blob: 1217c2cd66dd88fc5cf53556835a9d9c5a619947 [file] [log] [blame]
Christoph Hellwig3dcf60bc2019-04-30 14:42:43 -04001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 * 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>
Jens Axboe6728cb02008-01-31 13:03:55 +01007 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
8 * - July2000
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * 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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/kernel.h>
16#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/bio.h>
18#include <linux/blkdev.h>
Alan Stern52abca62020-12-08 21:29:51 -080019#include <linux/blk-pm.h>
Christoph Hellwigfe45e632021-09-20 14:33:27 +020020#include <linux/blk-integrity.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/highmem.h>
22#include <linux/mm.h>
Matthew Wilcox (Oracle)cee9a0c2020-06-01 21:46:07 -070023#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/kernel_stat.h>
25#include <linux/string.h>
26#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/completion.h>
28#include <linux/slab.h>
29#include <linux/swap.h>
30#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080031#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080032#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010033#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020034#include <linux/delay.h>
Tejun Heoaaf7c682012-04-19 16:29:22 -070035#include <linux/ratelimit.h>
Lin Ming6c954662013-03-23 11:42:26 +080036#include <linux/pm_runtime.h>
Max Gurtovoy54d4e6a2019-09-16 18:44:29 +030037#include <linux/t10-pi.h>
Omar Sandoval18fbda92017-01-31 14:53:20 -080038#include <linux/debugfs.h>
Howard McLauchlan30abb3a2018-02-06 14:05:39 -080039#include <linux/bpf.h>
Christoph Hellwig82d981d2021-11-23 19:53:12 +010040#include <linux/part_stat.h>
Ming Lei71ac8602020-05-14 16:45:09 +080041#include <linux/sched/sysctl.h>
Satya Tangiralaa892c8d2020-05-14 00:37:18 +000042#include <linux/blk-crypto.h>
Li Zefan55782132009-06-09 13:43:05 +080043
44#define CREATE_TRACE_POINTS
45#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Jens Axboe8324aa92008-01-29 14:51:59 +010047#include "blk.h"
Christoph Hellwig2aa77452021-11-23 19:53:08 +010048#include "blk-mq-sched.h"
Bart Van Asschebca6b062018-09-26 14:01:03 -070049#include "blk-pm.h"
Ming Lei672fdcf2022-02-11 18:11:49 +080050#include "blk-cgroup.h"
Jens Axboea7b36ee2021-10-05 09:11:56 -060051#include "blk-throttle.h"
Hongyu Jinf3c89982024-01-30 15:26:34 -050052#include "blk-ioprio.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010053
Omar Sandoval18fbda92017-01-31 14:53:20 -080054struct dentry *blk_debugfs_root;
Omar Sandoval18fbda92017-01-31 14:53:20 -080055
Mike Snitzerd07335e2010-11-16 12:52:38 +010056EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020057EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070058EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Keith Busch3291fa52014-04-28 12:30:52 -060059EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
NeilBrowncbae8d42012-12-14 20:49:27 +010060EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Chaitanya Kulkarnib357e4a2021-02-21 21:29:59 -080061EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_insert);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010062
Christoph Hellwig2bd85222022-11-14 05:26:36 +010063static DEFINE_IDA(blk_queue_ida);
Tejun Heoa73f7302011-12-14 00:33:37 +010064
Linus Torvalds1da177e2005-04-16 15:20:36 -070065/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * For queue allocation
67 */
Christoph Hellwig2bd85222022-11-14 05:26:36 +010068static struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 * Controlling structure to kblockd
72 */
Jens Axboeff856ba2006-01-09 16:02:34 +010073static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Bart Van Assche8814ce82018-03-07 17:10:04 -080075/**
76 * blk_queue_flag_set - atomically set a queue flag
77 * @flag: flag to be set
78 * @q: request queue
79 */
80void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
81{
Christoph Hellwig57d74df2018-11-14 17:02:07 +010082 set_bit(flag, &q->queue_flags);
Bart Van Assche8814ce82018-03-07 17:10:04 -080083}
84EXPORT_SYMBOL(blk_queue_flag_set);
85
86/**
87 * blk_queue_flag_clear - atomically clear a queue flag
88 * @flag: flag to be cleared
89 * @q: request queue
90 */
91void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
92{
Christoph Hellwig57d74df2018-11-14 17:02:07 +010093 clear_bit(flag, &q->queue_flags);
Bart Van Assche8814ce82018-03-07 17:10:04 -080094}
95EXPORT_SYMBOL(blk_queue_flag_clear);
96
Chaitanya Kulkarnie47bc4e2019-06-20 10:59:16 -070097#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
98static const char *const blk_op_name[] = {
99 REQ_OP_NAME(READ),
100 REQ_OP_NAME(WRITE),
101 REQ_OP_NAME(FLUSH),
102 REQ_OP_NAME(DISCARD),
103 REQ_OP_NAME(SECURE_ERASE),
104 REQ_OP_NAME(ZONE_RESET),
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700105 REQ_OP_NAME(ZONE_RESET_ALL),
Ajay Joshi6c1b1da2019-10-27 23:05:45 +0900106 REQ_OP_NAME(ZONE_OPEN),
107 REQ_OP_NAME(ZONE_CLOSE),
108 REQ_OP_NAME(ZONE_FINISH),
Keith Busch0512a752020-05-12 17:55:47 +0900109 REQ_OP_NAME(ZONE_APPEND),
Chaitanya Kulkarnie47bc4e2019-06-20 10:59:16 -0700110 REQ_OP_NAME(WRITE_ZEROES),
Chaitanya Kulkarnie47bc4e2019-06-20 10:59:16 -0700111 REQ_OP_NAME(DRV_IN),
112 REQ_OP_NAME(DRV_OUT),
113};
114#undef REQ_OP_NAME
115
116/**
117 * blk_op_str - Return string XXX in the REQ_OP_XXX.
118 * @op: REQ_OP_XXX.
119 *
120 * Description: Centralize block layer function to convert REQ_OP_XXX into
121 * string format. Useful in the debugging and tracing bio or request. For
122 * invalid REQ_OP_XXX it returns string "UNKNOWN".
123 */
Bart Van Assche77e7ffd2022-07-14 11:06:28 -0700124inline const char *blk_op_str(enum req_op op)
Chaitanya Kulkarnie47bc4e2019-06-20 10:59:16 -0700125{
126 const char *op_str = "UNKNOWN";
127
128 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
129 op_str = blk_op_name[op];
130
131 return op_str;
132}
133EXPORT_SYMBOL_GPL(blk_op_str);
134
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200135static const struct {
136 int errno;
137 const char *name;
138} blk_errors[] = {
139 [BLK_STS_OK] = { 0, "" },
140 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
141 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
142 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
143 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
144 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
Mike Christie7ba15082023-04-07 15:05:35 -0500145 [BLK_STS_RESV_CONFLICT] = { -EBADE, "reservation conflict" },
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200146 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
147 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
148 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
Ming Lei86ff7c22018-01-30 22:04:57 -0500149 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
Goldwyn Rodrigues03a07c92017-06-20 07:05:46 -0500150 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
Song Liu7d32c022022-02-03 11:28:26 -0800151 [BLK_STS_OFFLINE] = { -ENODEV, "device offline" },
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200152
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200153 /* device mapper special case, should not leak out: */
154 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
155
Keith Busch3b481d92020-09-24 13:53:28 -0700156 /* zone device specific errors */
157 [BLK_STS_ZONE_OPEN_RESOURCE] = { -ETOOMANYREFS, "open zones exceeded" },
158 [BLK_STS_ZONE_ACTIVE_RESOURCE] = { -EOVERFLOW, "active zones exceeded" },
159
Damien Le Moaldffc4802023-05-11 03:13:36 +0200160 /* Command duration limit device-side timeout */
161 [BLK_STS_DURATION_LIMIT] = { -ETIME, "duration limit exceeded" },
162
John Garry9da3d1e2024-06-20 12:53:54 +0000163 [BLK_STS_INVAL] = { -EINVAL, "invalid" },
164
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200165 /* everything else not covered above: */
166 [BLK_STS_IOERR] = { -EIO, "I/O" },
167};
168
169blk_status_t errno_to_blk_status(int errno)
170{
171 int i;
172
173 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
174 if (blk_errors[i].errno == errno)
175 return (__force blk_status_t)i;
176 }
177
178 return BLK_STS_IOERR;
179}
180EXPORT_SYMBOL_GPL(errno_to_blk_status);
181
182int blk_status_to_errno(blk_status_t status)
183{
184 int idx = (__force int)status;
185
Bart Van Assche34bd9c12017-06-21 10:55:46 -0700186 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200187 return -EIO;
188 return blk_errors[idx].errno;
189}
190EXPORT_SYMBOL_GPL(blk_status_to_errno);
191
Christoph Hellwig0d7a29a2021-11-17 07:14:03 +0100192const char *blk_status_to_str(blk_status_t status)
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200193{
194 int idx = (__force int)status;
195
Bart Van Assche34bd9c12017-06-21 10:55:46 -0700196 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
Christoph Hellwig0d7a29a2021-11-17 07:14:03 +0100197 return "<null>";
198 return blk_errors[idx].name;
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200199}
Kent Overstreet7ba37922023-08-13 14:26:34 -0400200EXPORT_SYMBOL_GPL(blk_status_to_str);
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200201
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202/**
203 * blk_sync_queue - cancel any pending callbacks on a queue
204 * @q: the queue
205 *
206 * Description:
207 * The block layer may perform asynchronous callback activity
208 * on a queue, such as calling the unplug function after a timeout.
