blob: 875e8d105067a2248b7829e487bd66c5f4c9281c [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>
17#include <linux/backing-dev.h>
18#include <linux/bio.h>
19#include <linux/blkdev.h>
Jens Axboe320ae512013-10-24 09:20:05 +010020#include <linux/blk-mq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/completion.h>
27#include <linux/slab.h>
28#include <linux/swap.h>
29#include <linux/writeback.h>
Andrew Mortonfaccbd4b2006-12-10 02:19:35 -080030#include <linux/task_io_accounting_ops.h>
Akinobu Mitac17bb492006-12-08 02:39:46 -080031#include <linux/fault-inject.h>
Jens Axboe73c10102011-03-08 13:19:51 +010032#include <linux/list_sort.h>
Tejun Heoe3c78ca2011-10-19 14:32:38 +020033#include <linux/delay.h>
Tejun Heoaaf7c682012-04-19 16:29:22 -070034#include <linux/ratelimit.h>
Lin Ming6c954662013-03-23 11:42:26 +080035#include <linux/pm_runtime.h>
Tejun Heoeea8f412015-05-22 17:13:17 -040036#include <linux/blk-cgroup.h>
Omar Sandoval18fbda92017-01-31 14:53:20 -080037#include <linux/debugfs.h>
Howard McLauchlan30abb3a2018-02-06 14:05:39 -080038#include <linux/bpf.h>
Johannes Weinerb8e24a92019-08-08 15:03:00 -040039#include <linux/psi.h>
Li Zefan55782132009-06-09 13:43:05 +080040
41#define CREATE_TRACE_POINTS
42#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Jens Axboe8324aa92008-01-29 14:51:59 +010044#include "blk.h"
Ming Lei43a5e4e2013-12-26 21:31:35 +080045#include "blk-mq.h"
Jens Axboebd166ef2017-01-17 06:03:22 -070046#include "blk-mq-sched.h"
Bart Van Asschebca6b062018-09-26 14:01:03 -070047#include "blk-pm.h"
Josef Bacikc1c80382018-07-03 11:14:59 -040048#include "blk-rq-qos.h"
Jens Axboe8324aa92008-01-29 14:51:59 +010049
Omar Sandoval18fbda92017-01-31 14:53:20 -080050#ifdef CONFIG_DEBUG_FS
51struct dentry *blk_debugfs_root;
52#endif
53
Mike Snitzerd07335e2010-11-16 12:52:38 +010054EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
Jun'ichi Nomurab0da3f02009-10-01 21:16:13 +020055EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
Linus Torvalds0a82a8d2013-04-18 09:00:26 -070056EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
Keith Busch3291fa52014-04-28 12:30:52 -060057EXPORT_TRACEPOINT_SYMBOL_GPL(block_split);
NeilBrowncbae8d42012-12-14 20:49:27 +010058EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
Ingo Molnar0bfc2452008-11-26 11:59:56 +010059
Tejun Heoa73f7302011-12-14 00:33:37 +010060DEFINE_IDA(blk_queue_ida);
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 * For queue allocation
64 */
Jens Axboe6728cb02008-01-31 13:03:55 +010065struct kmem_cache *blk_requestq_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 * Controlling structure to kblockd
69 */
Jens Axboeff856ba2006-01-09 16:02:34 +010070static struct workqueue_struct *kblockd_workqueue;
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Bart Van Assche8814ce82018-03-07 17:10:04 -080072/**
73 * blk_queue_flag_set - atomically set a queue flag
74 * @flag: flag to be set
75 * @q: request queue
76 */
77void blk_queue_flag_set(unsigned int flag, struct request_queue *q)
78{
Christoph Hellwig57d74df2018-11-14 17:02:07 +010079 set_bit(flag, &q->queue_flags);
Bart Van Assche8814ce82018-03-07 17:10:04 -080080}
81EXPORT_SYMBOL(blk_queue_flag_set);
82
83/**
84 * blk_queue_flag_clear - atomically clear a queue flag
85 * @flag: flag to be cleared
86 * @q: request queue
87 */
88void blk_queue_flag_clear(unsigned int flag, struct request_queue *q)
89{
Christoph Hellwig57d74df2018-11-14 17:02:07 +010090 clear_bit(flag, &q->queue_flags);
Bart Van Assche8814ce82018-03-07 17:10:04 -080091}
92EXPORT_SYMBOL(blk_queue_flag_clear);
93
94/**
95 * blk_queue_flag_test_and_set - atomically test and set a queue flag
96 * @flag: flag to be set
97 * @q: request queue
98 *
99 * Returns the previous value of @flag - 0 if the flag was not set and 1 if
100 * the flag was already set.
101 */
102bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q)
103{
Christoph Hellwig57d74df2018-11-14 17:02:07 +0100104 return test_and_set_bit(flag, &q->queue_flags);
Bart Van Assche8814ce82018-03-07 17:10:04 -0800105}
106EXPORT_SYMBOL_GPL(blk_queue_flag_test_and_set);
107
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200108void blk_rq_init(struct request_queue *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109{
FUJITA Tomonori1afb20f2008-04-25 12:26:28 +0200110 memset(rq, 0, sizeof(*rq));
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 INIT_LIST_HEAD(&rq->queuelist);
Jens Axboe63a71382008-02-08 12:41:03 +0100113 rq->q = q;
Tejun Heoa2dec7b2009-05-07 22:24:44 +0900114 rq->__sector = (sector_t) -1;
Jens Axboe2e662b62006-07-13 11:55:04 +0200115 INIT_HLIST_NODE(&rq->hash);
116 RB_CLEAR_NODE(&rq->rb_node);
Jens Axboe63a71382008-02-08 12:41:03 +0100117 rq->tag = -1;
Jens Axboebd166ef2017-01-17 06:03:22 -0700118 rq->internal_tag = -1;
Omar Sandoval522a7772018-05-09 02:08:53 -0700119 rq->start_time_ns = ktime_get_ns();
Jerome Marchand09e099d2011-01-05 16:57:38 +0100120 rq->part = NULL;
Josef Bacikb554db12019-03-07 21:37:18 +0000121 refcount_set(&rq->ref, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122}
FUJITA Tomonori2a4aa302008-04-29 09:54:36 +0200123EXPORT_SYMBOL(blk_rq_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Chaitanya Kulkarnie47bc4e2019-06-20 10:59:16 -0700125#define REQ_OP_NAME(name) [REQ_OP_##name] = #name
126static const char *const blk_op_name[] = {
127 REQ_OP_NAME(READ),
128 REQ_OP_NAME(WRITE),
129 REQ_OP_NAME(FLUSH),
130 REQ_OP_NAME(DISCARD),
131 REQ_OP_NAME(SECURE_ERASE),
132 REQ_OP_NAME(ZONE_RESET),
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700133 REQ_OP_NAME(ZONE_RESET_ALL),
Chaitanya Kulkarnie47bc4e2019-06-20 10:59:16 -0700134 REQ_OP_NAME(WRITE_SAME),
135 REQ_OP_NAME(WRITE_ZEROES),
136 REQ_OP_NAME(SCSI_IN),
137 REQ_OP_NAME(SCSI_OUT),
138 REQ_OP_NAME(DRV_IN),
139 REQ_OP_NAME(DRV_OUT),
140};
141#undef REQ_OP_NAME
142
143/**
144 * blk_op_str - Return string XXX in the REQ_OP_XXX.
145 * @op: REQ_OP_XXX.
146 *
147 * Description: Centralize block layer function to convert REQ_OP_XXX into
148 * string format. Useful in the debugging and tracing bio or request. For
149 * invalid REQ_OP_XXX it returns string "UNKNOWN".
150 */
151inline const char *blk_op_str(unsigned int op)
152{
153 const char *op_str = "UNKNOWN";
154
155 if (op < ARRAY_SIZE(blk_op_name) && blk_op_name[op])
156 op_str = blk_op_name[op];
157
158 return op_str;
159}
160EXPORT_SYMBOL_GPL(blk_op_str);
161
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200162static const struct {
163 int errno;
164 const char *name;
165} blk_errors[] = {
166 [BLK_STS_OK] = { 0, "" },
167 [BLK_STS_NOTSUPP] = { -EOPNOTSUPP, "operation not supported" },
168 [BLK_STS_TIMEOUT] = { -ETIMEDOUT, "timeout" },
169 [BLK_STS_NOSPC] = { -ENOSPC, "critical space allocation" },
170 [BLK_STS_TRANSPORT] = { -ENOLINK, "recoverable transport" },
171 [BLK_STS_TARGET] = { -EREMOTEIO, "critical target" },
172 [BLK_STS_NEXUS] = { -EBADE, "critical nexus" },
173 [BLK_STS_MEDIUM] = { -ENODATA, "critical medium" },
174 [BLK_STS_PROTECTION] = { -EILSEQ, "protection" },
175 [BLK_STS_RESOURCE] = { -ENOMEM, "kernel resource" },
Ming Lei86ff7c22018-01-30 22:04:57 -0500176 [BLK_STS_DEV_RESOURCE] = { -EBUSY, "device resource" },
Goldwyn Rodrigues03a07c92017-06-20 07:05:46 -0500177 [BLK_STS_AGAIN] = { -EAGAIN, "nonblocking retry" },
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200178
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200179 /* device mapper special case, should not leak out: */
180 [BLK_STS_DM_REQUEUE] = { -EREMCHG, "dm internal retry" },
181
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200182 /* everything else not covered above: */
183 [BLK_STS_IOERR] = { -EIO, "I/O" },
184};
185
186blk_status_t errno_to_blk_status(int errno)
187{
188 int i;
189
190 for (i = 0; i < ARRAY_SIZE(blk_errors); i++) {
191 if (blk_errors[i].errno == errno)
192 return (__force blk_status_t)i;
193 }
194
195 return BLK_STS_IOERR;
196}
197EXPORT_SYMBOL_GPL(errno_to_blk_status);
198
199int blk_status_to_errno(blk_status_t status)
200{
201 int idx = (__force int)status;
202
Bart Van Assche34bd9c12017-06-21 10:55:46 -0700203 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200204 return -EIO;
205 return blk_errors[idx].errno;
206}
207EXPORT_SYMBOL_GPL(blk_status_to_errno);
208
Christoph Hellwig178cc592019-06-20 10:59:15 -0700209static void print_req_error(struct request *req, blk_status_t status,
210 const char *caller)
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200211{
212 int idx = (__force int)status;
213
Bart Van Assche34bd9c12017-06-21 10:55:46 -0700214 if (WARN_ON_ONCE(idx >= ARRAY_SIZE(blk_errors)))
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200215 return;
216
Christoph Hellwig178cc592019-06-20 10:59:15 -0700217 printk_ratelimited(KERN_ERR
Chaitanya Kulkarnib0e51682019-06-20 10:59:18 -0700218 "%s: %s error, dev %s, sector %llu op 0x%x:(%s) flags 0x%x "
219 "phys_seg %u prio class %u\n",
Christoph Hellwig178cc592019-06-20 10:59:15 -0700220 caller, blk_errors[idx].name,
Chaitanya Kulkarnib0e51682019-06-20 10:59:18 -0700221 req->rq_disk ? req->rq_disk->disk_name : "?",
222 blk_rq_pos(req), req_op(req), blk_op_str(req_op(req)),
223 req->cmd_flags & ~REQ_OP_MASK,
224 req->nr_phys_segments,
225 IOPRIO_PRIO_CLASS(req->ioprio));
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200226}
227
NeilBrown5bb23a62007-09-27 12:46:13 +0200228static void req_bio_endio(struct request *rq, struct bio *bio,
Christoph Hellwig2a842ac2017-06-03 09:38:04 +0200229 unsigned int nbytes, blk_status_t error)
Tejun Heo797e7db2006-01-06 09:51:03 +0100230{
Mike Snitzer78d8e582015-06-26 10:01:13 -0400231 if (error)
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200232 bio->bi_status = error;
Tejun Heo797e7db2006-01-06 09:51:03 +0100233
Christoph Hellwige8064022016-10-20 15:12:13 +0200234 if (unlikely(rq->rq_flags & RQF_QUIET))
Jens Axboeb7c44ed2015-07-24 12:37:59 -0600235 bio_set_flag(bio, BIO_QUIET);
Tejun Heo143a87f2011-01-25 12:43:52 +0100236
Kent Overstreetf79ea412012-09-20 16:38:30 -0700237 bio_advance(bio, nbytes);
Tejun Heo143a87f2011-01-25 12:43:52 +0100238
239 /* don't actually finish bio if it's part of flush sequence */
Christoph Hellwige8064022016-10-20 15:12:13 +0200240 if (bio->bi_iter.bi_size == 0 && !(rq->rq_flags & RQF_FLUSH_SEQ))
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200241 bio_endio(bio);
Tejun Heo797e7db2006-01-06 09:51:03 +0100242}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244void blk_dump_rq_flags(struct request *rq, char *msg)
245{
Christoph Hellwigaebf5262017-01-31 16:57:31 +0100246 printk(KERN_INFO "%s: dev %s: flags=%llx\n", msg,
247 rq->rq_disk ? rq->rq_disk->disk_name : "?",
Jens Axboe59533162013-05-23 12:25:08 +0200248 (unsigned long long) rq->cmd_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Tejun Heo83096eb2009-05-07 22:24:39 +0900250 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
251 (unsigned long long)blk_rq_pos(rq),
252 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
Jens Axboeb4f42e22014-04-10 09:46:28 -0600253 printk(KERN_INFO " bio %p, biotail %p, len %u\n",
254 rq->bio, rq->biotail, blk_rq_bytes(rq));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256EXPORT_SYMBOL(blk_dump_rq_flags);
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258/**
259 * blk_sync_queue - cancel any pending callbacks on a queue
260 * @q: the queue
261 *
262 * Description:
263 * The block layer may perform asynchronous callback activity
264 * on a queue, such as calling the unplug function after a timeout.
