Doug / smarc-fsl-linux-kernel

Download as
Browse Code ยป

Commit cbae8d45d61f3a8c155caf267d01e5e0f0b2f4b7

Authored by NeilBrown
Committed by Jens Axboe
1 parent 0cfbcafcae
Exists in smarc-l5.0.0_1.0.0-ga and in 5 other branches imx_3.10.17_1.0.1_ga, imx_3.10.53_1.1.0_ga, imx_3.14.28_1.0.0_ga, smarc-imx_3.10.53_1.1.0_ga, smarc-imx_3.14.28_1.0.0_ga

block: export block_unplug tracepoint 

This allows stacked devices (like md/raid5) to provide blktrace
tracing, including unplug events.

Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Showing 1 changed file with 1 additions and 0 deletions Inline Diff

1 /* 1 /*
2 * Copyright (C) 1991, 1992 Linus Torvalds 2 * Copyright (C) 1991, 1992 Linus Torvalds
3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics 3 * Copyright (C) 1994, Karl Keyte: Added support for disk statistics
4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE 4 * Elevator latency, (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de> 5 * Queue request tables / lock, selectable elevator, Jens Axboe <axboe@suse.de>
6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au> 6 * kernel-doc documentation started by NeilBrown <neilb@cse.unsw.edu.au>
7 * - July2000 7 * - July2000
8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001 8 * bio rewrite, highmem i/o, etc, Jens Axboe <axboe@suse.de> - may 2001
9 */ 9 */
10 10
11 /* 11 /*
12 * This handles all read/write requests to block devices 12 * This handles all read/write requests to block devices
13 */ 13 */
14 #include <linux/kernel.h> 14 #include <linux/kernel.h>
15 #include <linux/module.h> 15 #include <linux/module.h>
16 #include <linux/backing-dev.h> 16 #include <linux/backing-dev.h>
17 #include <linux/bio.h> 17 #include <linux/bio.h>
18 #include <linux/blkdev.h> 18 #include <linux/blkdev.h>
19 #include <linux/highmem.h> 19 #include <linux/highmem.h>
20 #include <linux/mm.h> 20 #include <linux/mm.h>
21 #include <linux/kernel_stat.h> 21 #include <linux/kernel_stat.h>
22 #include <linux/string.h> 22 #include <linux/string.h>
23 #include <linux/init.h> 23 #include <linux/init.h>
24 #include <linux/completion.h> 24 #include <linux/completion.h>
25 #include <linux/slab.h> 25 #include <linux/slab.h>
26 #include <linux/swap.h> 26 #include <linux/swap.h>
27 #include <linux/writeback.h> 27 #include <linux/writeback.h>
28 #include <linux/task_io_accounting_ops.h> 28 #include <linux/task_io_accounting_ops.h>
29 #include <linux/fault-inject.h> 29 #include <linux/fault-inject.h>
30 #include <linux/list_sort.h> 30 #include <linux/list_sort.h>
31 #include <linux/delay.h> 31 #include <linux/delay.h>
32 #include <linux/ratelimit.h> 32 #include <linux/ratelimit.h>
33 33
34 #define CREATE_TRACE_POINTS 34 #define CREATE_TRACE_POINTS
35 #include <trace/events/block.h> 35 #include <trace/events/block.h>
36 36
37 #include "blk.h" 37 #include "blk.h"
38 #include "blk-cgroup.h" 38 #include "blk-cgroup.h"
39 39
40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap); 40 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap); 41 EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete); 42 EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_complete);
43 EXPORT_TRACEPOINT_SYMBOL_GPL(block_unplug);
43 44
44 DEFINE_IDA(blk_queue_ida); 45 DEFINE_IDA(blk_queue_ida);
45 46
46 /* 47 /*
47 * For the allocated request tables 48 * For the allocated request tables
48 */ 49 */
49 static struct kmem_cache *request_cachep; 50 static struct kmem_cache *request_cachep;
50 51
51 /* 52 /*
52 * For queue allocation 53 * For queue allocation
53 */ 54 */
54 struct kmem_cache *blk_requestq_cachep; 55 struct kmem_cache *blk_requestq_cachep;
55 56
56 /* 57 /*
57 * Controlling structure to kblockd 58 * Controlling structure to kblockd
58 */ 59 */
59 static struct workqueue_struct *kblockd_workqueue; 60 static struct workqueue_struct *kblockd_workqueue;
60 61
61 static void drive_stat_acct(struct request *rq, int new_io) 62 static void drive_stat_acct(struct request *rq, int new_io)
62 { 63 {
63 struct hd_struct *part; 64 struct hd_struct *part;
64 int rw = rq_data_dir(rq); 65 int rw = rq_data_dir(rq);
65 int cpu; 66 int cpu;
66 67
67 if (!blk_do_io_stat(rq)) 68 if (!blk_do_io_stat(rq))
68 return; 69 return;
69 70
70 cpu = part_stat_lock(); 71 cpu = part_stat_lock();
71 72
72 if (!new_io) { 73 if (!new_io) {
73 part = rq->part; 74 part = rq->part;
74 part_stat_inc(cpu, part, merges[rw]); 75 part_stat_inc(cpu, part, merges[rw]);
75 } else { 76 } else {
76 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq)); 77 part = disk_map_sector_rcu(rq->rq_disk, blk_rq_pos(rq));
77 if (!hd_struct_try_get(part)) { 78 if (!hd_struct_try_get(part)) {
78 /* 79 /*
79 * The partition is already being removed, 80 * The partition is already being removed,
80 * the request will be accounted on the disk only 81 * the request will be accounted on the disk only
81 * 82 *
82 * We take a reference on disk->part0 although that 83 * We take a reference on disk->part0 although that
83 * partition will never be deleted, so we can treat 84 * partition will never be deleted, so we can treat
84 * it as any other partition. 85 * it as any other partition.
85 */ 86 */
86 part = &rq->rq_disk->part0; 87 part = &rq->rq_disk->part0;
87 hd_struct_get(part); 88 hd_struct_get(part);
88 } 89 }
89 part_round_stats(cpu, part); 90 part_round_stats(cpu, part);
90 part_inc_in_flight(part, rw); 91 part_inc_in_flight(part, rw);
91 rq->part = part; 92 rq->part = part;
92 } 93 }
93 94
94 part_stat_unlock(); 95 part_stat_unlock();
95 } 96 }
96 97
97 void blk_queue_congestion_threshold(struct request_queue *q) 98 void blk_queue_congestion_threshold(struct request_queue *q)
98 { 99 {
99 int nr; 100 int nr;
100 101
101 nr = q->nr_requests - (q->nr_requests / 8) + 1; 102 nr = q->nr_requests - (q->nr_requests / 8) + 1;
102 if (nr > q->nr_requests) 103 if (nr > q->nr_requests)
103 nr = q->nr_requests; 104 nr = q->nr_requests;
104 q->nr_congestion_on = nr; 105 q->nr_congestion_on = nr;
105 106
106 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1; 107 nr = q->nr_requests - (q->nr_requests / 8) - (q->nr_requests / 16) - 1;
107 if (nr < 1) 108 if (nr < 1)
108 nr = 1; 109 nr = 1;
109 q->nr_congestion_off = nr; 110 q->nr_congestion_off = nr;
110 } 111 }
111 112
112 /** 113 /**
113 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info 114 * blk_get_backing_dev_info - get the address of a queue's backing_dev_info
114 * @bdev: device 115 * @bdev: device
115 * 116 *
116 * Locates the passed device's request queue and returns the address of its 117 * Locates the passed device's request queue and returns the address of its
117 * backing_dev_info 118 * backing_dev_info
118 * 119 *
119 * Will return NULL if the request queue cannot be located. 120 * Will return NULL if the request queue cannot be located.
120 */ 121 */
121 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev) 122 struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev)
122 { 123 {
123 struct backing_dev_info *ret = NULL; 124 struct backing_dev_info *ret = NULL;
124 struct request_queue *q = bdev_get_queue(bdev); 125 struct request_queue *q = bdev_get_queue(bdev);
125 126
126 if (q) 127 if (q)
127 ret = &q->backing_dev_info; 128 ret = &q->backing_dev_info;
128 return ret; 129 return ret;
129 } 130 }
130 EXPORT_SYMBOL(blk_get_backing_dev_info); 131 EXPORT_SYMBOL(blk_get_backing_dev_info);
131 132
132 void blk_rq_init(struct request_queue *q, struct request *rq) 133 void blk_rq_init(struct request_queue *q, struct request *rq)
133 { 134 {
134 memset(rq, 0, sizeof(*rq)); 135 memset(rq, 0, sizeof(*rq));
135 136
136 INIT_LIST_HEAD(&rq->queuelist); 137 INIT_LIST_HEAD(&rq->queuelist);
137 INIT_LIST_HEAD(&rq->timeout_list); 138 INIT_LIST_HEAD(&rq->timeout_list);
138 rq->cpu = -1; 139 rq->cpu = -1;
139 rq->q = q; 140 rq->q = q;
140 rq->__sector = (sector_t) -1; 141 rq->__sector = (sector_t) -1;
141 INIT_HLIST_NODE(&rq->hash); 142 INIT_HLIST_NODE(&rq->hash);
142 RB_CLEAR_NODE(&rq->rb_node); 143 RB_CLEAR_NODE(&rq->rb_node);
143 rq->cmd = rq->__cmd; 144 rq->cmd = rq->__cmd;
144 rq->cmd_len = BLK_MAX_CDB; 145 rq->cmd_len = BLK_MAX_CDB;
145 rq->tag = -1; 146 rq->tag = -1;
146 rq->ref_count = 1; 147 rq->ref_count = 1;
147 rq->start_time = jiffies; 148 rq->start_time = jiffies;
148 set_start_time_ns(rq); 149 set_start_time_ns(rq);
149 rq->part = NULL; 150 rq->part = NULL;
150 } 151 }
151 EXPORT_SYMBOL(blk_rq_init); 152 EXPORT_SYMBOL(blk_rq_init);
152 153
153 static void req_bio_endio(struct request *rq, struct bio *bio, 154 static void req_bio_endio(struct request *rq, struct bio *bio,
154 unsigned int nbytes, int error) 155 unsigned int nbytes, int error)
155 { 156 {
156 if (error) 157 if (error)
157 clear_bit(BIO_UPTODATE, &bio->bi_flags); 158 clear_bit(BIO_UPTODATE, &bio->bi_flags);
158 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) 159 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
159 error = -EIO; 160 error = -EIO;
160 161
161 if (unlikely(nbytes > bio->bi_size)) { 162 if (unlikely(nbytes > bio->bi_size)) {
162 printk(KERN_ERR "%s: want %u bytes done, %u left\n", 163 printk(KERN_ERR "%s: want %u bytes done, %u left\n",
163 __func__, nbytes, bio->bi_size); 164 __func__, nbytes, bio->bi_size);
164 nbytes = bio->bi_size; 165 nbytes = bio->bi_size;
165 } 166 }
166 167
167 if (unlikely(rq->cmd_flags & REQ_QUIET)) 168 if (unlikely(rq->cmd_flags & REQ_QUIET))
168 set_bit(BIO_QUIET, &bio->bi_flags); 169 set_bit(BIO_QUIET, &bio->bi_flags);
169 170
170 bio->bi_size -= nbytes; 171 bio->bi_size -= nbytes;
171 bio->bi_sector += (nbytes >> 9); 172 bio->bi_sector += (nbytes >> 9);
172 173
173 if (bio_integrity(bio)) 174 if (bio_integrity(bio))
174 bio_integrity_advance(bio, nbytes); 175 bio_integrity_advance(bio, nbytes);
175 176
176 /* don't actually finish bio if it's part of flush sequence */ 177 /* don't actually finish bio if it's part of flush sequence */
177 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ)) 178 if (bio->bi_size == 0 && !(rq->cmd_flags & REQ_FLUSH_SEQ))
178 bio_endio(bio, error); 179 bio_endio(bio, error);
179 } 180 }
180 181
181 void blk_dump_rq_flags(struct request *rq, char *msg) 182 void blk_dump_rq_flags(struct request *rq, char *msg)
182 { 183 {
183 int bit; 184 int bit;
184 185
185 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg, 186 printk(KERN_INFO "%s: dev %s: type=%x, flags=%x\n", msg,
186 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type, 187 rq->rq_disk ? rq->rq_disk->disk_name : "?", rq->cmd_type,
187 rq->cmd_flags); 188 rq->cmd_flags);
188 189
189 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n", 190 printk(KERN_INFO " sector %llu, nr/cnr %u/%u\n",
190 (unsigned long long)blk_rq_pos(rq), 191 (unsigned long long)blk_rq_pos(rq),
191 blk_rq_sectors(rq), blk_rq_cur_sectors(rq)); 192 blk_rq_sectors(rq), blk_rq_cur_sectors(rq));
192 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n", 193 printk(KERN_INFO " bio %p, biotail %p, buffer %p, len %u\n",
193 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq)); 194 rq->bio, rq->biotail, rq->buffer, blk_rq_bytes(rq));
194 195
195 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { 196 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
196 printk(KERN_INFO " cdb: "); 197 printk(KERN_INFO " cdb: ");
197 for (bit = 0; bit < BLK_MAX_CDB; bit++) 198 for (bit = 0; bit < BLK_MAX_CDB; bit++)
198 printk("%02x ", rq->cmd[bit]); 199 printk("%02x ", rq->cmd[bit]);
199 printk("\n"); 200 printk("\n");
200 } 201 }
201 } 202 }
202 EXPORT_SYMBOL(blk_dump_rq_flags); 203 EXPORT_SYMBOL(blk_dump_rq_flags);
203 204
204 static void blk_delay_work(struct work_struct *work) 205 static void blk_delay_work(struct work_struct *work)
205 { 206 {
206 struct request_queue *q; 207 struct request_queue *q;
207 208
208 q = container_of(work, struct request_queue, delay_work.work); 209 q = container_of(work, struct request_queue, delay_work.work);
209 spin_lock_irq(q->queue_lock); 210 spin_lock_irq(q->queue_lock);
210 __blk_run_queue(q); 211 __blk_run_queue(q);
211 spin_unlock_irq(q->queue_lock); 212 spin_unlock_irq(q->queue_lock);
212 } 213 }
213 214
214 /** 215 /**
215 * blk_delay_queue - restart queueing after defined interval 216 * blk_delay_queue - restart queueing after defined interval
216 * @q: The &struct request_queue in question 217 * @q: The &struct request_queue in question
217 * @msecs: Delay in msecs 218 * @msecs: Delay in msecs
218 * 219 *
219 * Description: 220 * Description:
220 * Sometimes queueing needs to be postponed for a little while, to allow 221 * Sometimes queueing needs to be postponed for a little while, to allow
221 * resources to come back. This function will make sure that queueing is 222 * resources to come back. This function will make sure that queueing is
222 * restarted around the specified time. Queue lock must be held. 223 * restarted around the specified time. Queue lock must be held.
223 */ 224 */
224 void blk_delay_queue(struct request_queue *q, unsigned long msecs) 225 void blk_delay_queue(struct request_queue *q, unsigned long msecs)
225 { 226 {
226 if (likely(!blk_queue_dead(q))) 227 if (likely(!blk_queue_dead(q)))
227 queue_delayed_work(kblockd_workqueue, &q->delay_work, 228 queue_delayed_work(kblockd_workqueue, &q->delay_work,
228 msecs_to_jiffies(msecs)); 229 msecs_to_jiffies(msecs));
229 } 230 }
230 EXPORT_SYMBOL(blk_delay_queue); 231 EXPORT_SYMBOL(blk_delay_queue);
231 232
232 /** 233 /**
233 * blk_start_queue - restart a previously stopped queue 234 * blk_start_queue - restart a previously stopped queue
234 * @q: The &struct request_queue in question 235 * @q: The &struct request_queue in question
235 * 236 *
236 * Description: 237 * Description:
237 * blk_start_queue() will clear the stop flag on the queue, and call 238 * blk_start_queue() will clear the stop flag on the queue, and call
238 * the request_fn for the queue if it was in a stopped state when 239 * the request_fn for the queue if it was in a stopped state when
239 * entered. Also see blk_stop_queue(). Queue lock must be held. 240 * entered. Also see blk_stop_queue(). Queue lock must be held.
240 **/ 241 **/
241 void blk_start_queue(struct request_queue *q) 242 void blk_start_queue(struct request_queue *q)
242 { 243 {
243 WARN_ON(!irqs_disabled()); 244 WARN_ON(!irqs_disabled());
244 245
245 queue_flag_clear(QUEUE_FLAG_STOPPED, q); 246 queue_flag_clear(QUEUE_FLAG_STOPPED, q);
246 __blk_run_queue(q); 247 __blk_run_queue(q);
247 } 248 }
248 EXPORT_SYMBOL(blk_start_queue); 249 EXPORT_SYMBOL(blk_start_queue);
249 250
250 /** 251 /**
251 * blk_stop_queue - stop a queue 252 * blk_stop_queue - stop a queue
252 * @q: The &struct request_queue in question 253 * @q: The &struct request_queue in question
253 * 254 *
254 * Description: 255 * Description:
255 * The Linux block layer assumes that a block driver will consume all 256 * The Linux block layer assumes that a block driver will consume all
256 * entries on the request queue when the request_fn strategy is called. 257 * entries on the request queue when the request_fn strategy is called.
257 * Often this will not happen, because of hardware limitations (queue 258 * Often this will not happen, because of hardware limitations (queue
258 * depth settings). If a device driver gets a 'queue full' response, 259 * depth settings). If a device driver gets a 'queue full' response,
259 * or if it simply chooses not to queue more I/O at one point, it can 260 * or if it simply chooses not to queue more I/O at one point, it can
260 * call this function to prevent the request_fn from being called until 261 * call this function to prevent the request_fn from being called until
261 * the driver has signalled it's ready to go again. This happens by calling 262 * the driver has signalled it's ready to go again. This happens by calling
262 * blk_start_queue() to restart queue operations. Queue lock must be held. 263 * blk_start_queue() to restart queue operations. Queue lock must be held.
263 **/ 264 **/
264 void blk_stop_queue(struct request_queue *q) 265 void blk_stop_queue(struct request_queue *q)
265 { 266 {
266 cancel_delayed_work(&q->delay_work); 267 cancel_delayed_work(&q->delay_work);
267 queue_flag_set(QUEUE_FLAG_STOPPED, q); 268 queue_flag_set(QUEUE_FLAG_STOPPED, q);
268 } 269 }
269 EXPORT_SYMBOL(blk_stop_queue); 270 EXPORT_SYMBOL(blk_stop_queue);
270 271
271 /** 272 /**
272 * blk_sync_queue - cancel any pending callbacks on a queue 273 * blk_sync_queue - cancel any pending callbacks on a queue
273 * @q: the queue 274 * @q: the queue
274 * 275 *
275 * Description: 276 * Description:
276 * The block layer may perform asynchronous callback activity 277 * The block layer may perform asynchronous callback activity
277 * on a queue, such as calling the unplug function after a timeout. 278 * on a queue, such as calling the unplug function after a timeout.
278 * A block device may call blk_sync_queue to ensure that any 279 * A block device may call blk_sync_queue to ensure that any
279 * such activity is cancelled, thus allowing it to release resources 280 * such activity is cancelled, thus allowing it to release resources
280 * that the callbacks might use. The caller must already have made sure 281 * that the callbacks might use. The caller must already have made sure
281 * that its ->make_request_fn will not re-add plugging prior to calling 282 * that its ->make_request_fn will not re-add plugging prior to calling
282 * this function. 283 * this function.
283 * 284 *
284 * This function does not cancel any asynchronous activity arising 285 * This function does not cancel any asynchronous activity arising
285 * out of elevator or throttling code. That would require elevaotor_exit() 286 * out of elevator or throttling code. That would require elevaotor_exit()
286 * and blkcg_exit_queue() to be called with queue lock initialized. 287 * and blkcg_exit_queue() to be called with queue lock initialized.
287 * 288 *
288 */ 289 */
289 void blk_sync_queue(struct request_queue *q) 290 void blk_sync_queue(struct request_queue *q)
290 { 291 {
291 del_timer_sync(&q->timeout); 292 del_timer_sync(&q->timeout);
292 cancel_delayed_work_sync(&q->delay_work); 293 cancel_delayed_work_sync(&q->delay_work);
293 } 294 }
294 EXPORT_SYMBOL(blk_sync_queue); 295 EXPORT_SYMBOL(blk_sync_queue);
295 296
296 /** 297 /**
297 * __blk_run_queue_uncond - run a queue whether or not it has been stopped 298 * __blk_run_queue_uncond - run a queue whether or not it has been stopped
298 * @q: The queue to run 299 * @q: The queue to run
299 * 300 *
300 * Description: 301 * Description:
301 * Invoke request handling on a queue if there are any pending requests. 302 * Invoke request handling on a queue if there are any pending requests.
302 * May be used to restart request handling after a request has completed. 303 * May be used to restart request handling after a request has completed.
303 * This variant runs the queue whether or not the queue has been 304 * This variant runs the queue whether or not the queue has been
304 * stopped. Must be called with the queue lock held and interrupts 305 * stopped. Must be called with the queue lock held and interrupts
305 * disabled. See also @blk_run_queue. 306 * disabled. See also @blk_run_queue.
306 */ 307 */
307 inline void __blk_run_queue_uncond(struct request_queue *q) 308 inline void __blk_run_queue_uncond(struct request_queue *q)
308 { 309 {
309 if (unlikely(blk_queue_dead(q))) 310 if (unlikely(blk_queue_dead(q)))
310 return; 311 return;
311 312
312 /* 313 /*
313 * Some request_fn implementations, e.g. scsi_request_fn(), unlock 314 * Some request_fn implementations, e.g. scsi_request_fn(), unlock
314 * the queue lock internally. As a result multiple threads may be 315 * the queue lock internally. As a result multiple threads may be
315 * running such a request function concurrently. Keep track of the 316 * running such a request function concurrently. Keep track of the
316 * number of active request_fn invocations such that blk_drain_queue() 317 * number of active request_fn invocations such that blk_drain_queue()
317 * can wait until all these request_fn calls have finished. 318 * can wait until all these request_fn calls have finished.
318 */ 319 */
319 q->request_fn_active++; 320 q->request_fn_active++;
320 q->request_fn(q); 321 q->request_fn(q);
321 q->request_fn_active--; 322 q->request_fn_active--;
322 } 323 }
323 324
324 /** 325 /**
325 * __blk_run_queue - run a single device queue 326 * __blk_run_queue - run a single device queue
326 * @q: The queue to run 327 * @q: The queue to run
327 * 328 *
328 * Description: 329 * Description:
329 * See @blk_run_queue. This variant must be called with the queue lock 330 * See @blk_run_queue. This variant must be called with the queue lock
330 * held and interrupts disabled. 331 * held and interrupts disabled.
331 */ 332 */
332 void __blk_run_queue(struct request_queue *q) 333 void __blk_run_queue(struct request_queue *q)
333 { 334 {
334 if (unlikely(blk_queue_stopped(q))) 335 if (unlikely(blk_queue_stopped(q)))
335 return; 336 return;
336 337
337 __blk_run_queue_uncond(q); 338 __blk_run_queue_uncond(q);
338 } 339 }
339 EXPORT_SYMBOL(__blk_run_queue); 340 EXPORT_SYMBOL(__blk_run_queue);
340 341
341 /** 342 /**
342 * blk_run_queue_async - run a single device queue in workqueue context 343 * blk_run_queue_async - run a single device queue in workqueue context
343 * @q: The queue to run 344 * @q: The queue to run
344 * 345 *
345 * Description: 346 * Description:
346 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf 347 * Tells kblockd to perform the equivalent of @blk_run_queue on behalf
347 * of us. The caller must hold the queue lock. 348 * of us. The caller must hold the queue lock.
348 */ 349 */
349 void blk_run_queue_async(struct request_queue *q) 350 void blk_run_queue_async(struct request_queue *q)
350 { 351 {
351 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q))) 352 if (likely(!blk_queue_stopped(q) && !blk_queue_dead(q)))
352 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0); 353 mod_delayed_work(kblockd_workqueue, &q->delay_work, 0);
353 } 354 }
354 EXPORT_SYMBOL(blk_run_queue_async); 355 EXPORT_SYMBOL(blk_run_queue_async);
355 356
356 /** 357 /**
357 * blk_run_queue - run a single device queue 358 * blk_run_queue - run a single device queue
358 * @q: The queue to run 359 * @q: The queue to run
359 * 360 *
360 * Description: 361 * Description:
361 * Invoke request handling on this queue, if it has pending work to do. 362 * Invoke request handling on this queue, if it has pending work to do.
362 * May be used to restart queueing when a request has completed. 363 * May be used to restart queueing when a request has completed.
363 */ 364 */
364 void blk_run_queue(struct request_queue *q) 365 void blk_run_queue(struct request_queue *q)
365 { 366 {
366 unsigned long flags; 367 unsigned long flags;
367 368
368 spin_lock_irqsave(q->queue_lock, flags); 369 spin_lock_irqsave(q->queue_lock, flags);
369 __blk_run_queue(q); 370 __blk_run_queue(q);
370 spin_unlock_irqrestore(q->queue_lock, flags); 371 spin_unlock_irqrestore(q->queue_lock, flags);
371 } 372 }
372 EXPORT_SYMBOL(blk_run_queue); 373 EXPORT_SYMBOL(blk_run_queue);
373 374
374 void blk_put_queue(struct request_queue *q) 375 void blk_put_queue(struct request_queue *q)
375 { 376 {
376 kobject_put(&q->kobj); 377 kobject_put(&q->kobj);
377 } 378 }
378 EXPORT_SYMBOL(blk_put_queue); 379 EXPORT_SYMBOL(blk_put_queue);
379 380
380 /** 381 /**
381 * __blk_drain_queue - drain requests from request_queue 382 * __blk_drain_queue - drain requests from request_queue
382 * @q: queue to drain 383 * @q: queue to drain
383 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV 384 * @drain_all: whether to drain all requests or only the ones w/ ELVPRIV
384 * 385 *
385 * Drain requests from @q. If @drain_all is set, all requests are drained. 386 * Drain requests from @q. If @drain_all is set, all requests are drained.
