Remove double removal of blktrace directory
[linux-3.10.git] / kernel / trace / blktrace.c
1 /*
2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
16  *
17  */
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/debugfs.h>
25 #include <linux/smp_lock.h>
26 #include <linux/time.h>
27 #include <linux/uaccess.h>
28
29 #include <trace/events/block.h>
30
31 #include "trace_output.h"
32
33 #ifdef CONFIG_BLK_DEV_IO_TRACE
34
35 static unsigned int blktrace_seq __read_mostly = 1;
36
37 static struct trace_array *blk_tr;
38 static bool blk_tracer_enabled __read_mostly;
39
40 /* Select an alternative, minimalistic output than the original one */
41 #define TRACE_BLK_OPT_CLASSIC   0x1
42
43 static struct tracer_opt blk_tracer_opts[] = {
44         /* Default disable the minimalistic output */
45         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
46         { }
47 };
48
49 static struct tracer_flags blk_tracer_flags = {
50         .val  = 0,
51         .opts = blk_tracer_opts,
52 };
53
54 /* Global reference count of probes */
55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
56
57 static void blk_register_tracepoints(void);
58 static void blk_unregister_tracepoints(void);
59
60 /*
61  * Send out a notify message.
62  */
63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
64                        const void *data, size_t len)
65 {
66         struct blk_io_trace *t;
67         struct ring_buffer_event *event = NULL;
68         int pc = 0;
69         int cpu = smp_processor_id();
70         bool blk_tracer = blk_tracer_enabled;
71
72         if (blk_tracer) {
73                 pc = preempt_count();
74                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
75                                                   sizeof(*t) + len,
76                                                   0, pc);
77                 if (!event)
78                         return;
79                 t = ring_buffer_event_data(event);
80                 goto record_it;
81         }
82
83         if (!bt->rchan)
84                 return;
85
86         t = relay_reserve(bt->rchan, sizeof(*t) + len);
87         if (t) {
88                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
89                 t->time = ktime_to_ns(ktime_get());
90 record_it:
91                 t->device = bt->dev;
92                 t->action = action;
93                 t->pid = pid;
94                 t->cpu = cpu;
95                 t->pdu_len = len;
96                 memcpy((void *) t + sizeof(*t), data, len);
97
98                 if (blk_tracer)
99                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
100         }
101 }
102
103 /*
104  * Send out a notify for this process, if we haven't done so since a trace
105  * started
106  */
107 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
108 {
109         tsk->btrace_seq = blktrace_seq;
110         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
111 }
112
113 static void trace_note_time(struct blk_trace *bt)
114 {
115         struct timespec now;
116         unsigned long flags;
117         u32 words[2];
118
119         getnstimeofday(&now);
120         words[0] = now.tv_sec;
121         words[1] = now.tv_nsec;
122
123         local_irq_save(flags);
124         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
125         local_irq_restore(flags);
126 }
127
128 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
129 {
130         int n;
131         va_list args;
132         unsigned long flags;
133         char *buf;
134
135         if (unlikely(bt->trace_state != Blktrace_running &&
136                      !blk_tracer_enabled))
137                 return;
138
139         local_irq_save(flags);
140         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
141         va_start(args, fmt);
142         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
143         va_end(args);
144
145         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
146         local_irq_restore(flags);
147 }
148 EXPORT_SYMBOL_GPL(__trace_note_message);
149
150 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
151                          pid_t pid)
152 {
153         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
154                 return 1;
155         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
156                 return 1;
157         if (bt->pid && pid != bt->pid)
158                 return 1;
159
160         return 0;
161 }
162
163 /*
164  * Data direction bit lookup
165  */
166 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
167                                  BLK_TC_ACT(BLK_TC_WRITE) };
168
169 /* The ilog2() calls fall out because they're constant */
170 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
171           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
172
173 /*
174  * The worker for the various blk_add_trace*() types. Fills out a
175  * blk_io_trace structure and places it in a per-cpu subbuffer.