265 * A block device may call blk_sync_queue to ensure that any
266 * such activity is cancelled, thus allowing it to release resources
Michael Opdenacker59c51592007-05-09 08:57:56 +0200267 * that the callbacks might use. The caller must already have made sure
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 * that its ->make_request_fn will not re-add plugging prior to calling
269 * this function.
270 *
Vivek Goyalda527772011-03-02 19:05:33 -0500271 * This function does not cancel any asynchronous activity arising
Masanari Iidada3dae52014-09-09 01:27:23 +0900272 * out of elevator or throttling code. That would require elevator_exit()
Tejun Heo5efd6112012-03-05 13:15:12 -0800273 * and blkcg_exit_queue() to be called with queue lock initialized.
Vivek Goyalda527772011-03-02 19:05:33 -0500274 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 */
276void blk_sync_queue(struct request_queue *q)
277{
Jens Axboe70ed28b2008-11-19 14:38:39 +0100278 del_timer_sync(&q->timeout);
Bart Van Assche4e9b6f22017-10-19 10:00:48 -0700279 cancel_work_sync(&q->timeout_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280}
281EXPORT_SYMBOL(blk_sync_queue);
282
283/**
Bart Van Asschecd84a622018-09-26 14:01:04 -0700284 * blk_set_pm_only - increment pm_only counter
Bart Van Asschec9254f22017-11-09 10:49:57 -0800285 * @q: request queue pointer
Bart Van Asschec9254f22017-11-09 10:49:57 -0800286 */
Bart Van Asschecd84a622018-09-26 14:01:04 -0700287void blk_set_pm_only(struct request_queue *q)
Bart Van Asschec9254f22017-11-09 10:49:57 -0800288{
Bart Van Asschecd84a622018-09-26 14:01:04 -0700289 atomic_inc(&q->pm_only);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800290}
Bart Van Asschecd84a622018-09-26 14:01:04 -0700291EXPORT_SYMBOL_GPL(blk_set_pm_only);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800292
Bart Van Asschecd84a622018-09-26 14:01:04 -0700293void blk_clear_pm_only(struct request_queue *q)
Bart Van Asschec9254f22017-11-09 10:49:57 -0800294{
Bart Van Asschecd84a622018-09-26 14:01:04 -0700295 int pm_only;
296
297 pm_only = atomic_dec_return(&q->pm_only);
298 WARN_ON_ONCE(pm_only < 0);
299 if (pm_only == 0)
300 wake_up_all(&q->mq_freeze_wq);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800301}
Bart Van Asschecd84a622018-09-26 14:01:04 -0700302EXPORT_SYMBOL_GPL(blk_clear_pm_only);
Bart Van Asschec9254f22017-11-09 10:49:57 -0800303
Jens Axboe165125e2007-07-24 09:28:11 +0200304void blk_put_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500305{
306 kobject_put(&q->kobj);
307}
Jens Axboed86e0e82011-05-27 07:44:43 +0200308EXPORT_SYMBOL(blk_put_queue);
Al Viro483f4af2006-03-18 18:34:37 -0500309
Jens Axboeaed3ea92014-12-22 14:04:42 -0700310void blk_set_queue_dying(struct request_queue *q)
311{
Bart Van Assche8814ce82018-03-07 17:10:04 -0800312 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700313
Ming Leid3cfb2a2017-03-27 20:06:58 +0800314 /*
315 * When queue DYING flag is set, we need to block new req
316 * entering queue, so we call blk_freeze_queue_start() to
317 * prevent I/O from crossing blk_queue_enter().
318 */
319 blk_freeze_queue_start(q);
320
Jens Axboe344e9ff2018-11-15 12:22:51 -0700321 if (queue_is_mq(q))
Jens Axboeaed3ea92014-12-22 14:04:42 -0700322 blk_mq_wake_waiters(q);
Ming Lei055f6e12017-11-09 10:49:53 -0800323
324 /* Make blk_queue_enter() reexamine the DYING flag. */
325 wake_up_all(&q->mq_freeze_wq);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700326}
327EXPORT_SYMBOL_GPL(blk_set_queue_dying);
328
Tejun Heod7325802012-03-05 13:14:58 -0800329/**
Tejun Heoc9a929d2011-10-19 14:42:16 +0200330 * blk_cleanup_queue - shutdown a request queue
331 * @q: request queue to shutdown
332 *
Bart Van Asschec246e802012-12-06 14:32:01 +0100333 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
334 * put it. All future requests will be failed immediately with -ENODEV.
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500335 */
Jens Axboe6728cb02008-01-31 13:03:55 +0100336void blk_cleanup_queue(struct request_queue *q)
Al Viro483f4af2006-03-18 18:34:37 -0500337{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100338 /* mark @q DYING, no new request or merges will be allowed afterwards */
Al Viro483f4af2006-03-18 18:34:37 -0500339 mutex_lock(&q->sysfs_lock);
Jens Axboeaed3ea92014-12-22 14:04:42 -0700340 blk_set_queue_dying(q);
Tejun Heo6ecf23a2012-03-05 13:14:59 -0800341
Christoph Hellwig57d74df2018-11-14 17:02:07 +0100342 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
343 blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
344 blk_queue_flag_set(QUEUE_FLAG_DYING, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200345 mutex_unlock(&q->sysfs_lock);
346
Bart Van Asschec246e802012-12-06 14:32:01 +0100347 /*
348 * Drain all requests queued before DYING marking. Set DEAD flag to
Bart Van Assche67ed8b72019-08-01 15:39:55 -0700349 * prevent that blk_mq_run_hw_queues() accesses the hardware queues
350 * after draining finished.
Bart Van Asschec246e802012-12-06 14:32:01 +0100351 */
Dan Williams3ef28e82015-10-21 13:20:12 -0400352 blk_freeze_queue(q);
Ming Leic57cdf72018-10-24 21:18:09 +0800353
354 rq_qos_exit(q);
355
Christoph Hellwig57d74df2018-11-14 17:02:07 +0100356 blk_queue_flag_set(QUEUE_FLAG_DEAD, q);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200357
Dan Williams5a48fc12015-10-21 13:20:23 -0400358 /* for synchronous bio-based driver finish in-flight integrity i/o */
359 blk_flush_integrity();
360
Tejun Heoc9a929d2011-10-19 14:42:16 +0200361 /* @q won't process any more request, flush async actions */
Jan Karadc3b17c2017-02-02 15:56:50 +0100362 del_timer_sync(&q->backing_dev_info->laptop_mode_wb_timer);
Tejun Heoc9a929d2011-10-19 14:42:16 +0200363 blk_sync_queue(q);
364
Jens Axboe344e9ff2018-11-15 12:22:51 -0700365 if (queue_is_mq(q))
Ming Leic7e2d942019-04-30 09:52:25 +0800366 blk_mq_exit_queue(q);
Jens Axboea1ce35f2018-10-29 10:23:51 -0600367
Ming Leic3e22192019-06-04 21:08:02 +0800368 /*
369 * In theory, request pool of sched_tags belongs to request queue.
370 * However, the current implementation requires tag_set for freeing
371 * requests, so free the pool now.
372 *
373 * Queue has become frozen, there can't be any in-queue requests, so
374 * it is safe to free requests now.
375 */
376 mutex_lock(&q->sysfs_lock);
377 if (q->elevator)
378 blk_mq_sched_free_requests(q);
379 mutex_unlock(&q->sysfs_lock);
380
Dan Williams3ef28e82015-10-21 13:20:12 -0400381 percpu_ref_exit(&q->q_usage_counter);
Bart Van Assche45a9c9d2014-12-09 16:57:48 +0100382
Tejun Heoc9a929d2011-10-19 14:42:16 +0200383 /* @q is and will stay empty, shutdown and put */
Al Viro483f4af2006-03-18 18:34:37 -0500384 blk_put_queue(q);
385}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386EXPORT_SYMBOL(blk_cleanup_queue);
387
Jens Axboe165125e2007-07-24 09:28:11 +0200388struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389{
Christoph Hellwig6d469642018-11-14 17:02:18 +0100390 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
Christoph Lameter19460892005-06-23 00:08:19 -0700391}
392EXPORT_SYMBOL(blk_alloc_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800394/**
395 * blk_queue_enter() - try to increase q->q_usage_counter
396 * @q: request queue pointer
397 * @flags: BLK_MQ_REQ_NOWAIT and/or BLK_MQ_REQ_PREEMPT
398 */
Bart Van Assche9a95e4e2017-11-09 10:49:59 -0800399int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags)
Dan Williams3ef28e82015-10-21 13:20:12 -0400400{
Bart Van Asschecd84a622018-09-26 14:01:04 -0700401 const bool pm = flags & BLK_MQ_REQ_PREEMPT;
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800402
Dan Williams3ef28e82015-10-21 13:20:12 -0400403 while (true) {
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800404 bool success = false;
Dan Williams3ef28e82015-10-21 13:20:12 -0400405
Bart Van Assche818e0fa2018-03-19 11:46:13 -0700406 rcu_read_lock();
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800407 if (percpu_ref_tryget_live(&q->q_usage_counter)) {
408 /*
Bart Van Asschecd84a622018-09-26 14:01:04 -0700409 * The code that increments the pm_only counter is
410 * responsible for ensuring that that counter is
411 * globally visible before the queue is unfrozen.
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800412 */
Bart Van Asschecd84a622018-09-26 14:01:04 -0700413 if (pm || !blk_queue_pm_only(q)) {
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800414 success = true;
415 } else {
416 percpu_ref_put(&q->q_usage_counter);
417 }
418 }
Bart Van Assche818e0fa2018-03-19 11:46:13 -0700419 rcu_read_unlock();
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800420
421 if (success)
Dan Williams3ef28e82015-10-21 13:20:12 -0400422 return 0;
423
Bart Van Assche3a0a5292017-11-09 10:49:58 -0800424 if (flags & BLK_MQ_REQ_NOWAIT)
Dan Williams3ef28e82015-10-21 13:20:12 -0400425 return -EBUSY;
426
Ming Lei5ed61d32017-03-27 20:06:56 +0800427 /*
Ming Lei1671d522017-03-27 20:06:57 +0800428 * read pair of barrier in blk_freeze_queue_start(),
Ming Lei5ed61d32017-03-27 20:06:56 +0800429 * we need to order reading __PERCPU_REF_DEAD flag of
Ming Leid3cfb2a2017-03-27 20:06:58 +0800430 * .q_usage_counter and reading .mq_freeze_depth or
431 * queue dying flag, otherwise the following wait may
432 * never return if the two reads are reordered.