386 * If not, only ELVPRIV requests are drained. The caller is responsible 387 * If not, only ELVPRIV requests are drained. The caller is responsible
387 * for ensuring that no new requests which need to be drained are queued. 388 * for ensuring that no new requests which need to be drained are queued.
388 */ 389 */
389 static void __blk_drain_queue(struct request_queue *q, bool drain_all) 390 static void __blk_drain_queue(struct request_queue *q, bool drain_all)
390 __releases(q->queue_lock) 391 __releases(q->queue_lock)
391 __acquires(q->queue_lock) 392 __acquires(q->queue_lock)
392 { 393 {
393 int i; 394 int i;
394 395
395 lockdep_assert_held(q->queue_lock); 396 lockdep_assert_held(q->queue_lock);
396 397
397 while (true) { 398 while (true) {
398 bool drain = false; 399 bool drain = false;
399 400
400 /* 401 /*
401 * The caller might be trying to drain @q before its 402 * The caller might be trying to drain @q before its
402 * elevator is initialized. 403 * elevator is initialized.
403 */ 404 */
404 if (q->elevator) 405 if (q->elevator)
405 elv_drain_elevator(q); 406 elv_drain_elevator(q);
406 407
407 blkcg_drain_queue(q); 408 blkcg_drain_queue(q);
408 409
409 /* 410 /*
410 * This function might be called on a queue which failed 411 * This function might be called on a queue which failed
411 * driver init after queue creation or is not yet fully 412 * driver init after queue creation or is not yet fully
412 * active yet. Some drivers (e.g. fd and loop) get unhappy 413 * active yet. Some drivers (e.g. fd and loop) get unhappy
413 * in such cases. Kick queue iff dispatch queue has 414 * in such cases. Kick queue iff dispatch queue has
414 * something on it and @q has request_fn set. 415 * something on it and @q has request_fn set.
415 */ 416 */
416 if (!list_empty(&q->queue_head) && q->request_fn) 417 if (!list_empty(&q->queue_head) && q->request_fn)
417 __blk_run_queue(q); 418 __blk_run_queue(q);
418 419
419 drain |= q->nr_rqs_elvpriv; 420 drain |= q->nr_rqs_elvpriv;
420 drain |= q->request_fn_active; 421 drain |= q->request_fn_active;
421 422
422 /* 423 /*
423 * Unfortunately, requests are queued at and tracked from 424 * Unfortunately, requests are queued at and tracked from
424 * multiple places and there's no single counter which can 425 * multiple places and there's no single counter which can
425 * be drained. Check all the queues and counters. 426 * be drained. Check all the queues and counters.
426 */ 427 */
427 if (drain_all) { 428 if (drain_all) {
428 drain |= !list_empty(&q->queue_head); 429 drain |= !list_empty(&q->queue_head);
429 for (i = 0; i < 2; i++) { 430 for (i = 0; i < 2; i++) {
430 drain |= q->nr_rqs[i]; 431 drain |= q->nr_rqs[i];
431 drain |= q->in_flight[i]; 432 drain |= q->in_flight[i];
432 drain |= !list_empty(&q->flush_queue[i]); 433 drain |= !list_empty(&q->flush_queue[i]);
433 } 434 }
434 } 435 }
435 436
436 if (!drain) 437 if (!drain)
437 break; 438 break;
438 439
439 spin_unlock_irq(q->queue_lock); 440 spin_unlock_irq(q->queue_lock);
440 441
441 msleep(10); 442 msleep(10);
442 443
443 spin_lock_irq(q->queue_lock); 444 spin_lock_irq(q->queue_lock);
444 } 445 }
445 446
446 /* 447 /*
447 * With queue marked dead, any woken up waiter will fail the 448 * With queue marked dead, any woken up waiter will fail the
448 * allocation path, so the wakeup chaining is lost and we're 449 * allocation path, so the wakeup chaining is lost and we're
449 * left with hung waiters. We need to wake up those waiters. 450 * left with hung waiters. We need to wake up those waiters.
450 */ 451 */
451 if (q->request_fn) { 452 if (q->request_fn) {
452 struct request_list *rl; 453 struct request_list *rl;
453 454
454 blk_queue_for_each_rl(rl, q) 455 blk_queue_for_each_rl(rl, q)
455 for (i = 0; i < ARRAY_SIZE(rl->wait); i++) 456 for (i = 0; i < ARRAY_SIZE(rl->wait); i++)
456 wake_up_all(&rl->wait[i]); 457 wake_up_all(&rl->wait[i]);
457 } 458 }
458 } 459 }
459 460
460 /** 461 /**
461 * blk_queue_bypass_start - enter queue bypass mode 462 * blk_queue_bypass_start - enter queue bypass mode
462 * @q: queue of interest 463 * @q: queue of interest
463 * 464 *
464 * In bypass mode, only the dispatch FIFO queue of @q is used. This 465 * In bypass mode, only the dispatch FIFO queue of @q is used. This
465 * function makes @q enter bypass mode and drains all requests which were 466 * function makes @q enter bypass mode and drains all requests which were
466 * throttled or issued before. On return, it's guaranteed that no request 467 * throttled or issued before. On return, it's guaranteed that no request
467 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true 468 * is being throttled or has ELVPRIV set and blk_queue_bypass() %true
468 * inside queue or RCU read lock. 469 * inside queue or RCU read lock.
469 */ 470 */
470 void blk_queue_bypass_start(struct request_queue *q) 471 void blk_queue_bypass_start(struct request_queue *q)
471 { 472 {
472 bool drain; 473 bool drain;
473 474
474 spin_lock_irq(q->queue_lock); 475 spin_lock_irq(q->queue_lock);
475 drain = !q->bypass_depth++; 476 drain = !q->bypass_depth++;
476 queue_flag_set(QUEUE_FLAG_BYPASS, q); 477 queue_flag_set(QUEUE_FLAG_BYPASS, q);
477 spin_unlock_irq(q->queue_lock); 478 spin_unlock_irq(q->queue_lock);
478 479
479 if (drain) { 480 if (drain) {
480 spin_lock_irq(q->queue_lock); 481 spin_lock_irq(q->queue_lock);
481 __blk_drain_queue(q, false); 482 __blk_drain_queue(q, false);
482 spin_unlock_irq(q->queue_lock); 483 spin_unlock_irq(q->queue_lock);
483 484
484 /* ensure blk_queue_bypass() is %true inside RCU read lock */ 485 /* ensure blk_queue_bypass() is %true inside RCU read lock */
485 synchronize_rcu(); 486 synchronize_rcu();
486 } 487 }
487 } 488 }
488 EXPORT_SYMBOL_GPL(blk_queue_bypass_start); 489 EXPORT_SYMBOL_GPL(blk_queue_bypass_start);
489 490
490 /** 491 /**
491 * blk_queue_bypass_end - leave queue bypass mode 492 * blk_queue_bypass_end - leave queue bypass mode
492 * @q: queue of interest 493 * @q: queue of interest
493 * 494 *
494 * Leave bypass mode and restore the normal queueing behavior. 495 * Leave bypass mode and restore the normal queueing behavior.
495 */ 496 */
496 void blk_queue_bypass_end(struct request_queue *q) 497 void blk_queue_bypass_end(struct request_queue *q)
497 { 498 {
498 spin_lock_irq(q->queue_lock); 499 spin_lock_irq(q->queue_lock);
499 if (!--q->bypass_depth) 500 if (!--q->bypass_depth)
500 queue_flag_clear(QUEUE_FLAG_BYPASS, q); 501 queue_flag_clear(QUEUE_FLAG_BYPASS, q);
501 WARN_ON_ONCE(q->bypass_depth < 0); 502 WARN_ON_ONCE(q->bypass_depth < 0);
502 spin_unlock_irq(q->queue_lock); 503 spin_unlock_irq(q->queue_lock);
503 } 504 }
504 EXPORT_SYMBOL_GPL(blk_queue_bypass_end); 505 EXPORT_SYMBOL_GPL(blk_queue_bypass_end);
505 506
506 /** 507 /**
507 * blk_cleanup_queue - shutdown a request queue 508 * blk_cleanup_queue - shutdown a request queue
508 * @q: request queue to shutdown 509 * @q: request queue to shutdown
509 * 510 *
510 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and 511 * Mark @q DYING, drain all pending requests, mark @q DEAD, destroy and
511 * put it. All future requests will be failed immediately with -ENODEV. 512 * put it. All future requests will be failed immediately with -ENODEV.
512 */ 513 */
513 void blk_cleanup_queue(struct request_queue *q) 514 void blk_cleanup_queue(struct request_queue *q)
514 { 515 {
515 spinlock_t *lock = q->queue_lock; 516 spinlock_t *lock = q->queue_lock;
516 517
517 /* mark @q DYING, no new request or merges will be allowed afterwards */ 518 /* mark @q DYING, no new request or merges will be allowed afterwards */
518 mutex_lock(&q->sysfs_lock); 519 mutex_lock(&q->sysfs_lock);
519 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q); 520 queue_flag_set_unlocked(QUEUE_FLAG_DYING, q);
520 spin_lock_irq(lock); 521 spin_lock_irq(lock);
521 522
522 /* 523 /*
523 * A dying queue is permanently in bypass mode till released. Note 524 * A dying queue is permanently in bypass mode till released. Note
524 * that, unlike blk_queue_bypass_start(), we aren't performing 525 * that, unlike blk_queue_bypass_start(), we aren't performing
525 * synchronize_rcu() after entering bypass mode to avoid the delay 526 * synchronize_rcu() after entering bypass mode to avoid the delay
526 * as some drivers create and destroy a lot of queues while 527 * as some drivers create and destroy a lot of queues while
527 * probing. This is still safe because blk_release_queue() will be 528 * probing. This is still safe because blk_release_queue() will be
528 * called only after the queue refcnt drops to zero and nothing, 529 * called only after the queue refcnt drops to zero and nothing,
529 * RCU or not, would be traversing the queue by then. 530 * RCU or not, would be traversing the queue by then.
530 */ 531 */
531 q->bypass_depth++; 532 q->bypass_depth++;
532 queue_flag_set(QUEUE_FLAG_BYPASS, q); 533 queue_flag_set(QUEUE_FLAG_BYPASS, q);
533 534
534 queue_flag_set(QUEUE_FLAG_NOMERGES, q); 535 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
535 queue_flag_set(QUEUE_FLAG_NOXMERGES, q); 536 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
536 queue_flag_set(QUEUE_FLAG_DYING, q); 537 queue_flag_set(QUEUE_FLAG_DYING, q);
537 spin_unlock_irq(lock); 538 spin_unlock_irq(lock);
538 mutex_unlock(&q->sysfs_lock); 539 mutex_unlock(&q->sysfs_lock);
539 540
540 /* 541 /*
541 * Drain all requests queued before DYING marking. Set DEAD flag to 542 * Drain all requests queued before DYING marking. Set DEAD flag to
542 * prevent that q->request_fn() gets invoked after draining finished. 543 * prevent that q->request_fn() gets invoked after draining finished.
543 */ 544 */
544 spin_lock_irq(lock); 545 spin_lock_irq(lock);
545 __blk_drain_queue(q, true); 546 __blk_drain_queue(q, true);
546 queue_flag_set(QUEUE_FLAG_DEAD, q); 547 queue_flag_set(QUEUE_FLAG_DEAD, q);
547 spin_unlock_irq(lock); 548 spin_unlock_irq(lock);
548 549
549 /* @q won't process any more request, flush async actions */ 550 /* @q won't process any more request, flush async actions */
550 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer); 551 del_timer_sync(&q->backing_dev_info.laptop_mode_wb_timer);
551 blk_sync_queue(q); 552 blk_sync_queue(q);
552 553
553 spin_lock_irq(lock); 554 spin_lock_irq(lock);
554 if (q->queue_lock != &q->__queue_lock) 555 if (q->queue_lock != &q->__queue_lock)
555 q->queue_lock = &q->__queue_lock; 556 q->queue_lock = &q->__queue_lock;
556 spin_unlock_irq(lock); 557 spin_unlock_irq(lock);
557 558
558 /* @q is and will stay empty, shutdown and put */ 559 /* @q is and will stay empty, shutdown and put */
559 blk_put_queue(q); 560 blk_put_queue(q);
560 } 561 }
561 EXPORT_SYMBOL(blk_cleanup_queue); 562 EXPORT_SYMBOL(blk_cleanup_queue);
562 563
563 int blk_init_rl(struct request_list *rl, struct request_queue *q, 564 int blk_init_rl(struct request_list *rl, struct request_queue *q,
564 gfp_t gfp_mask) 565 gfp_t gfp_mask)
565 { 566 {
566 if (unlikely(rl->rq_pool)) 567 if (unlikely(rl->rq_pool))
567 return 0; 568 return 0;
568 569
569 rl->q = q; 570 rl->q = q;
570 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0; 571 rl->count[BLK_RW_SYNC] = rl->count[BLK_RW_ASYNC] = 0;
571 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0; 572 rl->starved[BLK_RW_SYNC] = rl->starved[BLK_RW_ASYNC] = 0;
572 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]); 573 init_waitqueue_head(&rl->wait[BLK_RW_SYNC]);
573 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]); 574 init_waitqueue_head(&rl->wait[BLK_RW_ASYNC]);
574 575
575 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab, 576 rl->rq_pool = mempool_create_node(BLKDEV_MIN_RQ, mempool_alloc_slab,
576 mempool_free_slab, request_cachep, 577 mempool_free_slab, request_cachep,
577 gfp_mask, q->node); 578 gfp_mask, q->node);
578 if (!rl->rq_pool) 579 if (!rl->rq_pool)
579 return -ENOMEM; 580 return -ENOMEM;
580 581
581 return 0; 582 return 0;
582 } 583 }
583 584
584 void blk_exit_rl(struct request_list *rl) 585 void blk_exit_rl(struct request_list *rl)
585 { 586 {
586 if (rl->rq_pool) 587 if (rl->rq_pool)
587 mempool_destroy(rl->rq_pool); 588 mempool_destroy(rl->rq_pool);
588 } 589 }
589 590
590 struct request_queue *blk_alloc_queue(gfp_t gfp_mask) 591 struct request_queue *blk_alloc_queue(gfp_t gfp_mask)
591 { 592 {
592 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE); 593 return blk_alloc_queue_node(gfp_mask, NUMA_NO_NODE);
593 } 594 }
594 EXPORT_SYMBOL(blk_alloc_queue); 595 EXPORT_SYMBOL(blk_alloc_queue);
595 596
596 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id) 597 struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
597 { 598 {
598 struct request_queue *q; 599 struct request_queue *q;
599 int err; 600 int err;
600 601
601 q = kmem_cache_alloc_node(blk_requestq_cachep, 602 q = kmem_cache_alloc_node(blk_requestq_cachep,
602 gfp_mask | __GFP_ZERO, node_id); 603 gfp_mask | __GFP_ZERO, node_id);
603 if (!q) 604 if (!q)
604 return NULL; 605 return NULL;
605 606
606 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask); 607 q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
607 if (q->id < 0) 608 if (q->id < 0)
608 goto fail_q; 609 goto fail_q;
609 610
610 q->backing_dev_info.ra_pages = 611 q->backing_dev_info.ra_pages =
611 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE; 612 (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
612 q->backing_dev_info.state = 0; 613 q->backing_dev_info.state = 0;
613 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY; 614 q->backing_dev_info.capabilities = BDI_CAP_MAP_COPY;
614 q->backing_dev_info.name = "block"; 615 q->backing_dev_info.name = "block";
615 q->node = node_id; 616 q->node = node_id;
616 617
617 err = bdi_init(&q->backing_dev_info); 618 err = bdi_init(&q->backing_dev_info);
618 if (err) 619 if (err)
619 goto fail_id; 620 goto fail_id;
620 621
621 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer, 622 setup_timer(&q->backing_dev_info.laptop_mode_wb_timer,
622 laptop_mode_timer_fn, (unsigned long) q); 623 laptop_mode_timer_fn, (unsigned long) q);
623 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q); 624 setup_timer(&q->timeout, blk_rq_timed_out_timer, (unsigned long) q);
624 INIT_LIST_HEAD(&q->queue_head); 625 INIT_LIST_HEAD(&q->queue_head);
625 INIT_LIST_HEAD(&q->timeout_list); 626 INIT_LIST_HEAD(&q->timeout_list);
626 INIT_LIST_HEAD(&q->icq_list); 627 INIT_LIST_HEAD(&q->icq_list);
627 #ifdef CONFIG_BLK_CGROUP 628 #ifdef CONFIG_BLK_CGROUP
628 INIT_LIST_HEAD(&q->blkg_list); 629 INIT_LIST_HEAD(&q->blkg_list);
629 #endif 630 #endif
630 INIT_LIST_HEAD(&q->flush_queue[0]); 631 INIT_LIST_HEAD(&q->flush_queue[0]);
631 INIT_LIST_HEAD(&q->flush_queue[1]); 632 INIT_LIST_HEAD(&q->flush_queue[1]);
632 INIT_LIST_HEAD(&q->flush_data_in_flight); 633 INIT_LIST_HEAD(&q->flush_data_in_flight);
633 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work); 634 INIT_DELAYED_WORK(&q->delay_work, blk_delay_work);
634 635
635 kobject_init(&q->kobj, &blk_queue_ktype); 636 kobject_init(&q->kobj, &blk_queue_ktype);
636 637
637 mutex_init(&q->sysfs_lock); 638 mutex_init(&q->sysfs_lock);
638 spin_lock_init(&q->__queue_lock); 639 spin_lock_init(&q->__queue_lock);
639 640
640 /* 641 /*
641 * By default initialize queue_lock to internal lock and driver can 642 * By default initialize queue_lock to internal lock and driver can
642 * override it later if need be. 643 * override it later if need be.
643 */ 644 */
644 q->queue_lock = &q->__queue_lock; 645 q->queue_lock = &q->__queue_lock;
645 646
646 /* 647 /*
647 * A queue starts its life with bypass turned on to avoid 648 * A queue starts its life with bypass turned on to avoid
648 * unnecessary bypass on/off overhead and nasty surprises during 649 * unnecessary bypass on/off overhead and nasty surprises during
649 * init. The initial bypass will be finished when the queue is 650 * init. The initial bypass will be finished when the queue is
650 * registered by blk_register_queue(). 651 * registered by blk_register_queue().
651 */ 652 */
652 q->bypass_depth = 1; 653 q->bypass_depth = 1;
653 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); 654 __set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags);
654 655
655 if (blkcg_init_queue(q)) 656 if (blkcg_init_queue(q))
656 goto fail_id; 657 goto fail_id;
657 658
658 return q; 659 return q;
659 660
660 fail_id: 661 fail_id:
661 ida_simple_remove(&blk_queue_ida, q->id); 662 ida_simple_remove(&blk_queue_ida, q->id);
662 fail_q: 663 fail_q:
663 kmem_cache_free(blk_requestq_cachep, q); 664 kmem_cache_free(blk_requestq_cachep, q);
664 return NULL; 665 return NULL;
665 } 666 }
666 EXPORT_SYMBOL(blk_alloc_queue_node); 667 EXPORT_SYMBOL(blk_alloc_queue_node);
667 668
668 /** 669 /**
669 * blk_init_queue - prepare a request queue for use with a block device 670 * blk_init_queue - prepare a request queue for use with a block device
670 * @rfn: The function to be called to process requests that have been 671 * @rfn: The function to be called to process requests that have been
671 * placed on the queue. 672 * placed on the queue.
672 * @lock: Request queue spin lock 673 * @lock: Request queue spin lock
673 * 674 *
674 * Description: 675 * Description:
675 * If a block device wishes to use the standard request handling procedures, 676 * If a block device wishes to use the standard request handling procedures,
676 * which sorts requests and coalesces adjacent requests, then it must 677 * which sorts requests and coalesces adjacent requests, then it must
677 * call blk_init_queue(). The function @rfn will be called when there 678 * call blk_init_queue(). The function @rfn will be called when there
678 * are requests on the queue that need to be processed. If the device 679 * are requests on the queue that need to be processed. If the device
679 * supports plugging, then @rfn may not be called immediately when requests 680 * supports plugging, then @rfn may not be called immediately when requests
680 * are available on the queue, but may be called at some time later instead. 681 * are available on the queue, but may be called at some time later instead.
681 * Plugged queues are generally unplugged when a buffer belonging to one 682 * Plugged queues are generally unplugged when a buffer belonging to one
682 * of the requests on the queue is needed, or due to memory pressure. 683 * of the requests on the queue is needed, or due to memory pressure.
683 * 684 *
684 * @rfn is not required, or even expected, to remove all requests off the 685 * @rfn is not required, or even expected, to remove all requests off the
685 * queue, but only as many as it can handle at a time. If it does leave 686 * queue, but only as many as it can handle at a time. If it does leave
686 * requests on the queue, it is responsible for arranging that the requests 687 * requests on the queue, it is responsible for arranging that the requests
687 * get dealt with eventually. 688 * get dealt with eventually.
688 * 689 *
689 * The queue spin lock must be held while manipulating the requests on the 690 * The queue spin lock must be held while manipulating the requests on the
690 * request queue; this lock will be taken also from interrupt context, so irq 691 * request queue; this lock will be taken also from interrupt context, so irq
691 * disabling is needed for it. 692 * disabling is needed for it.
692 * 693 *
693 * Function returns a pointer to the initialized request queue, or %NULL if 694 * Function returns a pointer to the initialized request queue, or %NULL if
694 * it didn't succeed. 695 * it didn't succeed.
695 * 696 *
696 * Note: 697 * Note:
697 * blk_init_queue() must be paired with a blk_cleanup_queue() call 698 * blk_init_queue() must be paired with a blk_cleanup_queue() call
698 * when the block device is deactivated (such as at module unload). 699 * when the block device is deactivated (such as at module unload).
699 **/ 700 **/
700 701
701 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock) 702 struct request_queue *blk_init_queue(request_fn_proc *rfn, spinlock_t *lock)
702 { 703 {
703 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE); 704 return blk_init_queue_node(rfn, lock, NUMA_NO_NODE);
704 } 705 }
705 EXPORT_SYMBOL(blk_init_queue); 706 EXPORT_SYMBOL(blk_init_queue);
706 707
707 struct request_queue * 708 struct request_queue *
708 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id) 709 blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
709 { 710 {
710 struct request_queue *uninit_q, *q; 711 struct request_queue *uninit_q, *q;
711 712
712 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id); 713 uninit_q = blk_alloc_queue_node(GFP_KERNEL, node_id);
713 if (!uninit_q) 714 if (!uninit_q)
714 return NULL; 715 return NULL;
715 716
716 q = blk_init_allocated_queue(uninit_q, rfn, lock); 717 q = blk_init_allocated_queue(uninit_q, rfn, lock);
717 if (!q) 718 if (!q)
718 blk_cleanup_queue(uninit_q); 719 blk_cleanup_queue(uninit_q);
719 720
720 return q; 721 return q;
721 } 722 }
722 EXPORT_SYMBOL(blk_init_queue_node); 723 EXPORT_SYMBOL(blk_init_queue_node);
723 724
724 struct request_queue * 725 struct request_queue *
725 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn, 726 blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
726 spinlock_t *lock) 727 spinlock_t *lock)
727 { 728 {
728 if (!q) 729 if (!q)
729 return NULL; 730 return NULL;
730 731
731 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL)) 732 if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
732 return NULL; 733 return NULL;
733 734
734 q->request_fn = rfn; 735 q->request_fn = rfn;
735 q->prep_rq_fn = NULL; 736 q->prep_rq_fn = NULL;
736 q->unprep_rq_fn = NULL; 737 q->unprep_rq_fn = NULL;
737 q->queue_flags |= QUEUE_FLAG_DEFAULT; 738 q->queue_flags |= QUEUE_FLAG_DEFAULT;
738 739
739 /* Override internal queue lock with supplied lock pointer */ 740 /* Override internal queue lock with supplied lock pointer */
740 if (lock) 741 if (lock)
741 q->queue_lock = lock; 742 q->queue_lock = lock;
742 743
743 /* 744 /*
744 * This also sets hw/phys segments, boundary and size 745 * This also sets hw/phys segments, boundary and size
745 */ 746 */
746 blk_queue_make_request(q, blk_queue_bio); 747 blk_queue_make_request(q, blk_queue_bio);
747 748
748 q->sg_reserved_size = INT_MAX; 749 q->sg_reserved_size = INT_MAX;
749 750
750 /* init elevator */ 751 /* init elevator */
751 if (elevator_init(q, NULL)) 752 if (elevator_init(q, NULL))
752 return NULL; 753 return NULL;
753 return q; 754 return q;
754 } 755 }
755 EXPORT_SYMBOL(blk_init_allocated_queue); 756 EXPORT_SYMBOL(blk_init_allocated_queue);
756 757
757 bool blk_get_queue(struct request_queue *q) 758 bool blk_get_queue(struct request_queue *q)
758 { 759 {
759 if (likely(!blk_queue_dying(q))) { 760 if (likely(!blk_queue_dying(q))) {
760 __blk_get_queue(q); 761 __blk_get_queue(q);
761 return true; 762 return true;
762 } 763 }
763 764
764 return false; 765 return false;
765 } 766 }
766 EXPORT_SYMBOL(blk_get_queue); 767 EXPORT_SYMBOL(blk_get_queue);
767 768
768 static inline void blk_free_request(struct request_list *rl, struct request *rq) 769 static inline void blk_free_request(struct request_list *rl, struct request *rq)
769 { 770 {
770 if (rq->cmd_flags & REQ_ELVPRIV) { 771 if (rq->cmd_flags & REQ_ELVPRIV) {
771 elv_put_request(rl->q, rq); 772 elv_put_request(rl->q, rq);
772 if (rq->elv.icq) 773 if (rq->elv.icq)
773 put_io_context(rq->elv.icq->ioc); 774 put_io_context(rq->elv.icq->ioc);
774 } 775 }
775 776
776 mempool_free(rq, rl->rq_pool); 777 mempool_free(rq, rl->rq_pool);
777 } 778 }
778 779
779 /* 780 /*
780 * ioc_batching returns true if the ioc is a valid batching request and 781 * ioc_batching returns true if the ioc is a valid batching request and
781 * should be given priority access to a request. 782 * should be given priority access to a request.