209 * A block device may call blk_sync_queue to ensure that any
210 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200211 * that the callbacks might use. The caller must already have made sure
Christoph Hellwigc62b37d2020-07-01 10:59:43 +0200212 * that its ->submit_bio will not re-add plugging prior to calling
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 * this function.
214 *
Vivek Goyalda527772011-03-02 19:05:33 -0500215 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900216 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800217 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500218 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 */
220void blk_sync_queue(struct request_queue *q)
221{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100222 del_timer_sync(&q->timeout);
Bart Van Assche4e9b6f22017-10-19 10:00:48 -0700223 cancel_work_sync(&q->timeout_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225EXPORT_SYMBOL(blk_sync_queue);
226
227/**
Bart Van Asschecd84a622018-09-26 14:01:04 -0700228 * blk_set_pm_only - increment pm_only counter
Bart Van Asschec9254f22017-11-09 10:49:57 -0800229 * @q: request queue pointer
Bart Van Asschec9254f22017-11-09 10:49:57 -0800230 */
Bart Van Asschecd84a622018-09-26 14:01:04 -0700231void blk_set_pm_only(struct request_queue *q)
Bart Van Asschec9254f22017-11-09 10:49:57 -0800232{
Bart Van Asschecd84a622018-09-26 14:01:04 -0700233 atomic_inc(&q->pm_only);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800234}
Bart Van Asschecd84a622018-09-26 14:01:04 -0700235EXPORT_SYMBOL_GPL(blk_set_pm_only);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800236
Bart Van Asschecd84a622018-09-26 14:01:04 -0700237void blk_clear_pm_only(struct request_queue *q)
Bart Van Asschec9254f22017-11-09 10:49:57 -0800238{
Bart Van Asschecd84a622018-09-26 14:01:04 -0700239 int pm_only;
240
241 pm_only = atomic_dec_return(&q->pm_only);
242 WARN_ON_ONCE(pm_only < 0);
243 if (pm_only == 0)
244 wake_up_all(&q->mq_freeze_wq);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800245}
Bart Van Asschecd84a622018-09-26 14:01:04 -0700246EXPORT_SYMBOL_GPL(blk_clear_pm_only);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800247
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100248static void blk_free_queue_rcu(struct rcu_head *rcu_head)
249{
Ming Leid36a9ea2022-12-15 10:16:29 +0800250 struct request_queue *q = container_of(rcu_head,
251 struct request_queue, rcu_head);
252
253 percpu_ref_exit(&q->q_usage_counter);
254 kmem_cache_free(blk_requestq_cachep, q);
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100255}
256
257static void blk_free_queue(struct request_queue *q)
258{
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100259 blk_free_queue_stats(q->stats);
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100260 if (queue_is_mq(q))
261 blk_mq_release(q);
262
263 ida_free(&blk_queue_ida, q->id);
264 call_rcu(&q->rcu_head, blk_free_queue_rcu);
265}
266
Luis Chamberlainb5bd3572020-06-19 20:47:23 +0000267/**
268 * blk_put_queue - decrement the request_queue refcount
269 * @q: the request_queue structure to decrement the refcount for
270 *
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100271 * Decrements the refcount of the request_queue and free it when the refcount
272 * reaches 0.
Luis Chamberlainb5bd3572020-06-19 20:47:23 +0000273 */
Jens Axboe165125e2007-07-24 09:28:11 +0200274void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500275{
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100276 if (refcount_dec_and_test(&q->refs))
277 blk_free_queue(q);
Al Viro483f4af2006-03-18 18:34:37 -0500278}
Jens Axboed86e0e82011-05-27 07:44:43 +0200279EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500280
Christoph Hellwig8e141f92021-09-29 09:12:40 +0200281void blk_queue_start_drain(struct request_queue *q)
Jens Axboeaed3ea92014-12-22 14:04:42 -0700282{
Ming Leid3cfb2a2017-03-27 20:06:58 +0800283 /*
284 * When queue DYING flag is set, we need to block new req
285 * entering queue, so we call blk_freeze_queue_start() to
286 * prevent I/O from crossing blk_queue_enter().
287 */
288 blk_freeze_queue_start(q);
Jens Axboe344e9ff2018-11-15 12:22:51 -0700289 if (queue_is_mq(q))
Jens Axboeaed3ea92014-12-22 14:04:42 -0700290 blk_mq_wake_waiters(q);
Ming Lei055f6e12017-11-09 10:49:53 -0800291 /* Make blk_queue_enter() reexamine the DYING flag. */
292 wake_up_all(&q->mq_freeze_wq);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700293}
Christoph Hellwig8e141f92021-09-29 09:12:40 +0200294
Tejun Heod7325802012-03-05 13:14:58 -0800295/**
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800296 * blk_queue_enter() - try to increase q->q_usage_counter
297 * @q: request queue pointer
Bart Van Asschea4d34da2020-12-08 21:29:50 -0800298 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PM
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800299 */
Bart Van Assche9a95e4e2017-11-09 10:49:59 -0800300int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
Dan Williams3ef28e82015-10-21 13:20:12 -0400301{
Bart Van Asschea4d34da2020-12-08 21:29:50 -0800302 const bool pm = flags & BLK_MQ_REQ_PM;
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800303
Christoph Hellwig1f14a092021-09-29 09:12:38 +0200304 while (!blk_try_enter_queue(q, pm)) {
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800305 if (flags & BLK_MQ_REQ_NOWAIT)
Stefan Roesch56f99b82022-09-12 09:53:25 -0700306 return -EAGAIN;
Dan Williams3ef28e82015-10-21 13:20:12 -0400307
Ming Lei5ed61d32017-03-27 20:06:56 +0800308 /*
Christoph Hellwig1f14a092021-09-29 09:12:38 +0200309 * read pair of barrier in blk_freeze_queue_start(), we need to
310 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
311 * reading .mq_freeze_depth or queue dying flag, otherwise the
312 * following wait may never return if the two reads are
313 * reordered.
Ming Lei5ed61d32017-03-27 20:06:56 +0800314 */
315 smp_rmb();
Alan Jenkins1dc30392018-04-12 19:11:58 +0100316 wait_event(q->mq_freeze_wq,
Bob Liu7996a8b2019-05-21 11:25:55 +0800317 (!q->mq_freeze_depth &&
Alan Stern52abca62020-12-08 21:29:51 -0800318 blk_pm_resume_queue(pm, q)) ||
Alan Jenkins1dc30392018-04-12 19:11:58 +0100319 blk_queue_dying(q));
Dan Williams3ef28e82015-10-21 13:20:12 -0400320 if (blk_queue_dying(q))
321 return -ENODEV;
Dan Williams3ef28e82015-10-21 13:20:12 -0400322 }
Christoph Hellwig1f14a092021-09-29 09:12:38 +0200323
324 return 0;
Dan Williams3ef28e82015-10-21 13:20:12 -0400325}
326
Jens Axboec98cb5b2021-11-04 12:45:51 -0600327int __bio_queue_enter(struct request_queue *q, struct bio *bio)
Christoph Hellwigaccea322020-04-28 13:27:56 +0200328{
Christoph Hellwiga6741532021-09-29 09:12:39 +0200329 while (!blk_try_enter_queue(q, false)) {
Pavel Begunkoveab4e022021-10-14 15:03:29 +0100330 struct gendisk *disk = bio->bi_bdev->bd_disk;
331
Christoph Hellwiga6741532021-09-29 09:12:39 +0200332 if (bio->bi_opf & REQ_NOWAIT) {
Christoph Hellwig8e141f92021-09-29 09:12:40 +0200333 if (test_bit(GD_DEAD, &disk->state))
Christoph Hellwiga6741532021-09-29 09:12:39 +0200334 goto dead;
Christoph Hellwigaccea322020-04-28 13:27:56 +0200335 bio_wouldblock_error(bio);
Stefan Roesch56f99b82022-09-12 09:53:25 -0700336 return -EAGAIN;
Christoph Hellwiga6741532021-09-29 09:12:39 +0200337 }
338
339 /*
340 * read pair of barrier in blk_freeze_queue_start(), we need to
341 * order reading __PERCPU_REF_DEAD flag of .q_usage_counter and
342 * reading .mq_freeze_depth or queue dying flag, otherwise the
343 * following wait may never return if the two reads are
344 * reordered.