Ming Lei5ed61d32017-03-27 20:06:56 +0800433 */
434 smp_rmb();
435
Alan Jenkins1dc30392018-04-12 19:11:58 +0100436 wait_event(q->mq_freeze_wq,
Bob Liu7996a8b2019-05-21 11:25:55 +0800437 (!q->mq_freeze_depth &&
Bart Van Assche0d25bd02018-09-26 14:01:06 -0700438 (pm || (blk_pm_request_resume(q),
439 !blk_queue_pm_only(q)))) ||
Alan Jenkins1dc30392018-04-12 19:11:58 +0100440 blk_queue_dying(q));
Dan Williams3ef28e82015-10-21 13:20:12 -0400441 if (blk_queue_dying(q))
442 return -ENODEV;
Dan Williams3ef28e82015-10-21 13:20:12 -0400443 }
444}
445
446void blk_queue_exit(struct request_queue *q)
447{
448 percpu_ref_put(&q->q_usage_counter);
449}
450
451static void blk_queue_usage_counter_release(struct percpu_ref *ref)
452{
453 struct request_queue *q =
454 container_of(ref, struct request_queue, q_usage_counter);
455
456 wake_up_all(&q->mq_freeze_wq);
457}
458
Kees Cookbca237a2017-08-28 15:03:41 -0700459static void blk_rq_timed_out_timer(struct timer_list *t)
Christoph Hellwig287922e2015-10-30 20:57:30 +0800460{
Kees Cookbca237a2017-08-28 15:03:41 -0700461 struct request_queue *q = from_timer(q, t, timeout);
Christoph Hellwig287922e2015-10-30 20:57:30 +0800462
463 kblockd_schedule_work(&q->timeout_work);
464}
465
Tetsuo Handa2e3c18d2019-01-30 22:21:45 +0900466static void blk_timeout_work(struct work_struct *work)
467{
468}
469
Bart Van Assche498f6652018-02-28 10:15:32 -0800470/**
471 * blk_alloc_queue_node - allocate a request queue
472 * @gfp_mask: memory allocation flags
473 * @node_id: NUMA node to allocate memory from
Bart Van Assche498f6652018-02-28 10:15:32 -0800474 */
Christoph Hellwig6d469642018-11-14 17:02:18 +0100475struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
Christoph Lameter19460892005-06-23 00:08:19 -0700476{
Jens Axboe165125e2007-07-24 09:28:11 +0200477 struct request_queue *q;
Kent Overstreet338aa962018-05-20 18:25:47 -0400478 int ret;
Christoph Lameter19460892005-06-23 00:08:19 -0700479
Jens Axboe8324aa92008-01-29 14:51:59 +0100480 q = kmem_cache_alloc_node(blk_requestq_cachep,
Christoph Lameter94f60302007-07-17 04:03:29 -0700481 gfp_mask | __GFP_ZERO, node_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (!q)
483 return NULL;
484
Christoph Hellwigcbf62af2018-05-31 19:11:36 +0200485 q->last_merge = NULL;
Christoph Hellwigcbf62af2018-05-31 19:11:36 +0200486
Dan Carpenter00380a42012-03-23 09:58:54 +0100487 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
Tejun Heoa73f7302011-12-14 00:33:37 +0100488 if (q->id < 0)
Ming Lei3d2936f2014-05-27 23:35:14 +0800489 goto fail_q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100490
Kent Overstreet338aa962018-05-20 18:25:47 -0400491 ret = bioset_init(&q->bio_split, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
492 if (ret)
Kent Overstreet54efd502015-04-23 22:37:18 -0700493 goto fail_id;
494
Jan Karad03f6cd2017-02-02 15:56:51 +0100495 q->backing_dev_info = bdi_alloc_node(gfp_mask, node_id);
496 if (!q->backing_dev_info)
Kent Overstreet54efd502015-04-23 22:37:18 -0700497 goto fail_split;
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -0700498
Jens Axboea83b5762017-03-21 17:20:01 -0600499 q->stats = blk_alloc_queue_stats();
500 if (!q->stats)
501 goto fail_stats;
502
Nikolay Borisovb5420232019-03-11 23:28:13 -0700503 q->backing_dev_info->ra_pages = VM_READAHEAD_PAGES;
Jan Karadc3b17c2017-02-02 15:56:50 +0100504 q->backing_dev_info->capabilities = BDI_CAP_CGROUP_WRITEBACK;
505 q->backing_dev_info->name = "block";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 q->node = node_id;
507
Kees Cookbca237a2017-08-28 15:03:41 -0700508 timer_setup(&q->backing_dev_info->laptop_mode_wb_timer,
509 laptop_mode_timer_fn, 0);
510 timer_setup(&q->timeout, blk_rq_timed_out_timer, 0);
Tetsuo Handa2e3c18d2019-01-30 22:21:45 +0900511 INIT_WORK(&q->timeout_work, blk_timeout_work);
Tejun Heoa612fdd2011-12-14 00:33:41 +0100512 INIT_LIST_HEAD(&q->icq_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800513#ifdef CONFIG_BLK_CGROUP
Tejun Heoe8989fa2012-03-05 13:15:20 -0800514 INIT_LIST_HEAD(&q->blkg_list);
Tejun Heo4eef3042012-03-05 13:15:18 -0800515#endif
Al Viro483f4af2006-03-18 18:34:37 -0500516
Jens Axboe8324aa92008-01-29 14:51:59 +0100517 kobject_init(&q->kobj, &blk_queue_ktype);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Waiman Long5acb3cc2017-09-20 13:12:20 -0600519#ifdef CONFIG_BLK_DEV_IO_TRACE
520 mutex_init(&q->blk_trace_mutex);
521#endif
Al Viro483f4af2006-03-18 18:34:37 -0500522 mutex_init(&q->sysfs_lock);
Ming Leicecf5d82019-08-27 19:01:48 +0800523 mutex_init(&q->sysfs_dir_lock);
Christoph Hellwig0d945c12018-11-15 12:17:28 -0700524 spin_lock_init(&q->queue_lock);
Vivek Goyalc94a96a2011-03-02 19:04:42 -0500525
Jens Axboe320ae512013-10-24 09:20:05 +0100526 init_waitqueue_head(&q->mq_freeze_wq);
Bob Liu7996a8b2019-05-21 11:25:55 +0800527 mutex_init(&q->mq_freeze_lock);
Jens Axboe320ae512013-10-24 09:20:05 +0100528
Dan Williams3ef28e82015-10-21 13:20:12 -0400529 /*
530 * Init percpu_ref in atomic mode so that it's faster to shutdown.
531 * See blk_register_queue() for details.
532 */
533 if (percpu_ref_init(&q->q_usage_counter,
534 blk_queue_usage_counter_release,
535 PERCPU_REF_INIT_ATOMIC, GFP_KERNEL))
Mikulas Patockafff49962013-10-14 12:11:36 -0400536 goto fail_bdi;
Tejun Heof51b8022012-03-05 13:15:05 -0800537
Dan Williams3ef28e82015-10-21 13:20:12 -0400538 if (blkcg_init_queue(q))
539 goto fail_ref;
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 return q;
Tejun Heoa73f7302011-12-14 00:33:37 +0100542
Dan Williams3ef28e82015-10-21 13:20:12 -0400543fail_ref:
544 percpu_ref_exit(&q->q_usage_counter);
Mikulas Patockafff49962013-10-14 12:11:36 -0400545fail_bdi:
Jens Axboea83b5762017-03-21 17:20:01 -0600546 blk_free_queue_stats(q->stats);
547fail_stats:
Jan Karad03f6cd2017-02-02 15:56:51 +0100548 bdi_put(q->backing_dev_info);
Kent Overstreet54efd502015-04-23 22:37:18 -0700549fail_split:
Kent Overstreet338aa962018-05-20 18:25:47 -0400550 bioset_exit(&q->bio_split);
Tejun Heoa73f7302011-12-14 00:33:37 +0100551fail_id:
552 ida_simple_remove(&blk_queue_ida, q->id);
553fail_q:
554 kmem_cache_free(blk_requestq_cachep, q);
555 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
Christoph Lameter19460892005-06-23 00:08:19 -0700557EXPORT_SYMBOL(blk_alloc_queue_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
Tejun Heo09ac46c2011-12-14 00:33:38 +0100559bool blk_get_queue(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100561 if (likely(!blk_queue_dying(q))) {
Tejun Heo09ac46c2011-12-14 00:33:38 +0100562 __blk_get_queue(q);
563 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 }
565
Tejun Heo09ac46c2011-12-14 00:33:38 +0100566 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
Jens Axboed86e0e82011-05-27 07:44:43 +0200568EXPORT_SYMBOL(blk_get_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Bart Van Assche6a156742017-11-09 10:49:54 -0800570/**
Christoph Hellwigff005a02018-05-09 09:54:05 +0200571 * blk_get_request - allocate a request
Bart Van Assche6a156742017-11-09 10:49:54 -0800572 * @q: request queue to allocate a request for
573 * @op: operation (REQ_OP_*) and REQ_* flags, e.g. REQ_SYNC.
574 * @flags: BLK_MQ_REQ_* flags, e.g. BLK_MQ_REQ_NOWAIT.