782 */ 783 */
783 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc) 784 static inline int ioc_batching(struct request_queue *q, struct io_context *ioc)
784 { 785 {
785 if (!ioc) 786 if (!ioc)
786 return 0; 787 return 0;
787 788
788 /* 789 /*
789 * Make sure the process is able to allocate at least 1 request 790 * Make sure the process is able to allocate at least 1 request
790 * even if the batch times out, otherwise we could theoretically 791 * even if the batch times out, otherwise we could theoretically
791 * lose wakeups. 792 * lose wakeups.
792 */ 793 */
793 return ioc->nr_batch_requests == q->nr_batching || 794 return ioc->nr_batch_requests == q->nr_batching ||
794 (ioc->nr_batch_requests > 0 795 (ioc->nr_batch_requests > 0
795 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME)); 796 && time_before(jiffies, ioc->last_waited + BLK_BATCH_TIME));
796 } 797 }
797 798
798 /* 799 /*
799 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This 800 * ioc_set_batching sets ioc to be a new "batcher" if it is not one. This
800 * will cause the process to be a "batcher" on all queues in the system. This 801 * will cause the process to be a "batcher" on all queues in the system. This
801 * is the behaviour we want though - once it gets a wakeup it should be given 802 * is the behaviour we want though - once it gets a wakeup it should be given
802 * a nice run. 803 * a nice run.
803 */ 804 */
804 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc) 805 static void ioc_set_batching(struct request_queue *q, struct io_context *ioc)
805 { 806 {
806 if (!ioc || ioc_batching(q, ioc)) 807 if (!ioc || ioc_batching(q, ioc))
807 return; 808 return;
808 809
809 ioc->nr_batch_requests = q->nr_batching; 810 ioc->nr_batch_requests = q->nr_batching;
810 ioc->last_waited = jiffies; 811 ioc->last_waited = jiffies;
811 } 812 }
812 813
813 static void __freed_request(struct request_list *rl, int sync) 814 static void __freed_request(struct request_list *rl, int sync)
814 { 815 {
815 struct request_queue *q = rl->q; 816 struct request_queue *q = rl->q;
816 817
817 /* 818 /*
818 * bdi isn't aware of blkcg yet. As all async IOs end up root 819 * bdi isn't aware of blkcg yet. As all async IOs end up root
819 * blkcg anyway, just use root blkcg state. 820 * blkcg anyway, just use root blkcg state.
820 */ 821 */
821 if (rl == &q->root_rl && 822 if (rl == &q->root_rl &&
822 rl->count[sync] < queue_congestion_off_threshold(q)) 823 rl->count[sync] < queue_congestion_off_threshold(q))
823 blk_clear_queue_congested(q, sync); 824 blk_clear_queue_congested(q, sync);
824 825
825 if (rl->count[sync] + 1 <= q->nr_requests) { 826 if (rl->count[sync] + 1 <= q->nr_requests) {
826 if (waitqueue_active(&rl->wait[sync])) 827 if (waitqueue_active(&rl->wait[sync]))
827 wake_up(&rl->wait[sync]); 828 wake_up(&rl->wait[sync]);
828 829
829 blk_clear_rl_full(rl, sync); 830 blk_clear_rl_full(rl, sync);
830 } 831 }
831 } 832 }
832 833
833 /* 834 /*
834 * A request has just been released. Account for it, update the full and 835 * A request has just been released. Account for it, update the full and
835 * congestion status, wake up any waiters. Called under q->queue_lock. 836 * congestion status, wake up any waiters. Called under q->queue_lock.
836 */ 837 */
837 static void freed_request(struct request_list *rl, unsigned int flags) 838 static void freed_request(struct request_list *rl, unsigned int flags)
838 { 839 {
839 struct request_queue *q = rl->q; 840 struct request_queue *q = rl->q;
840 int sync = rw_is_sync(flags); 841 int sync = rw_is_sync(flags);
841 842
842 q->nr_rqs[sync]--; 843 q->nr_rqs[sync]--;
843 rl->count[sync]--; 844 rl->count[sync]--;
844 if (flags & REQ_ELVPRIV) 845 if (flags & REQ_ELVPRIV)
845 q->nr_rqs_elvpriv--; 846 q->nr_rqs_elvpriv--;
846 847
847 __freed_request(rl, sync); 848 __freed_request(rl, sync);
848 849
849 if (unlikely(rl->starved[sync ^ 1])) 850 if (unlikely(rl->starved[sync ^ 1]))
850 __freed_request(rl, sync ^ 1); 851 __freed_request(rl, sync ^ 1);
851 } 852 }
852 853
853 /* 854 /*
854 * Determine if elevator data should be initialized when allocating the 855 * Determine if elevator data should be initialized when allocating the
855 * request associated with @bio. 856 * request associated with @bio.
856 */ 857 */
857 static bool blk_rq_should_init_elevator(struct bio *bio) 858 static bool blk_rq_should_init_elevator(struct bio *bio)
858 { 859 {
859 if (!bio) 860 if (!bio)
860 return true; 861 return true;
861 862
862 /* 863 /*
863 * Flush requests do not use the elevator so skip initialization. 864 * Flush requests do not use the elevator so skip initialization.
864 * This allows a request to share the flush and elevator data. 865 * This allows a request to share the flush and elevator data.
865 */ 866 */
866 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) 867 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA))
867 return false; 868 return false;
868 869
869 return true; 870 return true;
870 } 871 }
871 872
872 /** 873 /**
873 * rq_ioc - determine io_context for request allocation 874 * rq_ioc - determine io_context for request allocation
874 * @bio: request being allocated is for this bio (can be %NULL) 875 * @bio: request being allocated is for this bio (can be %NULL)
875 * 876 *
876 * Determine io_context to use for request allocation for @bio. May return 877 * Determine io_context to use for request allocation for @bio. May return
877 * %NULL if %current->io_context doesn't exist. 878 * %NULL if %current->io_context doesn't exist.
878 */ 879 */
879 static struct io_context *rq_ioc(struct bio *bio) 880 static struct io_context *rq_ioc(struct bio *bio)
880 { 881 {
881 #ifdef CONFIG_BLK_CGROUP 882 #ifdef CONFIG_BLK_CGROUP
882 if (bio && bio->bi_ioc) 883 if (bio && bio->bi_ioc)
883 return bio->bi_ioc; 884 return bio->bi_ioc;
884 #endif 885 #endif
885 return current->io_context; 886 return current->io_context;
886 } 887 }
887 888
888 /** 889 /**
889 * __get_request - get a free request 890 * __get_request - get a free request
890 * @rl: request list to allocate from 891 * @rl: request list to allocate from
891 * @rw_flags: RW and SYNC flags 892 * @rw_flags: RW and SYNC flags
892 * @bio: bio to allocate request for (can be %NULL) 893 * @bio: bio to allocate request for (can be %NULL)
893 * @gfp_mask: allocation mask 894 * @gfp_mask: allocation mask
894 * 895 *
895 * Get a free request from @q. This function may fail under memory 896 * Get a free request from @q. This function may fail under memory
896 * pressure or if @q is dead. 897 * pressure or if @q is dead.
897 * 898 *
898 * Must be callled with @q->queue_lock held and, 899 * Must be callled with @q->queue_lock held and,
899 * Returns %NULL on failure, with @q->queue_lock held. 900 * Returns %NULL on failure, with @q->queue_lock held.
900 * Returns !%NULL on success, with @q->queue_lock *not held*. 901 * Returns !%NULL on success, with @q->queue_lock *not held*.
901 */ 902 */
902 static struct request *__get_request(struct request_list *rl, int rw_flags, 903 static struct request *__get_request(struct request_list *rl, int rw_flags,
903 struct bio *bio, gfp_t gfp_mask) 904 struct bio *bio, gfp_t gfp_mask)
904 { 905 {
905 struct request_queue *q = rl->q; 906 struct request_queue *q = rl->q;
906 struct request *rq; 907 struct request *rq;
907 struct elevator_type *et = q->elevator->type; 908 struct elevator_type *et = q->elevator->type;
908 struct io_context *ioc = rq_ioc(bio); 909 struct io_context *ioc = rq_ioc(bio);
909 struct io_cq *icq = NULL; 910 struct io_cq *icq = NULL;
910 const bool is_sync = rw_is_sync(rw_flags) != 0; 911 const bool is_sync = rw_is_sync(rw_flags) != 0;
911 int may_queue; 912 int may_queue;
912 913
913 if (unlikely(blk_queue_dying(q))) 914 if (unlikely(blk_queue_dying(q)))
914 return NULL; 915 return NULL;
915 916
916 may_queue = elv_may_queue(q, rw_flags); 917 may_queue = elv_may_queue(q, rw_flags);
917 if (may_queue == ELV_MQUEUE_NO) 918 if (may_queue == ELV_MQUEUE_NO)
918 goto rq_starved; 919 goto rq_starved;
919 920
920 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) { 921 if (rl->count[is_sync]+1 >= queue_congestion_on_threshold(q)) {
921 if (rl->count[is_sync]+1 >= q->nr_requests) { 922 if (rl->count[is_sync]+1 >= q->nr_requests) {
922 /* 923 /*
923 * The queue will fill after this allocation, so set 924 * The queue will fill after this allocation, so set
924 * it as full, and mark this process as "batching". 925 * it as full, and mark this process as "batching".
925 * This process will be allowed to complete a batch of 926 * This process will be allowed to complete a batch of
926 * requests, others will be blocked. 927 * requests, others will be blocked.
927 */ 928 */
928 if (!blk_rl_full(rl, is_sync)) { 929 if (!blk_rl_full(rl, is_sync)) {
929 ioc_set_batching(q, ioc); 930 ioc_set_batching(q, ioc);
930 blk_set_rl_full(rl, is_sync); 931 blk_set_rl_full(rl, is_sync);
931 } else { 932 } else {
932 if (may_queue != ELV_MQUEUE_MUST 933 if (may_queue != ELV_MQUEUE_MUST
933 && !ioc_batching(q, ioc)) { 934 && !ioc_batching(q, ioc)) {
934 /* 935 /*
935 * The queue is full and the allocating 936 * The queue is full and the allocating
936 * process is not a "batcher", and not 937 * process is not a "batcher", and not
937 * exempted by the IO scheduler 938 * exempted by the IO scheduler
938 */ 939 */
939 return NULL; 940 return NULL;
940 } 941 }
941 } 942 }
942 } 943 }
943 /* 944 /*
944 * bdi isn't aware of blkcg yet. As all async IOs end up 945 * bdi isn't aware of blkcg yet. As all async IOs end up
945 * root blkcg anyway, just use root blkcg state. 946 * root blkcg anyway, just use root blkcg state.
946 */ 947 */
947 if (rl == &q->root_rl) 948 if (rl == &q->root_rl)
948 blk_set_queue_congested(q, is_sync); 949 blk_set_queue_congested(q, is_sync);
949 } 950 }
950 951
951 /* 952 /*
952 * Only allow batching queuers to allocate up to 50% over the defined 953 * Only allow batching queuers to allocate up to 50% over the defined
953 * limit of requests, otherwise we could have thousands of requests 954 * limit of requests, otherwise we could have thousands of requests
954 * allocated with any setting of ->nr_requests 955 * allocated with any setting of ->nr_requests
955 */ 956 */
956 if (rl->count[is_sync] >= (3 * q->nr_requests / 2)) 957 if (rl->count[is_sync] >= (3 * q->nr_requests / 2))
957 return NULL; 958 return NULL;
958 959
959 q->nr_rqs[is_sync]++; 960 q->nr_rqs[is_sync]++;
960 rl->count[is_sync]++; 961 rl->count[is_sync]++;
961 rl->starved[is_sync] = 0; 962 rl->starved[is_sync] = 0;
962 963
963 /* 964 /*
964 * Decide whether the new request will be managed by elevator. If 965 * Decide whether the new request will be managed by elevator. If
965 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will 966 * so, mark @rw_flags and increment elvpriv. Non-zero elvpriv will
966 * prevent the current elevator from being destroyed until the new 967 * prevent the current elevator from being destroyed until the new
967 * request is freed. This guarantees icq's won't be destroyed and 968 * request is freed. This guarantees icq's won't be destroyed and
968 * makes creating new ones safe. 969 * makes creating new ones safe.
969 * 970 *
970 * Also, lookup icq while holding queue_lock. If it doesn't exist, 971 * Also, lookup icq while holding queue_lock. If it doesn't exist,
971 * it will be created after releasing queue_lock. 972 * it will be created after releasing queue_lock.
972 */ 973 */
973 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) { 974 if (blk_rq_should_init_elevator(bio) && !blk_queue_bypass(q)) {
974 rw_flags |= REQ_ELVPRIV; 975 rw_flags |= REQ_ELVPRIV;
975 q->nr_rqs_elvpriv++; 976 q->nr_rqs_elvpriv++;
976 if (et->icq_cache && ioc) 977 if (et->icq_cache && ioc)
977 icq = ioc_lookup_icq(ioc, q); 978 icq = ioc_lookup_icq(ioc, q);
978 } 979 }
979 980
980 if (blk_queue_io_stat(q)) 981 if (blk_queue_io_stat(q))
981 rw_flags |= REQ_IO_STAT; 982 rw_flags |= REQ_IO_STAT;
982 spin_unlock_irq(q->queue_lock); 983 spin_unlock_irq(q->queue_lock);
983 984
984 /* allocate and init request */ 985 /* allocate and init request */
985 rq = mempool_alloc(rl->rq_pool, gfp_mask); 986 rq = mempool_alloc(rl->rq_pool, gfp_mask);
986 if (!rq) 987 if (!rq)
987 goto fail_alloc; 988 goto fail_alloc;
988 989
989 blk_rq_init(q, rq); 990 blk_rq_init(q, rq);
990 blk_rq_set_rl(rq, rl); 991 blk_rq_set_rl(rq, rl);
991 rq->cmd_flags = rw_flags | REQ_ALLOCED; 992 rq->cmd_flags = rw_flags | REQ_ALLOCED;
992 993
993 /* init elvpriv */ 994 /* init elvpriv */
994 if (rw_flags & REQ_ELVPRIV) { 995 if (rw_flags & REQ_ELVPRIV) {
995 if (unlikely(et->icq_cache && !icq)) { 996 if (unlikely(et->icq_cache && !icq)) {
996 if (ioc) 997 if (ioc)
997 icq = ioc_create_icq(ioc, q, gfp_mask); 998 icq = ioc_create_icq(ioc, q, gfp_mask);
998 if (!icq) 999 if (!icq)
999 goto fail_elvpriv; 1000 goto fail_elvpriv;
1000 } 1001 }
1001 1002
1002 rq->elv.icq = icq; 1003 rq->elv.icq = icq;
1003 if (unlikely(elv_set_request(q, rq, bio, gfp_mask))) 1004 if (unlikely(elv_set_request(q, rq, bio, gfp_mask)))
1004 goto fail_elvpriv; 1005 goto fail_elvpriv;
1005 1006
1006 /* @rq->elv.icq holds io_context until @rq is freed */ 1007 /* @rq->elv.icq holds io_context until @rq is freed */
1007 if (icq) 1008 if (icq)
1008 get_io_context(icq->ioc); 1009 get_io_context(icq->ioc);
1009 } 1010 }
1010 out: 1011 out:
1011 /* 1012 /*
1012 * ioc may be NULL here, and ioc_batching will be false. That's 1013 * ioc may be NULL here, and ioc_batching will be false. That's
1013 * OK, if the queue is under the request limit then requests need 1014 * OK, if the queue is under the request limit then requests need
1014 * not count toward the nr_batch_requests limit. There will always 1015 * not count toward the nr_batch_requests limit. There will always
1015 * be some limit enforced by BLK_BATCH_TIME. 1016 * be some limit enforced by BLK_BATCH_TIME.
1016 */ 1017 */
1017 if (ioc_batching(q, ioc)) 1018 if (ioc_batching(q, ioc))
1018 ioc->nr_batch_requests--; 1019 ioc->nr_batch_requests--;
1019 1020
1020 trace_block_getrq(q, bio, rw_flags & 1); 1021 trace_block_getrq(q, bio, rw_flags & 1);
1021 return rq; 1022 return rq;
1022 1023
1023 fail_elvpriv: 1024 fail_elvpriv:
1024 /* 1025 /*
1025 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed 1026 * elvpriv init failed. ioc, icq and elvpriv aren't mempool backed
1026 * and may fail indefinitely under memory pressure and thus 1027 * and may fail indefinitely under memory pressure and thus
1027 * shouldn't stall IO. Treat this request as !elvpriv. This will 1028 * shouldn't stall IO. Treat this request as !elvpriv. This will
1028 * disturb iosched and blkcg but weird is bettern than dead. 1029 * disturb iosched and blkcg but weird is bettern than dead.
1029 */ 1030 */
1030 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n", 1031 printk_ratelimited(KERN_WARNING "%s: request aux data allocation failed, iosched may be disturbed\n",
1031 dev_name(q->backing_dev_info.dev)); 1032 dev_name(q->backing_dev_info.dev));
1032 1033
1033 rq->cmd_flags &= ~REQ_ELVPRIV; 1034 rq->cmd_flags &= ~REQ_ELVPRIV;
1034 rq->elv.icq = NULL; 1035 rq->elv.icq = NULL;
1035 1036
1036 spin_lock_irq(q->queue_lock); 1037 spin_lock_irq(q->queue_lock);
1037 q->nr_rqs_elvpriv--; 1038 q->nr_rqs_elvpriv--;
1038 spin_unlock_irq(q->queue_lock); 1039 spin_unlock_irq(q->queue_lock);
1039 goto out; 1040 goto out;
1040 1041
1041 fail_alloc: 1042 fail_alloc:
1042 /* 1043 /*
1043 * Allocation failed presumably due to memory. Undo anything we 1044 * Allocation failed presumably due to memory. Undo anything we
1044 * might have messed up. 1045 * might have messed up.
1045 * 1046 *
1046 * Allocating task should really be put onto the front of the wait 1047 * Allocating task should really be put onto the front of the wait
1047 * queue, but this is pretty rare. 1048 * queue, but this is pretty rare.
1048 */ 1049 */
1049 spin_lock_irq(q->queue_lock); 1050 spin_lock_irq(q->queue_lock);
1050 freed_request(rl, rw_flags); 1051 freed_request(rl, rw_flags);
1051 1052
1052 /* 1053 /*
1053 * in the very unlikely event that allocation failed and no 1054 * in the very unlikely event that allocation failed and no
1054 * requests for this direction was pending, mark us starved so that 1055 * requests for this direction was pending, mark us starved so that
1055 * freeing of a request in the other direction will notice 1056 * freeing of a request in the other direction will notice
1056 * us. another possible fix would be to split the rq mempool into 1057 * us. another possible fix would be to split the rq mempool into
1057 * READ and WRITE 1058 * READ and WRITE
1058 */ 1059 */
1059 rq_starved: 1060 rq_starved:
1060 if (unlikely(rl->count[is_sync] == 0)) 1061 if (unlikely(rl->count[is_sync] == 0))
1061 rl->starved[is_sync] = 1; 1062 rl->starved[is_sync] = 1;
1062 return NULL; 1063 return NULL;
1063 } 1064 }
1064 1065
1065 /** 1066 /**
1066 * get_request - get a free request 1067 * get_request - get a free request
1067 * @q: request_queue to allocate request from 1068 * @q: request_queue to allocate request from
1068 * @rw_flags: RW and SYNC flags 1069 * @rw_flags: RW and SYNC flags
1069 * @bio: bio to allocate request for (can be %NULL) 1070 * @bio: bio to allocate request for (can be %NULL)
1070 * @gfp_mask: allocation mask 1071 * @gfp_mask: allocation mask
1071 * 1072 *
1072 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this 1073 * Get a free request from @q. If %__GFP_WAIT is set in @gfp_mask, this
1073 * function keeps retrying under memory pressure and fails iff @q is dead. 1074 * function keeps retrying under memory pressure and fails iff @q is dead.
1074 * 1075 *
1075 * Must be callled with @q->queue_lock held and, 1076 * Must be callled with @q->queue_lock held and,
1076 * Returns %NULL on failure, with @q->queue_lock held. 1077 * Returns %NULL on failure, with @q->queue_lock held.
1077 * Returns !%NULL on success, with @q->queue_lock *not held*. 1078 * Returns !%NULL on success, with @q->queue_lock *not held*.
1078 */ 1079 */
1079 static struct request *get_request(struct request_queue *q, int rw_flags, 1080 static struct request *get_request(struct request_queue *q, int rw_flags,
1080 struct bio *bio, gfp_t gfp_mask) 1081 struct bio *bio, gfp_t gfp_mask)
1081 { 1082 {
1082 const bool is_sync = rw_is_sync(rw_flags) != 0; 1083 const bool is_sync = rw_is_sync(rw_flags) != 0;
1083 DEFINE_WAIT(wait); 1084 DEFINE_WAIT(wait);
1084 struct request_list *rl; 1085 struct request_list *rl;
1085 struct request *rq; 1086 struct request *rq;
1086 1087
1087 rl = blk_get_rl(q, bio); /* transferred to @rq on success */ 1088 rl = blk_get_rl(q, bio); /* transferred to @rq on success */
1088 retry: 1089 retry:
1089 rq = __get_request(rl, rw_flags, bio, gfp_mask); 1090 rq = __get_request(rl, rw_flags, bio, gfp_mask);
1090 if (rq) 1091 if (rq)
1091 return rq; 1092 return rq;
1092 1093
1093 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) { 1094 if (!(gfp_mask & __GFP_WAIT) || unlikely(blk_queue_dying(q))) {
1094 blk_put_rl(rl); 1095 blk_put_rl(rl);
1095 return NULL; 1096 return NULL;
1096 } 1097 }
1097 1098
1098 /* wait on @rl and retry */ 1099 /* wait on @rl and retry */
1099 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait, 1100 prepare_to_wait_exclusive(&rl->wait[is_sync], &wait,
1100 TASK_UNINTERRUPTIBLE); 1101 TASK_UNINTERRUPTIBLE);
1101 1102
1102 trace_block_sleeprq(q, bio, rw_flags & 1); 1103 trace_block_sleeprq(q, bio, rw_flags & 1);
1103 1104
1104 spin_unlock_irq(q->queue_lock); 1105 spin_unlock_irq(q->queue_lock);
1105 io_schedule(); 1106 io_schedule();
1106 1107
1107 /* 1108 /*
1108 * After sleeping, we become a "batching" process and will be able 1109 * After sleeping, we become a "batching" process and will be able
1109 * to allocate at least one request, and up to a big batch of them 1110 * to allocate at least one request, and up to a big batch of them
1110 * for a small period time. See ioc_batching, ioc_set_batching 1111 * for a small period time. See ioc_batching, ioc_set_batching
1111 */ 1112 */
1112 ioc_set_batching(q, current->io_context); 1113 ioc_set_batching(q, current->io_context);
1113 1114
1114 spin_lock_irq(q->queue_lock); 1115 spin_lock_irq(q->queue_lock);
1115 finish_wait(&rl->wait[is_sync], &wait); 1116 finish_wait(&rl->wait[is_sync], &wait);
1116 1117
1117 goto retry; 1118 goto retry;
1118 } 1119 }
1119 1120
1120 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) 1121 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
1121 { 1122 {
1122 struct request *rq; 1123 struct request *rq;
1123 1124
1124 BUG_ON(rw != READ && rw != WRITE); 1125 BUG_ON(rw != READ && rw != WRITE);
1125 1126
1126 /* create ioc upfront */ 1127 /* create ioc upfront */
1127 create_io_context(gfp_mask, q->node); 1128 create_io_context(gfp_mask, q->node);
1128 1129
1129 spin_lock_irq(q->queue_lock); 1130 spin_lock_irq(q->queue_lock);
1130 rq = get_request(q, rw, NULL, gfp_mask); 1131 rq = get_request(q, rw, NULL, gfp_mask);
1131 if (!rq) 1132 if (!rq)
1132 spin_unlock_irq(q->queue_lock); 1133 spin_unlock_irq(q->queue_lock);
1133 /* q->queue_lock is unlocked at this point */ 1134 /* q->queue_lock is unlocked at this point */
1134 1135
1135 return rq; 1136 return rq;
1136 } 1137 }
1137 EXPORT_SYMBOL(blk_get_request); 1138 EXPORT_SYMBOL(blk_get_request);
1138 1139
1139 /** 1140 /**
1140 * blk_make_request - given a bio, allocate a corresponding struct request. 1141 * blk_make_request - given a bio, allocate a corresponding struct request.
1141 * @q: target request queue 1142 * @q: target request queue
1142 * @bio: The bio describing the memory mappings that will be submitted for IO. 1143 * @bio: The bio describing the memory mappings that will be submitted for IO.
1143 * It may be a chained-bio properly constructed by block/bio layer. 1144 * It may be a chained-bio properly constructed by block/bio layer.
1144 * @gfp_mask: gfp flags to be used for memory allocation 1145 * @gfp_mask: gfp flags to be used for memory allocation
1145 * 1146 *
1146 * blk_make_request is the parallel of generic_make_request for BLOCK_PC 1147 * blk_make_request is the parallel of generic_make_request for BLOCK_PC
1147 * type commands. Where the struct request needs to be farther initialized by 1148 * type commands. Where the struct request needs to be farther initialized by
1148 * the caller. It is passed a &struct bio, which describes the memory info of 1149 * the caller. It is passed a &struct bio, which describes the memory info of
1149 * the I/O transfer. 1150 * the I/O transfer.