345 */
346 smp_rmb();
347 wait_event(q->mq_freeze_wq,
348 (!q->mq_freeze_depth &&
349 blk_pm_resume_queue(false, q)) ||
Christoph Hellwig8e141f92021-09-29 09:12:40 +0200350 test_bit(GD_DEAD, &disk->state));
351 if (test_bit(GD_DEAD, &disk->state))
Christoph Hellwiga6741532021-09-29 09:12:39 +0200352 goto dead;
Christoph Hellwigaccea322020-04-28 13:27:56 +0200353 }
354
Christoph Hellwiga6741532021-09-29 09:12:39 +0200355 return 0;
356dead:
357 bio_io_error(bio);
358 return -ENODEV;
Christoph Hellwigaccea322020-04-28 13:27:56 +0200359}
360
Dan Williams3ef28e82015-10-21 13:20:12 -0400361void blk_queue_exit(struct request_queue *q)
362{
363 percpu_ref_put(&q->q_usage_counter);
364}
365
366static void blk_queue_usage_counter_release(struct percpu_ref *ref)
367{
368 struct request_queue *q =
369 container_of(ref, struct request_queue, q_usage_counter);
370
371 wake_up_all(&q->mq_freeze_wq);
372}
373
Kees Cookbca237a2017-08-28 15:03:41 -0700374static void blk_rq_timed_out_timer(struct timer_list *t)
Christoph Hellwig287922e2015-10-30 20:57:30 +0800375{
Kees Cookbca237a2017-08-28 15:03:41 -0700376 struct request_queue *q = from_timer(q, t, timeout);
Christoph Hellwig287922e2015-10-30 20:57:30 +0800377
378 kblockd_schedule_work(&q->timeout_work);
379}
380
Tetsuo Handa2e3c18d2019-01-30 22:21:45 +0900381static void blk_timeout_work(struct work_struct *work)
382{
383}
384
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100385struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700386{
Jens Axboe165125e2007-07-24 09:28:11 +0200387 struct request_queue *q;
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100388 int error;
Christoph Lameter19460892005-06-23 00:08:19 -0700389
Christoph Hellwig80bd4a72022-11-01 16:00:47 +0100390 q = kmem_cache_alloc_node(blk_requestq_cachep, GFP_KERNEL | __GFP_ZERO,
391 node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 if (!q)
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100393 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
Christoph Hellwigcbf62af2018-05-31 19:11:36 +0200395 q->last_merge = NULL;
Christoph Hellwigcbf62af2018-05-31 19:11:36 +0200396
Bo Liu798f2a62022-06-15 04:18:16 -0400397 q->id = ida_alloc(&blk_queue_ida, GFP_KERNEL);
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100398 if (q->id < 0) {
399 error = q->id;
Christoph Hellwig80bd4a72022-11-01 16:00:47 +0100400 goto fail_q;
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100401 }
Tejun Heoa73f7302011-12-14 00:33:37 +0100402
Jens Axboea83b5762017-03-21 17:20:01 -0600403 q->stats = blk_alloc_queue_stats();
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100404 if (!q->stats) {
405 error = -ENOMEM;
Christoph Hellwig46754bd2022-07-27 12:22:57 -0400406 goto fail_id;
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100407 }
408
409 error = blk_set_default_limits(lim);
410 if (error)
411 goto fail_stats;
412 q->limits = *lim;
Jens Axboea83b5762017-03-21 17:20:01 -0600413
Mike Snitzer51514122011-11-23 10:59:13 +0100414 q->node = node_id;
Jens Axboe0989a022009-06-12 14:42:56 +0200415
John Garry079a2e32021-10-05 18:23:39 +0800416 atomic_set(&q->nr_active_requests_shared_tags, 0);
John Garrybccf5e22020-08-19 23:20:26 +0800417
Kees Cookbca237a2017-08-28 15:03:41 -0700418 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
Tetsuo Handa2e3c18d2019-01-30 22:21:45 +0900419 INIT_WORK(&q->timeout_work, blk_timeout_work);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100420 INIT_LIST_HEAD(&q->icq_list);
Al Viro483f4af2006-03-18 18:34:37 -0500421
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100422 refcount_set(&q->refs, 1);
Luis Chamberlain85e0cbb2020-06-19 20:47:30 +0000423 mutex_init(&q->debugfs_mutex);
Al Viro483f4af2006-03-18 18:34:37 -0500424 mutex_init(&q->sysfs_lock);
Ming Leicecf5d82019-08-27 19:01:48 +0800425 mutex_init(&q->sysfs_dir_lock);
Christoph Hellwigd690cb8a2024-02-13 08:34:14 +0100426 mutex_init(&q->limits_lock);
Yu Kuaia13bd912023-04-14 16:40:08 +0800427 mutex_init(&q->rq_qos_mutex);
Christoph Hellwig0d945c12018-11-15 12:17:28 -0700428 spin_lock_init(&q->queue_lock);
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500429
Jens Axboe320ae512013-10-24 09:20:05 +0100430 init_waitqueue_head(&q->mq_freeze_wq);
Bob Liu7996a8b2019-05-21 11:25:55 +0800431 mutex_init(&q->mq_freeze_lock);
Jens Axboe320ae512013-10-24 09:20:05 +0100432
Ming Lei8b8ace02024-04-07 20:59:10 +0800433 blkg_init_queue(q);
434
Dan Williams3ef28e82015-10-21 13:20:12 -0400435 /*
436 * Init percpu_ref in atomic mode so that it's faster to shutdown.
437 * See blk_register_queue() for details.
438 */
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100439 error = percpu_ref_init(&q->q_usage_counter,
Dan Williams3ef28e82015-10-21 13:20:12 -0400440 blk_queue_usage_counter_release,
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100441 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL);
442 if (error)
Christoph Hellwigedb08722021-08-09 16:17:43 +0200443 goto fail_stats;
Tejun Heof51b8022012-03-05 13:15:05 -0800444
John Garryd2a27962021-10-05 18:23:27 +0800445 q->nr_requests = BLKDEV_DEFAULT_RQ;
Christoph Hellwig3d745ea2020-03-27 09:30:11 +0100446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100448
Jens Axboea83b5762017-03-21 17:20:01 -0600449fail_stats:
Christoph Hellwigedb08722021-08-09 16:17:43 +0200450 blk_free_queue_stats(q->stats);
Tejun Heoa73f7302011-12-14 00:33:37 +0100451fail_id:
Bo Liu798f2a62022-06-15 04:18:16 -0400452 ida_free(&blk_queue_ida, q->id);
Tejun Heoa73f7302011-12-14 00:33:37 +0100453fail_q:
Christoph Hellwig80bd4a72022-11-01 16:00:47 +0100454 kmem_cache_free(blk_requestq_cachep, q);
Christoph Hellwigad751ba2024-02-13 08:34:18 +0100455 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
Luis Chamberlainb5bd3572020-06-19 20:47:23 +0000458/**
459 * blk_get_queue - increment the request_queue refcount
460 * @q: the request_queue structure to increment the refcount for
461 *
462 * Increment the refcount of the request_queue kobject.
Luis Chamberlain763b5892020-06-19 20:47:24 +0000463 *
464 * Context: Any context.