575 */
Christoph Hellwigff005a02018-05-09 09:54:05 +0200576struct request *blk_get_request(struct request_queue *q, unsigned int op,
577 blk_mq_req_flags_t flags)
Jens Axboe320ae512013-10-24 09:20:05 +0100578{
Bart Van Assched280bab2017-06-20 11:15:40 -0700579 struct request *req;
580
Bart Van Assche6a156742017-11-09 10:49:54 -0800581 WARN_ON_ONCE(op & REQ_NOWAIT);
Bart Van Assche1b6d65a2017-11-09 10:49:55 -0800582 WARN_ON_ONCE(flags & ~(BLK_MQ_REQ_NOWAIT | BLK_MQ_REQ_PREEMPT));
Bart Van Assche6a156742017-11-09 10:49:54 -0800583
Jens Axboea1ce35f2018-10-29 10:23:51 -0600584 req = blk_mq_alloc_request(q, op, flags);
585 if (!IS_ERR(req) && q->mq_ops->initialize_rq_fn)
586 q->mq_ops->initialize_rq_fn(req);
Bart Van Assched280bab2017-06-20 11:15:40 -0700587
588 return req;
Jens Axboe320ae512013-10-24 09:20:05 +0100589}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590EXPORT_SYMBOL(blk_get_request);
591
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592void blk_put_request(struct request *req)
593{
Jens Axboea1ce35f2018-10-29 10:23:51 -0600594 blk_mq_free_request(req);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596EXPORT_SYMBOL(blk_put_request);
597
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200598bool bio_attempt_back_merge(struct request *req, struct bio *bio,
599 unsigned int nr_segs)
Jens Axboe73c10102011-03-08 13:19:51 +0100600{
Jens Axboe1eff9d32016-08-05 15:35:16 -0600601 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +0100602
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200603 if (!ll_back_merge_fn(req, bio, nr_segs))
Jens Axboe73c10102011-03-08 13:19:51 +0100604 return false;
605
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200606 trace_block_bio_backmerge(req->q, req, bio);
Tejun Heod3e65ff2019-08-28 15:05:54 -0700607 rq_qos_merge(req->q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +0100608
609 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
610 blk_rq_set_mixed_merge(req);
611
612 req->biotail->bi_next = bio;
613 req->biotail = bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700614 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +0100615
Jens Axboe320ae512013-10-24 09:20:05 +0100616 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +0100617 return true;
618}
619
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200620bool bio_attempt_front_merge(struct request *req, struct bio *bio,
621 unsigned int nr_segs)
Jens Axboe73c10102011-03-08 13:19:51 +0100622{
Jens Axboe1eff9d32016-08-05 15:35:16 -0600623 const int ff = bio->bi_opf & REQ_FAILFAST_MASK;
Jens Axboe73c10102011-03-08 13:19:51 +0100624
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200625 if (!ll_front_merge_fn(req, bio, nr_segs))
Jens Axboe73c10102011-03-08 13:19:51 +0100626 return false;
627
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200628 trace_block_bio_frontmerge(req->q, req, bio);
Tejun Heod3e65ff2019-08-28 15:05:54 -0700629 rq_qos_merge(req->q, req, bio);
Jens Axboe73c10102011-03-08 13:19:51 +0100630
631 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
632 blk_rq_set_mixed_merge(req);
633
Jens Axboe73c10102011-03-08 13:19:51 +0100634 bio->bi_next = req->bio;
635 req->bio = bio;
636
Kent Overstreet4f024f32013-10-11 15:44:27 -0700637 req->__sector = bio->bi_iter.bi_sector;
638 req->__data_len += bio->bi_iter.bi_size;
Jens Axboe73c10102011-03-08 13:19:51 +0100639
Jens Axboe320ae512013-10-24 09:20:05 +0100640 blk_account_io_start(req, false);
Jens Axboe73c10102011-03-08 13:19:51 +0100641 return true;
642}
643
Christoph Hellwig1e739732017-02-08 14:46:49 +0100644bool bio_attempt_discard_merge(struct request_queue *q, struct request *req,
645 struct bio *bio)
646{
647 unsigned short segments = blk_rq_nr_discard_segments(req);
648
649 if (segments >= queue_max_discard_segments(q))
650 goto no_merge;
651 if (blk_rq_sectors(req) + bio_sectors(bio) >
652 blk_rq_get_max_sectors(req, blk_rq_pos(req)))
653 goto no_merge;
654
Tejun Heod3e65ff2019-08-28 15:05:54 -0700655 rq_qos_merge(q, req, bio);
656
Christoph Hellwig1e739732017-02-08 14:46:49 +0100657 req->biotail->bi_next = bio;
658 req->biotail = bio;
659 req->__data_len += bio->bi_iter.bi_size;
Christoph Hellwig1e739732017-02-08 14:46:49 +0100660 req->nr_phys_segments = segments + 1;
661
662 blk_account_io_start(req, false);
663 return true;
664no_merge:
665 req_set_nomerge(q, req);
666 return false;
667}
668
Tejun Heobd87b582011-10-19 14:33:08 +0200669/**
Jens Axboe320ae512013-10-24 09:20:05 +0100670 * blk_attempt_plug_merge - try to merge with %current's plugged list
Tejun Heobd87b582011-10-19 14:33:08 +0200671 * @q: request_queue new bio is being queued at
672 * @bio: new bio being queued
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200673 * @nr_segs: number of segments in @bio
Randy Dunlapccc26002015-10-30 18:36:16 -0700674 * @same_queue_rq: pointer to &struct request that gets filled in when
675 * another request associated with @q is found on the plug list
676 * (optional, may be %NULL)
Tejun Heobd87b582011-10-19 14:33:08 +0200677 *
678 * Determine whether @bio being queued on @q can be merged with a request
679 * on %current's plugged list. Returns %true if merge was successful,
680 * otherwise %false.
681 *
Tejun Heo07c2bd32012-02-08 09:19:42 +0100682 * Plugging coalesces IOs from the same issuer for the same purpose without
683 * going through @q->queue_lock. As such it's more of an issuing mechanism
684 * than scheduling, and the request, while may have elvpriv data, is not
685 * added on the elevator at this point. In addition, we don't have
686 * reliable access to the elevator outside queue lock. Only check basic
687 * merging parameters without querying the elevator.
Robert Elliottda41a582014-05-20 16:46:26 -0500688 *
689 * Caller must ensure !blk_queue_nomerges(q) beforehand.
Jens Axboe73c10102011-03-08 13:19:51 +0100690 */
Jens Axboe320ae512013-10-24 09:20:05 +0100691bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200692 unsigned int nr_segs, struct request **same_queue_rq)
Jens Axboe73c10102011-03-08 13:19:51 +0100693{
694 struct blk_plug *plug;
695 struct request *rq;
Shaohua Li92f399c2013-10-29 12:01:03 -0600696 struct list_head *plug_list;
Jens Axboe73c10102011-03-08 13:19:51 +0100697
Damien Le Moalb49773e72019-07-11 01:18:31 +0900698 plug = blk_mq_plug(q, bio);
Jens Axboe73c10102011-03-08 13:19:51 +0100699 if (!plug)
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100700 return false;
Jens Axboe73c10102011-03-08 13:19:51 +0100701
Jens Axboea1ce35f2018-10-29 10:23:51 -0600702 plug_list = &plug->mq_list;
Shaohua Li92f399c2013-10-29 12:01:03 -0600703
704 list_for_each_entry_reverse(rq, plug_list, queuelist) {
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100705 bool merged = false;
Jens Axboe73c10102011-03-08 13:19:51 +0100706
Jens Axboe5f0ed772018-11-23 22:04:33 -0700707 if (rq->q == q && same_queue_rq) {
Shaohua Li5b3f3412015-05-08 10:51:33 -0700708 /*
709 * Only blk-mq multiple hardware queues case checks the
710 * rq in the same queue, there should be only one such
711 * rq in a queue
712 **/
Jens Axboe5f0ed772018-11-23 22:04:33 -0700713 *same_queue_rq = rq;
Shaohua Li5b3f3412015-05-08 10:51:33 -0700714 }
Shaohua Li56ebdaf2011-08-24 16:04:34 +0200715
Tejun Heo07c2bd32012-02-08 09:19:42 +0100716 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
Jens Axboe73c10102011-03-08 13:19:51 +0100717 continue;
718
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100719 switch (blk_try_merge(rq, bio)) {
720 case ELEVATOR_BACK_MERGE:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200721 merged = bio_attempt_back_merge(rq, bio, nr_segs);
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100722 break;
723 case ELEVATOR_FRONT_MERGE:
Christoph Hellwig14ccb662019-06-06 12:29:01 +0200724 merged = bio_attempt_front_merge(rq, bio, nr_segs);
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100725 break;
Christoph Hellwig1e739732017-02-08 14:46:49 +0100726 case ELEVATOR_DISCARD_MERGE:
727 merged = bio_attempt_discard_merge(q, rq, bio);
728 break;
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100729 default:
730 break;
Jens Axboe73c10102011-03-08 13:19:51 +0100731 }
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100732
733 if (merged)
734 return true;
Jens Axboe73c10102011-03-08 13:19:51 +0100735 }
Christoph Hellwig34fe7c02017-02-08 14:46:48 +0100736
737 return false;
Jens Axboe73c10102011-03-08 13:19:51 +0100738}
739
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100740static void handle_bad_sector(struct bio *bio, sector_t maxsector)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
742 char b[BDEVNAME_SIZE];
743
744 printk(KERN_INFO "attempt to access beyond end of device\n");
Mike Christie6296b962016-06-05 14:32:21 -0500745 printk(KERN_INFO "%s: rw=%d, want=%Lu, limit=%Lu\n",
Christoph Hellwig74d46992017-08-23 19:10:32 +0200746 bio_devname(bio, b), bio->bi_opf,
Kent Overstreetf73a1c72012-09-25 15:05:12 -0700747 (unsigned long long)bio_end_sector(bio),
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100748 (long long)maxsector);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
750
Akinobu Mitac17bb492006-12-08 02:39:46 -0800751#ifdef CONFIG_FAIL_MAKE_REQUEST
752
753static DECLARE_FAULT_ATTR(fail_make_request);
754
755static int __init setup_fail_make_request(char *str)
756{
757 return setup_fault_attr(&fail_make_request, str);
758}
759__setup("fail_make_request=", setup_fail_make_request);
760
Akinobu Mitab2c9cd32011-07-26 16:09:03 -0700761static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -0800762{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -0700763 return part->make_it_fail && should_fail(&fail_make_request, bytes);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800764}
765
766static int __init fail_make_request_debugfs(void)
767{
Akinobu Mitadd48c082011-08-03 16:21:01 -0700768 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
769 NULL, &fail_make_request);
770
Duan Jiong21f9fcd2014-04-11 15:58:56 +0800771 return PTR_ERR_OR_ZERO(dir);
Akinobu Mitac17bb492006-12-08 02:39:46 -0800772}
773
774late_initcall(fail_make_request_debugfs);
775
776#else /* CONFIG_FAIL_MAKE_REQUEST */
777
Akinobu Mitab2c9cd32011-07-26 16:09:03 -0700778static inline bool should_fail_request(struct hd_struct *part,
779 unsigned int bytes)
Akinobu Mitac17bb492006-12-08 02:39:46 -0800780{
Akinobu Mitab2c9cd32011-07-26 16:09:03 -0700781 return false;
Akinobu Mitac17bb492006-12-08 02:39:46 -0800782}
783
784#endif /* CONFIG_FAIL_MAKE_REQUEST */
785
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100786static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
787{
Jens Axboeb089cfd2018-08-14 10:52:40 -0600788 const int op = bio_op(bio);
789
Mikulas Patocka8b2ded12018-09-05 16:14:36 -0600790 if (part->policy && op_is_write(op)) {
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100791 char b[BDEVNAME_SIZE];
792
Mikulas Patocka8b2ded12018-09-05 16:14:36 -0600793 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
794 return false;
795
Linus Torvaldsa32e2362018-08-03 12:22:09 -0700796 WARN_ONCE(1,
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100797 "generic_make_request: Trying to write "
798 "to read-only block-device %s (partno %d)\n",
799 bio_devname(bio, b), part->partno);
Linus Torvaldsa32e2362018-08-03 12:22:09 -0700800 /* Older lvm-tools actually trigger this */
801 return false;
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100802 }
803
804 return false;
805}
806
Howard McLauchlan30abb3a2018-02-06 14:05:39 -0800807static noinline int should_fail_bio(struct bio *bio)
808{
809 if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
810 return -EIO;
811 return 0;
812}
813ALLOW_ERROR_INJECTION(should_fail_bio, ERRNO);
814
Jens Axboec07e2b42007-07-18 13:27:58 +0200815/*
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100816 * Check whether this bio extends beyond the end of the device or partition.
817 * This may well happen - the kernel calls bread() without checking the size of
818 * the device, e.g., when mounting a file system.
819 */
820static inline int bio_check_eod(struct bio *bio, sector_t maxsector)
821{
822 unsigned int nr_sectors = bio_sectors(bio);
823
824 if (nr_sectors && maxsector &&
825 (nr_sectors > maxsector ||
826 bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
827 handle_bad_sector(bio, maxsector);
828 return -EIO;
829 }
830 return 0;
831}
832
833/*
Christoph Hellwig74d46992017-08-23 19:10:32 +0200834 * Remap block n of partition p to block n+start(p) of the disk.
835 */
836static inline int blk_partition_remap(struct bio *bio)
837{
838 struct hd_struct *p;
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100839 int ret = -EIO;
Christoph Hellwig74d46992017-08-23 19:10:32 +0200840
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100841 rcu_read_lock();
842 p = __disk_get_part(bio->bi_disk, bio->bi_partno);
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100843 if (unlikely(!p))
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100844 goto out;
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100845 if (unlikely(should_fail_request(p, bio->bi_iter.bi_size)))
846 goto out;
847 if (unlikely(bio_check_ro(bio, p)))
848 goto out;
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100849
Christoph Hellwig74d46992017-08-23 19:10:32 +0200850 /*
851 * Zone reset does not include bi_size so bio_sectors() is always 0.
852 * Include a test for the reset op code and perform the remap if needed.