176  */
177 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
178                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
179 {
180         struct task_struct *tsk = current;
181         struct ring_buffer_event *event = NULL;
182         struct blk_io_trace *t;
183         unsigned long flags = 0;
184         unsigned long *sequence;
185         pid_t pid;
186         int cpu, pc = 0;
187         bool blk_tracer = blk_tracer_enabled;
188
189         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
190                 return;
191
192         what |= ddir_act[rw & WRITE];
193         what |= MASK_TC_BIT(rw, BARRIER);
194         what |= MASK_TC_BIT(rw, SYNCIO);
195         what |= MASK_TC_BIT(rw, AHEAD);
196         what |= MASK_TC_BIT(rw, META);
197         what |= MASK_TC_BIT(rw, DISCARD);
198
199         pid = tsk->pid;
200         if (act_log_check(bt, what, sector, pid))
201                 return;
202         cpu = raw_smp_processor_id();
203
204         if (blk_tracer) {
205                 tracing_record_cmdline(current);
206
207                 pc = preempt_count();
208                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
209                                                   sizeof(*t) + pdu_len,
210                                                   0, pc);
211                 if (!event)
212                         return;
213                 t = ring_buffer_event_data(event);
214                 goto record_it;
215         }
216
217         /*
218          * A word about the locking here - we disable interrupts to reserve
219          * some space in the relay per-cpu buffer, to prevent an irq
220          * from coming in and stepping on our toes.
221          */
222         local_irq_save(flags);
223
224         if (unlikely(tsk->btrace_seq != blktrace_seq))
225                 trace_note_tsk(bt, tsk);
226
227         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
228         if (t) {
229                 sequence = per_cpu_ptr(bt->sequence, cpu);
230
231                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
232                 t->sequence = ++(*sequence);
233                 t->time = ktime_to_ns(ktime_get());
234 record_it:
235                 /*
236                  * These two are not needed in ftrace as they are in the
237                  * generic trace_entry, filled by tracing_generic_entry_update,
238                  * but for the trace_event->bin() synthesizer benefit we do it
239                  * here too.
240                  */
241                 t->cpu = cpu;
242                 t->pid = pid;
243
244                 t->sector = sector;
245                 t->bytes = bytes;
246                 t->action = what;
247                 t->device = bt->dev;
248                 t->error = error;
249                 t->pdu_len = pdu_len;
250
251                 if (pdu_len)
252                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
253
254                 if (blk_tracer) {
255                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
256                         return;
257                 }
258         }
259
260         local_irq_restore(flags);
261 }
262
263 static struct dentry *blk_tree_root;
264 static DEFINE_MUTEX(blk_tree_mutex);
265
266 static void blk_trace_free(struct blk_trace *bt)
267 {
268         debugfs_remove(bt->msg_file);
269         debugfs_remove(bt->dropped_file);
270         relay_close(bt->rchan);
271         debugfs_remove(bt->dir);
272         free_percpu(bt->sequence);
273         free_percpu(bt->msg_data);
274         kfree(bt);
275 }
276
277 static void blk_trace_cleanup(struct blk_trace *bt)
278 {
279         blk_trace_free(bt);
280         if (atomic_dec_and_test(&blk_probes_ref))
281                 blk_unregister_tracepoints();
282 }
283
284 int blk_trace_remove(struct request_queue *q)
285 {
286         struct blk_trace *bt;
287
288         bt = xchg(&q->blk_trace, NULL);
289         if (!bt)
290                 return -EINVAL;
291
292         if (bt->trace_state != Blktrace_running)
293                 blk_trace_cleanup(bt);
294
295         return 0;
296 }
297 EXPORT_SYMBOL_GPL(blk_trace_remove);
298
299 static int blk_dropped_open(struct inode *inode, struct file *filp)
300 {
301         filp->private_data = inode->i_private;
302
303         return 0;
304 }
305
306 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
307                                 size_t count, loff_t *ppos)
308 {
309         struct blk_trace *bt = filp->private_data;
310         char buf[16];
311
312         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
313
314         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
315 }
316
317 static const struct file_operations blk_dropped_fops = {
318         .owner =        THIS_MODULE,
319         .open =         blk_dropped_open,
320         .read =         blk_dropped_read,
321 };
322
323 static int blk_msg_open(struct inode *inode, struct file *filp)
324 {
325         filp->private_data = inode->i_private;
326
327         return 0;
328 }
329
330 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
331                                 size_t count, loff_t *ppos)
332 {
333         char *msg;
334         struct blk_trace *bt;
335
336         if (count >= BLK_TN_MAX_MSG)
337                 return -EINVAL;
338
339         msg = kmalloc(count + 1, GFP_KERNEL);
340         if (msg == NULL)
341                 return -ENOMEM;
342
343         if (copy_from_user(msg, buffer, count)) {
344                 kfree(msg);
345                 return -EFAULT;
346         }
347
348         msg[count] = '\0';
349         bt = filp->private_data;
350         __trace_note_message(bt, "%s", msg);
351         kfree(msg);
352
353         return count;
354 }
355
356 static const struct file_operations blk_msg_fops = {
357         .owner =        THIS_MODULE,
358         .open =         blk_msg_open,
359         .write =        blk_msg_write,
360 };
361
362 /*
363  * Keep track of how many times we encountered a full subbuffer, to aid
364  * the user space app in telling how many lost events there were.