1150 * 1151 *
1151 * The caller of blk_make_request must make sure that bi_io_vec 1152 * The caller of blk_make_request must make sure that bi_io_vec
1152 * are set to describe the memory buffers. That bio_data_dir() will return 1153 * are set to describe the memory buffers. That bio_data_dir() will return
1153 * the needed direction of the request. (And all bio's in the passed bio-chain 1154 * the needed direction of the request. (And all bio's in the passed bio-chain
1154 * are properly set accordingly) 1155 * are properly set accordingly)
1155 * 1156 *
1156 * If called under none-sleepable conditions, mapped bio buffers must not 1157 * If called under none-sleepable conditions, mapped bio buffers must not
1157 * need bouncing, by calling the appropriate masked or flagged allocator, 1158 * need bouncing, by calling the appropriate masked or flagged allocator,
1158 * suitable for the target device. Otherwise the call to blk_queue_bounce will 1159 * suitable for the target device. Otherwise the call to blk_queue_bounce will
1159 * BUG. 1160 * BUG.
1160 * 1161 *
1161 * WARNING: When allocating/cloning a bio-chain, careful consideration should be 1162 * WARNING: When allocating/cloning a bio-chain, careful consideration should be
1162 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for 1163 * given to how you allocate bios. In particular, you cannot use __GFP_WAIT for
1163 * anything but the first bio in the chain. Otherwise you risk waiting for IO 1164 * anything but the first bio in the chain. Otherwise you risk waiting for IO
1164 * completion of a bio that hasn't been submitted yet, thus resulting in a 1165 * completion of a bio that hasn't been submitted yet, thus resulting in a
1165 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead 1166 * deadlock. Alternatively bios should be allocated using bio_kmalloc() instead
1166 * of bio_alloc(), as that avoids the mempool deadlock. 1167 * of bio_alloc(), as that avoids the mempool deadlock.
1167 * If possible a big IO should be split into smaller parts when allocation 1168 * If possible a big IO should be split into smaller parts when allocation
1168 * fails. Partial allocation should not be an error, or you risk a live-lock. 1169 * fails. Partial allocation should not be an error, or you risk a live-lock.
1169 */ 1170 */
1170 struct request *blk_make_request(struct request_queue *q, struct bio *bio, 1171 struct request *blk_make_request(struct request_queue *q, struct bio *bio,
1171 gfp_t gfp_mask) 1172 gfp_t gfp_mask)
1172 { 1173 {
1173 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask); 1174 struct request *rq = blk_get_request(q, bio_data_dir(bio), gfp_mask);
1174 1175
1175 if (unlikely(!rq)) 1176 if (unlikely(!rq))
1176 return ERR_PTR(-ENOMEM); 1177 return ERR_PTR(-ENOMEM);
1177 1178
1178 for_each_bio(bio) { 1179 for_each_bio(bio) {
1179 struct bio *bounce_bio = bio; 1180 struct bio *bounce_bio = bio;
1180 int ret; 1181 int ret;
1181 1182
1182 blk_queue_bounce(q, &bounce_bio); 1183 blk_queue_bounce(q, &bounce_bio);
1183 ret = blk_rq_append_bio(q, rq, bounce_bio); 1184 ret = blk_rq_append_bio(q, rq, bounce_bio);
1184 if (unlikely(ret)) { 1185 if (unlikely(ret)) {
1185 blk_put_request(rq); 1186 blk_put_request(rq);
1186 return ERR_PTR(ret); 1187 return ERR_PTR(ret);
1187 } 1188 }
1188 } 1189 }
1189 1190
1190 return rq; 1191 return rq;
1191 } 1192 }
1192 EXPORT_SYMBOL(blk_make_request); 1193 EXPORT_SYMBOL(blk_make_request);
1193 1194
1194 /** 1195 /**
1195 * blk_requeue_request - put a request back on queue 1196 * blk_requeue_request - put a request back on queue
1196 * @q: request queue where request should be inserted 1197 * @q: request queue where request should be inserted
1197 * @rq: request to be inserted 1198 * @rq: request to be inserted
1198 * 1199 *
1199 * Description: 1200 * Description:
1200 * Drivers often keep queueing requests until the hardware cannot accept 1201 * Drivers often keep queueing requests until the hardware cannot accept
1201 * more, when that condition happens we need to put the request back 1202 * more, when that condition happens we need to put the request back
1202 * on the queue. Must be called with queue lock held. 1203 * on the queue. Must be called with queue lock held.
1203 */ 1204 */
1204 void blk_requeue_request(struct request_queue *q, struct request *rq) 1205 void blk_requeue_request(struct request_queue *q, struct request *rq)
1205 { 1206 {
1206 blk_delete_timer(rq); 1207 blk_delete_timer(rq);
1207 blk_clear_rq_complete(rq); 1208 blk_clear_rq_complete(rq);
1208 trace_block_rq_requeue(q, rq); 1209 trace_block_rq_requeue(q, rq);
1209 1210
1210 if (blk_rq_tagged(rq)) 1211 if (blk_rq_tagged(rq))
1211 blk_queue_end_tag(q, rq); 1212 blk_queue_end_tag(q, rq);
1212 1213
1213 BUG_ON(blk_queued_rq(rq)); 1214 BUG_ON(blk_queued_rq(rq));
1214 1215
1215 elv_requeue_request(q, rq); 1216 elv_requeue_request(q, rq);
1216 } 1217 }
1217 EXPORT_SYMBOL(blk_requeue_request); 1218 EXPORT_SYMBOL(blk_requeue_request);
1218 1219
1219 static void add_acct_request(struct request_queue *q, struct request *rq, 1220 static void add_acct_request(struct request_queue *q, struct request *rq,
1220 int where) 1221 int where)
1221 { 1222 {
1222 drive_stat_acct(rq, 1); 1223 drive_stat_acct(rq, 1);
1223 __elv_add_request(q, rq, where); 1224 __elv_add_request(q, rq, where);
1224 } 1225 }
1225 1226
1226 static void part_round_stats_single(int cpu, struct hd_struct *part, 1227 static void part_round_stats_single(int cpu, struct hd_struct *part,
1227 unsigned long now) 1228 unsigned long now)
1228 { 1229 {
1229 if (now == part->stamp) 1230 if (now == part->stamp)
1230 return; 1231 return;
1231 1232
1232 if (part_in_flight(part)) { 1233 if (part_in_flight(part)) {
1233 __part_stat_add(cpu, part, time_in_queue, 1234 __part_stat_add(cpu, part, time_in_queue,
1234 part_in_flight(part) * (now - part->stamp)); 1235 part_in_flight(part) * (now - part->stamp));
1235 __part_stat_add(cpu, part, io_ticks, (now - part->stamp)); 1236 __part_stat_add(cpu, part, io_ticks, (now - part->stamp));
1236 } 1237 }
1237 part->stamp = now; 1238 part->stamp = now;
1238 } 1239 }
1239 1240
1240 /** 1241 /**
1241 * part_round_stats() - Round off the performance stats on a struct disk_stats. 1242 * part_round_stats() - Round off the performance stats on a struct disk_stats.
1242 * @cpu: cpu number for stats access 1243 * @cpu: cpu number for stats access
1243 * @part: target partition 1244 * @part: target partition
1244 * 1245 *
1245 * The average IO queue length and utilisation statistics are maintained 1246 * The average IO queue length and utilisation statistics are maintained
1246 * by observing the current state of the queue length and the amount of 1247 * by observing the current state of the queue length and the amount of
1247 * time it has been in this state for. 1248 * time it has been in this state for.
1248 * 1249 *
1249 * Normally, that accounting is done on IO completion, but that can result 1250 * Normally, that accounting is done on IO completion, but that can result
1250 * in more than a second's worth of IO being accounted for within any one 1251 * in more than a second's worth of IO being accounted for within any one
1251 * second, leading to >100% utilisation. To deal with that, we call this 1252 * second, leading to >100% utilisation. To deal with that, we call this
1252 * function to do a round-off before returning the results when reading 1253 * function to do a round-off before returning the results when reading
1253 * /proc/diskstats. This accounts immediately for all queue usage up to 1254 * /proc/diskstats. This accounts immediately for all queue usage up to
1254 * the current jiffies and restarts the counters again. 1255 * the current jiffies and restarts the counters again.
1255 */ 1256 */
1256 void part_round_stats(int cpu, struct hd_struct *part) 1257 void part_round_stats(int cpu, struct hd_struct *part)
1257 { 1258 {
1258 unsigned long now = jiffies; 1259 unsigned long now = jiffies;
1259 1260
1260 if (part->partno) 1261 if (part->partno)
1261 part_round_stats_single(cpu, &part_to_disk(part)->part0, now); 1262 part_round_stats_single(cpu, &part_to_disk(part)->part0, now);
1262 part_round_stats_single(cpu, part, now); 1263 part_round_stats_single(cpu, part, now);
1263 } 1264 }
1264 EXPORT_SYMBOL_GPL(part_round_stats); 1265 EXPORT_SYMBOL_GPL(part_round_stats);
1265 1266
1266 /* 1267 /*
1267 * queue lock must be held 1268 * queue lock must be held
1268 */ 1269 */
1269 void __blk_put_request(struct request_queue *q, struct request *req) 1270 void __blk_put_request(struct request_queue *q, struct request *req)
1270 { 1271 {
1271 if (unlikely(!q)) 1272 if (unlikely(!q))
1272 return; 1273 return;
1273 if (unlikely(--req->ref_count)) 1274 if (unlikely(--req->ref_count))
1274 return; 1275 return;
1275 1276
1276 elv_completed_request(q, req); 1277 elv_completed_request(q, req);
1277 1278
1278 /* this is a bio leak */ 1279 /* this is a bio leak */
1279 WARN_ON(req->bio != NULL); 1280 WARN_ON(req->bio != NULL);
1280 1281
1281 /* 1282 /*
1282 * Request may not have originated from ll_rw_blk. if not, 1283 * Request may not have originated from ll_rw_blk. if not,
1283 * it didn't come out of our reserved rq pools 1284 * it didn't come out of our reserved rq pools
1284 */ 1285 */
1285 if (req->cmd_flags & REQ_ALLOCED) { 1286 if (req->cmd_flags & REQ_ALLOCED) {
1286 unsigned int flags = req->cmd_flags; 1287 unsigned int flags = req->cmd_flags;
1287 struct request_list *rl = blk_rq_rl(req); 1288 struct request_list *rl = blk_rq_rl(req);
1288 1289
1289 BUG_ON(!list_empty(&req->queuelist)); 1290 BUG_ON(!list_empty(&req->queuelist));
1290 BUG_ON(!hlist_unhashed(&req->hash)); 1291 BUG_ON(!hlist_unhashed(&req->hash));
1291 1292
1292 blk_free_request(rl, req); 1293 blk_free_request(rl, req);
1293 freed_request(rl, flags); 1294 freed_request(rl, flags);
1294 blk_put_rl(rl); 1295 blk_put_rl(rl);
1295 } 1296 }
1296 } 1297 }
1297 EXPORT_SYMBOL_GPL(__blk_put_request); 1298 EXPORT_SYMBOL_GPL(__blk_put_request);
1298 1299
1299 void blk_put_request(struct request *req) 1300 void blk_put_request(struct request *req)
1300 { 1301 {
1301 unsigned long flags; 1302 unsigned long flags;
1302 struct request_queue *q = req->q; 1303 struct request_queue *q = req->q;
1303 1304
1304 spin_lock_irqsave(q->queue_lock, flags); 1305 spin_lock_irqsave(q->queue_lock, flags);
1305 __blk_put_request(q, req); 1306 __blk_put_request(q, req);
1306 spin_unlock_irqrestore(q->queue_lock, flags); 1307 spin_unlock_irqrestore(q->queue_lock, flags);
1307 } 1308 }
1308 EXPORT_SYMBOL(blk_put_request); 1309 EXPORT_SYMBOL(blk_put_request);
1309 1310
1310 /** 1311 /**
1311 * blk_add_request_payload - add a payload to a request 1312 * blk_add_request_payload - add a payload to a request
1312 * @rq: request to update 1313 * @rq: request to update
1313 * @page: page backing the payload 1314 * @page: page backing the payload
1314 * @len: length of the payload. 1315 * @len: length of the payload.
1315 * 1316 *
1316 * This allows to later add a payload to an already submitted request by 1317 * This allows to later add a payload to an already submitted request by
1317 * a block driver. The driver needs to take care of freeing the payload 1318 * a block driver. The driver needs to take care of freeing the payload
1318 * itself. 1319 * itself.
1319 * 1320 *
1320 * Note that this is a quite horrible hack and nothing but handling of 1321 * Note that this is a quite horrible hack and nothing but handling of
1321 * discard requests should ever use it. 1322 * discard requests should ever use it.
1322 */ 1323 */
1323 void blk_add_request_payload(struct request *rq, struct page *page, 1324 void blk_add_request_payload(struct request *rq, struct page *page,
1324 unsigned int len) 1325 unsigned int len)
1325 { 1326 {
1326 struct bio *bio = rq->bio; 1327 struct bio *bio = rq->bio;
1327 1328
1328 bio->bi_io_vec->bv_page = page; 1329 bio->bi_io_vec->bv_page = page;
1329 bio->bi_io_vec->bv_offset = 0; 1330 bio->bi_io_vec->bv_offset = 0;
1330 bio->bi_io_vec->bv_len = len; 1331 bio->bi_io_vec->bv_len = len;
1331 1332
1332 bio->bi_size = len; 1333 bio->bi_size = len;
1333 bio->bi_vcnt = 1; 1334 bio->bi_vcnt = 1;
1334 bio->bi_phys_segments = 1; 1335 bio->bi_phys_segments = 1;
1335 1336
1336 rq->__data_len = rq->resid_len = len; 1337 rq->__data_len = rq->resid_len = len;
1337 rq->nr_phys_segments = 1; 1338 rq->nr_phys_segments = 1;
1338 rq->buffer = bio_data(bio); 1339 rq->buffer = bio_data(bio);
1339 } 1340 }
1340 EXPORT_SYMBOL_GPL(blk_add_request_payload); 1341 EXPORT_SYMBOL_GPL(blk_add_request_payload);
1341 1342
1342 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req, 1343 static bool bio_attempt_back_merge(struct request_queue *q, struct request *req,
1343 struct bio *bio) 1344 struct bio *bio)
1344 { 1345 {
1345 const int ff = bio->bi_rw & REQ_FAILFAST_MASK; 1346 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1346 1347
1347 if (!ll_back_merge_fn(q, req, bio)) 1348 if (!ll_back_merge_fn(q, req, bio))
1348 return false; 1349 return false;
1349 1350
1350 trace_block_bio_backmerge(q, bio); 1351 trace_block_bio_backmerge(q, bio);
1351 1352
1352 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 1353 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1353 blk_rq_set_mixed_merge(req); 1354 blk_rq_set_mixed_merge(req);
1354 1355
1355 req->biotail->bi_next = bio; 1356 req->biotail->bi_next = bio;
1356 req->biotail = bio; 1357 req->biotail = bio;
1357 req->__data_len += bio->bi_size; 1358 req->__data_len += bio->bi_size;
1358 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); 1359 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1359 1360
1360 drive_stat_acct(req, 0); 1361 drive_stat_acct(req, 0);
1361 return true; 1362 return true;
1362 } 1363 }
1363 1364
1364 static bool bio_attempt_front_merge(struct request_queue *q, 1365 static bool bio_attempt_front_merge(struct request_queue *q,
1365 struct request *req, struct bio *bio) 1366 struct request *req, struct bio *bio)
1366 { 1367 {
1367 const int ff = bio->bi_rw & REQ_FAILFAST_MASK; 1368 const int ff = bio->bi_rw & REQ_FAILFAST_MASK;
1368 1369
1369 if (!ll_front_merge_fn(q, req, bio)) 1370 if (!ll_front_merge_fn(q, req, bio))
1370 return false; 1371 return false;
1371 1372
1372 trace_block_bio_frontmerge(q, bio); 1373 trace_block_bio_frontmerge(q, bio);
1373 1374
1374 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff) 1375 if ((req->cmd_flags & REQ_FAILFAST_MASK) != ff)
1375 blk_rq_set_mixed_merge(req); 1376 blk_rq_set_mixed_merge(req);
1376 1377
1377 bio->bi_next = req->bio; 1378 bio->bi_next = req->bio;
1378 req->bio = bio; 1379 req->bio = bio;
1379 1380
1380 /* 1381 /*
1381 * may not be valid. if the low level driver said 1382 * may not be valid. if the low level driver said
1382 * it didn't need a bounce buffer then it better 1383 * it didn't need a bounce buffer then it better
1383 * not touch req->buffer either... 1384 * not touch req->buffer either...
1384 */ 1385 */
1385 req->buffer = bio_data(bio); 1386 req->buffer = bio_data(bio);
1386 req->__sector = bio->bi_sector; 1387 req->__sector = bio->bi_sector;
1387 req->__data_len += bio->bi_size; 1388 req->__data_len += bio->bi_size;
1388 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio)); 1389 req->ioprio = ioprio_best(req->ioprio, bio_prio(bio));
1389 1390
1390 drive_stat_acct(req, 0); 1391 drive_stat_acct(req, 0);
1391 return true; 1392 return true;
1392 } 1393 }
1393 1394
1394 /** 1395 /**
1395 * attempt_plug_merge - try to merge with %current's plugged list 1396 * attempt_plug_merge - try to merge with %current's plugged list
1396 * @q: request_queue new bio is being queued at 1397 * @q: request_queue new bio is being queued at
1397 * @bio: new bio being queued 1398 * @bio: new bio being queued
1398 * @request_count: out parameter for number of traversed plugged requests 1399 * @request_count: out parameter for number of traversed plugged requests
1399 * 1400 *
1400 * Determine whether @bio being queued on @q can be merged with a request 1401 * Determine whether @bio being queued on @q can be merged with a request
1401 * on %current's plugged list. Returns %true if merge was successful, 1402 * on %current's plugged list. Returns %true if merge was successful,
1402 * otherwise %false. 1403 * otherwise %false.
1403 * 1404 *
1404 * Plugging coalesces IOs from the same issuer for the same purpose without 1405 * Plugging coalesces IOs from the same issuer for the same purpose without
1405 * going through @q->queue_lock. As such it's more of an issuing mechanism 1406 * going through @q->queue_lock. As such it's more of an issuing mechanism
1406 * than scheduling, and the request, while may have elvpriv data, is not 1407 * than scheduling, and the request, while may have elvpriv data, is not
1407 * added on the elevator at this point. In addition, we don't have 1408 * added on the elevator at this point. In addition, we don't have
1408 * reliable access to the elevator outside queue lock. Only check basic 1409 * reliable access to the elevator outside queue lock. Only check basic
1409 * merging parameters without querying the elevator. 1410 * merging parameters without querying the elevator.
1410 */ 1411 */
1411 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio, 1412 static bool attempt_plug_merge(struct request_queue *q, struct bio *bio,
1412 unsigned int *request_count) 1413 unsigned int *request_count)
1413 { 1414 {
1414 struct blk_plug *plug; 1415 struct blk_plug *plug;
1415 struct request *rq; 1416 struct request *rq;
1416 bool ret = false; 1417 bool ret = false;
1417 1418
1418 plug = current->plug; 1419 plug = current->plug;
1419 if (!plug) 1420 if (!plug)
1420 goto out; 1421 goto out;
1421 *request_count = 0; 1422 *request_count = 0;
1422 1423
1423 list_for_each_entry_reverse(rq, &plug->list, queuelist) { 1424 list_for_each_entry_reverse(rq, &plug->list, queuelist) {
1424 int el_ret; 1425 int el_ret;
1425 1426
1426 if (rq->q == q) 1427 if (rq->q == q)
1427 (*request_count)++; 1428 (*request_count)++;
1428 1429
1429 if (rq->q != q || !blk_rq_merge_ok(rq, bio)) 1430 if (rq->q != q || !blk_rq_merge_ok(rq, bio))
1430 continue; 1431 continue;
1431 1432
1432 el_ret = blk_try_merge(rq, bio); 1433 el_ret = blk_try_merge(rq, bio);
1433 if (el_ret == ELEVATOR_BACK_MERGE) { 1434 if (el_ret == ELEVATOR_BACK_MERGE) {
1434 ret = bio_attempt_back_merge(q, rq, bio); 1435 ret = bio_attempt_back_merge(q, rq, bio);
1435 if (ret) 1436 if (ret)
1436 break; 1437 break;
1437 } else if (el_ret == ELEVATOR_FRONT_MERGE) { 1438 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1438 ret = bio_attempt_front_merge(q, rq, bio); 1439 ret = bio_attempt_front_merge(q, rq, bio);
1439 if (ret) 1440 if (ret)
1440 break; 1441 break;
1441 } 1442 }
1442 } 1443 }
1443 out: 1444 out:
1444 return ret; 1445 return ret;
1445 } 1446 }
1446 1447
1447 void init_request_from_bio(struct request *req, struct bio *bio) 1448 void init_request_from_bio(struct request *req, struct bio *bio)
1448 { 1449 {
1449 req->cmd_type = REQ_TYPE_FS; 1450 req->cmd_type = REQ_TYPE_FS;
1450 1451
1451 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK; 1452 req->cmd_flags |= bio->bi_rw & REQ_COMMON_MASK;
1452 if (bio->bi_rw & REQ_RAHEAD) 1453 if (bio->bi_rw & REQ_RAHEAD)
1453 req->cmd_flags |= REQ_FAILFAST_MASK; 1454 req->cmd_flags |= REQ_FAILFAST_MASK;
1454 1455
1455 req->errors = 0; 1456 req->errors = 0;
1456 req->__sector = bio->bi_sector; 1457 req->__sector = bio->bi_sector;
1457 req->ioprio = bio_prio(bio); 1458 req->ioprio = bio_prio(bio);
1458 blk_rq_bio_prep(req->q, req, bio); 1459 blk_rq_bio_prep(req->q, req, bio);
1459 } 1460 }
1460 1461
1461 void blk_queue_bio(struct request_queue *q, struct bio *bio) 1462 void blk_queue_bio(struct request_queue *q, struct bio *bio)
1462 { 1463 {
1463 const bool sync = !!(bio->bi_rw & REQ_SYNC); 1464 const bool sync = !!(bio->bi_rw & REQ_SYNC);
1464 struct blk_plug *plug; 1465 struct blk_plug *plug;
1465 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT; 1466 int el_ret, rw_flags, where = ELEVATOR_INSERT_SORT;
1466 struct request *req; 1467 struct request *req;
1467 unsigned int request_count = 0; 1468 unsigned int request_count = 0;
1468 1469
1469 /* 1470 /*
1470 * low level driver can indicate that it wants pages above a 1471 * low level driver can indicate that it wants pages above a
1471 * certain limit bounced to low memory (ie for highmem, or even 1472 * certain limit bounced to low memory (ie for highmem, or even
1472 * ISA dma in theory) 1473 * ISA dma in theory)
1473 */ 1474 */
1474 blk_queue_bounce(q, &bio); 1475 blk_queue_bounce(q, &bio);
1475 1476
1476 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) { 1477 if (bio->bi_rw & (REQ_FLUSH | REQ_FUA)) {
1477 spin_lock_irq(q->queue_lock); 1478 spin_lock_irq(q->queue_lock);
1478 where = ELEVATOR_INSERT_FLUSH; 1479 where = ELEVATOR_INSERT_FLUSH;
1479 goto get_rq; 1480 goto get_rq;
1480 } 1481 }
1481 1482
1482 /* 1483 /*
1483 * Check if we can merge with the plugged list before grabbing 1484 * Check if we can merge with the plugged list before grabbing
1484 * any locks. 1485 * any locks.
1485 */ 1486 */
1486 if (attempt_plug_merge(q, bio, &request_count)) 1487 if (attempt_plug_merge(q, bio, &request_count))
1487 return; 1488 return;
1488 1489
1489 spin_lock_irq(q->queue_lock); 1490 spin_lock_irq(q->queue_lock);
1490 1491
1491 el_ret = elv_merge(q, &req, bio); 1492 el_ret = elv_merge(q, &req, bio);
1492 if (el_ret == ELEVATOR_BACK_MERGE) { 1493 if (el_ret == ELEVATOR_BACK_MERGE) {
1493 if (bio_attempt_back_merge(q, req, bio)) { 1494 if (bio_attempt_back_merge(q, req, bio)) {
1494 elv_bio_merged(q, req, bio); 1495 elv_bio_merged(q, req, bio);
1495 if (!attempt_back_merge(q, req)) 1496 if (!attempt_back_merge(q, req))
1496 elv_merged_request(q, req, el_ret); 1497 elv_merged_request(q, req, el_ret);
1497 goto out_unlock; 1498 goto out_unlock;
1498 } 1499 }
1499 } else if (el_ret == ELEVATOR_FRONT_MERGE) { 1500 } else if (el_ret == ELEVATOR_FRONT_MERGE) {
1500 if (bio_attempt_front_merge(q, req, bio)) { 1501 if (bio_attempt_front_merge(q, req, bio)) {
1501 elv_bio_merged(q, req, bio); 1502 elv_bio_merged(q, req, bio);
1502 if (!attempt_front_merge(q, req)) 1503 if (!attempt_front_merge(q, req))
1503 elv_merged_request(q, req, el_ret); 1504 elv_merged_request(q, req, el_ret);
1504 goto out_unlock; 1505 goto out_unlock;
1505 } 1506 }
1506 } 1507 }
1507 1508
1508 get_rq: 1509 get_rq:
1509 /* 1510 /*
1510 * This sync check and mask will be re-done in init_request_from_bio(), 1511 * This sync check and mask will be re-done in init_request_from_bio(),
1511 * but we need to set it earlier to expose the sync flag to the 1512 * but we need to set it earlier to expose the sync flag to the
1512 * rq allocator and io schedulers. 1513 * rq allocator and io schedulers.
1513 */ 1514 */
1514 rw_flags = bio_data_dir(bio); 1515 rw_flags = bio_data_dir(bio);
1515 if (sync) 1516 if (sync)
1516 rw_flags |= REQ_SYNC; 1517 rw_flags |= REQ_SYNC;
1517 1518
1518 /* 1519 /*
1519 * Grab a free request. This is might sleep but can not fail. 1520 * Grab a free request. This is might sleep but can not fail.
1520 * Returns with the queue unlocked. 1521 * Returns with the queue unlocked.