853 */
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100854 if (bio_sectors(bio) || bio_op(bio) == REQ_OP_ZONE_RESET) {
855 if (bio_check_eod(bio, part_nr_sects_read(p)))
856 goto out;
857 bio->bi_iter.bi_sector += p->start_sect;
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100858 trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
859 bio->bi_iter.bi_sector - p->start_sect);
860 }
Hannes Reineckec04fa442018-06-07 10:29:44 +0200861 bio->bi_partno = 0;
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100862 ret = 0;
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100863out:
Christoph Hellwig74d46992017-08-23 19:10:32 +0200864 rcu_read_unlock();
Christoph Hellwig74d46992017-08-23 19:10:32 +0200865 return ret;
866}
867
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200868static noinline_for_stack bool
869generic_make_request_checks(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
Jens Axboe165125e2007-07-24 09:28:11 +0200871 struct request_queue *q;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200872 int nr_sectors = bio_sectors(bio);
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200873 blk_status_t status = BLK_STS_IOERR;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200874 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
876 might_sleep();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Christoph Hellwig74d46992017-08-23 19:10:32 +0200878 q = bio->bi_disk->queue;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200879 if (unlikely(!q)) {
880 printk(KERN_ERR
881 "generic_make_request: Trying to access "
882 "nonexistent block-device %s (%Lu)\n",
Christoph Hellwig74d46992017-08-23 19:10:32 +0200883 bio_devname(bio, b), (long long)bio->bi_iter.bi_sector);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200884 goto end_io;
885 }
886
Goldwyn Rodrigues03a07c92017-06-20 07:05:46 -0500887 /*
888 * For a REQ_NOWAIT based request, return -EOPNOTSUPP
889 * if queue is not a request based queue.
890 */
Jens Axboe344e9ff2018-11-15 12:22:51 -0700891 if ((bio->bi_opf & REQ_NOWAIT) && !queue_is_mq(q))
Goldwyn Rodrigues03a07c92017-06-20 07:05:46 -0500892 goto not_supported;
893
Howard McLauchlan30abb3a2018-02-06 14:05:39 -0800894 if (should_fail_bio(bio))
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200895 goto end_io;
896
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100897 if (bio->bi_partno) {
898 if (unlikely(blk_partition_remap(bio)))
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100899 goto end_io;
900 } else {
Christoph Hellwig52c5e622018-03-14 16:56:53 +0100901 if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
902 goto end_io;
903 if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
Ilya Dryomov721c7fc2018-01-11 14:09:11 +0100904 goto end_io;
905 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200907 /*
908 * Filter flush bio's early so that make_request based
909 * drivers without flush support don't have to worry
910 * about them.
911 */
Jens Axboef3a8ab72017-01-27 09:08:23 -0700912 if (op_is_flush(bio->bi_opf) &&
Jens Axboec888a8f2016-04-13 13:33:19 -0600913 !test_bit(QUEUE_FLAG_WC, &q->queue_flags)) {
Jens Axboe1eff9d32016-08-05 15:35:16 -0600914 bio->bi_opf &= ~(REQ_PREFLUSH | REQ_FUA);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200915 if (!nr_sectors) {
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200916 status = BLK_STS_OK;
Tejun Heoa7384672008-11-28 13:32:03 +0900917 goto end_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 }
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200919 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
Christoph Hellwigd04c4062018-12-14 17:21:22 +0100921 if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
922 bio->bi_opf &= ~REQ_HIPRI;
923
Christoph Hellwig288dab82016-06-09 16:00:36 +0200924 switch (bio_op(bio)) {
925 case REQ_OP_DISCARD:
926 if (!blk_queue_discard(q))
927 goto not_supported;
928 break;
929 case REQ_OP_SECURE_ERASE:
930 if (!blk_queue_secure_erase(q))
931 goto not_supported;
932 break;
933 case REQ_OP_WRITE_SAME:
Christoph Hellwig74d46992017-08-23 19:10:32 +0200934 if (!q->limits.max_write_same_sectors)
Christoph Hellwig288dab82016-06-09 16:00:36 +0200935 goto not_supported;
Nicolai Stange58886782016-12-04 14:56:39 +0100936 break;
Shaun Tancheff2d253442016-10-18 15:40:32 +0900937 case REQ_OP_ZONE_RESET:
Christoph Hellwig74d46992017-08-23 19:10:32 +0200938 if (!blk_queue_is_zoned(q))
Shaun Tancheff2d253442016-10-18 15:40:32 +0900939 goto not_supported;
Christoph Hellwig288dab82016-06-09 16:00:36 +0200940 break;
Chaitanya Kulkarni6e33dbf2019-08-01 10:26:36 -0700941 case REQ_OP_ZONE_RESET_ALL:
942 if (!blk_queue_is_zoned(q) || !blk_queue_zone_resetall(q))
943 goto not_supported;
944 break;
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800945 case REQ_OP_WRITE_ZEROES:
Christoph Hellwig74d46992017-08-23 19:10:32 +0200946 if (!q->limits.max_write_zeroes_sectors)
Chaitanya Kulkarnia6f07882016-11-30 12:28:59 -0800947 goto not_supported;
948 break;
Christoph Hellwig288dab82016-06-09 16:00:36 +0200949 default:
950 break;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200951 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952
Tejun Heo7f4b35d2012-06-04 20:40:56 -0700953 /*
954 * Various block parts want %current->io_context and lazy ioc
955 * allocation ends up trading a lot of pain for a small amount of
956 * memory. Just allocate it upfront. This may fail and block
957 * layer knows how to live with it.
958 */
959 create_io_context(GFP_ATOMIC, q->node);
960
Tejun Heoae118892015-08-18 14:55:20 -0700961 if (!blkcg_bio_issue_check(q, bio))
962 return false;
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200963
NeilBrownfbbaf702017-04-07 09:40:52 -0600964 if (!bio_flagged(bio, BIO_TRACE_COMPLETION)) {
965 trace_block_bio_queue(q, bio);
966 /* Now that enqueuing has been traced, we need to trace
967 * completion as well.
968 */
969 bio_set_flag(bio, BIO_TRACE_COMPLETION);
970 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200971 return true;
Tejun Heoa7384672008-11-28 13:32:03 +0900972
Christoph Hellwig288dab82016-06-09 16:00:36 +0200973not_supported:
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200974 status = BLK_STS_NOTSUPP;
Tejun Heoa7384672008-11-28 13:32:03 +0900975end_io:
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200976 bio->bi_status = status;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200977 bio_endio(bio);
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200978 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979}
980
Christoph Hellwig27a84d52011-09-15 14:01:40 +0200981/**
982 * generic_make_request - hand a buffer to its device driver for I/O
983 * @bio: The bio describing the location in memory and on the device.
984 *
985 * generic_make_request() is used to make I/O requests of block
986 * devices. It is passed a &struct bio, which describes the I/O that needs
987 * to be done.
988 *
989 * generic_make_request() does not return any status. The
990 * success/failure status of the request, along with notification of
991 * completion, is delivered asynchronously through the bio->bi_end_io
992 * function described (one day) else where.
993 *
994 * The caller of generic_make_request must make sure that bi_io_vec
995 * are set to describe the memory buffer, and that bi_dev and bi_sector are
996 * set to describe the device address, and the
997 * bi_end_io and optionally bi_private are set to describe how
998 * completion notification should be signaled.
999 *
1000 * generic_make_request and the drivers it calls may use bi_next if this
1001 * bio happens to be merged with someone else, and may resubmit the bio to
1002 * a lower device by calling into generic_make_request recursively, which
1003 * means the bio should NOT be touched after the call to ->make_request_fn.
Neil Brownd89d8792007-05-01 09:53:42 +02001004 */
Jens Axboedece1632015-11-05 10:41:16 -07001005blk_qc_t generic_make_request(struct bio *bio)
Neil Brownd89d8792007-05-01 09:53:42 +02001006{
NeilBrownf5fe1b52017-03-10 17:00:47 +11001007 /*
1008 * bio_list_on_stack[0] contains bios submitted by the current
1009 * make_request_fn.
1010 * bio_list_on_stack[1] contains bios that were submitted before
1011 * the current make_request_fn, but that haven't been processed
1012 * yet.
1013 */
1014 struct bio_list bio_list_on_stack[2];
Jens Axboedece1632015-11-05 10:41:16 -07001015 blk_qc_t ret = BLK_QC_T_NONE;
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001016
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001017 if (!generic_make_request_checks(bio))
Jens Axboedece1632015-11-05 10:41:16 -07001018 goto out;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001019
1020 /*
1021 * We only want one ->make_request_fn to be active at a time, else
1022 * stack usage with stacked devices could be a problem. So use
1023 * current->bio_list to keep a list of requests submited by a
1024 * make_request_fn function. current->bio_list is also used as a
1025 * flag to say if generic_make_request is currently active in this
1026 * task or not. If it is NULL, then no make_request is active. If
1027 * it is non-NULL, then a make_request is active, and new requests
1028 * should be added at the tail
1029 */
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001030 if (current->bio_list) {
NeilBrownf5fe1b52017-03-10 17:00:47 +11001031 bio_list_add(&current->bio_list[0], bio);
Jens Axboedece1632015-11-05 10:41:16 -07001032 goto out;
Neil Brownd89d8792007-05-01 09:53:42 +02001033 }
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001034
Neil Brownd89d8792007-05-01 09:53:42 +02001035 /* following loop may be a bit non-obvious, and so deserves some
1036 * explanation.
1037 * Before entering the loop, bio->bi_next is NULL (as all callers
1038 * ensure that) so we have a list with a single bio.
1039 * We pretend that we have just taken it off a longer list, so
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001040 * we assign bio_list to a pointer to the bio_list_on_stack,
1041 * thus initialising the bio_list of new bios to be
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001042 * added. ->make_request() may indeed add some more bios
Neil Brownd89d8792007-05-01 09:53:42 +02001043 * through a recursive call to generic_make_request. If it
1044 * did, we find a non-NULL value in bio_list and re-enter the loop
1045 * from the top. In this case we really did just take the bio
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001046 * of the top of the list (no pretending) and so remove it from
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001047 * bio_list, and call into ->make_request() again.
Neil Brownd89d8792007-05-01 09:53:42 +02001048 */
1049 BUG_ON(bio->bi_next);
NeilBrownf5fe1b52017-03-10 17:00:47 +11001050 bio_list_init(&bio_list_on_stack[0]);
1051 current->bio_list = bio_list_on_stack;
Neil Brownd89d8792007-05-01 09:53:42 +02001052 do {
Ming Leife200862019-05-15 11:03:09 +08001053 struct request_queue *q = bio->bi_disk->queue;
1054 blk_mq_req_flags_t flags = bio->bi_opf & REQ_NOWAIT ?
1055 BLK_MQ_REQ_NOWAIT : 0;
Christoph Hellwig27a84d52011-09-15 14:01:40 +02001056
Ming Leife200862019-05-15 11:03:09 +08001057 if (likely(blk_queue_enter(q, flags) == 0)) {
NeilBrown79bd9952017-03-08 07:38:05 +11001058 struct bio_list lower, same;
1059
1060 /* Create a fresh bio_list for all subordinate requests */
NeilBrownf5fe1b52017-03-10 17:00:47 +11001061 bio_list_on_stack[1] = bio_list_on_stack[0];
1062 bio_list_init(&bio_list_on_stack[0]);
Jens Axboedece1632015-11-05 10:41:16 -07001063 ret = q->make_request_fn(q, bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04001064
Ming Leife200862019-05-15 11:03:09 +08001065 blk_queue_exit(q);
1066
NeilBrown79bd9952017-03-08 07:38:05 +11001067 /* sort new bios into those for a lower level
1068 * and those for the same level
1069 */
1070 bio_list_init(&lower);
1071 bio_list_init(&same);
NeilBrownf5fe1b52017-03-10 17:00:47 +11001072 while ((bio = bio_list_pop(&bio_list_on_stack[0])) != NULL)
Christoph Hellwig74d46992017-08-23 19:10:32 +02001073 if (q == bio->bi_disk->queue)
NeilBrown79bd9952017-03-08 07:38:05 +11001074 bio_list_add(&same, bio);
1075 else
1076 bio_list_add(&lower, bio);
1077 /* now assemble so we handle the lowest level first */
NeilBrownf5fe1b52017-03-10 17:00:47 +11001078 bio_list_merge(&bio_list_on_stack[0], &lower);
1079 bio_list_merge(&bio_list_on_stack[0], &same);
1080 bio_list_merge(&bio_list_on_stack[0], &bio_list_on_stack[1]);
Dan Williams3ef28e82015-10-21 13:20:12 -04001081 } else {
Goldwyn Rodrigues03a07c92017-06-20 07:05:46 -05001082 if (unlikely(!blk_queue_dying(q) &&
1083 (bio->bi_opf & REQ_NOWAIT)))
1084 bio_wouldblock_error(bio);
1085 else
1086 bio_io_error(bio);
Dan Williams3ef28e82015-10-21 13:20:12 -04001087 }
NeilBrownf5fe1b52017-03-10 17:00:47 +11001088 bio = bio_list_pop(&bio_list_on_stack[0]);
Neil Brownd89d8792007-05-01 09:53:42 +02001089 } while (bio);
Akinobu Mitabddd87c2010-02-23 08:55:42 +01001090 current->bio_list = NULL; /* deactivate */
Jens Axboedece1632015-11-05 10:41:16 -07001091
1092out:
1093 return ret;
Neil Brownd89d8792007-05-01 09:53:42 +02001094}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095EXPORT_SYMBOL(generic_make_request);
1096
1097/**
Christoph Hellwigf421e1d92017-11-02 21:29:50 +03001098 * direct_make_request - hand a buffer directly to its device driver for I/O
1099 * @bio: The bio describing the location in memory and on the device.