Luis Chamberlainb5bd3572020-06-19 20:47:23 +0000465 */
Tejun Heo09ac46c2011-12-14 00:33:38 +0100466bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
Christoph Hellwig828b5f02022-07-21 08:34:32 +0200468 if (unlikely(blk_queue_dying(q)))
469 return false;
Christoph Hellwig2bd85222022-11-14 05:26:36 +0100470 refcount_inc(&q->refs);
Christoph Hellwig828b5f02022-07-21 08:34:32 +0200471 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472}
Jens Axboed86e0e82011-05-27 07:44:43 +0200473EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
Akinobu Mitac17bb492006-12-08 02:39:46 -0800475#ifdef CONFIG_FAIL_MAKE_REQUEST
476
477static DECLARE_FAULT_ATTR(fail_make_request);
478
479static int __init setup_fail_make_request(char *str)
480{
481 return setup_fault_attr(&fail_make_request, str);
482}
483__setup("fail_make_request=", setup_fail_make_request);
484
Christoph Hellwig06c8c692021-11-17 07:13:58 +0100485bool should_fail_request(struct block_device *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -0800486{
Al Viro811ba892024-04-28 00:15:07 -0400487 return bdev_test_flag(part, BD_MAKE_IT_FAIL) &&
488 should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800489}
490
491static int __init fail_make_request_debugfs(void)
492{
Akinobu Mitadd48c082011-08-03 16:21:01 -0700493 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
494 NULL, &fail_make_request);
495
Duan Jiong21f9fcd2014-04-11 15:58:56 +0800496 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800497}
498
499late_initcall(fail_make_request_debugfs);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800500#endif /* CONFIG_FAIL_MAKE_REQUEST */
501
Miaohe Linbdb7d422022-09-05 18:27:54 +0800502static inline void bio_check_ro(struct bio *bio)
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100503{
Christoph Hellwig2f9f6222021-01-24 11:02:35 +0100504 if (op_is_write(bio_op(bio)) && bdev_read_only(bio->bi_bdev)) {
Mikulas Patocka8b2ded12018-09-05 16:14:36 -0600505 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
Miaohe Linbdb7d422022-09-05 18:27:54 +0800506 return;
Yu Kuai67d995e2023-11-28 20:30:27 +0800507
Al Viro49a43da2024-04-12 01:24:27 -0400508 if (bdev_test_flag(bio->bi_bdev, BD_RO_WARNED))
Yu Kuai67d995e2023-11-28 20:30:27 +0800509 return;
510
Al Viro49a43da2024-04-12 01:24:27 -0400511 bdev_set_flag(bio->bi_bdev, BD_RO_WARNED);
512
Yu Kuai67d995e2023-11-28 20:30:27 +0800513 /*
514 * Use ioctl to set underlying disk of raid/dm to read-only
515 * will trigger this.
516 */
517 pr_warn("Trying to write to read-only block-device %pg\n",
518 bio->bi_bdev);
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100519 }
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100520}
521
Howard McLauchlan30abb3a2018-02-06 14:05:39 -0800522static noinline int should_fail_bio(struct bio *bio)
523{
Christoph Hellwig309dca302021-01-24 11:02:34 +0100524 if (should_fail_request(bdev_whole(bio->bi_bdev), bio->bi_iter.bi_size))
Howard McLauchlan30abb3a2018-02-06 14:05:39 -0800525 return -EIO;
526 return 0;
527}
528ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
529
Jens Axboec07e2b42007-07-18 13:27:58 +0200530/*
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100531 * Check whether this bio extends beyond the end of the device or partition.
532 * This may well happen - the kernel calls bread() without checking the size of
533 * the device, e.g., when mounting a file system.
534 */
Christoph Hellwig2f9f6222021-01-24 11:02:35 +0100535static inline int bio_check_eod(struct bio *bio)
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100536{
Christoph Hellwig2f9f6222021-01-24 11:02:35 +0100537 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100538 unsigned int nr_sectors = bio_sectors(bio);
539
Christoph Hellwig3eb96942023-05-24 08:05:38 +0200540 if (nr_sectors &&
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100541 (nr_sectors > maxsector ||
542 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
Christoph Hellwigad740782022-03-04 19:00:57 +0100543 pr_info_ratelimited("%s: attempt to access beyond end of device\n"
Christoph Hellwig069adba2022-05-04 07:33:55 -0700544 "%pg: rw=%d, sector=%llu, nr_sectors = %u limit=%llu\n",
545 current->comm, bio->bi_bdev, bio->bi_opf,
546 bio->bi_iter.bi_sector, nr_sectors, maxsector);
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100547 return -EIO;
548 }
549 return 0;
550}
551
552/*
Christoph Hellwig74d46992017-08-23 19:10:32 +0200553 * Remap block n of partition p to block n+start(p) of the disk.
554 */
Christoph Hellwig2f9f6222021-01-24 11:02:35 +0100555static int blk_partition_remap(struct bio *bio)
Christoph Hellwig74d46992017-08-23 19:10:32 +0200556{
Christoph Hellwig309dca302021-01-24 11:02:34 +0100557 struct block_device *p = bio->bi_bdev;
Christoph Hellwig74d46992017-08-23 19:10:32 +0200558
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100559 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
Christoph Hellwig2f9f6222021-01-24 11:02:35 +0100560 return -EIO;
Damien Le Moal5eac3eb2019-11-11 11:39:25 +0900561 if (bio_sectors(bio)) {
Christoph Hellwig8446fe92020-11-24 09:36:54 +0100562 bio->bi_iter.bi_sector += p->bd_start_sect;
Christoph Hellwig1c02fca2020-12-03 17:21:38 +0100563 trace_block_bio_remap(bio, p->bd_dev,
Christoph Hellwig29ff57c2020-11-24 09:34:24 +0100564 bio->bi_iter.bi_sector -
Christoph Hellwig8446fe92020-11-24 09:36:54 +0100565 p->bd_start_sect);
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100566 }
Christoph Hellwig30c5d342021-01-24 11:02:36 +0100567 bio_set_flag(bio, BIO_REMAPPED);
Christoph Hellwig2f9f6222021-01-24 11:02:35 +0100568 return 0;
Christoph Hellwig74d46992017-08-23 19:10:32 +0200569}
570
Keith Busch0512a752020-05-12 17:55:47 +0900571/*
572 * Check write append to a zoned block device.
573 */
574static inline blk_status_t blk_check_zone_append(struct request_queue *q,
575 struct bio *bio)
576{
Keith Busch0512a752020-05-12 17:55:47 +0900577 int nr_sectors = bio_sectors(bio);
578
579 /* Only applicable to zoned block devices */
Christoph Hellwigedd1dbc2022-07-06 09:03:37 +0200580 if (!bdev_is_zoned(bio->bi_bdev))
Keith Busch0512a752020-05-12 17:55:47 +0900581 return BLK_STS_NOTSUPP;
582
583 /* The bio sector must point to the start of a sequential zone */
Damien Le Moalbca150f2024-04-08 10:41:23 +0900584 if (!bdev_is_zone_start(bio->bi_bdev, bio->bi_iter.bi_sector))
Keith Busch0512a752020-05-12 17:55:47 +0900585 return BLK_STS_IOERR;
586
587 /*
588 * Not allowed to cross zone boundaries. Otherwise, the BIO will be
589 * split and could result in non-contiguous sectors being written in
590 * different zones.
591 */
592 if (nr_sectors > q->limits.chunk_sectors)
593 return BLK_STS_IOERR;
594
595 /* Make sure the BIO is small enough and will not get split */
Damien Le Moalccdbf0a2024-04-08 10:41:09 +0900596 if (nr_sectors > queue_max_zone_append_sectors(q))
Keith Busch0512a752020-05-12 17:55:47 +0900597 return BLK_STS_IOERR;
598
599 bio->bi_opf |= REQ_NOMERGE;
600
601 return BLK_STS_OK;
602}
603
Christoph Hellwig3e087732021-10-12 13:12:24 +0200604static void __submit_bio(struct bio *bio)
Christoph Hellwigac7c5672020-05-16 20:28:01 +0200605{
Yu Kuai9a428912024-05-22 04:03:08 +0800606 /* If plug is not used, add new plug here to cache nsecs time. */
607 struct blk_plug plug;
608
Ming Lei7f36b7d2022-02-16 12:45:08 +0800609 if (unlikely(!blk_crypto_bio_prep(&bio)))
610 return;
611
Yu Kuai9a428912024-05-22 04:03:08 +0800612 blk_start_plug(&plug);
613
Al Viroac2b6f92024-04-12 01:21:45 -0400614 if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO)) {
Christoph Hellwig3e087732021-10-12 13:12:24 +0200615 blk_mq_submit_bio(bio);
Ming Lei7f36b7d2022-02-16 12:45:08 +0800616 } else if (likely(bio_queue_enter(bio) == 0)) {
Jens Axboe9f4107b2023-04-14 07:32:02 -0600617 struct gendisk *disk = bio->bi_bdev->bd_disk;
618
Ming Lei7f36b7d2022-02-16 12:45:08 +0800619 disk->fops->submit_bio(bio);
620 blk_queue_exit(disk->queue);
621 }
Yu Kuai9a428912024-05-22 04:03:08 +0800622
623 blk_finish_plug(&plug);
Christoph Hellwigac7c5672020-05-16 20:28:01 +0200624}
625
Christoph Hellwig566acf22020-07-01 10:59:45 +0200626/*
627 * The loop in this function may be a bit non-obvious, and so deserves some
628 * explanation:
629 *
630 * - Before entering the loop, bio->bi_next is NULL (as all callers ensure
631 * that), so we have a list with a single bio.