1521 */ 1522 */
1522 req = get_request(q, rw_flags, bio, GFP_NOIO); 1523 req = get_request(q, rw_flags, bio, GFP_NOIO);
1523 if (unlikely(!req)) { 1524 if (unlikely(!req)) {
1524 bio_endio(bio, -ENODEV); /* @q is dead */ 1525 bio_endio(bio, -ENODEV); /* @q is dead */
1525 goto out_unlock; 1526 goto out_unlock;
1526 } 1527 }
1527 1528
1528 /* 1529 /*
1529 * After dropping the lock and possibly sleeping here, our request 1530 * After dropping the lock and possibly sleeping here, our request
1530 * may now be mergeable after it had proven unmergeable (above). 1531 * may now be mergeable after it had proven unmergeable (above).
1531 * We don't worry about that case for efficiency. It won't happen 1532 * We don't worry about that case for efficiency. It won't happen
1532 * often, and the elevators are able to handle it. 1533 * often, and the elevators are able to handle it.
1533 */ 1534 */
1534 init_request_from_bio(req, bio); 1535 init_request_from_bio(req, bio);
1535 1536
1536 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags)) 1537 if (test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags))
1537 req->cpu = raw_smp_processor_id(); 1538 req->cpu = raw_smp_processor_id();
1538 1539
1539 plug = current->plug; 1540 plug = current->plug;
1540 if (plug) { 1541 if (plug) {
1541 /* 1542 /*
1542 * If this is the first request added after a plug, fire 1543 * If this is the first request added after a plug, fire
1543 * of a plug trace. If others have been added before, check 1544 * of a plug trace. If others have been added before, check
1544 * if we have multiple devices in this plug. If so, make a 1545 * if we have multiple devices in this plug. If so, make a
1545 * note to sort the list before dispatch. 1546 * note to sort the list before dispatch.
1546 */ 1547 */
1547 if (list_empty(&plug->list)) 1548 if (list_empty(&plug->list))
1548 trace_block_plug(q); 1549 trace_block_plug(q);
1549 else { 1550 else {
1550 if (!plug->should_sort) { 1551 if (!plug->should_sort) {
1551 struct request *__rq; 1552 struct request *__rq;
1552 1553
1553 __rq = list_entry_rq(plug->list.prev); 1554 __rq = list_entry_rq(plug->list.prev);
1554 if (__rq->q != q) 1555 if (__rq->q != q)
1555 plug->should_sort = 1; 1556 plug->should_sort = 1;
1556 } 1557 }
1557 if (request_count >= BLK_MAX_REQUEST_COUNT) { 1558 if (request_count >= BLK_MAX_REQUEST_COUNT) {
1558 blk_flush_plug_list(plug, false); 1559 blk_flush_plug_list(plug, false);
1559 trace_block_plug(q); 1560 trace_block_plug(q);
1560 } 1561 }
1561 } 1562 }
1562 list_add_tail(&req->queuelist, &plug->list); 1563 list_add_tail(&req->queuelist, &plug->list);
1563 drive_stat_acct(req, 1); 1564 drive_stat_acct(req, 1);
1564 } else { 1565 } else {
1565 spin_lock_irq(q->queue_lock); 1566 spin_lock_irq(q->queue_lock);
1566 add_acct_request(q, req, where); 1567 add_acct_request(q, req, where);
1567 __blk_run_queue(q); 1568 __blk_run_queue(q);
1568 out_unlock: 1569 out_unlock:
1569 spin_unlock_irq(q->queue_lock); 1570 spin_unlock_irq(q->queue_lock);
1570 } 1571 }
1571 } 1572 }
1572 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */ 1573 EXPORT_SYMBOL_GPL(blk_queue_bio); /* for device mapper only */
1573 1574
1574 /* 1575 /*
1575 * If bio->bi_dev is a partition, remap the location 1576 * If bio->bi_dev is a partition, remap the location
1576 */ 1577 */
1577 static inline void blk_partition_remap(struct bio *bio) 1578 static inline void blk_partition_remap(struct bio *bio)
1578 { 1579 {
1579 struct block_device *bdev = bio->bi_bdev; 1580 struct block_device *bdev = bio->bi_bdev;
1580 1581
1581 if (bio_sectors(bio) && bdev != bdev->bd_contains) { 1582 if (bio_sectors(bio) && bdev != bdev->bd_contains) {
1582 struct hd_struct *p = bdev->bd_part; 1583 struct hd_struct *p = bdev->bd_part;
1583 1584
1584 bio->bi_sector += p->start_sect; 1585 bio->bi_sector += p->start_sect;
1585 bio->bi_bdev = bdev->bd_contains; 1586 bio->bi_bdev = bdev->bd_contains;
1586 1587
1587 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio, 1588 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev), bio,
1588 bdev->bd_dev, 1589 bdev->bd_dev,
1589 bio->bi_sector - p->start_sect); 1590 bio->bi_sector - p->start_sect);
1590 } 1591 }
1591 } 1592 }
1592 1593
1593 static void handle_bad_sector(struct bio *bio) 1594 static void handle_bad_sector(struct bio *bio)
1594 { 1595 {
1595 char b[BDEVNAME_SIZE]; 1596 char b[BDEVNAME_SIZE];
1596 1597
1597 printk(KERN_INFO "attempt to access beyond end of device\n"); 1598 printk(KERN_INFO "attempt to access beyond end of device\n");
1598 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n", 1599 printk(KERN_INFO "%s: rw=%ld, want=%Lu, limit=%Lu\n",
1599 bdevname(bio->bi_bdev, b), 1600 bdevname(bio->bi_bdev, b),
1600 bio->bi_rw, 1601 bio->bi_rw,
1601 (unsigned long long)bio->bi_sector + bio_sectors(bio), 1602 (unsigned long long)bio->bi_sector + bio_sectors(bio),
1602 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); 1603 (long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
1603 1604
1604 set_bit(BIO_EOF, &bio->bi_flags); 1605 set_bit(BIO_EOF, &bio->bi_flags);
1605 } 1606 }
1606 1607
1607 #ifdef CONFIG_FAIL_MAKE_REQUEST 1608 #ifdef CONFIG_FAIL_MAKE_REQUEST
1608 1609
1609 static DECLARE_FAULT_ATTR(fail_make_request); 1610 static DECLARE_FAULT_ATTR(fail_make_request);
1610 1611
1611 static int __init setup_fail_make_request(char *str) 1612 static int __init setup_fail_make_request(char *str)
1612 { 1613 {
1613 return setup_fault_attr(&fail_make_request, str); 1614 return setup_fault_attr(&fail_make_request, str);
1614 } 1615 }
1615 __setup("fail_make_request=", setup_fail_make_request); 1616 __setup("fail_make_request=", setup_fail_make_request);
1616 1617
1617 static bool should_fail_request(struct hd_struct *part, unsigned int bytes) 1618 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
1618 { 1619 {
1619 return part->make_it_fail && should_fail(&fail_make_request, bytes); 1620 return part->make_it_fail && should_fail(&fail_make_request, bytes);
1620 } 1621 }
1621 1622
1622 static int __init fail_make_request_debugfs(void) 1623 static int __init fail_make_request_debugfs(void)
1623 { 1624 {
1624 struct dentry *dir = fault_create_debugfs_attr("fail_make_request", 1625 struct dentry *dir = fault_create_debugfs_attr("fail_make_request",
1625 NULL, &fail_make_request); 1626 NULL, &fail_make_request);
1626 1627
1627 return IS_ERR(dir) ? PTR_ERR(dir) : 0; 1628 return IS_ERR(dir) ? PTR_ERR(dir) : 0;
1628 } 1629 }
1629 1630
1630 late_initcall(fail_make_request_debugfs); 1631 late_initcall(fail_make_request_debugfs);
1631 1632
1632 #else /* CONFIG_FAIL_MAKE_REQUEST */ 1633 #else /* CONFIG_FAIL_MAKE_REQUEST */
1633 1634
1634 static inline bool should_fail_request(struct hd_struct *part, 1635 static inline bool should_fail_request(struct hd_struct *part,
1635 unsigned int bytes) 1636 unsigned int bytes)
1636 { 1637 {
1637 return false; 1638 return false;
1638 } 1639 }
1639 1640
1640 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 1641 #endif /* CONFIG_FAIL_MAKE_REQUEST */
1641 1642
1642 /* 1643 /*
1643 * Check whether this bio extends beyond the end of the device. 1644 * Check whether this bio extends beyond the end of the device.
1644 */ 1645 */
1645 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors) 1646 static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
1646 { 1647 {
1647 sector_t maxsector; 1648 sector_t maxsector;
1648 1649
1649 if (!nr_sectors) 1650 if (!nr_sectors)
1650 return 0; 1651 return 0;
1651 1652
1652 /* Test device or partition size, when known. */ 1653 /* Test device or partition size, when known. */
1653 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; 1654 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
1654 if (maxsector) { 1655 if (maxsector) {
1655 sector_t sector = bio->bi_sector; 1656 sector_t sector = bio->bi_sector;
1656 1657
1657 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) { 1658 if (maxsector < nr_sectors || maxsector - nr_sectors < sector) {
1658 /* 1659 /*
1659 * This may well happen - the kernel calls bread() 1660 * This may well happen - the kernel calls bread()
1660 * without checking the size of the device, e.g., when 1661 * without checking the size of the device, e.g., when
1661 * mounting a device. 1662 * mounting a device.
1662 */ 1663 */
1663 handle_bad_sector(bio); 1664 handle_bad_sector(bio);
1664 return 1; 1665 return 1;
1665 } 1666 }
1666 } 1667 }
1667 1668
1668 return 0; 1669 return 0;
1669 } 1670 }
1670 1671
1671 static noinline_for_stack bool 1672 static noinline_for_stack bool
1672 generic_make_request_checks(struct bio *bio) 1673 generic_make_request_checks(struct bio *bio)
1673 { 1674 {
1674 struct request_queue *q; 1675 struct request_queue *q;
1675 int nr_sectors = bio_sectors(bio); 1676 int nr_sectors = bio_sectors(bio);
1676 int err = -EIO; 1677 int err = -EIO;
1677 char b[BDEVNAME_SIZE]; 1678 char b[BDEVNAME_SIZE];
1678 struct hd_struct *part; 1679 struct hd_struct *part;
1679 1680
1680 might_sleep(); 1681 might_sleep();
1681 1682
1682 if (bio_check_eod(bio, nr_sectors)) 1683 if (bio_check_eod(bio, nr_sectors))
1683 goto end_io; 1684 goto end_io;
1684 1685
1685 q = bdev_get_queue(bio->bi_bdev); 1686 q = bdev_get_queue(bio->bi_bdev);
1686 if (unlikely(!q)) { 1687 if (unlikely(!q)) {
1687 printk(KERN_ERR 1688 printk(KERN_ERR
1688 "generic_make_request: Trying to access " 1689 "generic_make_request: Trying to access "
1689 "nonexistent block-device %s (%Lu)\n", 1690 "nonexistent block-device %s (%Lu)\n",
1690 bdevname(bio->bi_bdev, b), 1691 bdevname(bio->bi_bdev, b),
1691 (long long) bio->bi_sector); 1692 (long long) bio->bi_sector);
1692 goto end_io; 1693 goto end_io;
1693 } 1694 }
1694 1695
1695 if (likely(bio_is_rw(bio) && 1696 if (likely(bio_is_rw(bio) &&
1696 nr_sectors > queue_max_hw_sectors(q))) { 1697 nr_sectors > queue_max_hw_sectors(q))) {
1697 printk(KERN_ERR "bio too big device %s (%u > %u)\n", 1698 printk(KERN_ERR "bio too big device %s (%u > %u)\n",
1698 bdevname(bio->bi_bdev, b), 1699 bdevname(bio->bi_bdev, b),
1699 bio_sectors(bio), 1700 bio_sectors(bio),
1700 queue_max_hw_sectors(q)); 1701 queue_max_hw_sectors(q));
1701 goto end_io; 1702 goto end_io;
1702 } 1703 }
1703 1704
1704 part = bio->bi_bdev->bd_part; 1705 part = bio->bi_bdev->bd_part;
1705 if (should_fail_request(part, bio->bi_size) || 1706 if (should_fail_request(part, bio->bi_size) ||
1706 should_fail_request(&part_to_disk(part)->part0, 1707 should_fail_request(&part_to_disk(part)->part0,
1707 bio->bi_size)) 1708 bio->bi_size))
1708 goto end_io; 1709 goto end_io;
1709 1710
1710 /* 1711 /*
1711 * If this device has partitions, remap block n 1712 * If this device has partitions, remap block n
1712 * of partition p to block n+start(p) of the disk. 1713 * of partition p to block n+start(p) of the disk.
1713 */ 1714 */
1714 blk_partition_remap(bio); 1715 blk_partition_remap(bio);
1715 1716
1716 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) 1717 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio))
1717 goto end_io; 1718 goto end_io;
1718 1719
1719 if (bio_check_eod(bio, nr_sectors)) 1720 if (bio_check_eod(bio, nr_sectors))
1720 goto end_io; 1721 goto end_io;
1721 1722
1722 /* 1723 /*
1723 * Filter flush bio's early so that make_request based 1724 * Filter flush bio's early so that make_request based
1724 * drivers without flush support don't have to worry 1725 * drivers without flush support don't have to worry
1725 * about them. 1726 * about them.
1726 */ 1727 */
1727 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) { 1728 if ((bio->bi_rw & (REQ_FLUSH | REQ_FUA)) && !q->flush_flags) {
1728 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA); 1729 bio->bi_rw &= ~(REQ_FLUSH | REQ_FUA);
1729 if (!nr_sectors) { 1730 if (!nr_sectors) {
1730 err = 0; 1731 err = 0;
1731 goto end_io; 1732 goto end_io;
1732 } 1733 }
1733 } 1734 }
1734 1735
1735 if ((bio->bi_rw & REQ_DISCARD) && 1736 if ((bio->bi_rw & REQ_DISCARD) &&
1736 (!blk_queue_discard(q) || 1737 (!blk_queue_discard(q) ||
1737 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) { 1738 ((bio->bi_rw & REQ_SECURE) && !blk_queue_secdiscard(q)))) {
1738 err = -EOPNOTSUPP; 1739 err = -EOPNOTSUPP;
1739 goto end_io; 1740 goto end_io;
1740 } 1741 }
1741 1742
1742 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) { 1743 if (bio->bi_rw & REQ_WRITE_SAME && !bdev_write_same(bio->bi_bdev)) {
1743 err = -EOPNOTSUPP; 1744 err = -EOPNOTSUPP;
1744 goto end_io; 1745 goto end_io;
1745 } 1746 }
1746 1747
1747 /* 1748 /*
1748 * Various block parts want %current->io_context and lazy ioc 1749 * Various block parts want %current->io_context and lazy ioc
1749 * allocation ends up trading a lot of pain for a small amount of 1750 * allocation ends up trading a lot of pain for a small amount of
1750 * memory. Just allocate it upfront. This may fail and block 1751 * memory. Just allocate it upfront. This may fail and block
1751 * layer knows how to live with it. 1752 * layer knows how to live with it.
1752 */ 1753 */
1753 create_io_context(GFP_ATOMIC, q->node); 1754 create_io_context(GFP_ATOMIC, q->node);
1754 1755
1755 if (blk_throtl_bio(q, bio)) 1756 if (blk_throtl_bio(q, bio))
1756 return false; /* throttled, will be resubmitted later */ 1757 return false; /* throttled, will be resubmitted later */
1757 1758
1758 trace_block_bio_queue(q, bio); 1759 trace_block_bio_queue(q, bio);
1759 return true; 1760 return true;
1760 1761
1761 end_io: 1762 end_io:
1762 bio_endio(bio, err); 1763 bio_endio(bio, err);
1763 return false; 1764 return false;
1764 } 1765 }
1765 1766
1766 /** 1767 /**
1767 * generic_make_request - hand a buffer to its device driver for I/O 1768 * generic_make_request - hand a buffer to its device driver for I/O
1768 * @bio: The bio describing the location in memory and on the device. 1769 * @bio: The bio describing the location in memory and on the device.
1769 * 1770 *
1770 * generic_make_request() is used to make I/O requests of block 1771 * generic_make_request() is used to make I/O requests of block
1771 * devices. It is passed a &struct bio, which describes the I/O that needs 1772 * devices. It is passed a &struct bio, which describes the I/O that needs
1772 * to be done. 1773 * to be done.
1773 * 1774 *
1774 * generic_make_request() does not return any status. The 1775 * generic_make_request() does not return any status. The
1775 * success/failure status of the request, along with notification of 1776 * success/failure status of the request, along with notification of
1776 * completion, is delivered asynchronously through the bio->bi_end_io 1777 * completion, is delivered asynchronously through the bio->bi_end_io
1777 * function described (one day) else where. 1778 * function described (one day) else where.
1778 * 1779 *
1779 * The caller of generic_make_request must make sure that bi_io_vec 1780 * The caller of generic_make_request must make sure that bi_io_vec
1780 * are set to describe the memory buffer, and that bi_dev and bi_sector are 1781 * are set to describe the memory buffer, and that bi_dev and bi_sector are
1781 * set to describe the device address, and the 1782 * set to describe the device address, and the
1782 * bi_end_io and optionally bi_private are set to describe how 1783 * bi_end_io and optionally bi_private are set to describe how
1783 * completion notification should be signaled. 1784 * completion notification should be signaled.
1784 * 1785 *
1785 * generic_make_request and the drivers it calls may use bi_next if this 1786 * generic_make_request and the drivers it calls may use bi_next if this
1786 * bio happens to be merged with someone else, and may resubmit the bio to 1787 * bio happens to be merged with someone else, and may resubmit the bio to
1787 * a lower device by calling into generic_make_request recursively, which 1788 * a lower device by calling into generic_make_request recursively, which
1788 * means the bio should NOT be touched after the call to ->make_request_fn. 1789 * means the bio should NOT be touched after the call to ->make_request_fn.
1789 */ 1790 */
1790 void generic_make_request(struct bio *bio) 1791 void generic_make_request(struct bio *bio)
1791 { 1792 {
1792 struct bio_list bio_list_on_stack; 1793 struct bio_list bio_list_on_stack;
1793 1794
1794 if (!generic_make_request_checks(bio)) 1795 if (!generic_make_request_checks(bio))
1795 return; 1796 return;
1796 1797
1797 /* 1798 /*
1798 * We only want one ->make_request_fn to be active at a time, else 1799 * We only want one ->make_request_fn to be active at a time, else
1799 * stack usage with stacked devices could be a problem. So use 1800 * stack usage with stacked devices could be a problem. So use
1800 * current->bio_list to keep a list of requests submited by a 1801 * current->bio_list to keep a list of requests submited by a
1801 * make_request_fn function. current->bio_list is also used as a 1802 * make_request_fn function. current->bio_list is also used as a
1802 * flag to say if generic_make_request is currently active in this 1803 * flag to say if generic_make_request is currently active in this
1803 * task or not. If it is NULL, then no make_request is active. If 1804 * task or not. If it is NULL, then no make_request is active. If
1804 * it is non-NULL, then a make_request is active, and new requests 1805 * it is non-NULL, then a make_request is active, and new requests
1805 * should be added at the tail 1806 * should be added at the tail
1806 */ 1807 */
1807 if (current->bio_list) { 1808 if (current->bio_list) {
1808 bio_list_add(current->bio_list, bio); 1809 bio_list_add(current->bio_list, bio);
1809 return; 1810 return;
1810 } 1811 }
1811 1812
1812 /* following loop may be a bit non-obvious, and so deserves some 1813 /* following loop may be a bit non-obvious, and so deserves some
1813 * explanation. 1814 * explanation.
1814 * Before entering the loop, bio->bi_next is NULL (as all callers 1815 * Before entering the loop, bio->bi_next is NULL (as all callers
1815 * ensure that) so we have a list with a single bio. 1816 * ensure that) so we have a list with a single bio.
1816 * We pretend that we have just taken it off a longer list, so 1817 * We pretend that we have just taken it off a longer list, so
1817 * we assign bio_list to a pointer to the bio_list_on_stack, 1818 * we assign bio_list to a pointer to the bio_list_on_stack,
1818 * thus initialising the bio_list of new bios to be 1819 * thus initialising the bio_list of new bios to be
1819 * added. ->make_request() may indeed add some more bios 1820 * added. ->make_request() may indeed add some more bios
1820 * through a recursive call to generic_make_request. If it 1821 * through a recursive call to generic_make_request. If it
1821 * did, we find a non-NULL value in bio_list and re-enter the loop 1822 * did, we find a non-NULL value in bio_list and re-enter the loop
1822 * from the top. In this case we really did just take the bio 1823 * from the top. In this case we really did just take the bio
1823 * of the top of the list (no pretending) and so remove it from 1824 * of the top of the list (no pretending) and so remove it from
1824 * bio_list, and call into ->make_request() again. 1825 * bio_list, and call into ->make_request() again.
1825 */ 1826 */
1826 BUG_ON(bio->bi_next); 1827 BUG_ON(bio->bi_next);
1827 bio_list_init(&bio_list_on_stack); 1828 bio_list_init(&bio_list_on_stack);
1828 current->bio_list = &bio_list_on_stack; 1829 current->bio_list = &bio_list_on_stack;
1829 do { 1830 do {
1830 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1831 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1831 1832
1832 q->make_request_fn(q, bio); 1833 q->make_request_fn(q, bio);
1833 1834
1834 bio = bio_list_pop(current->bio_list); 1835 bio = bio_list_pop(current->bio_list);
1835 } while (bio); 1836 } while (bio);
1836 current->bio_list = NULL; /* deactivate */ 1837 current->bio_list = NULL; /* deactivate */
1837 } 1838 }
1838 EXPORT_SYMBOL(generic_make_request); 1839 EXPORT_SYMBOL(generic_make_request);
1839 1840
1840 /** 1841 /**
1841 * submit_bio - submit a bio to the block device layer for I/O 1842 * submit_bio - submit a bio to the block device layer for I/O
1842 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 1843 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
1843 * @bio: The &struct bio which describes the I/O 1844 * @bio: The &struct bio which describes the I/O
1844 * 1845 *
1845 * submit_bio() is very similar in purpose to generic_make_request(), and 1846 * submit_bio() is very similar in purpose to generic_make_request(), and
1846 * uses that function to do most of the work. Both are fairly rough 1847 * uses that function to do most of the work. Both are fairly rough
1847 * interfaces; @bio must be presetup and ready for I/O. 1848 * interfaces; @bio must be presetup and ready for I/O.
1848 * 1849 *
1849 */ 1850 */
1850 void submit_bio(int rw, struct bio *bio) 1851 void submit_bio(int rw, struct bio *bio)
1851 { 1852 {
1852 bio->bi_rw |= rw; 1853 bio->bi_rw |= rw;
1853 1854
1854 /* 1855 /*
1855 * If it's a regular read/write or a barrier with data attached, 1856 * If it's a regular read/write or a barrier with data attached,
1856 * go through the normal accounting stuff before submission. 1857 * go through the normal accounting stuff before submission.
1857 */ 1858 */
1858 if (bio_has_data(bio)) { 1859 if (bio_has_data(bio)) {
1859 unsigned int count; 1860 unsigned int count;
1860 1861
1861 if (unlikely(rw & REQ_WRITE_SAME)) 1862 if (unlikely(rw & REQ_WRITE_SAME))
1862 count = bdev_logical_block_size(bio->bi_bdev) >> 9; 1863 count = bdev_logical_block_size(bio->bi_bdev) >> 9;
1863 else 1864 else
1864 count = bio_sectors(bio); 1865 count = bio_sectors(bio);
1865 1866
1866 if (rw & WRITE) { 1867 if (rw & WRITE) {
1867 count_vm_events(PGPGOUT, count); 1868 count_vm_events(PGPGOUT, count);
1868 } else { 1869 } else {
1869 task_io_account_read(bio->bi_size); 1870 task_io_account_read(bio->bi_size);
1870 count_vm_events(PGPGIN, count); 1871 count_vm_events(PGPGIN, count);
1871 } 1872 }
1872 1873
1873 if (unlikely(block_dump)) { 1874 if (unlikely(block_dump)) {
1874 char b[BDEVNAME_SIZE]; 1875 char b[BDEVNAME_SIZE];
1875 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", 1876 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n",
1876 current->comm, task_pid_nr(current), 1877 current->comm, task_pid_nr(current),
1877 (rw & WRITE) ? "WRITE" : "READ", 1878 (rw & WRITE) ? "WRITE" : "READ",
1878 (unsigned long long)bio->bi_sector, 1879 (unsigned long long)bio->bi_sector,
1879 bdevname(bio->bi_bdev, b), 1880 bdevname(bio->bi_bdev, b),
1880 count); 1881 count);
1881 } 1882 }
1882 } 1883 }
1883 1884
1884 generic_make_request(bio); 1885 generic_make_request(bio);
1885 } 1886 }
1886 EXPORT_SYMBOL(submit_bio); 1887 EXPORT_SYMBOL(submit_bio);
1887 1888
1888 /** 1889 /**
1889 * blk_rq_check_limits - Helper function to check a request for the queue limit 1890 * blk_rq_check_limits - Helper function to check a request for the queue limit
1890 * @q: the queue 1891 * @q: the queue
1891 * @rq: the request being checked 1892 * @rq: the request being checked
1892 * 1893 *
1893 * Description: 1894 * Description:
1894 * @rq may have been made based on weaker limitations of upper-level queues 1895 * @rq may have been made based on weaker limitations of upper-level queues
1895 * in request stacking drivers, and it may violate the limitation of @q. 1896 * in request stacking drivers, and it may violate the limitation of @q.
1896 * Since the block layer and the underlying device driver trust @rq 1897 * Since the block layer and the underlying device driver trust @rq
1897 * after it is inserted to @q, it should be checked against @q before 1898 * after it is inserted to @q, it should be checked against @q before
1898 * the insertion using this generic function. 1899 * the insertion using this generic function.
1899 * 1900 *
1900 * This function should also be useful for request stacking drivers 1901 * This function should also be useful for request stacking drivers
1901 * in some cases below, so export this function. 1902 * in some cases below, so export this function.