1100 *
1101 * This function behaves like generic_make_request(), but does not protect
1102 * against recursion. Must only be used if the called driver is known
1103 * to not call generic_make_request (or direct_make_request) again from
1104 * its make_request function. (Calling direct_make_request again from
1105 * a workqueue is perfectly fine as that doesn't recurse).
1106 */
1107blk_qc_t direct_make_request(struct bio *bio)
1108{
1109 struct request_queue *q = bio->bi_disk->queue;
1110 bool nowait = bio->bi_opf & REQ_NOWAIT;
1111 blk_qc_t ret;
1112
1113 if (!generic_make_request_checks(bio))
1114 return BLK_QC_T_NONE;
1115
Bart Van Assche3a0a5292017-11-09 10:49:58 -08001116 if (unlikely(blk_queue_enter(q, nowait ? BLK_MQ_REQ_NOWAIT : 0))) {
Christoph Hellwigf421e1d92017-11-02 21:29:50 +03001117 if (nowait && !blk_queue_dying(q))
1118 bio->bi_status = BLK_STS_AGAIN;
1119 else
1120 bio->bi_status = BLK_STS_IOERR;
1121 bio_endio(bio);
1122 return BLK_QC_T_NONE;
1123 }
1124
1125 ret = q->make_request_fn(q, bio);
1126 blk_queue_exit(q);
1127 return ret;
1128}
1129EXPORT_SYMBOL_GPL(direct_make_request);
1130
1131/**
Randy Dunlap710027a2008-08-19 20:13:11 +02001132 * submit_bio - submit a bio to the block device layer for I/O
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 * @bio: The &struct bio which describes the I/O
1134 *
1135 * submit_bio() is very similar in purpose to generic_make_request(), and
1136 * uses that function to do most of the work. Both are fairly rough
Randy Dunlap710027a2008-08-19 20:13:11 +02001137 * interfaces; @bio must be presetup and ready for I/O.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 *
1139 */
Mike Christie4e49ea42016-06-05 14:31:41 -05001140blk_qc_t submit_bio(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141{
Johannes Weinerb8e24a92019-08-08 15:03:00 -04001142 bool workingset_read = false;
1143 unsigned long pflags;
1144 blk_qc_t ret;
1145
Tejun Heod3f77df2019-06-27 13:39:52 -07001146 if (blkcg_punt_bio_submit(bio))
1147 return BLK_QC_T_NONE;
1148
Jens Axboebf2de6f2007-09-27 13:01:25 +02001149 /*
1150 * If it's a regular read/write or a barrier with data attached,
1151 * go through the normal accounting stuff before submission.
1152 */
Martin K. Petersene2a60da2012-09-18 12:19:25 -04001153 if (bio_has_data(bio)) {
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001154 unsigned int count;
1155
Mike Christie95fe6c12016-06-05 14:31:48 -05001156 if (unlikely(bio_op(bio) == REQ_OP_WRITE_SAME))
Jiufei Xue7c5a0dc2018-02-27 20:10:03 +08001157 count = queue_logical_block_size(bio->bi_disk->queue) >> 9;
Martin K. Petersen4363ac72012-09-18 12:19:27 -04001158 else
1159 count = bio_sectors(bio);
1160
Mike Christiea8ebb052016-06-05 14:31:45 -05001161 if (op_is_write(bio_op(bio))) {
Jens Axboebf2de6f2007-09-27 13:01:25 +02001162 count_vm_events(PGPGOUT, count);
1163 } else {
Johannes Weinerb8e24a92019-08-08 15:03:00 -04001164 if (bio_flagged(bio, BIO_WORKINGSET))
1165 workingset_read = true;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001166 task_io_account_read(bio->bi_iter.bi_size);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001167 count_vm_events(PGPGIN, count);
1168 }
1169
1170 if (unlikely(block_dump)) {
1171 char b[BDEVNAME_SIZE];
San Mehat8dcbdc72010-09-14 08:48:01 +02001172 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -07001173 current->comm, task_pid_nr(current),
Mike Christiea8ebb052016-06-05 14:31:45 -05001174 op_is_write(bio_op(bio)) ? "WRITE" : "READ",
Kent Overstreet4f024f32013-10-11 15:44:27 -07001175 (unsigned long long)bio->bi_iter.bi_sector,
Christoph Hellwig74d46992017-08-23 19:10:32 +02001176 bio_devname(bio, b), count);
Jens Axboebf2de6f2007-09-27 13:01:25 +02001177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 }
1179
Johannes Weinerb8e24a92019-08-08 15:03:00 -04001180 /*
1181 * If we're reading data that is part of the userspace
1182 * workingset, count submission time as memory stall. When the
1183 * device is congested, or the submitting cgroup IO-throttled,
1184 * submission can be a significant part of overall IO time.
1185 */
1186 if (workingset_read)
1187 psi_memstall_enter(&pflags);
1188
1189 ret = generic_make_request(bio);
1190
1191 if (workingset_read)
1192 psi_memstall_leave(&pflags);
1193
1194 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196EXPORT_SYMBOL(submit_bio);
1197
Jens Axboe1052b8a2018-11-26 08:21:49 -07001198/**
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01001199 * blk_cloned_rq_check_limits - Helper function to check a cloned request
1200 * for new the queue limits
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001201 * @q: the queue
1202 * @rq: the request being checked
1203 *
1204 * Description:
1205 * @rq may have been made based on weaker limitations of upper-level queues
1206 * in request stacking drivers, and it may violate the limitation of @q.
1207 * Since the block layer and the underlying device driver trust @rq
1208 * after it is inserted to @q, it should be checked against @q before
1209 * the insertion using this generic function.
1210 *
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001211 * Request stacking drivers like request-based dm may change the queue
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01001212 * limits when retrying requests on other queues. Those requests need
1213 * to be checked against the new queue limits again during dispatch.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001214 */
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01001215static int blk_cloned_rq_check_limits(struct request_queue *q,
1216 struct request *rq)
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001217{
Mike Christie8fe0d472016-06-05 14:32:15 -05001218 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, req_op(rq))) {
John Pittman61939b12019-05-23 17:49:39 -04001219 printk(KERN_ERR "%s: over max size limit. (%u > %u)\n",
1220 __func__, blk_rq_sectors(rq),
1221 blk_queue_get_max_sectors(q, req_op(rq)));
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001222 return -EIO;
1223 }
1224
1225 /*
1226 * queue's settings related to segment counting like q->bounce_pfn
1227 * may differ from that of other stacking queues.
1228 * Recalculate it to check the request correctly on this queue's
1229 * limitation.
1230 */
Christoph Hellwige9cd19c2019-06-06 12:29:02 +02001231 rq->nr_phys_segments = blk_recalc_rq_segments(rq);
Martin K. Petersen8a783622010-02-26 00:20:39 -05001232 if (rq->nr_phys_segments > queue_max_segments(q)) {
John Pittman61939b12019-05-23 17:49:39 -04001233 printk(KERN_ERR "%s: over max segments limit. (%hu > %hu)\n",
1234 __func__, rq->nr_phys_segments, queue_max_segments(q));
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001235 return -EIO;
1236 }
1237
1238 return 0;
1239}
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001240
1241/**
1242 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1243 * @q: the queue to submit the request
1244 * @rq: the request being queued
1245 */
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001246blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *rq)
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001247{
Hannes Reineckebf4e6b42015-11-26 08:46:57 +01001248 if (blk_cloned_rq_check_limits(q, rq))
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001249 return BLK_STS_IOERR;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001250
Akinobu Mitab2c9cd32011-07-26 16:09:03 -07001251 if (rq->rq_disk &&
1252 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001253 return BLK_STS_IOERR;
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001254
Jens Axboea1ce35f2018-10-29 10:23:51 -06001255 if (blk_queue_io_stat(q))
1256 blk_account_io_start(rq, true);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001257
1258 /*
Jens Axboea1ce35f2018-10-29 10:23:51 -06001259 * Since we have a scheduler attached on the top device,
1260 * bypass a potential scheduler on the bottom device for
1261 * insert.
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001262 */
Bart Van Asschefd9c40f2019-04-04 10:08:43 -07001263 return blk_mq_request_issue_directly(rq, true);
Kiyoshi Ueda82124d62008-09-18 10:45:38 -04001264}
1265EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1266
Tejun Heo80a761f2009-07-03 17:48:17 +09001267/**
1268 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1269 * @rq: request to examine
1270 *
1271 * Description:
1272 * A request could be merge of IOs which require different failure
1273 * handling. This function determines the number of bytes which
1274 * can be failed from the beginning of the request without
1275 * crossing into area which need to be retried further.
1276 *
1277 * Return:
1278 * The number of bytes to fail.
Tejun Heo80a761f2009-07-03 17:48:17 +09001279 */
1280unsigned int blk_rq_err_bytes(const struct request *rq)
1281{
1282 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1283 unsigned int bytes = 0;
1284 struct bio *bio;
1285
Christoph Hellwige8064022016-10-20 15:12:13 +02001286 if (!(rq->rq_flags & RQF_MIXED_MERGE))
Tejun Heo80a761f2009-07-03 17:48:17 +09001287 return blk_rq_bytes(rq);
1288
1289 /*
1290 * Currently the only 'mixing' which can happen is between
1291 * different fastfail types. We can safely fail portions
1292 * which have all the failfast bits that the first one has -
1293 * the ones which are at least as eager to fail as the first
1294 * one.
1295 */
1296 for (bio = rq->bio; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -06001297 if ((bio->bi_opf & ff) != ff)
Tejun Heo80a761f2009-07-03 17:48:17 +09001298 break;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001299 bytes += bio->bi_iter.bi_size;
Tejun Heo80a761f2009-07-03 17:48:17 +09001300 }
1301
1302 /* this could lead to infinite loop */
1303 BUG_ON(blk_rq_bytes(rq) && !bytes);
1304 return bytes;
1305}
1306EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
1307
Jens Axboe320ae512013-10-24 09:20:05 +01001308void blk_account_io_completion(struct request *req, unsigned int bytes)
Jens Axboebc58ba92009-01-23 10:54:44 +01001309{
Jens Axboec2553b52009-04-24 08:10:11 +02001310 if (blk_do_io_stat(req)) {
Michael Callahanddcf35d2018-07-18 04:47:39 -07001311 const int sgrp = op_stat_group(req_op(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001312 struct hd_struct *part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001313
Mike Snitzer112f1582018-12-06 11:41:18 -05001314 part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001315 part = req->part;
Mike Snitzer112f1582018-12-06 11:41:18 -05001316 part_stat_add(part, sectors[sgrp], bytes >> 9);
Jens Axboebc58ba92009-01-23 10:54:44 +01001317 part_stat_unlock();
1318 }
1319}
1320
Omar Sandoval522a7772018-05-09 02:08:53 -07001321void blk_account_io_done(struct request *req, u64 now)
Jens Axboebc58ba92009-01-23 10:54:44 +01001322{
Jens Axboebc58ba92009-01-23 10:54:44 +01001323 /*
Tejun Heodd4c1332010-09-03 11:56:16 +02001324 * Account IO completion. flush_rq isn't accounted as a
1325 * normal IO on queueing nor completion. Accounting the
1326 * containing request is enough.