632 * - We pretend that we have just taken it off a longer list, so we assign
633 * bio_list to a pointer to the bio_list_on_stack, thus initialising the
634 * bio_list of new bios to be added. ->submit_bio() may indeed add some more
635 * bios through a recursive call to submit_bio_noacct. If it did, we find a
636 * non-NULL value in bio_list and re-enter the loop from the top.
637 * - In this case we really did just take the bio of the top of the list (no
638 * pretending) and so remove it from bio_list, and call into ->submit_bio()
639 * again.
640 *
641 * bio_list_on_stack[0] contains bios submitted by the current ->submit_bio.
642 * bio_list_on_stack[1] contains bios that were submitted before the current
Ming Lei69fe0f22022-03-04 21:08:03 -0500643 * ->submit_bio, but that haven't been processed yet.
Christoph Hellwig566acf22020-07-01 10:59:45 +0200644 */
Christoph Hellwig3e087732021-10-12 13:12:24 +0200645static void __submit_bio_noacct(struct bio *bio)
Christoph Hellwig566acf22020-07-01 10:59:45 +0200646{
647 struct bio_list bio_list_on_stack[2];
Christoph Hellwig566acf22020-07-01 10:59:45 +0200648
649 BUG_ON(bio->bi_next);
650
651 bio_list_init(&bio_list_on_stack[0]);
652 current->bio_list = bio_list_on_stack;
653
654 do {
Pavel Begunkoveab4e022021-10-14 15:03:29 +0100655 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
Christoph Hellwig566acf22020-07-01 10:59:45 +0200656 struct bio_list lower, same;
657
Christoph Hellwig566acf22020-07-01 10:59:45 +0200658 /*
659 * Create a fresh bio_list for all subordinate requests.
660 */
661 bio_list_on_stack[1] = bio_list_on_stack[0];
662 bio_list_init(&bio_list_on_stack[0]);
663
Christoph Hellwig3e087732021-10-12 13:12:24 +0200664 __submit_bio(bio);
Christoph Hellwig566acf22020-07-01 10:59:45 +0200665
666 /*
667 * Sort new bios into those for a lower level and those for the
668 * same level.
669 */
670 bio_list_init(&lower);
671 bio_list_init(&same);
672 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
Pavel Begunkoveab4e022021-10-14 15:03:29 +0100673 if (q == bdev_get_queue(bio->bi_bdev))
Christoph Hellwig566acf22020-07-01 10:59:45 +0200674 bio_list_add(&same, bio);
675 else
676 bio_list_add(&lower, bio);
677
678 /*
679 * Now assemble so we handle the lowest level first.
680 */
681 bio_list_merge(&bio_list_on_stack[0], &lower);
682 bio_list_merge(&bio_list_on_stack[0], &same);
683 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
684 } while ((bio = bio_list_pop(&bio_list_on_stack[0])));
685
686 current->bio_list = NULL;
Christoph Hellwig566acf22020-07-01 10:59:45 +0200687}
688
Christoph Hellwig3e087732021-10-12 13:12:24 +0200689static void __submit_bio_noacct_mq(struct bio *bio)
Christoph Hellwigff93ea02020-07-01 10:59:46 +0200690{
Christoph Hellwig7c792f32020-07-02 21:21:25 +0200691 struct bio_list bio_list[2] = { };
Christoph Hellwigff93ea02020-07-01 10:59:46 +0200692
Christoph Hellwig7c792f32020-07-02 21:21:25 +0200693 current->bio_list = bio_list;
Christoph Hellwigff93ea02020-07-01 10:59:46 +0200694
695 do {
Christoph Hellwig3e087732021-10-12 13:12:24 +0200696 __submit_bio(bio);
Christoph Hellwig7c792f32020-07-02 21:21:25 +0200697 } while ((bio = bio_list_pop(&bio_list[0])));
Christoph Hellwigff93ea02020-07-01 10:59:46 +0200698
699 current->bio_list = NULL;
Christoph Hellwigff93ea02020-07-01 10:59:46 +0200700}
701
Ming Lei3f98c752022-02-16 12:45:10 +0800702void submit_bio_noacct_nocheck(struct bio *bio)
703{
Jinke Han0f7c8f02023-02-16 11:22:50 +0800704 blk_cgroup_bio_start(bio);
705 blkcg_bio_issue_init(bio);
706
707 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
708 trace_block_bio_queue(bio);
709 /*
710 * Now that enqueuing has been traced, we need to trace
711 * completion as well.
712 */
713 bio_set_flag(bio, BIO_TRACE_COMPLETION);
714 }
715
Ming Lei3f98c752022-02-16 12:45:10 +0800716 /*
717 * We only want one ->submit_bio to be active at a time, else stack
718 * usage with stacked devices could be a problem. Use current->bio_list
719 * to collect a list of requests submited by a ->submit_bio method while
720 * it is active, and then process them after it returned.
721 */
722 if (current->bio_list)
723 bio_list_add(&current->bio_list[0], bio);
Al Viroac2b6f92024-04-12 01:21:45 -0400724 else if (!bdev_test_flag(bio->bi_bdev, BD_HAS_SUBMIT_BIO))
Ming Lei3f98c752022-02-16 12:45:10 +0800725 __submit_bio_noacct_mq(bio);
726 else
727 __submit_bio_noacct(bio);
728}
729
John Garry9da3d1e2024-06-20 12:53:54 +0000730static blk_status_t blk_validate_atomic_write_op_size(struct request_queue *q,
731 struct bio *bio)
732{
733 if (bio->bi_iter.bi_size > queue_atomic_write_unit_max_bytes(q))
734 return BLK_STS_INVAL;
735
736 if (bio->bi_iter.bi_size % queue_atomic_write_unit_min_bytes(q))
737 return BLK_STS_INVAL;
738
739 return BLK_STS_OK;
740}
741
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200742/**
Christoph Hellwiged00aab2020-07-01 10:59:44 +0200743 * submit_bio_noacct - re-submit a bio to the block device layer for I/O
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200744 * @bio: The bio describing the location in memory and on the device.
745 *
Christoph Hellwig3fdd4082020-04-28 13:27:53 +0200746 * This is a version of submit_bio() that shall only be used for I/O that is
747 * resubmitted to lower level drivers by stacking block drivers. All file
748 * systems and other upper level users of the block layer should use
749 * submit_bio() instead.
Neil Brownd89d8792007-05-01 09:53:42 +0200750 */
Christoph Hellwig3e087732021-10-12 13:12:24 +0200751void submit_bio_noacct(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752{
753 struct block_device *bdev = bio->bi_bdev;
754 struct request_queue *q = bdev_get_queue(bdev);
755 blk_status_t status = BLK_STS_IOERR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 might_sleep();
758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 /*
760 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
761 * if queue does not support NOWAIT.
NeilBrown5ddfe962006-10-30 22:07:21 -0800762 */
Christoph Hellwig568ec932022-09-27 09:58:15 +0200763 if ((bio->bi_opf & REQ_NOWAIT) && !bdev_nowait(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 goto not_supported;
765
766 if (should_fail_bio(bio))
767 goto end_io;
Miaohe Linbdb7d422022-09-05 18:27:54 +0800768 bio_check_ro(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 if (!bio_flagged(bio, BIO_REMAPPED)) {
770 if (unlikely(bio_check_eod(bio)))
771 goto end_io;
Al Viro3f9b8fb2024-04-12 00:54:19 -0400772 if (bdev_is_partition(bdev) &&
773 unlikely(blk_partition_remap(bio)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 goto end_io;
775 }
776
777 /*
778 * Filter flush bio's early so that bio based drivers without flush
779 * support don't have to worry about them.
780 */
Christoph Hellwigb4a6bb32022-11-02 00:09:03 -0700781 if (op_is_flush(bio->bi_opf)) {
782 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_WRITE &&
783 bio_op(bio) != REQ_OP_ZONE_APPEND))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 goto end_io;
Christoph Hellwig1122c0c2024-06-17 08:04:40 +0200785 if (!bdev_write_cache(bdev)) {
Christoph Hellwigb4a6bb32022-11-02 00:09:03 -0700786 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
787 if (!bio_sectors(bio)) {
788 status = BLK_STS_OK;
789 goto end_io;
790 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792 }
793
hexue73e59d32024-07-18 15:08:17 +0800794 if (!(q->limits.features & BLK_FEAT_POLL) &&
795 (bio->bi_opf & REQ_POLLED)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 bio_clear_polled(bio);
hexue73e59d32024-07-18 15:08:17 +0800797 goto not_supported;
798 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800 switch (bio_op(bio)) {
Christoph Hellwig1c042f82023-12-21 08:05:38 +0100801 case REQ_OP_READ:
802 case REQ_OP_WRITE:
John Garry9da3d1e2024-06-20 12:53:54 +0000803 if (bio->bi_opf & REQ_ATOMIC) {
804 status = blk_validate_atomic_write_op_size(q, bio);
805 if (status != BLK_STS_OK)
806 goto end_io;
807 }
Christoph Hellwig1c042f82023-12-21 08:05:38 +0100808 break;
809 case REQ_OP_FLUSH:
810 /*
811 * REQ_OP_FLUSH can't be submitted through bios, it is only
812 * synthetized in struct request by the flush state machine.