1902 * Request stacking drivers like request-based dm may change the queue 1903 * Request stacking drivers like request-based dm may change the queue
1903 * limits while requests are in the queue (e.g. dm's table swapping). 1904 * limits while requests are in the queue (e.g. dm's table swapping).
1904 * Such request stacking drivers should check those requests agaist 1905 * Such request stacking drivers should check those requests agaist
1905 * the new queue limits again when they dispatch those requests, 1906 * the new queue limits again when they dispatch those requests,
1906 * although such checkings are also done against the old queue limits 1907 * although such checkings are also done against the old queue limits
1907 * when submitting requests. 1908 * when submitting requests.
1908 */ 1909 */
1909 int blk_rq_check_limits(struct request_queue *q, struct request *rq) 1910 int blk_rq_check_limits(struct request_queue *q, struct request *rq)
1910 { 1911 {
1911 if (!rq_mergeable(rq)) 1912 if (!rq_mergeable(rq))
1912 return 0; 1913 return 0;
1913 1914
1914 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) { 1915 if (blk_rq_sectors(rq) > blk_queue_get_max_sectors(q, rq->cmd_flags)) {
1915 printk(KERN_ERR "%s: over max size limit.\n", __func__); 1916 printk(KERN_ERR "%s: over max size limit.\n", __func__);
1916 return -EIO; 1917 return -EIO;
1917 } 1918 }
1918 1919
1919 /* 1920 /*
1920 * queue's settings related to segment counting like q->bounce_pfn 1921 * queue's settings related to segment counting like q->bounce_pfn
1921 * may differ from that of other stacking queues. 1922 * may differ from that of other stacking queues.
1922 * Recalculate it to check the request correctly on this queue's 1923 * Recalculate it to check the request correctly on this queue's
1923 * limitation. 1924 * limitation.
1924 */ 1925 */
1925 blk_recalc_rq_segments(rq); 1926 blk_recalc_rq_segments(rq);
1926 if (rq->nr_phys_segments > queue_max_segments(q)) { 1927 if (rq->nr_phys_segments > queue_max_segments(q)) {
1927 printk(KERN_ERR "%s: over max segments limit.\n", __func__); 1928 printk(KERN_ERR "%s: over max segments limit.\n", __func__);
1928 return -EIO; 1929 return -EIO;
1929 } 1930 }
1930 1931
1931 return 0; 1932 return 0;
1932 } 1933 }
1933 EXPORT_SYMBOL_GPL(blk_rq_check_limits); 1934 EXPORT_SYMBOL_GPL(blk_rq_check_limits);
1934 1935
1935 /** 1936 /**
1936 * blk_insert_cloned_request - Helper for stacking drivers to submit a request 1937 * blk_insert_cloned_request - Helper for stacking drivers to submit a request
1937 * @q: the queue to submit the request 1938 * @q: the queue to submit the request
1938 * @rq: the request being queued 1939 * @rq: the request being queued
1939 */ 1940 */
1940 int blk_insert_cloned_request(struct request_queue *q, struct request *rq) 1941 int blk_insert_cloned_request(struct request_queue *q, struct request *rq)
1941 { 1942 {
1942 unsigned long flags; 1943 unsigned long flags;
1943 int where = ELEVATOR_INSERT_BACK; 1944 int where = ELEVATOR_INSERT_BACK;
1944 1945
1945 if (blk_rq_check_limits(q, rq)) 1946 if (blk_rq_check_limits(q, rq))
1946 return -EIO; 1947 return -EIO;
1947 1948
1948 if (rq->rq_disk && 1949 if (rq->rq_disk &&
1949 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq))) 1950 should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
1950 return -EIO; 1951 return -EIO;
1951 1952
1952 spin_lock_irqsave(q->queue_lock, flags); 1953 spin_lock_irqsave(q->queue_lock, flags);
1953 if (unlikely(blk_queue_dying(q))) { 1954 if (unlikely(blk_queue_dying(q))) {
1954 spin_unlock_irqrestore(q->queue_lock, flags); 1955 spin_unlock_irqrestore(q->queue_lock, flags);
1955 return -ENODEV; 1956 return -ENODEV;
1956 } 1957 }
1957 1958
1958 /* 1959 /*
1959 * Submitting request must be dequeued before calling this function 1960 * Submitting request must be dequeued before calling this function
1960 * because it will be linked to another request_queue 1961 * because it will be linked to another request_queue
1961 */ 1962 */
1962 BUG_ON(blk_queued_rq(rq)); 1963 BUG_ON(blk_queued_rq(rq));
1963 1964
1964 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA)) 1965 if (rq->cmd_flags & (REQ_FLUSH|REQ_FUA))
1965 where = ELEVATOR_INSERT_FLUSH; 1966 where = ELEVATOR_INSERT_FLUSH;
1966 1967
1967 add_acct_request(q, rq, where); 1968 add_acct_request(q, rq, where);
1968 if (where == ELEVATOR_INSERT_FLUSH) 1969 if (where == ELEVATOR_INSERT_FLUSH)
1969 __blk_run_queue(q); 1970 __blk_run_queue(q);
1970 spin_unlock_irqrestore(q->queue_lock, flags); 1971 spin_unlock_irqrestore(q->queue_lock, flags);
1971 1972
1972 return 0; 1973 return 0;
1973 } 1974 }
1974 EXPORT_SYMBOL_GPL(blk_insert_cloned_request); 1975 EXPORT_SYMBOL_GPL(blk_insert_cloned_request);
1975 1976
1976 /** 1977 /**
1977 * blk_rq_err_bytes - determine number of bytes till the next failure boundary 1978 * blk_rq_err_bytes - determine number of bytes till the next failure boundary
1978 * @rq: request to examine 1979 * @rq: request to examine
1979 * 1980 *
1980 * Description: 1981 * Description:
1981 * A request could be merge of IOs which require different failure 1982 * A request could be merge of IOs which require different failure
1982 * handling. This function determines the number of bytes which 1983 * handling. This function determines the number of bytes which
1983 * can be failed from the beginning of the request without 1984 * can be failed from the beginning of the request without
1984 * crossing into area which need to be retried further. 1985 * crossing into area which need to be retried further.
1985 * 1986 *
1986 * Return: 1987 * Return:
1987 * The number of bytes to fail. 1988 * The number of bytes to fail.
1988 * 1989 *
1989 * Context: 1990 * Context:
1990 * queue_lock must be held. 1991 * queue_lock must be held.
1991 */ 1992 */
1992 unsigned int blk_rq_err_bytes(const struct request *rq) 1993 unsigned int blk_rq_err_bytes(const struct request *rq)
1993 { 1994 {
1994 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK; 1995 unsigned int ff = rq->cmd_flags & REQ_FAILFAST_MASK;
1995 unsigned int bytes = 0; 1996 unsigned int bytes = 0;
1996 struct bio *bio; 1997 struct bio *bio;
1997 1998
1998 if (!(rq->cmd_flags & REQ_MIXED_MERGE)) 1999 if (!(rq->cmd_flags & REQ_MIXED_MERGE))
1999 return blk_rq_bytes(rq); 2000 return blk_rq_bytes(rq);
2000 2001
2001 /* 2002 /*
2002 * Currently the only 'mixing' which can happen is between 2003 * Currently the only 'mixing' which can happen is between
2003 * different fastfail types. We can safely fail portions 2004 * different fastfail types. We can safely fail portions
2004 * which have all the failfast bits that the first one has - 2005 * which have all the failfast bits that the first one has -
2005 * the ones which are at least as eager to fail as the first 2006 * the ones which are at least as eager to fail as the first
2006 * one. 2007 * one.
2007 */ 2008 */
2008 for (bio = rq->bio; bio; bio = bio->bi_next) { 2009 for (bio = rq->bio; bio; bio = bio->bi_next) {
2009 if ((bio->bi_rw & ff) != ff) 2010 if ((bio->bi_rw & ff) != ff)
2010 break; 2011 break;
2011 bytes += bio->bi_size; 2012 bytes += bio->bi_size;
2012 } 2013 }
2013 2014
2014 /* this could lead to infinite loop */ 2015 /* this could lead to infinite loop */
2015 BUG_ON(blk_rq_bytes(rq) && !bytes); 2016 BUG_ON(blk_rq_bytes(rq) && !bytes);
2016 return bytes; 2017 return bytes;
2017 } 2018 }
2018 EXPORT_SYMBOL_GPL(blk_rq_err_bytes); 2019 EXPORT_SYMBOL_GPL(blk_rq_err_bytes);
2019 2020
2020 static void blk_account_io_completion(struct request *req, unsigned int bytes) 2021 static void blk_account_io_completion(struct request *req, unsigned int bytes)
2021 { 2022 {
2022 if (blk_do_io_stat(req)) { 2023 if (blk_do_io_stat(req)) {
2023 const int rw = rq_data_dir(req); 2024 const int rw = rq_data_dir(req);
2024 struct hd_struct *part; 2025 struct hd_struct *part;
2025 int cpu; 2026 int cpu;
2026 2027
2027 cpu = part_stat_lock(); 2028 cpu = part_stat_lock();
2028 part = req->part; 2029 part = req->part;
2029 part_stat_add(cpu, part, sectors[rw], bytes >> 9); 2030 part_stat_add(cpu, part, sectors[rw], bytes >> 9);
2030 part_stat_unlock(); 2031 part_stat_unlock();
2031 } 2032 }
2032 } 2033 }
2033 2034
2034 static void blk_account_io_done(struct request *req) 2035 static void blk_account_io_done(struct request *req)
2035 { 2036 {
2036 /* 2037 /*
2037 * Account IO completion. flush_rq isn't accounted as a 2038 * Account IO completion. flush_rq isn't accounted as a
2038 * normal IO on queueing nor completion. Accounting the 2039 * normal IO on queueing nor completion. Accounting the
2039 * containing request is enough. 2040 * containing request is enough.
2040 */ 2041 */
2041 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) { 2042 if (blk_do_io_stat(req) && !(req->cmd_flags & REQ_FLUSH_SEQ)) {
2042 unsigned long duration = jiffies - req->start_time; 2043 unsigned long duration = jiffies - req->start_time;
2043 const int rw = rq_data_dir(req); 2044 const int rw = rq_data_dir(req);
2044 struct hd_struct *part; 2045 struct hd_struct *part;
2045 int cpu; 2046 int cpu;
2046 2047
2047 cpu = part_stat_lock(); 2048 cpu = part_stat_lock();
2048 part = req->part; 2049 part = req->part;
2049 2050
2050 part_stat_inc(cpu, part, ios[rw]); 2051 part_stat_inc(cpu, part, ios[rw]);
2051 part_stat_add(cpu, part, ticks[rw], duration); 2052 part_stat_add(cpu, part, ticks[rw], duration);
2052 part_round_stats(cpu, part); 2053 part_round_stats(cpu, part);
2053 part_dec_in_flight(part, rw); 2054 part_dec_in_flight(part, rw);
2054 2055
2055 hd_struct_put(part); 2056 hd_struct_put(part);
2056 part_stat_unlock(); 2057 part_stat_unlock();
2057 } 2058 }
2058 } 2059 }
2059 2060
2060 /** 2061 /**
2061 * blk_peek_request - peek at the top of a request queue 2062 * blk_peek_request - peek at the top of a request queue
2062 * @q: request queue to peek at 2063 * @q: request queue to peek at
2063 * 2064 *
2064 * Description: 2065 * Description:
2065 * Return the request at the top of @q. The returned request 2066 * Return the request at the top of @q. The returned request
2066 * should be started using blk_start_request() before LLD starts 2067 * should be started using blk_start_request() before LLD starts
2067 * processing it. 2068 * processing it.
2068 * 2069 *
2069 * Return: 2070 * Return:
2070 * Pointer to the request at the top of @q if available. Null 2071 * Pointer to the request at the top of @q if available. Null
2071 * otherwise. 2072 * otherwise.
2072 * 2073 *
2073 * Context: 2074 * Context:
2074 * queue_lock must be held. 2075 * queue_lock must be held.
2075 */ 2076 */
2076 struct request *blk_peek_request(struct request_queue *q) 2077 struct request *blk_peek_request(struct request_queue *q)
2077 { 2078 {
2078 struct request *rq; 2079 struct request *rq;
2079 int ret; 2080 int ret;
2080 2081
2081 while ((rq = __elv_next_request(q)) != NULL) { 2082 while ((rq = __elv_next_request(q)) != NULL) {
2082 if (!(rq->cmd_flags & REQ_STARTED)) { 2083 if (!(rq->cmd_flags & REQ_STARTED)) {
2083 /* 2084 /*
2084 * This is the first time the device driver 2085 * This is the first time the device driver
2085 * sees this request (possibly after 2086 * sees this request (possibly after
2086 * requeueing). Notify IO scheduler. 2087 * requeueing). Notify IO scheduler.
2087 */ 2088 */
2088 if (rq->cmd_flags & REQ_SORTED) 2089 if (rq->cmd_flags & REQ_SORTED)
2089 elv_activate_rq(q, rq); 2090 elv_activate_rq(q, rq);
2090 2091
2091 /* 2092 /*
2092 * just mark as started even if we don't start 2093 * just mark as started even if we don't start
2093 * it, a request that has been delayed should 2094 * it, a request that has been delayed should
2094 * not be passed by new incoming requests 2095 * not be passed by new incoming requests
2095 */ 2096 */
2096 rq->cmd_flags |= REQ_STARTED; 2097 rq->cmd_flags |= REQ_STARTED;
2097 trace_block_rq_issue(q, rq); 2098 trace_block_rq_issue(q, rq);
2098 } 2099 }
2099 2100
2100 if (!q->boundary_rq || q->boundary_rq == rq) { 2101 if (!q->boundary_rq || q->boundary_rq == rq) {
2101 q->end_sector = rq_end_sector(rq); 2102 q->end_sector = rq_end_sector(rq);
2102 q->boundary_rq = NULL; 2103 q->boundary_rq = NULL;
2103 } 2104 }
2104 2105
2105 if (rq->cmd_flags & REQ_DONTPREP) 2106 if (rq->cmd_flags & REQ_DONTPREP)
2106 break; 2107 break;
2107 2108
2108 if (q->dma_drain_size && blk_rq_bytes(rq)) { 2109 if (q->dma_drain_size && blk_rq_bytes(rq)) {
2109 /* 2110 /*
2110 * make sure space for the drain appears we 2111 * make sure space for the drain appears we
2111 * know we can do this because max_hw_segments 2112 * know we can do this because max_hw_segments
2112 * has been adjusted to be one fewer than the 2113 * has been adjusted to be one fewer than the
2113 * device can handle 2114 * device can handle
2114 */ 2115 */
2115 rq->nr_phys_segments++; 2116 rq->nr_phys_segments++;
2116 } 2117 }
2117 2118
2118 if (!q->prep_rq_fn) 2119 if (!q->prep_rq_fn)
2119 break; 2120 break;
2120 2121
2121 ret = q->prep_rq_fn(q, rq); 2122 ret = q->prep_rq_fn(q, rq);
2122 if (ret == BLKPREP_OK) { 2123 if (ret == BLKPREP_OK) {
2123 break; 2124 break;
2124 } else if (ret == BLKPREP_DEFER) { 2125 } else if (ret == BLKPREP_DEFER) {
2125 /* 2126 /*
2126 * the request may have been (partially) prepped. 2127 * the request may have been (partially) prepped.
2127 * we need to keep this request in the front to 2128 * we need to keep this request in the front to
2128 * avoid resource deadlock. REQ_STARTED will 2129 * avoid resource deadlock. REQ_STARTED will
2129 * prevent other fs requests from passing this one. 2130 * prevent other fs requests from passing this one.
2130 */ 2131 */
2131 if (q->dma_drain_size && blk_rq_bytes(rq) && 2132 if (q->dma_drain_size && blk_rq_bytes(rq) &&
2132 !(rq->cmd_flags & REQ_DONTPREP)) { 2133 !(rq->cmd_flags & REQ_DONTPREP)) {
2133 /* 2134 /*
2134 * remove the space for the drain we added 2135 * remove the space for the drain we added
2135 * so that we don't add it again 2136 * so that we don't add it again
2136 */ 2137 */
2137 --rq->nr_phys_segments; 2138 --rq->nr_phys_segments;
2138 } 2139 }
2139 2140
2140 rq = NULL; 2141 rq = NULL;
2141 break; 2142 break;
2142 } else if (ret == BLKPREP_KILL) { 2143 } else if (ret == BLKPREP_KILL) {
2143 rq->cmd_flags |= REQ_QUIET; 2144 rq->cmd_flags |= REQ_QUIET;
2144 /* 2145 /*
2145 * Mark this request as started so we don't trigger 2146 * Mark this request as started so we don't trigger
2146 * any debug logic in the end I/O path. 2147 * any debug logic in the end I/O path.
2147 */ 2148 */
2148 blk_start_request(rq); 2149 blk_start_request(rq);
2149 __blk_end_request_all(rq, -EIO); 2150 __blk_end_request_all(rq, -EIO);
2150 } else { 2151 } else {
2151 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); 2152 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
2152 break; 2153 break;
2153 } 2154 }
2154 } 2155 }
2155 2156
2156 return rq; 2157 return rq;
2157 } 2158 }
2158 EXPORT_SYMBOL(blk_peek_request); 2159 EXPORT_SYMBOL(blk_peek_request);
2159 2160
2160 void blk_dequeue_request(struct request *rq) 2161 void blk_dequeue_request(struct request *rq)
2161 { 2162 {
2162 struct request_queue *q = rq->q; 2163 struct request_queue *q = rq->q;
2163 2164
2164 BUG_ON(list_empty(&rq->queuelist)); 2165 BUG_ON(list_empty(&rq->queuelist));
2165 BUG_ON(ELV_ON_HASH(rq)); 2166 BUG_ON(ELV_ON_HASH(rq));
2166 2167
2167 list_del_init(&rq->queuelist); 2168 list_del_init(&rq->queuelist);
2168 2169
2169 /* 2170 /*
2170 * the time frame between a request being removed from the lists 2171 * the time frame between a request being removed from the lists
2171 * and to it is freed is accounted as io that is in progress at 2172 * and to it is freed is accounted as io that is in progress at
2172 * the driver side. 2173 * the driver side.
2173 */ 2174 */
2174 if (blk_account_rq(rq)) { 2175 if (blk_account_rq(rq)) {
2175 q->in_flight[rq_is_sync(rq)]++; 2176 q->in_flight[rq_is_sync(rq)]++;
2176 set_io_start_time_ns(rq); 2177 set_io_start_time_ns(rq);
2177 } 2178 }
2178 } 2179 }
2179 2180
2180 /** 2181 /**
2181 * blk_start_request - start request processing on the driver 2182 * blk_start_request - start request processing on the driver
2182 * @req: request to dequeue 2183 * @req: request to dequeue
2183 * 2184 *
2184 * Description: 2185 * Description:
2185 * Dequeue @req and start timeout timer on it. This hands off the 2186 * Dequeue @req and start timeout timer on it. This hands off the
2186 * request to the driver. 2187 * request to the driver.
2187 * 2188 *
2188 * Block internal functions which don't want to start timer should 2189 * Block internal functions which don't want to start timer should
2189 * call blk_dequeue_request(). 2190 * call blk_dequeue_request().
2190 * 2191 *
2191 * Context: 2192 * Context:
2192 * queue_lock must be held. 2193 * queue_lock must be held.
2193 */ 2194 */
2194 void blk_start_request(struct request *req) 2195 void blk_start_request(struct request *req)
2195 { 2196 {
2196 blk_dequeue_request(req); 2197 blk_dequeue_request(req);
2197 2198
2198 /* 2199 /*
2199 * We are now handing the request to the hardware, initialize 2200 * We are now handing the request to the hardware, initialize
2200 * resid_len to full count and add the timeout handler. 2201 * resid_len to full count and add the timeout handler.
2201 */ 2202 */
2202 req->resid_len = blk_rq_bytes(req); 2203 req->resid_len = blk_rq_bytes(req);
2203 if (unlikely(blk_bidi_rq(req))) 2204 if (unlikely(blk_bidi_rq(req)))
2204 req->next_rq->resid_len = blk_rq_bytes(req->next_rq); 2205 req->next_rq->resid_len = blk_rq_bytes(req->next_rq);
2205 2206
2206 blk_add_timer(req); 2207 blk_add_timer(req);
2207 } 2208 }
2208 EXPORT_SYMBOL(blk_start_request); 2209 EXPORT_SYMBOL(blk_start_request);
2209 2210
2210 /** 2211 /**
2211 * blk_fetch_request - fetch a request from a request queue 2212 * blk_fetch_request - fetch a request from a request queue
2212 * @q: request queue to fetch a request from 2213 * @q: request queue to fetch a request from
2213 * 2214 *
2214 * Description: 2215 * Description:
2215 * Return the request at the top of @q. The request is started on 2216 * Return the request at the top of @q. The request is started on
2216 * return and LLD can start processing it immediately. 2217 * return and LLD can start processing it immediately.
2217 * 2218 *
2218 * Return: 2219 * Return:
2219 * Pointer to the request at the top of @q if available. Null 2220 * Pointer to the request at the top of @q if available. Null
2220 * otherwise. 2221 * otherwise.
2221 * 2222 *
2222 * Context: 2223 * Context:
2223 * queue_lock must be held. 2224 * queue_lock must be held.
2224 */ 2225 */
2225 struct request *blk_fetch_request(struct request_queue *q) 2226 struct request *blk_fetch_request(struct request_queue *q)
2226 { 2227 {
2227 struct request *rq; 2228 struct request *rq;
2228 2229
2229 rq = blk_peek_request(q); 2230 rq = blk_peek_request(q);
2230 if (rq) 2231 if (rq)
2231 blk_start_request(rq); 2232 blk_start_request(rq);
2232 return rq; 2233 return rq;
2233 } 2234 }
2234 EXPORT_SYMBOL(blk_fetch_request); 2235 EXPORT_SYMBOL(blk_fetch_request);
2235 2236
2236 /** 2237 /**
2237 * blk_update_request - Special helper function for request stacking drivers 2238 * blk_update_request - Special helper function for request stacking drivers
2238 * @req: the request being processed 2239 * @req: the request being processed
2239 * @error: %0 for success, < %0 for error 2240 * @error: %0 for success, < %0 for error
2240 * @nr_bytes: number of bytes to complete @req 2241 * @nr_bytes: number of bytes to complete @req
2241 * 2242 *
2242 * Description: 2243 * Description:
2243 * Ends I/O on a number of bytes attached to @req, but doesn't complete 2244 * Ends I/O on a number of bytes attached to @req, but doesn't complete
2244 * the request structure even if @req doesn't have leftover. 2245 * the request structure even if @req doesn't have leftover.
2245 * If @req has leftover, sets it up for the next range of segments. 2246 * If @req has leftover, sets it up for the next range of segments.
2246 * 2247 *
2247 * This special helper function is only for request stacking drivers 2248 * This special helper function is only for request stacking drivers
2248 * (e.g. request-based dm) so that they can handle partial completion. 2249 * (e.g. request-based dm) so that they can handle partial completion.
2249 * Actual device drivers should use blk_end_request instead. 2250 * Actual device drivers should use blk_end_request instead.
2250 * 2251 *
2251 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees 2252 * Passing the result of blk_rq_bytes() as @nr_bytes guarantees
2252 * %false return from this function. 2253 * %false return from this function.
2253 * 2254 *
2254 * Return: 2255 * Return:
2255 * %false - this request doesn't have any more data 2256 * %false - this request doesn't have any more data
2256 * %true - this request has more data 2257 * %true - this request has more data
2257 **/ 2258 **/
2258 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes) 2259 bool blk_update_request(struct request *req, int error, unsigned int nr_bytes)
2259 { 2260 {
2260 int total_bytes, bio_nbytes, next_idx = 0; 2261 int total_bytes, bio_nbytes, next_idx = 0;
2261 struct bio *bio; 2262 struct bio *bio;
2262 2263
2263 if (!req->bio) 2264 if (!req->bio)
2264 return false; 2265 return false;
2265 2266
2266 trace_block_rq_complete(req->q, req); 2267 trace_block_rq_complete(req->q, req);
2267 2268
2268 /* 2269 /*
2269 * For fs requests, rq is just carrier of independent bio's 2270 * For fs requests, rq is just carrier of independent bio's
2270 * and each partial completion should be handled separately. 2271 * and each partial completion should be handled separately.
2271 * Reset per-request error on each partial completion. 2272 * Reset per-request error on each partial completion.
2272 * 2273 *
2273 * TODO: tj: This is too subtle. It would be better to let 2274 * TODO: tj: This is too subtle. It would be better to let
2274 * low level drivers do what they see fit. 2275 * low level drivers do what they see fit.
2275 */ 2276 */
2276 if (req->cmd_type == REQ_TYPE_FS) 2277 if (req->cmd_type == REQ_TYPE_FS)
2277 req->errors = 0; 2278 req->errors = 0;
2278 2279
2279 if (error && req->cmd_type == REQ_TYPE_FS && 2280 if (error && req->cmd_type == REQ_TYPE_FS &&
2280 !(req->cmd_flags & REQ_QUIET)) { 2281 !(req->cmd_flags & REQ_QUIET)) {
2281 char *error_type; 2282 char *error_type;
2282 2283
2283 switch (error) { 2284 switch (error) {
2284 case -ENOLINK: 2285 case -ENOLINK:
2285 error_type = "recoverable transport"; 2286 error_type = "recoverable transport";
2286 break; 2287 break;
2287 case -EREMOTEIO: 2288 case -EREMOTEIO:
2288 error_type = "critical target"; 2289 error_type = "critical target";
2289 break; 2290 break;
2290 case -EBADE: 2291 case -EBADE:
2291 error_type = "critical nexus"; 2292 error_type = "critical nexus";
2292 break; 2293 break;
2293 case -EIO: 2294 case -EIO:
2294 default: 2295 default:
2295 error_type = "I/O"; 2296 error_type = "I/O";
2296 break; 2297 break;
2297 } 2298 }
2298 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n", 2299 printk_ratelimited(KERN_ERR "end_request: %s error, dev %s, sector %llu\n",
2299 error_type, req->rq_disk ? 2300 error_type, req->rq_disk ?