Jens Axboebc58ba92009-01-23 10:54:44 +01001327 */
Christoph Hellwige8064022016-10-20 15:12:13 +02001328 if (blk_do_io_stat(req) && !(req->rq_flags & RQF_FLUSH_SEQ)) {
Michael Callahanddcf35d2018-07-18 04:47:39 -07001329 const int sgrp = op_stat_group(req_op(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001330 struct hd_struct *part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001331
Mike Snitzer112f1582018-12-06 11:41:18 -05001332 part_stat_lock();
Jerome Marchand09e099d2011-01-05 16:57:38 +01001333 part = req->part;
Jens Axboebc58ba92009-01-23 10:54:44 +01001334
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001335 update_io_ticks(part, jiffies);
Mike Snitzer112f1582018-12-06 11:41:18 -05001336 part_stat_inc(part, ios[sgrp]);
1337 part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001338 part_stat_add(part, time_in_queue, nsecs_to_jiffies64(now - req->start_time_ns));
Michael Callahanddcf35d2018-07-18 04:47:39 -07001339 part_dec_in_flight(req->q, part, rq_data_dir(req));
Jens Axboebc58ba92009-01-23 10:54:44 +01001340
Jens Axboe6c23a962011-01-07 08:43:37 +01001341 hd_struct_put(part);
Jens Axboebc58ba92009-01-23 10:54:44 +01001342 part_stat_unlock();
1343 }
1344}
1345
Jens Axboe320ae512013-10-24 09:20:05 +01001346void blk_account_io_start(struct request *rq, bool new_io)
1347{
1348 struct hd_struct *part;
1349 int rw = rq_data_dir(rq);
Jens Axboe320ae512013-10-24 09:20:05 +01001350
1351 if (!blk_do_io_stat(rq))
1352 return;
1353
Mike Snitzer112f1582018-12-06 11:41:18 -05001354 part_stat_lock();
Jens Axboe320ae512013-10-24 09:20:05 +01001355
1356 if (!new_io) {
1357 part = rq->part;
Mike Snitzer112f1582018-12-06 11:41:18 -05001358 part_stat_inc(part, merges[rw]);
Jens Axboe320ae512013-10-24 09:20:05 +01001359 } else {
1360 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
1361 if (!hd_struct_try_get(part)) {
1362 /*
1363 * The partition is already being removed,
1364 * the request will be accounted on the disk only
1365 *
1366 * We take a reference on disk->part0 although that
1367 * partition will never be deleted, so we can treat
1368 * it as any other partition.
1369 */
1370 part = &rq->rq_disk->part0;
1371 hd_struct_get(part);
1372 }
Jens Axboed62e26b2017-06-30 21:55:08 -06001373 part_inc_in_flight(rq->q, part, rw);
Jens Axboe320ae512013-10-24 09:20:05 +01001374 rq->part = part;
1375 }
1376
Mikulas Patocka5b18b5a2018-12-06 11:41:19 -05001377 update_io_ticks(part, jiffies);
1378
Jens Axboe320ae512013-10-24 09:20:05 +01001379 part_stat_unlock();
1380}
1381
Christoph Hellwigef71de82017-11-02 21:29:51 +03001382/*
1383 * Steal bios from a request and add them to a bio list.
1384 * The request must not have been partially completed before.
1385 */
1386void blk_steal_bios(struct bio_list *list, struct request *rq)
1387{
1388 if (rq->bio) {
1389 if (list->tail)
1390 list->tail->bi_next = rq->bio;
1391 else
1392 list->head = rq->bio;
1393 list->tail = rq->biotail;
1394
1395 rq->bio = NULL;
1396 rq->biotail = NULL;
1397 }
1398
1399 rq->__data_len = 0;
1400}
1401EXPORT_SYMBOL_GPL(blk_steal_bios);
1402
Tejun Heo9934c8c2009-05-08 11:54:16 +09001403/**
Tejun Heo2e60e022009-04-23 11:05:18 +09001404 * blk_update_request - Special helper function for request stacking drivers
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001405 * @req: the request being processed
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001406 * @error: block status code
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001407 * @nr_bytes: number of bytes to complete @req
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001408 *
1409 * Description:
Randy Dunlap8ebf9752009-06-11 20:00:41 -07001410 * Ends I/O on a number of bytes attached to @req, but doesn't complete
1411 * the request structure even if @req doesn't have leftover.
1412 * If @req has leftover, sets it up for the next range of segments.
Tejun Heo2e60e022009-04-23 11:05:18 +09001413 *
1414 * This special helper function is only for request stacking drivers
1415 * (e.g. request-based dm) so that they can handle partial completion.
Pavel Begunkov3a211b72019-05-23 18:43:11 +03001416 * Actual device drivers should use blk_mq_end_request instead.
Tejun Heo2e60e022009-04-23 11:05:18 +09001417 *
1418 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
1419 * %false return from this function.
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001420 *
Bart Van Assche1954e9a2018-06-27 13:09:05 -07001421 * Note:
1422 * The RQF_SPECIAL_PAYLOAD flag is ignored on purpose in both
1423 * blk_rq_bytes() and in blk_update_request().
1424 *
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001425 * Return:
Tejun Heo2e60e022009-04-23 11:05:18 +09001426 * %false - this request doesn't have any more data
1427 * %true - this request has more data
Kiyoshi Ueda3bcddea2007-12-11 17:52:28 -05001428 **/
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001429bool blk_update_request(struct request *req, blk_status_t error,
1430 unsigned int nr_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431{
Kent Overstreetf79ea412012-09-20 16:38:30 -07001432 int total_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001434 trace_block_rq_complete(req, blk_status_to_errno(error), nr_bytes);
Hannes Reinecke4a0efdc2014-10-01 14:32:31 +02001435
Tejun Heo2e60e022009-04-23 11:05:18 +09001436 if (!req->bio)
1437 return false;
1438
Christoph Hellwig2a842ac2017-06-03 09:38:04 +02001439 if (unlikely(error && !blk_rq_is_passthrough(req) &&
1440 !(req->rq_flags & RQF_QUIET)))
Christoph Hellwig178cc592019-06-20 10:59:15 -07001441 print_req_error(req, error, __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Jens Axboebc58ba92009-01-23 10:54:44 +01001443 blk_account_io_completion(req, nr_bytes);
Jens Axboed72d9042005-11-01 08:35:42 +01001444
Kent Overstreetf79ea412012-09-20 16:38:30 -07001445 total_bytes = 0;
1446 while (req->bio) {
1447 struct bio *bio = req->bio;
Kent Overstreet4f024f32013-10-11 15:44:27 -07001448 unsigned bio_bytes = min(bio->bi_iter.bi_size, nr_bytes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Bart Van Assche9c24c102018-06-19 10:26:40 -07001450 if (bio_bytes == bio->bi_iter.bi_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 req->bio = bio->bi_next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
NeilBrownfbbaf702017-04-07 09:40:52 -06001453 /* Completion has already been traced */
1454 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
Kent Overstreetf79ea412012-09-20 16:38:30 -07001455 req_bio_endio(req, bio, bio_bytes, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
Kent Overstreetf79ea412012-09-20 16:38:30 -07001457 total_bytes += bio_bytes;
1458 nr_bytes -= bio_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Kent Overstreetf79ea412012-09-20 16:38:30 -07001460 if (!nr_bytes)
1461 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 }
1463
1464 /*
1465 * completely done
1466 */
Tejun Heo2e60e022009-04-23 11:05:18 +09001467 if (!req->bio) {
1468 /*
1469 * Reset counters so that the request stacking driver
1470 * can find how many bytes remain in the request
1471 * later.
1472 */
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001473 req->__data_len = 0;
Tejun Heo2e60e022009-04-23 11:05:18 +09001474 return false;
1475 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001477 req->__data_len -= total_bytes;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001478
1479 /* update sector only for requests with clear definition of sector */
Christoph Hellwig57292b52017-01-31 16:57:29 +01001480 if (!blk_rq_is_passthrough(req))
Tejun Heoa2dec7b2009-05-07 22:24:44 +09001481 req->__sector += total_bytes >> 9;
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001482
Tejun Heo80a761f2009-07-03 17:48:17 +09001483 /* mixed attributes always follow the first bio */
Christoph Hellwige8064022016-10-20 15:12:13 +02001484 if (req->rq_flags & RQF_MIXED_MERGE) {
Tejun Heo80a761f2009-07-03 17:48:17 +09001485 req->cmd_flags &= ~REQ_FAILFAST_MASK;
Jens Axboe1eff9d32016-08-05 15:35:16 -06001486 req->cmd_flags |= req->bio->bi_opf & REQ_FAILFAST_MASK;
Tejun Heo80a761f2009-07-03 17:48:17 +09001487 }
1488
Christoph Hellwiged6565e2017-05-11 12:34:38 +02001489 if (!(req->rq_flags & RQF_SPECIAL_PAYLOAD)) {
1490 /*
1491 * If total number of sectors is less than the first segment
1492 * size, something has gone terribly wrong.
1493 */
1494 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
1495 blk_dump_rq_flags(req, "request botched");
1496 req->__data_len = blk_rq_cur_bytes(req);
1497 }
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001498
Christoph Hellwiged6565e2017-05-11 12:34:38 +02001499 /* recalculate the number of segments */
Christoph Hellwige9cd19c2019-06-06 12:29:02 +02001500 req->nr_phys_segments = blk_recalc_rq_segments(req);
Christoph Hellwiged6565e2017-05-11 12:34:38 +02001501 }
Tejun Heo2e46e8b2009-05-07 22:24:41 +09001502
Tejun Heo2e60e022009-04-23 11:05:18 +09001503 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504}
Tejun Heo2e60e022009-04-23 11:05:18 +09001505EXPORT_SYMBOL_GPL(blk_update_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001507#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1508/**
1509 * rq_flush_dcache_pages - Helper function to flush all pages in a request
1510 * @rq: the request to be flushed
1511 *
1512 * Description:
1513 * Flush all pages in @rq.
1514 */
1515void rq_flush_dcache_pages(struct request *rq)
1516{
1517 struct req_iterator iter;
Kent Overstreet79886132013-11-23 17:19:00 -08001518 struct bio_vec bvec;
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001519
1520 rq_for_each_segment(bvec, rq, iter)
Kent Overstreet79886132013-11-23 17:19:00 -08001521 flush_dcache_page(bvec.bv_page);
Ilya Loginov2d4dc892009-11-26 09:16:19 +01001522}
1523EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
1524#endif
1525
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02001526/**
1527 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
1528 * @q : the queue of the device being checked
1529 *
1530 * Description:
1531 * Check if underlying low-level drivers of a device are busy.
1532 * If the drivers want to export their busy state, they must set own
1533 * exporting function using blk_queue_lld_busy() first.
1534 *
1535 * Basically, this function is used only by request stacking drivers
1536 * to stop dispatching requests to underlying devices when underlying
1537 * devices are busy. This behavior helps more I/O merging on the queue
1538 * of the request stacking driver and prevents I/O throughput regression
1539 * on burst I/O load.
1540 *
1541 * Return:
1542 * 0 - Not busy (The request stacking driver should dispatch request)
1543 * 1 - Busy (The request stacking driver should stop dispatching request)
1544 */
1545int blk_lld_busy(struct request_queue *q)
1546{
Jens Axboe344e9ff2018-11-15 12:22:51 -07001547 if (queue_is_mq(q) && q->mq_ops->busy)
Jens Axboe9ba20522018-10-29 10:15:10 -06001548 return q->mq_ops->busy(q);
Kiyoshi Uedaef9e3fa2008-10-01 16:12:15 +02001549
1550 return 0;
1551}
1552EXPORT_SYMBOL_GPL(blk_lld_busy);
1553
Mike Snitzer78d8e582015-06-26 10:01:13 -04001554/**
1555 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
1556 * @rq: the clone request to be cleaned up
1557 *
1558 * Description:
1559 * Free all bios in @rq for a cloned request.