813 */
814 goto not_supported;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 case REQ_OP_DISCARD:
Christoph Hellwig70200572022-04-15 06:52:55 +0200816 if (!bdev_max_discard_sectors(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 goto not_supported;
NeilBrown5ddfe962006-10-30 22:07:21 -0800818 break;
Jens Axboe2056a782006-03-23 20:00:26 +0100819 case REQ_OP_SECURE_ERASE:
Christoph Hellwig44abff2c2022-04-15 06:52:57 +0200820 if (!bdev_max_secure_erase_sectors(bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 goto not_supported;
822 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 case REQ_OP_ZONE_APPEND:
824 status = blk_check_zone_append(q, bio);
825 if (status != BLK_STS_OK)
826 goto end_io;
827 break;
Christoph Hellwig1c042f82023-12-21 08:05:38 +0100828 case REQ_OP_WRITE_ZEROES:
829 if (!q->limits.max_write_zeroes_sectors)
830 goto not_supported;
831 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 case REQ_OP_ZONE_RESET:
833 case REQ_OP_ZONE_OPEN:
834 case REQ_OP_ZONE_CLOSE:
835 case REQ_OP_ZONE_FINISH:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 case REQ_OP_ZONE_RESET_ALL:
Damien Le Moalf2a7bea2024-07-04 14:28:15 +0900837 if (!bdev_is_zoned(bio->bi_bdev))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 goto not_supported;
839 break;
Christoph Hellwig1c042f82023-12-21 08:05:38 +0100840 case REQ_OP_DRV_IN:
841 case REQ_OP_DRV_OUT:
842 /*
843 * Driver private operations are only used with passthrough
844 * requests.
845 */
846 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 default:
Christoph Hellwig1c042f82023-12-21 08:05:38 +0100848 goto not_supported;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 }
850
851 if (blk_throtl_bio(bio))
Ming Leia6506282022-02-16 12:45:07 +0800852 return;
Ming Lei3f98c752022-02-16 12:45:10 +0800853 submit_bio_noacct_nocheck(bio);
Ming Leid24c6702022-02-16 12:45:11 +0800854 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Neil Brownd89d8792007-05-01 09:53:42 +0200856not_supported:
857 status = BLK_STS_NOTSUPP;
858end_io:
859 bio->bi_status = status;
860 bio_endio(bio);
Neil Brownd89d8792007-05-01 09:53:42 +0200861}
Christoph Hellwiged00aab2020-07-01 10:59:44 +0200862EXPORT_SYMBOL(submit_bio_noacct);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Hongyu Jinf3c89982024-01-30 15:26:34 -0500864static void bio_set_ioprio(struct bio *bio)
865{
866 /* Nobody set ioprio so far? Initialize it based on task's nice value */
867 if (IOPRIO_PRIO_CLASS(bio->bi_ioprio) == IOPRIO_CLASS_NONE)
868 bio->bi_ioprio = get_current_ioprio();
869 blkcg_set_ioprio(bio);
870}
871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872/**
Randy Dunlap710027a2008-08-19 20:13:11 +0200873 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 * @bio: The &struct bio which describes the I/O
875 *
Christoph Hellwig3fdd4082020-04-28 13:27:53 +0200876 * submit_bio() is used to submit I/O requests to block devices. It is passed a
877 * fully set up &struct bio that describes the I/O that needs to be done. The
Christoph Hellwig309dca302021-01-24 11:02:34 +0100878 * bio will be send to the device described by the bi_bdev field.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 *
Christoph Hellwig3fdd4082020-04-28 13:27:53 +0200880 * The success/failure status of the request, along with notification of
881 * completion, is delivered asynchronously through the ->bi_end_io() callback
Ping-Xiang Chene8848082022-09-14 00:42:37 -0700882 * in @bio. The bio must NOT be touched by the caller until ->bi_end_io() has
Christoph Hellwig3fdd4082020-04-28 13:27:53 +0200883 * been called.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 */
Christoph Hellwig3e087732021-10-12 13:12:24 +0200885void submit_bio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Christoph Hellwiga3e76892022-05-16 08:36:54 +0200887 if (bio_op(bio) == REQ_OP_READ) {
888 task_io_account_read(bio->bi_iter.bi_size);
889 count_vm_events(PGPGIN, bio_sectors(bio));
890 } else if (bio_op(bio) == REQ_OP_WRITE) {
891 count_vm_events(PGPGOUT, bio_sectors(bio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 }
893
Hongyu Jinf3c89982024-01-30 15:26:34 -0500894 bio_set_ioprio(bio);
Christoph Hellwig3e087732021-10-12 13:12:24 +0200895 submit_bio_noacct(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897EXPORT_SYMBOL(submit_bio);
898
Jens Axboe1052b8a2018-11-26 08:21:49 -0700899/**
Christoph Hellwig3e087732021-10-12 13:12:24 +0200900 * bio_poll - poll for BIO completions
901 * @bio: bio to poll for
Yang Guange30028a2021-11-26 00:20:55 +0800902 * @iob: batches of IO
Christoph Hellwig3e087732021-10-12 13:12:24 +0200903 * @flags: BLK_POLL_* flags that control the behavior
904 *
905 * Poll for completions on queue associated with the bio. Returns number of
906 * completed entries found.
907 *
908 * Note: the caller must either be the context that submitted @bio, or
909 * be in a RCU critical section to prevent freeing of @bio.
910 */
Jens Axboe5a72e892021-10-12 09:24:29 -0600911int bio_poll(struct bio *bio, struct io_comp_batch *iob, unsigned int flags)
Christoph Hellwig3e087732021-10-12 13:12:24 +0200912{
Christoph Hellwig3e087732021-10-12 13:12:24 +0200913 blk_qc_t cookie = READ_ONCE(bio->bi_cookie);
Jens Axboe310726c2023-02-24 10:01:19 -0700914 struct block_device *bdev;
915 struct request_queue *q;
Ming Lei69fe0f22022-03-04 21:08:03 -0500916 int ret = 0;
Christoph Hellwig3e087732021-10-12 13:12:24 +0200917
Jens Axboe310726c2023-02-24 10:01:19 -0700918 bdev = READ_ONCE(bio->bi_bdev);
919 if (!bdev)
920 return 0;
921
922 q = bdev_get_queue(bdev);
Christoph Hellwig8023e142024-06-17 08:04:48 +0200923 if (cookie == BLK_QC_T_NONE || !(q->limits.features & BLK_FEAT_POLL))
Christoph Hellwig3e087732021-10-12 13:12:24 +0200924 return 0;
925
Christoph Hellwigaa8dccc2022-01-27 08:05:49 +0100926 blk_flush_plug(current->plug, false);
Christoph Hellwig3e087732021-10-12 13:12:24 +0200927
Jens Axboe33391ee2023-01-20 07:51:07 -0700928 /*
929 * We need to be able to enter a frozen queue, similar to how
930 * timeouts also need to do that. If that is blocked, then we can
931 * have pending IO when a queue freeze is started, and then the
932 * wait for the freeze to finish will wait for polled requests to
933 * timeout as the poller is preventer from entering the queue and
934 * completing them. As long as we prevent new IO from being queued,
935 * that should be all that matters.
936 */
937 if (!percpu_ref_tryget(&q->q_usage_counter))
Christoph Hellwig3e087732021-10-12 13:12:24 +0200938 return 0;
Ming Lei69fe0f22022-03-04 21:08:03 -0500939 if (queue_is_mq(q)) {
Jens Axboe5a72e892021-10-12 09:24:29 -0600940 ret = blk_mq_poll(q, cookie, iob, flags);
Ming Lei69fe0f22022-03-04 21:08:03 -0500941 } else {
942 struct gendisk *disk = q->disk;
943
944 if (disk && disk->fops->poll_bio)
945 ret = disk->fops->poll_bio(bio, iob, flags);
946 }
Christoph Hellwig3e087732021-10-12 13:12:24 +0200947 blk_queue_exit(q);
948 return ret;
949}
950EXPORT_SYMBOL_GPL(bio_poll);
951
952/*
953 * Helper to implement file_operations.iopoll. Requires the bio to be stored
954 * in iocb->private, and cleared before freeing the bio.