2300 req->rq_disk->disk_name : "?", 2301 req->rq_disk->disk_name : "?",
2301 (unsigned long long)blk_rq_pos(req)); 2302 (unsigned long long)blk_rq_pos(req));
2302 2303
2303 } 2304 }
2304 2305
2305 blk_account_io_completion(req, nr_bytes); 2306 blk_account_io_completion(req, nr_bytes);
2306 2307
2307 total_bytes = bio_nbytes = 0; 2308 total_bytes = bio_nbytes = 0;
2308 while ((bio = req->bio) != NULL) { 2309 while ((bio = req->bio) != NULL) {
2309 int nbytes; 2310 int nbytes;
2310 2311
2311 if (nr_bytes >= bio->bi_size) { 2312 if (nr_bytes >= bio->bi_size) {
2312 req->bio = bio->bi_next; 2313 req->bio = bio->bi_next;
2313 nbytes = bio->bi_size; 2314 nbytes = bio->bi_size;
2314 req_bio_endio(req, bio, nbytes, error); 2315 req_bio_endio(req, bio, nbytes, error);
2315 next_idx = 0; 2316 next_idx = 0;
2316 bio_nbytes = 0; 2317 bio_nbytes = 0;
2317 } else { 2318 } else {
2318 int idx = bio->bi_idx + next_idx; 2319 int idx = bio->bi_idx + next_idx;
2319 2320
2320 if (unlikely(idx >= bio->bi_vcnt)) { 2321 if (unlikely(idx >= bio->bi_vcnt)) {
2321 blk_dump_rq_flags(req, "__end_that"); 2322 blk_dump_rq_flags(req, "__end_that");
2322 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n", 2323 printk(KERN_ERR "%s: bio idx %d >= vcnt %d\n",
2323 __func__, idx, bio->bi_vcnt); 2324 __func__, idx, bio->bi_vcnt);
2324 break; 2325 break;
2325 } 2326 }
2326 2327
2327 nbytes = bio_iovec_idx(bio, idx)->bv_len; 2328 nbytes = bio_iovec_idx(bio, idx)->bv_len;
2328 BIO_BUG_ON(nbytes > bio->bi_size); 2329 BIO_BUG_ON(nbytes > bio->bi_size);
2329 2330
2330 /* 2331 /*
2331 * not a complete bvec done 2332 * not a complete bvec done
2332 */ 2333 */
2333 if (unlikely(nbytes > nr_bytes)) { 2334 if (unlikely(nbytes > nr_bytes)) {
2334 bio_nbytes += nr_bytes; 2335 bio_nbytes += nr_bytes;
2335 total_bytes += nr_bytes; 2336 total_bytes += nr_bytes;
2336 break; 2337 break;
2337 } 2338 }
2338 2339
2339 /* 2340 /*
2340 * advance to the next vector 2341 * advance to the next vector
2341 */ 2342 */
2342 next_idx++; 2343 next_idx++;
2343 bio_nbytes += nbytes; 2344 bio_nbytes += nbytes;
2344 } 2345 }
2345 2346
2346 total_bytes += nbytes; 2347 total_bytes += nbytes;
2347 nr_bytes -= nbytes; 2348 nr_bytes -= nbytes;
2348 2349
2349 bio = req->bio; 2350 bio = req->bio;
2350 if (bio) { 2351 if (bio) {
2351 /* 2352 /*
2352 * end more in this run, or just return 'not-done' 2353 * end more in this run, or just return 'not-done'
2353 */ 2354 */
2354 if (unlikely(nr_bytes <= 0)) 2355 if (unlikely(nr_bytes <= 0))
2355 break; 2356 break;
2356 } 2357 }
2357 } 2358 }
2358 2359
2359 /* 2360 /*
2360 * completely done 2361 * completely done
2361 */ 2362 */
2362 if (!req->bio) { 2363 if (!req->bio) {
2363 /* 2364 /*
2364 * Reset counters so that the request stacking driver 2365 * Reset counters so that the request stacking driver
2365 * can find how many bytes remain in the request 2366 * can find how many bytes remain in the request
2366 * later. 2367 * later.
2367 */ 2368 */
2368 req->__data_len = 0; 2369 req->__data_len = 0;
2369 return false; 2370 return false;
2370 } 2371 }
2371 2372
2372 /* 2373 /*
2373 * if the request wasn't completed, update state 2374 * if the request wasn't completed, update state
2374 */ 2375 */
2375 if (bio_nbytes) { 2376 if (bio_nbytes) {
2376 req_bio_endio(req, bio, bio_nbytes, error); 2377 req_bio_endio(req, bio, bio_nbytes, error);
2377 bio->bi_idx += next_idx; 2378 bio->bi_idx += next_idx;
2378 bio_iovec(bio)->bv_offset += nr_bytes; 2379 bio_iovec(bio)->bv_offset += nr_bytes;
2379 bio_iovec(bio)->bv_len -= nr_bytes; 2380 bio_iovec(bio)->bv_len -= nr_bytes;
2380 } 2381 }
2381 2382
2382 req->__data_len -= total_bytes; 2383 req->__data_len -= total_bytes;
2383 req->buffer = bio_data(req->bio); 2384 req->buffer = bio_data(req->bio);
2384 2385
2385 /* update sector only for requests with clear definition of sector */ 2386 /* update sector only for requests with clear definition of sector */
2386 if (req->cmd_type == REQ_TYPE_FS) 2387 if (req->cmd_type == REQ_TYPE_FS)
2387 req->__sector += total_bytes >> 9; 2388 req->__sector += total_bytes >> 9;
2388 2389
2389 /* mixed attributes always follow the first bio */ 2390 /* mixed attributes always follow the first bio */
2390 if (req->cmd_flags & REQ_MIXED_MERGE) { 2391 if (req->cmd_flags & REQ_MIXED_MERGE) {
2391 req->cmd_flags &= ~REQ_FAILFAST_MASK; 2392 req->cmd_flags &= ~REQ_FAILFAST_MASK;
2392 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK; 2393 req->cmd_flags |= req->bio->bi_rw & REQ_FAILFAST_MASK;
2393 } 2394 }
2394 2395
2395 /* 2396 /*
2396 * If total number of sectors is less than the first segment 2397 * If total number of sectors is less than the first segment
2397 * size, something has gone terribly wrong. 2398 * size, something has gone terribly wrong.
2398 */ 2399 */
2399 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) { 2400 if (blk_rq_bytes(req) < blk_rq_cur_bytes(req)) {
2400 blk_dump_rq_flags(req, "request botched"); 2401 blk_dump_rq_flags(req, "request botched");
2401 req->__data_len = blk_rq_cur_bytes(req); 2402 req->__data_len = blk_rq_cur_bytes(req);
2402 } 2403 }
2403 2404
2404 /* recalculate the number of segments */ 2405 /* recalculate the number of segments */
2405 blk_recalc_rq_segments(req); 2406 blk_recalc_rq_segments(req);
2406 2407
2407 return true; 2408 return true;
2408 } 2409 }
2409 EXPORT_SYMBOL_GPL(blk_update_request); 2410 EXPORT_SYMBOL_GPL(blk_update_request);
2410 2411
2411 static bool blk_update_bidi_request(struct request *rq, int error, 2412 static bool blk_update_bidi_request(struct request *rq, int error,
2412 unsigned int nr_bytes, 2413 unsigned int nr_bytes,
2413 unsigned int bidi_bytes) 2414 unsigned int bidi_bytes)
2414 { 2415 {
2415 if (blk_update_request(rq, error, nr_bytes)) 2416 if (blk_update_request(rq, error, nr_bytes))
2416 return true; 2417 return true;
2417 2418
2418 /* Bidi request must be completed as a whole */ 2419 /* Bidi request must be completed as a whole */
2419 if (unlikely(blk_bidi_rq(rq)) && 2420 if (unlikely(blk_bidi_rq(rq)) &&
2420 blk_update_request(rq->next_rq, error, bidi_bytes)) 2421 blk_update_request(rq->next_rq, error, bidi_bytes))
2421 return true; 2422 return true;
2422 2423
2423 if (blk_queue_add_random(rq->q)) 2424 if (blk_queue_add_random(rq->q))
2424 add_disk_randomness(rq->rq_disk); 2425 add_disk_randomness(rq->rq_disk);
2425 2426
2426 return false; 2427 return false;
2427 } 2428 }
2428 2429
2429 /** 2430 /**
2430 * blk_unprep_request - unprepare a request 2431 * blk_unprep_request - unprepare a request
2431 * @req: the request 2432 * @req: the request
2432 * 2433 *
2433 * This function makes a request ready for complete resubmission (or 2434 * This function makes a request ready for complete resubmission (or
2434 * completion). It happens only after all error handling is complete, 2435 * completion). It happens only after all error handling is complete,
2435 * so represents the appropriate moment to deallocate any resources 2436 * so represents the appropriate moment to deallocate any resources
2436 * that were allocated to the request in the prep_rq_fn. The queue 2437 * that were allocated to the request in the prep_rq_fn. The queue
2437 * lock is held when calling this. 2438 * lock is held when calling this.
2438 */ 2439 */
2439 void blk_unprep_request(struct request *req) 2440 void blk_unprep_request(struct request *req)
2440 { 2441 {
2441 struct request_queue *q = req->q; 2442 struct request_queue *q = req->q;
2442 2443
2443 req->cmd_flags &= ~REQ_DONTPREP; 2444 req->cmd_flags &= ~REQ_DONTPREP;
2444 if (q->unprep_rq_fn) 2445 if (q->unprep_rq_fn)
2445 q->unprep_rq_fn(q, req); 2446 q->unprep_rq_fn(q, req);
2446 } 2447 }
2447 EXPORT_SYMBOL_GPL(blk_unprep_request); 2448 EXPORT_SYMBOL_GPL(blk_unprep_request);
2448 2449
2449 /* 2450 /*
2450 * queue lock must be held 2451 * queue lock must be held
2451 */ 2452 */
2452 static void blk_finish_request(struct request *req, int error) 2453 static void blk_finish_request(struct request *req, int error)
2453 { 2454 {
2454 if (blk_rq_tagged(req)) 2455 if (blk_rq_tagged(req))
2455 blk_queue_end_tag(req->q, req); 2456 blk_queue_end_tag(req->q, req);
2456 2457
2457 BUG_ON(blk_queued_rq(req)); 2458 BUG_ON(blk_queued_rq(req));
2458 2459
2459 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS) 2460 if (unlikely(laptop_mode) && req->cmd_type == REQ_TYPE_FS)
2460 laptop_io_completion(&req->q->backing_dev_info); 2461 laptop_io_completion(&req->q->backing_dev_info);
2461 2462
2462 blk_delete_timer(req); 2463 blk_delete_timer(req);
2463 2464
2464 if (req->cmd_flags & REQ_DONTPREP) 2465 if (req->cmd_flags & REQ_DONTPREP)
2465 blk_unprep_request(req); 2466 blk_unprep_request(req);
2466 2467
2467 2468
2468 blk_account_io_done(req); 2469 blk_account_io_done(req);
2469 2470
2470 if (req->end_io) 2471 if (req->end_io)
2471 req->end_io(req, error); 2472 req->end_io(req, error);
2472 else { 2473 else {
2473 if (blk_bidi_rq(req)) 2474 if (blk_bidi_rq(req))
2474 __blk_put_request(req->next_rq->q, req->next_rq); 2475 __blk_put_request(req->next_rq->q, req->next_rq);
2475 2476
2476 __blk_put_request(req->q, req); 2477 __blk_put_request(req->q, req);
2477 } 2478 }
2478 } 2479 }
2479 2480
2480 /** 2481 /**
2481 * blk_end_bidi_request - Complete a bidi request 2482 * blk_end_bidi_request - Complete a bidi request
2482 * @rq: the request to complete 2483 * @rq: the request to complete
2483 * @error: %0 for success, < %0 for error 2484 * @error: %0 for success, < %0 for error
2484 * @nr_bytes: number of bytes to complete @rq 2485 * @nr_bytes: number of bytes to complete @rq
2485 * @bidi_bytes: number of bytes to complete @rq->next_rq 2486 * @bidi_bytes: number of bytes to complete @rq->next_rq
2486 * 2487 *
2487 * Description: 2488 * Description:
2488 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq. 2489 * Ends I/O on a number of bytes attached to @rq and @rq->next_rq.
2489 * Drivers that supports bidi can safely call this member for any 2490 * Drivers that supports bidi can safely call this member for any
2490 * type of request, bidi or uni. In the later case @bidi_bytes is 2491 * type of request, bidi or uni. In the later case @bidi_bytes is
2491 * just ignored. 2492 * just ignored.
2492 * 2493 *
2493 * Return: 2494 * Return:
2494 * %false - we are done with this request 2495 * %false - we are done with this request
2495 * %true - still buffers pending for this request 2496 * %true - still buffers pending for this request
2496 **/ 2497 **/
2497 static bool blk_end_bidi_request(struct request *rq, int error, 2498 static bool blk_end_bidi_request(struct request *rq, int error,
2498 unsigned int nr_bytes, unsigned int bidi_bytes) 2499 unsigned int nr_bytes, unsigned int bidi_bytes)
2499 { 2500 {
2500 struct request_queue *q = rq->q; 2501 struct request_queue *q = rq->q;
2501 unsigned long flags; 2502 unsigned long flags;
2502 2503
2503 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2504 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2504 return true; 2505 return true;
2505 2506
2506 spin_lock_irqsave(q->queue_lock, flags); 2507 spin_lock_irqsave(q->queue_lock, flags);
2507 blk_finish_request(rq, error); 2508 blk_finish_request(rq, error);
2508 spin_unlock_irqrestore(q->queue_lock, flags); 2509 spin_unlock_irqrestore(q->queue_lock, flags);
2509 2510
2510 return false; 2511 return false;
2511 } 2512 }
2512 2513
2513 /** 2514 /**
2514 * __blk_end_bidi_request - Complete a bidi request with queue lock held 2515 * __blk_end_bidi_request - Complete a bidi request with queue lock held
2515 * @rq: the request to complete 2516 * @rq: the request to complete
2516 * @error: %0 for success, < %0 for error 2517 * @error: %0 for success, < %0 for error
2517 * @nr_bytes: number of bytes to complete @rq 2518 * @nr_bytes: number of bytes to complete @rq
2518 * @bidi_bytes: number of bytes to complete @rq->next_rq 2519 * @bidi_bytes: number of bytes to complete @rq->next_rq
2519 * 2520 *
2520 * Description: 2521 * Description:
2521 * Identical to blk_end_bidi_request() except that queue lock is 2522 * Identical to blk_end_bidi_request() except that queue lock is
2522 * assumed to be locked on entry and remains so on return. 2523 * assumed to be locked on entry and remains so on return.
2523 * 2524 *
2524 * Return: 2525 * Return:
2525 * %false - we are done with this request 2526 * %false - we are done with this request
2526 * %true - still buffers pending for this request 2527 * %true - still buffers pending for this request
2527 **/ 2528 **/
2528 bool __blk_end_bidi_request(struct request *rq, int error, 2529 bool __blk_end_bidi_request(struct request *rq, int error,
2529 unsigned int nr_bytes, unsigned int bidi_bytes) 2530 unsigned int nr_bytes, unsigned int bidi_bytes)
2530 { 2531 {
2531 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes)) 2532 if (blk_update_bidi_request(rq, error, nr_bytes, bidi_bytes))
2532 return true; 2533 return true;
2533 2534
2534 blk_finish_request(rq, error); 2535 blk_finish_request(rq, error);
2535 2536
2536 return false; 2537 return false;
2537 } 2538 }
2538 2539
2539 /** 2540 /**
2540 * blk_end_request - Helper function for drivers to complete the request. 2541 * blk_end_request - Helper function for drivers to complete the request.
2541 * @rq: the request being processed 2542 * @rq: the request being processed
2542 * @error: %0 for success, < %0 for error 2543 * @error: %0 for success, < %0 for error
2543 * @nr_bytes: number of bytes to complete 2544 * @nr_bytes: number of bytes to complete
2544 * 2545 *
2545 * Description: 2546 * Description:
2546 * Ends I/O on a number of bytes attached to @rq. 2547 * Ends I/O on a number of bytes attached to @rq.
2547 * If @rq has leftover, sets it up for the next range of segments. 2548 * If @rq has leftover, sets it up for the next range of segments.
2548 * 2549 *
2549 * Return: 2550 * Return:
2550 * %false - we are done with this request 2551 * %false - we are done with this request
2551 * %true - still buffers pending for this request 2552 * %true - still buffers pending for this request
2552 **/ 2553 **/
2553 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2554 bool blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2554 { 2555 {
2555 return blk_end_bidi_request(rq, error, nr_bytes, 0); 2556 return blk_end_bidi_request(rq, error, nr_bytes, 0);
2556 } 2557 }
2557 EXPORT_SYMBOL(blk_end_request); 2558 EXPORT_SYMBOL(blk_end_request);
2558 2559
2559 /** 2560 /**
2560 * blk_end_request_all - Helper function for drives to finish the request. 2561 * blk_end_request_all - Helper function for drives to finish the request.
2561 * @rq: the request to finish 2562 * @rq: the request to finish
2562 * @error: %0 for success, < %0 for error 2563 * @error: %0 for success, < %0 for error
2563 * 2564 *
2564 * Description: 2565 * Description:
2565 * Completely finish @rq. 2566 * Completely finish @rq.
2566 */ 2567 */
2567 void blk_end_request_all(struct request *rq, int error) 2568 void blk_end_request_all(struct request *rq, int error)
2568 { 2569 {
2569 bool pending; 2570 bool pending;
2570 unsigned int bidi_bytes = 0; 2571 unsigned int bidi_bytes = 0;
2571 2572
2572 if (unlikely(blk_bidi_rq(rq))) 2573 if (unlikely(blk_bidi_rq(rq)))
2573 bidi_bytes = blk_rq_bytes(rq->next_rq); 2574 bidi_bytes = blk_rq_bytes(rq->next_rq);
2574 2575
2575 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2576 pending = blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2576 BUG_ON(pending); 2577 BUG_ON(pending);
2577 } 2578 }
2578 EXPORT_SYMBOL(blk_end_request_all); 2579 EXPORT_SYMBOL(blk_end_request_all);
2579 2580
2580 /** 2581 /**
2581 * blk_end_request_cur - Helper function to finish the current request chunk. 2582 * blk_end_request_cur - Helper function to finish the current request chunk.
2582 * @rq: the request to finish the current chunk for 2583 * @rq: the request to finish the current chunk for
2583 * @error: %0 for success, < %0 for error 2584 * @error: %0 for success, < %0 for error
2584 * 2585 *
2585 * Description: 2586 * Description:
2586 * Complete the current consecutively mapped chunk from @rq. 2587 * Complete the current consecutively mapped chunk from @rq.
2587 * 2588 *
2588 * Return: 2589 * Return:
2589 * %false - we are done with this request 2590 * %false - we are done with this request
2590 * %true - still buffers pending for this request 2591 * %true - still buffers pending for this request
2591 */ 2592 */
2592 bool blk_end_request_cur(struct request *rq, int error) 2593 bool blk_end_request_cur(struct request *rq, int error)
2593 { 2594 {
2594 return blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2595 return blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2595 } 2596 }
2596 EXPORT_SYMBOL(blk_end_request_cur); 2597 EXPORT_SYMBOL(blk_end_request_cur);
2597 2598
2598 /** 2599 /**
2599 * blk_end_request_err - Finish a request till the next failure boundary. 2600 * blk_end_request_err - Finish a request till the next failure boundary.
2600 * @rq: the request to finish till the next failure boundary for 2601 * @rq: the request to finish till the next failure boundary for
2601 * @error: must be negative errno 2602 * @error: must be negative errno
2602 * 2603 *
2603 * Description: 2604 * Description:
2604 * Complete @rq till the next failure boundary. 2605 * Complete @rq till the next failure boundary.
2605 * 2606 *
2606 * Return: 2607 * Return:
2607 * %false - we are done with this request 2608 * %false - we are done with this request
2608 * %true - still buffers pending for this request 2609 * %true - still buffers pending for this request
2609 */ 2610 */
2610 bool blk_end_request_err(struct request *rq, int error) 2611 bool blk_end_request_err(struct request *rq, int error)
2611 { 2612 {
2612 WARN_ON(error >= 0); 2613 WARN_ON(error >= 0);
2613 return blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2614 return blk_end_request(rq, error, blk_rq_err_bytes(rq));
2614 } 2615 }
2615 EXPORT_SYMBOL_GPL(blk_end_request_err); 2616 EXPORT_SYMBOL_GPL(blk_end_request_err);
2616 2617
2617 /** 2618 /**
2618 * __blk_end_request - Helper function for drivers to complete the request. 2619 * __blk_end_request - Helper function for drivers to complete the request.
2619 * @rq: the request being processed 2620 * @rq: the request being processed
2620 * @error: %0 for success, < %0 for error 2621 * @error: %0 for success, < %0 for error
2621 * @nr_bytes: number of bytes to complete 2622 * @nr_bytes: number of bytes to complete
2622 * 2623 *
2623 * Description: 2624 * Description:
2624 * Must be called with queue lock held unlike blk_end_request(). 2625 * Must be called with queue lock held unlike blk_end_request().
2625 * 2626 *
2626 * Return: 2627 * Return:
2627 * %false - we are done with this request 2628 * %false - we are done with this request
2628 * %true - still buffers pending for this request 2629 * %true - still buffers pending for this request
2629 **/ 2630 **/
2630 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes) 2631 bool __blk_end_request(struct request *rq, int error, unsigned int nr_bytes)
2631 { 2632 {
2632 return __blk_end_bidi_request(rq, error, nr_bytes, 0); 2633 return __blk_end_bidi_request(rq, error, nr_bytes, 0);
2633 } 2634 }
2634 EXPORT_SYMBOL(__blk_end_request); 2635 EXPORT_SYMBOL(__blk_end_request);
2635 2636
2636 /** 2637 /**
2637 * __blk_end_request_all - Helper function for drives to finish the request. 2638 * __blk_end_request_all - Helper function for drives to finish the request.
2638 * @rq: the request to finish 2639 * @rq: the request to finish
2639 * @error: %0 for success, < %0 for error 2640 * @error: %0 for success, < %0 for error
2640 * 2641 *
2641 * Description: 2642 * Description:
2642 * Completely finish @rq. Must be called with queue lock held. 2643 * Completely finish @rq. Must be called with queue lock held.
2643 */ 2644 */
2644 void __blk_end_request_all(struct request *rq, int error) 2645 void __blk_end_request_all(struct request *rq, int error)
2645 { 2646 {
2646 bool pending; 2647 bool pending;
2647 unsigned int bidi_bytes = 0; 2648 unsigned int bidi_bytes = 0;
2648 2649
2649 if (unlikely(blk_bidi_rq(rq))) 2650 if (unlikely(blk_bidi_rq(rq)))
2650 bidi_bytes = blk_rq_bytes(rq->next_rq); 2651 bidi_bytes = blk_rq_bytes(rq->next_rq);
2651 2652
2652 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes); 2653 pending = __blk_end_bidi_request(rq, error, blk_rq_bytes(rq), bidi_bytes);
2653 BUG_ON(pending); 2654 BUG_ON(pending);
2654 } 2655 }
2655 EXPORT_SYMBOL(__blk_end_request_all); 2656 EXPORT_SYMBOL(__blk_end_request_all);
2656 2657
2657 /** 2658 /**
2658 * __blk_end_request_cur - Helper function to finish the current request chunk. 2659 * __blk_end_request_cur - Helper function to finish the current request chunk.
2659 * @rq: the request to finish the current chunk for 2660 * @rq: the request to finish the current chunk for
2660 * @error: %0 for success, < %0 for error 2661 * @error: %0 for success, < %0 for error
2661 * 2662 *
2662 * Description: 2663 * Description:
2663 * Complete the current consecutively mapped chunk from @rq. Must 2664 * Complete the current consecutively mapped chunk from @rq. Must
2664 * be called with queue lock held. 2665 * be called with queue lock held.
2665 * 2666 *
2666 * Return: 2667 * Return:
2667 * %false - we are done with this request 2668 * %false - we are done with this request
2668 * %true - still buffers pending for this request 2669 * %true - still buffers pending for this request
2669 */ 2670 */
2670 bool __blk_end_request_cur(struct request *rq, int error) 2671 bool __blk_end_request_cur(struct request *rq, int error)
2671 { 2672 {
2672 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq)); 2673 return __blk_end_request(rq, error, blk_rq_cur_bytes(rq));
2673 } 2674 }
2674 EXPORT_SYMBOL(__blk_end_request_cur); 2675 EXPORT_SYMBOL(__blk_end_request_cur);
2675 2676
2676 /** 2677 /**
2677 * __blk_end_request_err - Finish a request till the next failure boundary. 2678 * __blk_end_request_err - Finish a request till the next failure boundary.
2678 * @rq: the request to finish till the next failure boundary for 2679 * @rq: the request to finish till the next failure boundary for
2679 * @error: must be negative errno 2680 * @error: must be negative errno
2680 * 2681 *
2681 * Description: 2682 * Description:
2682 * Complete @rq till the next failure boundary. Must be called 2683 * Complete @rq till the next failure boundary. Must be called
2683 * with queue lock held. 2684 * with queue lock held.