1560 */
1561void blk_rq_unprep_clone(struct request *rq)
1562{
1563 struct bio *bio;
1564
1565 while ((bio = rq->bio) != NULL) {
1566 rq->bio = bio->bi_next;
1567
1568 bio_put(bio);
1569 }
1570}
1571EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
1572
1573/*
1574 * Copy attributes of the original request to the clone request.
1575 * The actual data parts (e.g. ->cmd, ->sense) are not copied.
1576 */
1577static void __blk_rq_prep_clone(struct request *dst, struct request *src)
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02001578{
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02001579 dst->__sector = blk_rq_pos(src);
1580 dst->__data_len = blk_rq_bytes(src);
Bart Van Assche297ba572018-06-27 12:55:18 -07001581 if (src->rq_flags & RQF_SPECIAL_PAYLOAD) {
1582 dst->rq_flags |= RQF_SPECIAL_PAYLOAD;
1583 dst->special_vec = src->special_vec;
1584 }
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02001585 dst->nr_phys_segments = src->nr_phys_segments;
1586 dst->ioprio = src->ioprio;
1587 dst->extra_len = src->extra_len;
Mike Snitzer78d8e582015-06-26 10:01:13 -04001588}
1589
1590/**
1591 * blk_rq_prep_clone - Helper function to setup clone request
1592 * @rq: the request to be setup
1593 * @rq_src: original request to be cloned
1594 * @bs: bio_set that bios for clone are allocated from
1595 * @gfp_mask: memory allocation mask for bio
1596 * @bio_ctr: setup function to be called for each clone bio.
1597 * Returns %0 for success, non %0 for failure.
1598 * @data: private data to be passed to @bio_ctr
1599 *
1600 * Description:
1601 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
1602 * The actual data parts of @rq_src (e.g. ->cmd, ->sense)
1603 * are not copied, and copying such parts is the caller's responsibility.
1604 * Also, pages which the original bios are pointing to are not copied
1605 * and the cloned bios just point same pages.
1606 * So cloned bios must be completed before original bios, which means
1607 * the caller must complete @rq before @rq_src.
1608 */
1609int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1610 struct bio_set *bs, gfp_t gfp_mask,
1611 int (*bio_ctr)(struct bio *, struct bio *, void *),
1612 void *data)
1613{
1614 struct bio *bio, *bio_src;
1615
1616 if (!bs)
Kent Overstreetf4f81542018-05-08 21:33:52 -04001617 bs = &fs_bio_set;
Mike Snitzer78d8e582015-06-26 10:01:13 -04001618
1619 __rq_for_each_bio(bio_src, rq_src) {
1620 bio = bio_clone_fast(bio_src, gfp_mask, bs);
1621 if (!bio)
1622 goto free_and_out;
1623
1624 if (bio_ctr && bio_ctr(bio, bio_src, data))
1625 goto free_and_out;
1626
1627 if (rq->bio) {
1628 rq->biotail->bi_next = bio;
1629 rq->biotail = bio;
1630 } else
1631 rq->bio = rq->biotail = bio;
1632 }
1633
1634 __blk_rq_prep_clone(rq, rq_src);
1635
1636 return 0;
1637
1638free_and_out:
1639 if (bio)
1640 bio_put(bio);
1641 blk_rq_unprep_clone(rq);
1642
1643 return -ENOMEM;
Kiyoshi Uedab0fd2712009-06-11 13:10:16 +02001644}
1645EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
1646
Jens Axboe59c3d452014-04-08 09:15:35 -06001647int kblockd_schedule_work(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
1649 return queue_work(kblockd_workqueue, work);
1650}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651EXPORT_SYMBOL(kblockd_schedule_work);
1652
Jens Axboeee63cfa2016-08-24 15:52:48 -06001653int kblockd_schedule_work_on(int cpu, struct work_struct *work)
1654{
1655 return queue_work_on(cpu, kblockd_workqueue, work);
1656}
1657EXPORT_SYMBOL(kblockd_schedule_work_on);
1658
Jens Axboe818cd1c2017-04-10 09:54:55 -06001659int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork,
1660 unsigned long delay)
1661{
1662 return mod_delayed_work_on(cpu, kblockd_workqueue, dwork, delay);
1663}
1664EXPORT_SYMBOL(kblockd_mod_delayed_work_on);
1665
Suresh Jayaraman75df7132011-09-21 10:00:16 +02001666/**
1667 * blk_start_plug - initialize blk_plug and track it inside the task_struct
1668 * @plug: The &struct blk_plug that needs to be initialized
1669 *
1670 * Description:
Jeff Moyer40405852019-01-08 16:57:34 -05001671 * blk_start_plug() indicates to the block layer an intent by the caller
1672 * to submit multiple I/O requests in a batch. The block layer may use
1673 * this hint to defer submitting I/Os from the caller until blk_finish_plug()
1674 * is called. However, the block layer may choose to submit requests
1675 * before a call to blk_finish_plug() if the number of queued I/Os
1676 * exceeds %BLK_MAX_REQUEST_COUNT, or if the size of the I/O is larger than
1677 * %BLK_PLUG_FLUSH_SIZE. The queued I/Os may also be submitted early if
1678 * the task schedules (see below).
1679 *
Suresh Jayaraman75df7132011-09-21 10:00:16 +02001680 * Tracking blk_plug inside the task_struct will help with auto-flushing the
1681 * pending I/O should the task end up blocking between blk_start_plug() and
1682 * blk_finish_plug(). This is important from a performance perspective, but
1683 * also ensures that we don't deadlock. For instance, if the task is blocking
1684 * for a memory allocation, memory reclaim could end up wanting to free a
1685 * page belonging to that request that is currently residing in our private
1686 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
1687 * this kind of deadlock.
1688 */
Jens Axboe73c10102011-03-08 13:19:51 +01001689void blk_start_plug(struct blk_plug *plug)
1690{
1691 struct task_struct *tsk = current;
1692
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07001693 /*
1694 * If this is a nested plug, don't actually assign it.
1695 */
1696 if (tsk->plug)
1697 return;
1698
Jens Axboe320ae512013-10-24 09:20:05 +01001699 INIT_LIST_HEAD(&plug->mq_list);
NeilBrown048c9372011-04-18 09:52:22 +02001700 INIT_LIST_HEAD(&plug->cb_list);
Jens Axboe5f0ed772018-11-23 22:04:33 -07001701 plug->rq_count = 0;
Jens Axboece5b0092018-11-27 17:13:56 -07001702 plug->multiple_queues = false;
Jens Axboe5f0ed772018-11-23 22:04:33 -07001703
Jens Axboe73c10102011-03-08 13:19:51 +01001704 /*
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07001705 * Store ordering should not be needed here, since a potential
1706 * preempt will imply a full memory barrier
Jens Axboe73c10102011-03-08 13:19:51 +01001707 */
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07001708 tsk->plug = plug;
Jens Axboe73c10102011-03-08 13:19:51 +01001709}
1710EXPORT_SYMBOL(blk_start_plug);
1711
NeilBrown74018dc2012-07-31 09:08:15 +02001712static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
NeilBrown048c9372011-04-18 09:52:22 +02001713{
1714 LIST_HEAD(callbacks);
1715
Shaohua Li2a7d5552012-07-31 09:08:15 +02001716 while (!list_empty(&plug->cb_list)) {
1717 list_splice_init(&plug->cb_list, &callbacks);
NeilBrown048c9372011-04-18 09:52:22 +02001718
Shaohua Li2a7d5552012-07-31 09:08:15 +02001719 while (!list_empty(&callbacks)) {
1720 struct blk_plug_cb *cb = list_first_entry(&callbacks,
NeilBrown048c9372011-04-18 09:52:22 +02001721 struct blk_plug_cb,
1722 list);
Shaohua Li2a7d5552012-07-31 09:08:15 +02001723 list_del(&cb->list);
NeilBrown74018dc2012-07-31 09:08:15 +02001724 cb->callback(cb, from_schedule);
Shaohua Li2a7d5552012-07-31 09:08:15 +02001725 }
NeilBrown048c9372011-04-18 09:52:22 +02001726 }
1727}
1728
NeilBrown9cbb1752012-07-31 09:08:14 +02001729struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
1730 int size)
1731{
1732 struct blk_plug *plug = current->plug;
1733 struct blk_plug_cb *cb;
1734
1735 if (!plug)
1736 return NULL;
1737
1738 list_for_each_entry(cb, &plug->cb_list, list)
1739 if (cb->callback == unplug && cb->data == data)
1740 return cb;
1741
1742 /* Not currently on the callback list */
1743 BUG_ON(size < sizeof(*cb));
1744 cb = kzalloc(size, GFP_ATOMIC);
1745 if (cb) {
1746 cb->data = data;
1747 cb->callback = unplug;
1748 list_add(&cb->list, &plug->cb_list);
1749 }
1750 return cb;
1751}
1752EXPORT_SYMBOL(blk_check_plugged);
1753
Jens Axboe49cac012011-04-16 13:51:05 +02001754void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
Jens Axboe73c10102011-03-08 13:19:51 +01001755{
NeilBrown74018dc2012-07-31 09:08:15 +02001756 flush_plug_callbacks(plug, from_schedule);
Jens Axboe320ae512013-10-24 09:20:05 +01001757
1758 if (!list_empty(&plug->mq_list))
1759 blk_mq_flush_plug_list(plug, from_schedule);
Jens Axboe73c10102011-03-08 13:19:51 +01001760}
Jens Axboe73c10102011-03-08 13:19:51 +01001761
Jeff Moyer40405852019-01-08 16:57:34 -05001762/**
1763 * blk_finish_plug - mark the end of a batch of submitted I/O
1764 * @plug: The &struct blk_plug passed to blk_start_plug()
1765 *
1766 * Description:
1767 * Indicate that a batch of I/O submissions is complete. This function
1768 * must be paired with an initial call to blk_start_plug(). The intent
1769 * is to allow the block layer to optimize I/O submission. See the
1770 * documentation for blk_start_plug() for more information.
1771 */
Jens Axboe73c10102011-03-08 13:19:51 +01001772void blk_finish_plug(struct blk_plug *plug)
1773{
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07001774 if (plug != current->plug)
1775 return;
Jens Axboef6603782011-04-15 15:49:07 +02001776 blk_flush_plug_list(plug, false);
Christoph Hellwig88b996c2011-04-15 15:20:10 +02001777
Shaohua Lidd6cf3e2015-05-08 10:51:28 -07001778 current->plug = NULL;
Jens Axboe73c10102011-03-08 13:19:51 +01001779}
1780EXPORT_SYMBOL(blk_finish_plug);
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782int __init blk_dev_init(void)
1783{
Christoph Hellwigef295ec2016-10-28 08:48:16 -06001784 BUILD_BUG_ON(REQ_OP_LAST >= (1 << REQ_OP_BITS));
1785 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
Maninder Singh0762b232015-07-07 12:41:07 +05301786 FIELD_SIZEOF(struct request, cmd_flags));
Christoph Hellwigef295ec2016-10-28 08:48:16 -06001787 BUILD_BUG_ON(REQ_OP_BITS + REQ_FLAG_BITS > 8 *
1788 FIELD_SIZEOF(struct bio, bi_opf));
Nikanth Karthikesan9eb55b02009-04-27 14:53:54 +02001789
Tejun Heo89b90be2011-01-03 15:01:47 +01001790 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
1791 kblockd_workqueue = alloc_workqueue("kblockd",
Matias Bjørling28747fc2014-06-11 23:43:54 +02001792 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 if (!kblockd_workqueue)
1794 panic("Failed to create kblockd\n");
1795
Ilya Dryomovc2789bd2015-11-20 22:16:46 +01001796 blk_requestq_cachep = kmem_cache_create("request_queue",
Jens Axboe165125e2007-07-24 09:28:11 +02001797 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798
Omar Sandoval18fbda92017-01-31 14:53:20 -08001799#ifdef CONFIG_DEBUG_FS
1800 blk_debugfs_root = debugfs_create_dir("block", NULL);
1801#endif
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 return 0;
1804}