365  */
366 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
367                                      void *prev_subbuf, size_t prev_padding)
368 {
369         struct blk_trace *bt;
370
371         if (!relay_buf_full(buf))
372                 return 1;
373
374         bt = buf->chan->private_data;
375         atomic_inc(&bt->dropped);
376         return 0;
377 }
378
379 static int blk_remove_buf_file_callback(struct dentry *dentry)
380 {
381         debugfs_remove(dentry);
382
383         return 0;
384 }
385
386 static struct dentry *blk_create_buf_file_callback(const char *filename,
387                                                    struct dentry *parent,
388                                                    int mode,
389                                                    struct rchan_buf *buf,
390                                                    int *is_global)
391 {
392         return debugfs_create_file(filename, mode, parent, buf,
393                                         &relay_file_operations);
394 }
395
396 static struct rchan_callbacks blk_relay_callbacks = {
397         .subbuf_start           = blk_subbuf_start_callback,
398         .create_buf_file        = blk_create_buf_file_callback,
399         .remove_buf_file        = blk_remove_buf_file_callback,
400 };
401
402 static void blk_trace_setup_lba(struct blk_trace *bt,
403                                 struct block_device *bdev)
404 {
405         struct hd_struct *part = NULL;
406
407         if (bdev)
408                 part = bdev->bd_part;
409
410         if (part) {
411                 bt->start_lba = part->start_sect;
412                 bt->end_lba = part->start_sect + part->nr_sects;
413         } else {
414                 bt->start_lba = 0;
415                 bt->end_lba = -1ULL;
416         }
417 }
418
419 /*
420  * Setup everything required to start tracing
421  */
422 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
423                        struct block_device *bdev,
424                        struct blk_user_trace_setup *buts)
425 {
426         struct blk_trace *old_bt, *bt = NULL;
427         struct dentry *dir = NULL;
428         int ret, i;
429
430         if (!buts->buf_size || !buts->buf_nr)
431                 return -EINVAL;
432
433         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
434         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
435
436         /*
437          * some device names have larger paths - convert the slashes
438          * to underscores for this to work as expected
439          */
440         for (i = 0; i < strlen(buts->name); i++)
441                 if (buts->name[i] == '/')
442                         buts->name[i] = '_';
443
444         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
445         if (!bt)
446                 return -ENOMEM;
447
448         ret = -ENOMEM;
449         bt->sequence = alloc_percpu(unsigned long);
450         if (!bt->sequence)
451                 goto err;
452
453         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
454         if (!bt->msg_data)
455                 goto err;
456
457         ret = -ENOENT;
458
459         mutex_lock(&blk_tree_mutex);
460         if (!blk_tree_root) {
461                 blk_tree_root = debugfs_create_dir("block", NULL);
462                 if (!blk_tree_root) {
463                         mutex_unlock(&blk_tree_mutex);
464                         goto err;
465                 }
466         }
467         mutex_unlock(&blk_tree_mutex);
468
469         dir = debugfs_create_dir(buts->name, blk_tree_root);
470
471         if (!dir)
472                 goto err;
473
474         bt->dir = dir;
475         bt->dev = dev;
476         atomic_set(&bt->dropped, 0);
477
478         ret = -EIO;
479         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
480                                                &blk_dropped_fops);
481         if (!bt->dropped_file)
482                 goto err;
483
484         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
485         if (!bt->msg_file)
486                 goto err;
487
488         bt->rchan = relay_open("trace", dir, buts->buf_size,
489                                 buts->buf_nr, &blk_relay_callbacks, bt);
490         if (!bt->rchan)
491                 goto err;
492
493         bt->act_mask = buts->act_mask;
494         if (!bt->act_mask)
495                 bt->act_mask = (u16) -1;