2684 * 2685 *
2685 * Return: 2686 * Return:
2686 * %false - we are done with this request 2687 * %false - we are done with this request
2687 * %true - still buffers pending for this request 2688 * %true - still buffers pending for this request
2688 */ 2689 */
2689 bool __blk_end_request_err(struct request *rq, int error) 2690 bool __blk_end_request_err(struct request *rq, int error)
2690 { 2691 {
2691 WARN_ON(error >= 0); 2692 WARN_ON(error >= 0);
2692 return __blk_end_request(rq, error, blk_rq_err_bytes(rq)); 2693 return __blk_end_request(rq, error, blk_rq_err_bytes(rq));
2693 } 2694 }
2694 EXPORT_SYMBOL_GPL(__blk_end_request_err); 2695 EXPORT_SYMBOL_GPL(__blk_end_request_err);
2695 2696
2696 void blk_rq_bio_prep(struct request_queue *q, struct request *rq, 2697 void blk_rq_bio_prep(struct request_queue *q, struct request *rq,
2697 struct bio *bio) 2698 struct bio *bio)
2698 { 2699 {
2699 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */ 2700 /* Bit 0 (R/W) is identical in rq->cmd_flags and bio->bi_rw */
2700 rq->cmd_flags |= bio->bi_rw & REQ_WRITE; 2701 rq->cmd_flags |= bio->bi_rw & REQ_WRITE;
2701 2702
2702 if (bio_has_data(bio)) { 2703 if (bio_has_data(bio)) {
2703 rq->nr_phys_segments = bio_phys_segments(q, bio); 2704 rq->nr_phys_segments = bio_phys_segments(q, bio);
2704 rq->buffer = bio_data(bio); 2705 rq->buffer = bio_data(bio);
2705 } 2706 }
2706 rq->__data_len = bio->bi_size; 2707 rq->__data_len = bio->bi_size;
2707 rq->bio = rq->biotail = bio; 2708 rq->bio = rq->biotail = bio;
2708 2709
2709 if (bio->bi_bdev) 2710 if (bio->bi_bdev)
2710 rq->rq_disk = bio->bi_bdev->bd_disk; 2711 rq->rq_disk = bio->bi_bdev->bd_disk;
2711 } 2712 }
2712 2713
2713 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 2714 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
2714 /** 2715 /**
2715 * rq_flush_dcache_pages - Helper function to flush all pages in a request 2716 * rq_flush_dcache_pages - Helper function to flush all pages in a request
2716 * @rq: the request to be flushed 2717 * @rq: the request to be flushed
2717 * 2718 *
2718 * Description: 2719 * Description:
2719 * Flush all pages in @rq. 2720 * Flush all pages in @rq.
2720 */ 2721 */
2721 void rq_flush_dcache_pages(struct request *rq) 2722 void rq_flush_dcache_pages(struct request *rq)
2722 { 2723 {
2723 struct req_iterator iter; 2724 struct req_iterator iter;
2724 struct bio_vec *bvec; 2725 struct bio_vec *bvec;
2725 2726
2726 rq_for_each_segment(bvec, rq, iter) 2727 rq_for_each_segment(bvec, rq, iter)
2727 flush_dcache_page(bvec->bv_page); 2728 flush_dcache_page(bvec->bv_page);
2728 } 2729 }
2729 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages); 2730 EXPORT_SYMBOL_GPL(rq_flush_dcache_pages);
2730 #endif 2731 #endif
2731 2732
2732 /** 2733 /**
2733 * blk_lld_busy - Check if underlying low-level drivers of a device are busy 2734 * blk_lld_busy - Check if underlying low-level drivers of a device are busy
2734 * @q : the queue of the device being checked 2735 * @q : the queue of the device being checked
2735 * 2736 *
2736 * Description: 2737 * Description:
2737 * Check if underlying low-level drivers of a device are busy. 2738 * Check if underlying low-level drivers of a device are busy.
2738 * If the drivers want to export their busy state, they must set own 2739 * If the drivers want to export their busy state, they must set own
2739 * exporting function using blk_queue_lld_busy() first. 2740 * exporting function using blk_queue_lld_busy() first.
2740 * 2741 *
2741 * Basically, this function is used only by request stacking drivers 2742 * Basically, this function is used only by request stacking drivers
2742 * to stop dispatching requests to underlying devices when underlying 2743 * to stop dispatching requests to underlying devices when underlying
2743 * devices are busy. This behavior helps more I/O merging on the queue 2744 * devices are busy. This behavior helps more I/O merging on the queue
2744 * of the request stacking driver and prevents I/O throughput regression 2745 * of the request stacking driver and prevents I/O throughput regression
2745 * on burst I/O load. 2746 * on burst I/O load.
2746 * 2747 *
2747 * Return: 2748 * Return:
2748 * 0 - Not busy (The request stacking driver should dispatch request) 2749 * 0 - Not busy (The request stacking driver should dispatch request)
2749 * 1 - Busy (The request stacking driver should stop dispatching request) 2750 * 1 - Busy (The request stacking driver should stop dispatching request)
2750 */ 2751 */
2751 int blk_lld_busy(struct request_queue *q) 2752 int blk_lld_busy(struct request_queue *q)
2752 { 2753 {
2753 if (q->lld_busy_fn) 2754 if (q->lld_busy_fn)
2754 return q->lld_busy_fn(q); 2755 return q->lld_busy_fn(q);
2755 2756
2756 return 0; 2757 return 0;
2757 } 2758 }
2758 EXPORT_SYMBOL_GPL(blk_lld_busy); 2759 EXPORT_SYMBOL_GPL(blk_lld_busy);
2759 2760
2760 /** 2761 /**
2761 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request 2762 * blk_rq_unprep_clone - Helper function to free all bios in a cloned request
2762 * @rq: the clone request to be cleaned up 2763 * @rq: the clone request to be cleaned up
2763 * 2764 *
2764 * Description: 2765 * Description:
2765 * Free all bios in @rq for a cloned request. 2766 * Free all bios in @rq for a cloned request.
2766 */ 2767 */
2767 void blk_rq_unprep_clone(struct request *rq) 2768 void blk_rq_unprep_clone(struct request *rq)
2768 { 2769 {
2769 struct bio *bio; 2770 struct bio *bio;
2770 2771
2771 while ((bio = rq->bio) != NULL) { 2772 while ((bio = rq->bio) != NULL) {
2772 rq->bio = bio->bi_next; 2773 rq->bio = bio->bi_next;
2773 2774
2774 bio_put(bio); 2775 bio_put(bio);
2775 } 2776 }
2776 } 2777 }
2777 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone); 2778 EXPORT_SYMBOL_GPL(blk_rq_unprep_clone);
2778 2779
2779 /* 2780 /*
2780 * Copy attributes of the original request to the clone request. 2781 * Copy attributes of the original request to the clone request.
2781 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied. 2782 * The actual data parts (e.g. ->cmd, ->buffer, ->sense) are not copied.
2782 */ 2783 */
2783 static void __blk_rq_prep_clone(struct request *dst, struct request *src) 2784 static void __blk_rq_prep_clone(struct request *dst, struct request *src)
2784 { 2785 {
2785 dst->cpu = src->cpu; 2786 dst->cpu = src->cpu;
2786 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE; 2787 dst->cmd_flags = (src->cmd_flags & REQ_CLONE_MASK) | REQ_NOMERGE;
2787 dst->cmd_type = src->cmd_type; 2788 dst->cmd_type = src->cmd_type;
2788 dst->__sector = blk_rq_pos(src); 2789 dst->__sector = blk_rq_pos(src);
2789 dst->__data_len = blk_rq_bytes(src); 2790 dst->__data_len = blk_rq_bytes(src);
2790 dst->nr_phys_segments = src->nr_phys_segments; 2791 dst->nr_phys_segments = src->nr_phys_segments;
2791 dst->ioprio = src->ioprio; 2792 dst->ioprio = src->ioprio;
2792 dst->extra_len = src->extra_len; 2793 dst->extra_len = src->extra_len;
2793 } 2794 }
2794 2795
2795 /** 2796 /**
2796 * blk_rq_prep_clone - Helper function to setup clone request 2797 * blk_rq_prep_clone - Helper function to setup clone request
2797 * @rq: the request to be setup 2798 * @rq: the request to be setup
2798 * @rq_src: original request to be cloned 2799 * @rq_src: original request to be cloned
2799 * @bs: bio_set that bios for clone are allocated from 2800 * @bs: bio_set that bios for clone are allocated from
2800 * @gfp_mask: memory allocation mask for bio 2801 * @gfp_mask: memory allocation mask for bio
2801 * @bio_ctr: setup function to be called for each clone bio. 2802 * @bio_ctr: setup function to be called for each clone bio.
2802 * Returns %0 for success, non %0 for failure. 2803 * Returns %0 for success, non %0 for failure.
2803 * @data: private data to be passed to @bio_ctr 2804 * @data: private data to be passed to @bio_ctr
2804 * 2805 *
2805 * Description: 2806 * Description:
2806 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq. 2807 * Clones bios in @rq_src to @rq, and copies attributes of @rq_src to @rq.
2807 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense) 2808 * The actual data parts of @rq_src (e.g. ->cmd, ->buffer, ->sense)
2808 * are not copied, and copying such parts is the caller's responsibility. 2809 * are not copied, and copying such parts is the caller's responsibility.
2809 * Also, pages which the original bios are pointing to are not copied 2810 * Also, pages which the original bios are pointing to are not copied
2810 * and the cloned bios just point same pages. 2811 * and the cloned bios just point same pages.
2811 * So cloned bios must be completed before original bios, which means 2812 * So cloned bios must be completed before original bios, which means
2812 * the caller must complete @rq before @rq_src. 2813 * the caller must complete @rq before @rq_src.
2813 */ 2814 */
2814 int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 2815 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
2815 struct bio_set *bs, gfp_t gfp_mask, 2816 struct bio_set *bs, gfp_t gfp_mask,
2816 int (*bio_ctr)(struct bio *, struct bio *, void *), 2817 int (*bio_ctr)(struct bio *, struct bio *, void *),
2817 void *data) 2818 void *data)
2818 { 2819 {
2819 struct bio *bio, *bio_src; 2820 struct bio *bio, *bio_src;
2820 2821
2821 if (!bs) 2822 if (!bs)
2822 bs = fs_bio_set; 2823 bs = fs_bio_set;
2823 2824
2824 blk_rq_init(NULL, rq); 2825 blk_rq_init(NULL, rq);
2825 2826
2826 __rq_for_each_bio(bio_src, rq_src) { 2827 __rq_for_each_bio(bio_src, rq_src) {
2827 bio = bio_clone_bioset(bio_src, gfp_mask, bs); 2828 bio = bio_clone_bioset(bio_src, gfp_mask, bs);
2828 if (!bio) 2829 if (!bio)
2829 goto free_and_out; 2830 goto free_and_out;
2830 2831
2831 if (bio_ctr && bio_ctr(bio, bio_src, data)) 2832 if (bio_ctr && bio_ctr(bio, bio_src, data))
2832 goto free_and_out; 2833 goto free_and_out;
2833 2834
2834 if (rq->bio) { 2835 if (rq->bio) {
2835 rq->biotail->bi_next = bio; 2836 rq->biotail->bi_next = bio;
2836 rq->biotail = bio; 2837 rq->biotail = bio;
2837 } else 2838 } else
2838 rq->bio = rq->biotail = bio; 2839 rq->bio = rq->biotail = bio;
2839 } 2840 }
2840 2841
2841 __blk_rq_prep_clone(rq, rq_src); 2842 __blk_rq_prep_clone(rq, rq_src);
2842 2843
2843 return 0; 2844 return 0;
2844 2845
2845 free_and_out: 2846 free_and_out:
2846 if (bio) 2847 if (bio)
2847 bio_put(bio); 2848 bio_put(bio);
2848 blk_rq_unprep_clone(rq); 2849 blk_rq_unprep_clone(rq);
2849 2850
2850 return -ENOMEM; 2851 return -ENOMEM;
2851 } 2852 }
2852 EXPORT_SYMBOL_GPL(blk_rq_prep_clone); 2853 EXPORT_SYMBOL_GPL(blk_rq_prep_clone);
2853 2854
2854 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work) 2855 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work)
2855 { 2856 {
2856 return queue_work(kblockd_workqueue, work); 2857 return queue_work(kblockd_workqueue, work);
2857 } 2858 }
2858 EXPORT_SYMBOL(kblockd_schedule_work); 2859 EXPORT_SYMBOL(kblockd_schedule_work);
2859 2860
2860 int kblockd_schedule_delayed_work(struct request_queue *q, 2861 int kblockd_schedule_delayed_work(struct request_queue *q,
2861 struct delayed_work *dwork, unsigned long delay) 2862 struct delayed_work *dwork, unsigned long delay)
2862 { 2863 {
2863 return queue_delayed_work(kblockd_workqueue, dwork, delay); 2864 return queue_delayed_work(kblockd_workqueue, dwork, delay);
2864 } 2865 }
2865 EXPORT_SYMBOL(kblockd_schedule_delayed_work); 2866 EXPORT_SYMBOL(kblockd_schedule_delayed_work);
2866 2867
2867 #define PLUG_MAGIC 0x91827364 2868 #define PLUG_MAGIC 0x91827364
2868 2869
2869 /** 2870 /**
2870 * blk_start_plug - initialize blk_plug and track it inside the task_struct 2871 * blk_start_plug - initialize blk_plug and track it inside the task_struct
2871 * @plug: The &struct blk_plug that needs to be initialized 2872 * @plug: The &struct blk_plug that needs to be initialized
2872 * 2873 *
2873 * Description: 2874 * Description:
2874 * Tracking blk_plug inside the task_struct will help with auto-flushing the 2875 * Tracking blk_plug inside the task_struct will help with auto-flushing the
2875 * pending I/O should the task end up blocking between blk_start_plug() and 2876 * pending I/O should the task end up blocking between blk_start_plug() and
2876 * blk_finish_plug(). This is important from a performance perspective, but 2877 * blk_finish_plug(). This is important from a performance perspective, but
2877 * also ensures that we don't deadlock. For instance, if the task is blocking 2878 * also ensures that we don't deadlock. For instance, if the task is blocking
2878 * for a memory allocation, memory reclaim could end up wanting to free a 2879 * for a memory allocation, memory reclaim could end up wanting to free a
2879 * page belonging to that request that is currently residing in our private 2880 * page belonging to that request that is currently residing in our private
2880 * plug. By flushing the pending I/O when the process goes to sleep, we avoid 2881 * plug. By flushing the pending I/O when the process goes to sleep, we avoid
2881 * this kind of deadlock. 2882 * this kind of deadlock.
2882 */ 2883 */
2883 void blk_start_plug(struct blk_plug *plug) 2884 void blk_start_plug(struct blk_plug *plug)
2884 { 2885 {
2885 struct task_struct *tsk = current; 2886 struct task_struct *tsk = current;
2886 2887
2887 plug->magic = PLUG_MAGIC; 2888 plug->magic = PLUG_MAGIC;
2888 INIT_LIST_HEAD(&plug->list); 2889 INIT_LIST_HEAD(&plug->list);
2889 INIT_LIST_HEAD(&plug->cb_list); 2890 INIT_LIST_HEAD(&plug->cb_list);
2890 plug->should_sort = 0; 2891 plug->should_sort = 0;
2891 2892
2892 /* 2893 /*
2893 * If this is a nested plug, don't actually assign it. It will be 2894 * If this is a nested plug, don't actually assign it. It will be
2894 * flushed on its own. 2895 * flushed on its own.
2895 */ 2896 */
2896 if (!tsk->plug) { 2897 if (!tsk->plug) {
2897 /* 2898 /*
2898 * Store ordering should not be needed here, since a potential 2899 * Store ordering should not be needed here, since a potential
2899 * preempt will imply a full memory barrier 2900 * preempt will imply a full memory barrier
2900 */ 2901 */
2901 tsk->plug = plug; 2902 tsk->plug = plug;
2902 } 2903 }
2903 } 2904 }
2904 EXPORT_SYMBOL(blk_start_plug); 2905 EXPORT_SYMBOL(blk_start_plug);
2905 2906
2906 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b) 2907 static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
2907 { 2908 {
2908 struct request *rqa = container_of(a, struct request, queuelist); 2909 struct request *rqa = container_of(a, struct request, queuelist);
2909 struct request *rqb = container_of(b, struct request, queuelist); 2910 struct request *rqb = container_of(b, struct request, queuelist);
2910 2911
2911 return !(rqa->q < rqb->q || 2912 return !(rqa->q < rqb->q ||
2912 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb))); 2913 (rqa->q == rqb->q && blk_rq_pos(rqa) < blk_rq_pos(rqb)));
2913 } 2914 }
2914 2915
2915 /* 2916 /*
2916 * If 'from_schedule' is true, then postpone the dispatch of requests 2917 * If 'from_schedule' is true, then postpone the dispatch of requests
2917 * until a safe kblockd context. We due this to avoid accidental big 2918 * until a safe kblockd context. We due this to avoid accidental big
2918 * additional stack usage in driver dispatch, in places where the originally 2919 * additional stack usage in driver dispatch, in places where the originally
2919 * plugger did not intend it. 2920 * plugger did not intend it.
2920 */ 2921 */
2921 static void queue_unplugged(struct request_queue *q, unsigned int depth, 2922 static void queue_unplugged(struct request_queue *q, unsigned int depth,
2922 bool from_schedule) 2923 bool from_schedule)
2923 __releases(q->queue_lock) 2924 __releases(q->queue_lock)
2924 { 2925 {
2925 trace_block_unplug(q, depth, !from_schedule); 2926 trace_block_unplug(q, depth, !from_schedule);
2926 2927
2927 if (from_schedule) 2928 if (from_schedule)
2928 blk_run_queue_async(q); 2929 blk_run_queue_async(q);
2929 else 2930 else
2930 __blk_run_queue(q); 2931 __blk_run_queue(q);
2931 spin_unlock(q->queue_lock); 2932 spin_unlock(q->queue_lock);
2932 } 2933 }
2933 2934
2934 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule) 2935 static void flush_plug_callbacks(struct blk_plug *plug, bool from_schedule)
2935 { 2936 {
2936 LIST_HEAD(callbacks); 2937 LIST_HEAD(callbacks);
2937 2938
2938 while (!list_empty(&plug->cb_list)) { 2939 while (!list_empty(&plug->cb_list)) {
2939 list_splice_init(&plug->cb_list, &callbacks); 2940 list_splice_init(&plug->cb_list, &callbacks);
2940 2941
2941 while (!list_empty(&callbacks)) { 2942 while (!list_empty(&callbacks)) {
2942 struct blk_plug_cb *cb = list_first_entry(&callbacks, 2943 struct blk_plug_cb *cb = list_first_entry(&callbacks,
2943 struct blk_plug_cb, 2944 struct blk_plug_cb,
2944 list); 2945 list);
2945 list_del(&cb->list); 2946 list_del(&cb->list);
2946 cb->callback(cb, from_schedule); 2947 cb->callback(cb, from_schedule);
2947 } 2948 }
2948 } 2949 }
2949 } 2950 }
2950 2951
2951 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data, 2952 struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, void *data,
2952 int size) 2953 int size)
2953 { 2954 {
2954 struct blk_plug *plug = current->plug; 2955 struct blk_plug *plug = current->plug;
2955 struct blk_plug_cb *cb; 2956 struct blk_plug_cb *cb;
2956 2957
2957 if (!plug) 2958 if (!plug)
2958 return NULL; 2959 return NULL;
2959 2960
2960 list_for_each_entry(cb, &plug->cb_list, list) 2961 list_for_each_entry(cb, &plug->cb_list, list)
2961 if (cb->callback == unplug && cb->data == data) 2962 if (cb->callback == unplug && cb->data == data)
2962 return cb; 2963 return cb;
2963 2964
2964 /* Not currently on the callback list */ 2965 /* Not currently on the callback list */
2965 BUG_ON(size < sizeof(*cb)); 2966 BUG_ON(size < sizeof(*cb));
2966 cb = kzalloc(size, GFP_ATOMIC); 2967 cb = kzalloc(size, GFP_ATOMIC);
2967 if (cb) { 2968 if (cb) {
2968 cb->data = data; 2969 cb->data = data;
2969 cb->callback = unplug; 2970 cb->callback = unplug;
2970 list_add(&cb->list, &plug->cb_list); 2971 list_add(&cb->list, &plug->cb_list);
2971 } 2972 }
2972 return cb; 2973 return cb;
2973 } 2974 }
2974 EXPORT_SYMBOL(blk_check_plugged); 2975 EXPORT_SYMBOL(blk_check_plugged);
2975 2976
2976 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule) 2977 void blk_flush_plug_list(struct blk_plug *plug, bool from_schedule)
2977 { 2978 {
2978 struct request_queue *q; 2979 struct request_queue *q;
2979 unsigned long flags; 2980 unsigned long flags;
2980 struct request *rq; 2981 struct request *rq;
2981 LIST_HEAD(list); 2982 LIST_HEAD(list);
2982 unsigned int depth; 2983 unsigned int depth;
2983 2984
2984 BUG_ON(plug->magic != PLUG_MAGIC); 2985 BUG_ON(plug->magic != PLUG_MAGIC);
2985 2986
2986 flush_plug_callbacks(plug, from_schedule); 2987 flush_plug_callbacks(plug, from_schedule);
2987 if (list_empty(&plug->list)) 2988 if (list_empty(&plug->list))
2988 return; 2989 return;
2989 2990
2990 list_splice_init(&plug->list, &list); 2991 list_splice_init(&plug->list, &list);
2991 2992
2992 if (plug->should_sort) { 2993 if (plug->should_sort) {
2993 list_sort(NULL, &list, plug_rq_cmp); 2994 list_sort(NULL, &list, plug_rq_cmp);
2994 plug->should_sort = 0; 2995 plug->should_sort = 0;
2995 } 2996 }
2996 2997
2997 q = NULL; 2998 q = NULL;
2998 depth = 0; 2999 depth = 0;
2999 3000
3000 /* 3001 /*
3001 * Save and disable interrupts here, to avoid doing it for every 3002 * Save and disable interrupts here, to avoid doing it for every
3002 * queue lock we have to take. 3003 * queue lock we have to take.
3003 */ 3004 */
3004 local_irq_save(flags); 3005 local_irq_save(flags);
3005 while (!list_empty(&list)) { 3006 while (!list_empty(&list)) {
3006 rq = list_entry_rq(list.next); 3007 rq = list_entry_rq(list.next);
3007 list_del_init(&rq->queuelist); 3008 list_del_init(&rq->queuelist);
3008 BUG_ON(!rq->q); 3009 BUG_ON(!rq->q);
3009 if (rq->q != q) { 3010 if (rq->q != q) {
3010 /* 3011 /*
3011 * This drops the queue lock 3012 * This drops the queue lock
3012 */ 3013 */
3013 if (q) 3014 if (q)
3014 queue_unplugged(q, depth, from_schedule); 3015 queue_unplugged(q, depth, from_schedule);
3015 q = rq->q; 3016 q = rq->q;
3016 depth = 0; 3017 depth = 0;
3017 spin_lock(q->queue_lock); 3018 spin_lock(q->queue_lock);
3018 } 3019 }
3019 3020
3020 /* 3021 /*
3021 * Short-circuit if @q is dead 3022 * Short-circuit if @q is dead
3022 */ 3023 */
3023 if (unlikely(blk_queue_dying(q))) { 3024 if (unlikely(blk_queue_dying(q))) {
3024 __blk_end_request_all(rq, -ENODEV); 3025 __blk_end_request_all(rq, -ENODEV);
3025 continue; 3026 continue;
3026 } 3027 }
3027 3028
3028 /* 3029 /*
3029 * rq is already accounted, so use raw insert 3030 * rq is already accounted, so use raw insert
3030 */ 3031 */
3031 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA)) 3032 if (rq->cmd_flags & (REQ_FLUSH | REQ_FUA))
3032 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH); 3033 __elv_add_request(q, rq, ELEVATOR_INSERT_FLUSH);
3033 else 3034 else
3034 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE); 3035 __elv_add_request(q, rq, ELEVATOR_INSERT_SORT_MERGE);
3035 3036
3036 depth++; 3037 depth++;
3037 } 3038 }
3038 3039
3039 /* 3040 /*
3040 * This drops the queue lock 3041 * This drops the queue lock
3041 */ 3042 */
3042 if (q) 3043 if (q)
3043 queue_unplugged(q, depth, from_schedule); 3044 queue_unplugged(q, depth, from_schedule);
3044 3045
3045 local_irq_restore(flags); 3046 local_irq_restore(flags);
3046 } 3047 }
3047 3048
3048 void blk_finish_plug(struct blk_plug *plug) 3049 void blk_finish_plug(struct blk_plug *plug)
3049 { 3050 {
3050 blk_flush_plug_list(plug, false); 3051 blk_flush_plug_list(plug, false);
3051 3052
3052 if (plug == current->plug) 3053 if (plug == current->plug)
3053 current->plug = NULL; 3054 current->plug = NULL;
3054 } 3055 }
3055 EXPORT_SYMBOL(blk_finish_plug); 3056 EXPORT_SYMBOL(blk_finish_plug);
3056 3057
3057 int __init blk_dev_init(void) 3058 int __init blk_dev_init(void)
3058 { 3059 {
3059 BUILD_BUG_ON(__REQ_NR_BITS > 8 * 3060 BUILD_BUG_ON(__REQ_NR_BITS > 8 *
3060 sizeof(((struct request *)0)->cmd_flags)); 3061 sizeof(((struct request *)0)->cmd_flags));
3061 3062
3062 /* used for unplugging and affects IO latency/throughput - HIGHPRI */ 3063 /* used for unplugging and affects IO latency/throughput - HIGHPRI */
3063 kblockd_workqueue = alloc_workqueue("kblockd", 3064 kblockd_workqueue = alloc_workqueue("kblockd",
3064 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0); 3065 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
3065 if (!kblockd_workqueue) 3066 if (!kblockd_workqueue)
3066 panic("Failed to create kblockd\n"); 3067 panic("Failed to create kblockd\n");
3067 3068
3068 request_cachep = kmem_cache_create("blkdev_requests", 3069 request_cachep = kmem_cache_create("blkdev_requests",
3069 sizeof(struct request), 0, SLAB_PANIC, NULL); 3070 sizeof(struct request), 0, SLAB_PANIC, NULL);
3070 3071
3071 blk_requestq_cachep = kmem_cache_create("blkdev_queue", 3072 blk_requestq_cachep = kmem_cache_create("blkdev_queue",
3072 sizeof(struct request_queue), 0, SLAB_PANIC, NULL); 3073 sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
3073 3074
3074 return 0; 3075 return 0;
3075 } 3076 }
3076 3077