd_workqueue, work); |
| 1615 | } |
| 1616 | EXPORT_SYMBOL(kblockd_schedule_work); |
| 1617 | |
| 1618 | int kblockd_schedule_work_on(int cpu, struct work_struct *work) |
| 1619 | { |
| 1620 | return queue_work_on(cpu, kblockd_workqueue, work); |
| 1621 | } |
| 1622 | EXPORT_SYMBOL(kblockd_schedule_work_on); |
| 1623 | |
| 1624 | int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, |
| 1625 | unsigned long delay) |
| 1626 | { |
| 1627 | return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay); |
| 1628 | } |
| 1629 | EXPORT_SYMBOL(kblockd_mod_delayed_work_on); |
| 1630 | |
| 1631 | /** |
| 1632 | * blk_start_plug - initialize blk_plug and track it inside the task_struct |
| 1633 | * @plug: The &struct blk_plug that needs to be initialized |
| 1634 | * |
| 1635 | * Description: |
| 1636 | * blk_start_plug() indicates to the block layer an intent by the caller |
| 1637 | * to submit multiple I/O requests in a batch. The block layer may use |
| 1638 | * this hint to defer submitting I/Os from the caller until blk_finish_plug() |
| 1639 | * is called. However, the block layer may choose to submit requests |
| 1640 | * before a call to blk_finish_plug() if the number of queued I/Os |
| 1641 | * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than |
| 1642 | * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if |
| 1643 | * the task schedules (see below). |
| 1644 | * |
| 1645 | * Tracking blk_plug inside the task_struct will help with auto-flushing the |
| 1646 | * pending I/O should the task end up blocking between blk_start_plug() and |
| 1647 | * blk_finish_plug(). This is important from a performance perspective, but |
| 1648 | * also ensures that we don't deadlock. For instance, if the task is blocking |
| 1649 | * for a memory allocation, memory reclaim could end up wanting to free a |
| 1650 | * page belonging to that request that is currently residing in our private |
| 1651 | * plug. By flushing the pending I/O when the process goes to sleep, we avoid |
| 1652 | * this kind of deadlock. |
| 1653 | */ |
| 1654 | void blk_start_plug(struct blk_plug *plug) |
| 1655 | { |
| 1656 | struct task_struct *tsk = current; |
| 1657 | |
| 1658 | /* |
| 1659 | * If this is a nested plug, don't actually assign it. |
| 1660 | */ |
| 1661 | if (tsk->plug) |
| 1662 | return; |
| 1663 | |
| 1664 | INIT_LIST_HEAD(&plug->mq_list); |
| 1665 | INIT_LIST_HEAD(&plug->cb_list); |
| 1666 | plug->rq_count = 0; |
| 1667 | plug->multiple_queues = false; |
| 1668 | |
| 1669 | /* |
| 1670 | * Store ordering should not be needed here, since a potential |
| 1671 | * preempt will imply a full memory barrier |
| 1672 | */ |
| 1673 | tsk->plug = plug; |
| 1674 | } |
| 1675 | EXPORT_SYMBOL(blk_start_plug); |
| 1676 | |
| 1677 | static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) |
| 1678 | { |
| 1679 | LIST_HEAD(callbacks); |
| 1680 | |
| 1681 | while (!list_empty(&plug->cb_list)) { |
| 1682 | list_splice_init(&plug->cb_list, &callbacks); |
| 1683 | |
| 1684 | while (!list_empty(&callbacks)) { |
| 1685 | struct blk_plug_cb *cb = list_first_entry(&callbacks, |
| 1686 | struct blk_plug_cb, |
| 1687 | list); |
| 1688 | list_del(&cb->list); |
| 1689 | cb->callback(cb, from_schedule); |
| 1690 | } |
| 1691 | } |
| 1692 | } |
| 1693 | |
| 1694 | struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, |
| 1695 | int size) |
| 1696 | { |
| 1697 | struct blk_plug *plug = current->plug; |
| 1698 | struct blk_plug_cb *cb; |
| 1699 | |
| 1700 | if (!plug) |
| 1701 | return NULL; |
| 1702 | |
| 1703 | list_for_each_entry(cb, &plug->cb_list, list) |
| 1704 | if (cb->callback == unplug && cb->data == data) |
| 1705 | return cb; |
| 1706 | |
| 1707 | /* Not currently on the callback list */ |
| 1708 | BUG_ON(size < sizeof(*cb)); |
| 1709 | cb = kzalloc(size, GFP_ATOMIC); |
| 1710 | if (cb) { |
| 1711 | cb->data = data; |
| 1712 | cb->callback = unplug; |
| 1713 | list_add(&cb->list, &plug->cb_list); |
| 1714 | } |
| 1715 | return cb; |
| 1716 | } |
| 1717 | EXPORT_SYMBOL(blk_check_plugged); |
| 1718 | |
| 1719 | void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) |
| 1720 | { |
| 1721 | flush_plug_callbacks(plug, from_schedule); |
| 1722 | |
| 1723 | if (!list_empty(&plug->mq_list)) |
| 1724 | blk_mq_flush_plug_list(plug, from_schedule); |
| 1725 | } |
| 1726 | |
| 1727 | /** |
| 1728 | * blk_finish_plug - mark the end of a batch of submitted I/O |
| 1729 | * @plug: The &struct blk_plug passed to blk_start_plug() |
| 1730 | * |
| 1731 | * Description: |
| 1732 | * Indicate that a batch of I/O submissions is complete. This function |
| 1733 | * must be paired with an initial call to blk_start_plug(). The intent |
| 1734 | * is to allow the block layer to optimize I/O submission. See the |
| 1735 | * documentation for blk_start_plug() for more information. |
| 1736 | */ |
| 1737 | void blk_finish_plug(struct blk_plug *plug) |
| 1738 | { |
| 1739 | if (plug != current->plug) |
| 1740 | return; |
| 1741 | blk_flush_plug_list(plug, false); |
| 1742 | |
| 1743 | current->plug = NULL; |
| 1744 | } |
| 1745 | EXPORT_SYMBOL(blk_finish_plug); |
| 1746 | |
| 1747 | int __init blk_dev_init(void) |
| 1748 | { |
| 1749 | BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS)); |
| 1750 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
| 1751 | FIELD_SIZEOF(struct request, cmd_flags)); |
| 1752 | BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 * |
| 1753 | FIELD_SIZEOF(struct bio, bi_opf)); |
| 1754 | |
| 1755 | /* used for unplugging and affects IO latency/throughput - HIGHPRI */ |
| 1756 | kblockd_workqueue = alloc_workqueue("kblockd", |
| 1757 | WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); |
| 1758 | if (!kblockd_workqueue) |
| 1759 | panic("Failed to create kblockd\n"); |
| 1760 | |
| 1761 | blk_requestq_cachep = kmem_cache_create("request_queue", |
| 1762 | sizeof(struct request_queue), 0, SLAB_PANIC, NULL); |
| 1763 | |
| 1764 | #ifdef CONFIG_DEBUG_FS |
| 1765 | blk_debugfs_root = debugfs_create_dir("block", NULL); |
| 1766 | #endif |
| 1767 | |
| 1768 | return 0; |
| 1769 | } |