955 */
Jens Axboe5a72e892021-10-12 09:24:29 -0600956int iocb_bio_iopoll(struct kiocb *kiocb, struct io_comp_batch *iob,
957 unsigned int flags)
Christoph Hellwig3e087732021-10-12 13:12:24 +0200958{
959 struct bio *bio;
960 int ret = 0;
961
962 /*
963 * Note: the bio cache only uses SLAB_TYPESAFE_BY_RCU, so bio can
964 * point to a freshly allocated bio at this point. If that happens
965 * we have a few cases to consider:
966 *
967 * 1) the bio is beeing initialized and bi_bdev is NULL. We can just
968 * simply nothing in this case
969 * 2) the bio points to a not poll enabled device. bio_poll will catch
970 * this and return 0
971 * 3) the bio points to a poll capable device, including but not
972 * limited to the one that the original bio pointed to. In this
973 * case we will call into the actual poll method and poll for I/O,
974 * even if we don't need to, but it won't cause harm either.
975 *
976 * For cases 2) and 3) above the RCU grace period ensures that bi_bdev
977 * is still allocated. Because partitions hold a reference to the whole
978 * device bdev and thus disk, the disk is also still valid. Grabbing
979 * a reference to the queue in bio_poll() ensures the hctxs and requests
980 * are still valid as well.
981 */
982 rcu_read_lock();
983 bio = READ_ONCE(kiocb->private);
Jens Axboe310726c2023-02-24 10:01:19 -0700984 if (bio)
Jens Axboe5a72e892021-10-12 09:24:29 -0600985 ret = bio_poll(bio, iob, flags);
Christoph Hellwig3e087732021-10-12 13:12:24 +0200986 rcu_read_unlock();
987
988 return ret;
989}
990EXPORT_SYMBOL_GPL(iocb_bio_iopoll);
991
Christoph Hellwig450b7872021-11-17 07:14:01 +0100992void update_io_ticks(struct block_device *part, unsigned long now, bool end)
Christoph Hellwig9123bf62020-05-27 07:24:13 +0200993{
994 unsigned long stamp;
995again:
Christoph Hellwig8446fe92020-11-24 09:36:54 +0100996 stamp = READ_ONCE(part->bd_stamp);
Yu Kuai99dc4222024-05-09 20:37:16 +0800997 if (unlikely(time_after(now, stamp)) &&
998 likely(try_cmpxchg(&part->bd_stamp, &stamp, now)) &&
999 (end || part_in_flight(part)))
1000 __part_stat_add(part, io_ticks, now - stamp);
1001
Al Viro3f9b8fb2024-04-12 00:54:19 -04001002 if (bdev_is_partition(part)) {
Christoph Hellwig8446fe92020-11-24 09:36:54 +01001003 part = bdev_whole(part);
Christoph Hellwig9123bf62020-05-27 07:24:13 +02001004 goto again;
1005 }
1006}
1007
Yu Kuai5f275712023-02-23 17:12:26 +08001008unsigned long bdev_start_io_acct(struct block_device *bdev, enum req_op op,
Ming Lei5f0614a2022-04-17 22:27:13 -04001009 unsigned long start_time)
Christoph Hellwig956d5102020-05-27 07:24:04 +02001010{
Christoph Hellwig956d5102020-05-27 07:24:04 +02001011 part_stat_lock();
Ming Lei5f0614a2022-04-17 22:27:13 -04001012 update_io_ticks(bdev, start_time, false);
Ming Lei5f0614a2022-04-17 22:27:13 -04001013 part_stat_local_inc(bdev, in_flight[op_is_write(op)]);
Christoph Hellwig956d5102020-05-27 07:24:04 +02001014 part_stat_unlock();
1015
Mike Snitzere45c47d2022-01-28 10:58:39 -05001016 return start_time;
Christoph Hellwig956d5102020-05-27 07:24:04 +02001017}
Ming Lei5f0614a2022-04-17 22:27:13 -04001018EXPORT_SYMBOL(bdev_start_io_acct);
Song Liu7b264102020-08-31 15:27:23 -07001019
Christoph Hellwig99dfc432021-01-24 11:02:37 +01001020/**
1021 * bio_start_io_acct - start I/O accounting for bio based drivers
1022 * @bio: bio to start account for
1023 *
1024 * Returns the start time that should be passed back to bio_end_io_acct().
1025 */
1026unsigned long bio_start_io_acct(struct bio *bio)
Song Liu7b264102020-08-31 15:27:23 -07001027{
Yu Kuai5f275712023-02-23 17:12:26 +08001028 return bdev_start_io_acct(bio->bi_bdev, bio_op(bio), jiffies);
Song Liu7b264102020-08-31 15:27:23 -07001029}
Christoph Hellwig99dfc432021-01-24 11:02:37 +01001030EXPORT_SYMBOL_GPL(bio_start_io_acct);
Song Liu7b264102020-08-31 15:27:23 -07001031
Bart Van Assche77e7ffd2022-07-14 11:06:28 -07001032void bdev_end_io_acct(struct block_device *bdev, enum req_op op,
Yu Kuai5f275712023-02-23 17:12:26 +08001033 unsigned int sectors, unsigned long start_time)
Christoph Hellwig956d5102020-05-27 07:24:04 +02001034{
Christoph Hellwig956d5102020-05-27 07:24:04 +02001035 const int sgrp = op_stat_group(op);
1036 unsigned long now = READ_ONCE(jiffies);
1037 unsigned long duration = now - start_time;
1038
1039 part_stat_lock();
Ming Lei5f0614a2022-04-17 22:27:13 -04001040 update_io_ticks(bdev, now, true);
Yu Kuai5f275712023-02-23 17:12:26 +08001041 part_stat_inc(bdev, ios[sgrp]);
1042 part_stat_add(bdev, sectors[sgrp], sectors);
Ming Lei5f0614a2022-04-17 22:27:13 -04001043 part_stat_add(bdev, nsecs[sgrp], jiffies_to_nsecs(duration));
1044 part_stat_local_dec(bdev, in_flight[op_is_write(op)]);
Christoph Hellwig956d5102020-05-27 07:24:04 +02001045 part_stat_unlock();
1046}
Ming Lei5f0614a2022-04-17 22:27:13 -04001047EXPORT_SYMBOL(bdev_end_io_acct);
Song Liu7b264102020-08-31 15:27:23 -07001048
Christoph Hellwig99dfc432021-01-24 11:02:37 +01001049void bio_end_io_acct_remapped(struct bio *bio, unsigned long start_time,
Ming Lei5f0614a2022-04-17 22:27:13 -04001050 struct block_device *orig_bdev)
Song Liu7b264102020-08-31 15:27:23 -07001051{
Yu Kuai5f275712023-02-23 17:12:26 +08001052 bdev_end_io_acct(orig_bdev, bio_op(bio), bio_sectors(bio), start_time);
Song Liu7b264102020-08-31 15:27:23 -07001053}
Christoph Hellwig99dfc432021-01-24 11:02:37 +01001054EXPORT_SYMBOL_GPL(bio_end_io_acct_remapped);
Song Liu7b264102020-08-31 15:27:23 -07001055
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02001056/**
1057 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1058 * @q : the queue of the device being checked
1059 *
1060 * Description:
1061 * Check if underlying low-level drivers of a device are busy.
1062 * If the drivers want to export their busy state, they must set own
1063 * exporting function using blk_queue_lld_busy() first.
1064 *
1065 * Basically, this function is used only by request stacking drivers
1066 * to stop dispatching requests to underlying devices when underlying
1067 * devices are busy. This behavior helps more I/O merging on the queue
1068 * of the request stacking driver and prevents I/O throughput regression
1069 * on burst I/O load.
1070 *
1071 * Return:
1072 * 0 - Not busy (The request stacking driver should dispatch request)
1073 * 1 - Busy (The request stacking driver should stop dispatching request)
1074 */
1075int blk_lld_busy(struct request_queue *q)
1076{
Jens Axboe344e9ff2018-11-15 12:22:51 -07001077 if (queue_is_mq(q) && q->mq_ops->busy)
Jens Axboe9ba20522018-10-29 10:15:10 -06001078 return q->mq_ops->busy(q);
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02001079
1080 return 0;
1081}
1082EXPORT_SYMBOL_GPL(blk_lld_busy);
1083
Jens Axboe59c3d452014-04-08 09:15:35 -06001084int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
1086 return queue_work(kblockd_workqueue, work);
1087}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088EXPORT_SYMBOL(kblockd_schedule_work);
1089
Jens Axboe818cd1c2017-04-10 09:54:55 -06001090int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1091 unsigned long delay)
1092{
1093 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1094}
1095EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1096
Jens Axboe47c122e2021-10-06 06:34:11 -06001097void blk_start_plug_nr_ios(struct blk_plug *plug, unsigned short nr_ios)
1098{
1099 struct task_struct *tsk = current;
1100
1101 /*
1102 * If this is a nested plug, don't actually assign it.
1103 */
1104 if (tsk->plug)
1105 return;
1106
Jens Axboeda4c8c32024-01-15 14:46:03 -07001107 plug->cur_ktime = 0;
Jens Axboebc490f82021-10-18 10:12:12 -06001108 plug->mq_list = NULL;
Jens Axboe47c122e2021-10-06 06:34:11 -06001109 plug->cached_rq = NULL;
1110 plug->nr_ios = min_t(unsigned short, nr_ios, BLK_MAX_REQUEST_COUNT);
1111 plug->rq_count = 0;
1112 plug->multiple_queues = false;
Jens Axboedc5fc3612021-10-19 06:02:30 -06001113 plug->has_elevator = false;
Jens Axboe47c122e2021-10-06 06:34:11 -06001114 INIT_LIST_HEAD(&plug->cb_list);
1115
1116 /*
1117 * Store ordering should not be needed here, since a potential
1118 * preempt will imply a full memory barrier
1119 */
1120 tsk->plug = plug;
1121}
1122
Suresh Jayaraman75df7132011-09-21 10:00:16 +02001123/**
1124 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1125 * @plug: The &struct blk_plug that needs to be initialized
1126 *
1127 * Description:
Jeff Moyer40405852019-01-08 16:57:34 -05001128 * blk_start_plug() indicates to the block layer an intent by the caller
1129 * to submit multiple I/O requests in a batch. The block layer may use
1130 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1131 * is called. However, the block layer may choose to submit requests
1132 * before a call to blk_finish_plug() if the number of queued I/Os
1133 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1134 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1135 * the task schedules (see below).
1136 *
Suresh Jayaraman75df7132011-09-21 10:00:16 +02001137 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1138 * pending I/O should the task end up blocking between blk_start_plug() and
1139 * blk_finish_plug(). This is important from a performance perspective, but
1140 * also ensures that we don't deadlock. For instance, if the task is blocking
1141 * for a memory allocation, memory reclaim could end up wanting to free a
1142 * page belonging to that request that is currently residing in our private
1143 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1144 * this kind of deadlock.
1145 */
Jens Axboe73c10102011-03-08 13:19:51 +01001146void blk_start_plug(struct blk_plug *plug)
1147{
Jens Axboe47c122e2021-10-06 06:34:11 -06001148 blk_start_plug_nr_ios(plug, 1);
Jens Axboe73c10102011-03-08 13:19:51 +01001149}
1150EXPORT_SYMBOL(blk_start_plug);
1151
NeilBrown74018dc2012-07-31 09:08:15 +02001152static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02001153{
1154 LIST_HEAD(callbacks);
1155
Shaohua Li2a7d5552012-07-31 09:08:15 +02001156 while (!list_empty(&plug->cb_list)) {
1157 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02001158
Shaohua Li2a7d5552012-07-31 09:08:15 +02001159 while (!list_empty(&callbacks)) {
1160 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02001161 struct blk_plug_cb,
1162 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02001163 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02001164 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02001165 }
NeilBrown048c9372011-04-18 09:52:22 +02001166 }
1167}
1168
NeilBrown9cbb1752012-07-31 09:08:14 +02001169struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1170 int size)
1171{
1172 struct blk_plug *plug = current->plug;
1173 struct blk_plug_cb *cb;
1174
1175 if (!plug)
1176 return NULL;
1177
1178 list_for_each_entry(cb, &plug->cb_list, list)
1179 if (cb->callback == unplug && cb->data == data)
1180 return cb;
1181
1182 /* Not currently on the callback list */
1183 BUG_ON(size < sizeof(*cb));
1184 cb = kzalloc(size, GFP_ATOMIC);
1185 if (cb) {
1186 cb->data = data;
1187 cb->callback = unplug;
1188 list_add(&cb->list, &plug->cb_list);
1189 }
1190 return cb;
1191}
1192EXPORT_SYMBOL(blk_check_plugged);
1193
Christoph Hellwigaa8dccc2022-01-27 08:05:49 +01001194void __blk_flush_plug(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01001195{
Pavel Begunkovb6004552021-10-20 16:41:18 +02001196 if (!list_empty(&plug->cb_list))
1197 flush_plug_callbacks(plug, from_schedule);
Ross Lagerwall70904262023-07-14 11:11:06 +01001198 blk_mq_flush_plug_list(plug, from_schedule);
Jens Axboec5fc7b92021-11-03 05:49:07 -06001199 /*
1200 * Unconditionally flush out cached requests, even if the unplug
1201 * event came from schedule. Since we know hold references to the
1202 * queue for cached requests, we don't want a blocked task holding
1203 * up a queue freeze/quiesce event.
1204 */
1205 if (unlikely(!rq_list_empty(plug->cached_rq)))
Jens Axboe47c122e2021-10-06 06:34:11 -06001206 blk_mq_free_plug_rqs(plug);
Jens Axboe06b23f922024-01-16 09:18:39 -07001207
Yu Kuai3ec48482024-04-11 11:23:48 +08001208 plug->cur_ktime = 0;
Jens Axboe06b23f922024-01-16 09:18:39 -07001209 current->flags &= ~PF_BLOCK_TS;
Jens Axboe73c10102011-03-08 13:19:51 +01001210}
Jens Axboe73c10102011-03-08 13:19:51 +01001211
Jeff Moyer40405852019-01-08 16:57:34 -05001212/**
1213 * blk_finish_plug - mark the end of a batch of submitted I/O
1214 * @plug: The &struct blk_plug passed to blk_start_plug()
1215 *
1216 * Description:
1217 * Indicate that a batch of I/O submissions is complete. This function
1218 * must be paired with an initial call to blk_start_plug(). The intent
1219 * is to allow the block layer to optimize I/O submission. See the
1220 * documentation for blk_start_plug() for more information.
1221 */
Jens Axboe73c10102011-03-08 13:19:51 +01001222void blk_finish_plug(struct blk_plug *plug)
1223{
Christoph Hellwig008f75a2021-10-20 16:41:19 +02001224 if (plug == current->plug) {
Christoph Hellwigaa8dccc2022-01-27 08:05:49 +01001225 __blk_flush_plug(plug, false);
Christoph Hellwig008f75a2021-10-20 16:41:19 +02001226 current->plug = NULL;
1227 }
Jens Axboe73c10102011-03-08 13:19:51 +01001228}
1229EXPORT_SYMBOL(blk_finish_plug);
1230
Ming Lei71ac8602020-05-14 16:45:09 +08001231void blk_io_schedule(void)
1232{
1233 /* Prevent hang_check timer from firing at us during very long I/O */
1234 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;
1235
1236 if (timeout)
1237 io_schedule_timeout(timeout);
1238 else
1239 io_schedule();
1240}
1241EXPORT_SYMBOL_GPL(blk_io_schedule);
1242
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243int __init blk_dev_init(void)
1244{
Bart Van Assche16458cf2022-07-14 11:06:32 -07001245 BUILD_BUG_ON((__force u32)REQ_OP_LAST >= (1 << REQ_OP_BITS));
Christoph Hellwigef295ec2016-10-28 08:48:16 -06001246 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
Pankaj Bharadiyac5936422019-12-09 10:31:43 -08001247 sizeof_field(struct request, cmd_flags));
Christoph Hellwigef295ec2016-10-28 08:48:16 -06001248 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
Pankaj Bharadiyac5936422019-12-09 10:31:43 -08001249 sizeof_field(struct bio, bi_opf));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02001250
Tejun Heo89b90be2011-01-03 15:01:47 +01001251 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1252 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02001253 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 if (!kblockd_workqueue)
1255 panic("Failed to create kblockd\n");
1256
Kunwu Chan48ff13a2024-01-31 17:43:23 +08001257 blk_requestq_cachep = KMEM_CACHE(request_queue, SLAB_PANIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
Omar Sandoval18fbda92017-01-31 14:53:20 -08001259 blk_debugfs_root = debugfs_create_dir("block", NULL);
Omar Sandoval18fbda92017-01-31 14:53:20 -08001260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 return 0;
1262}