496
497         blk_trace_setup_lba(bt, bdev);
498
499         /* overwrite with user settings */
500         if (buts->start_lba)
501                 bt->start_lba = buts->start_lba;
502         if (buts->end_lba)
503                 bt->end_lba = buts->end_lba;
504
505         bt->pid = buts->pid;
506         bt->trace_state = Blktrace_setup;
507
508         ret = -EBUSY;
509         old_bt = xchg(&q->blk_trace, bt);
510         if (old_bt) {
511                 (void) xchg(&q->blk_trace, old_bt);
512                 goto err;
513         }
514
515         if (atomic_inc_return(&blk_probes_ref) == 1)
516                 blk_register_tracepoints();
517
518         return 0;
519 err:
520         blk_trace_free(bt);
521         return ret;
522 }
523
524 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
525                     struct block_device *bdev,
526                     char __user *arg)
527 {
528         struct blk_user_trace_setup buts;
529         int ret;
530
531         ret = copy_from_user(&buts, arg, sizeof(buts));
532         if (ret)
533                 return -EFAULT;
534
535         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
536         if (ret)
537                 return ret;
538
539         if (copy_to_user(arg, &buts, sizeof(buts)))
540                 return -EFAULT;
541
542         return 0;
543 }
544 EXPORT_SYMBOL_GPL(blk_trace_setup);
545
546 int blk_trace_startstop(struct request_queue *q, int start)
547 {
548         int ret;
549         struct blk_trace *bt = q->blk_trace;
550
551         if (bt == NULL)
552                 return -EINVAL;
553
554         /*
555          * For starting a trace, we can transition from a setup or stopped
556          * trace. For stopping a trace, the state must be running
557          */
558         ret = -EINVAL;
559         if (start) {
560                 if (bt->trace_state == Blktrace_setup ||
561                     bt->trace_state == Blktrace_stopped) {
562                         blktrace_seq++;
563                         smp_mb();
564                         bt->trace_state = Blktrace_running;
565
566                         trace_note_time(bt);
567                         ret = 0;
568                 }
569         } else {
570                 if (bt->trace_state == Blktrace_running) {
571                         bt->trace_state = Blktrace_stopped;
572                         relay_flush(bt->rchan);
573                         ret = 0;
574                 }
575         }
576
577         return ret;
578 }
579 EXPORT_SYMBOL_GPL(blk_trace_startstop);
580
581 /**
582  * blk_trace_ioctl: - handle the ioctls associated with tracing
583  * @bdev:       the block device
584  * @cmd:        the ioctl cmd
585  * @arg:        the argument data, if any
586  *
587  **/
588 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
589 {
590         struct request_queue *q;
591         int ret, start = 0;
592         char b[BDEVNAME_SIZE];
593
594         q = bdev_get_queue(bdev);
595         if (!q)
596                 return -ENXIO;
597
598         mutex_lock(&bdev->bd_mutex);
599
600         switch (cmd) {
601         case BLKTRACESETUP:
602                 bdevname(bdev, b);
603                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
604                 break;
605         case BLKTRACESTART:
606                 start = 1;
607         case BLKTRACESTOP:
608                 ret = blk_trace_startstop(q, start);
609                 break;
610         case BLKTRACETEARDOWN:
611                 ret = blk_trace_remove(q);
612                 break;
613         default:
614                 ret = -ENOTTY;
615                 break;
616         }
617
618         mutex_unlock(&bdev->bd_mutex);
619         return ret;
620 }
621
622 /**
623  * blk_trace_shutdown: - stop and cleanup trace structures
624  * @q:    the request queue associated with the device
625  *
626  **/
627 void blk_trace_shutdown(struct request_queue *q)
628 {
629         if (q->blk_trace) {
630                 blk_trace_startstop(q, 0);
631                 blk_trace_remove(q);
632         }
633 }
634
635 /*
636  * blktrace probes
637  */
638
639 /**
640  * blk_add_trace_rq - Add a trace for a request oriented action
641  * @q:          queue the io is for
642  * @rq:         the source request
643  * @what:       the action
644  *
645  * Description:
646  *     Records an action against a request. Will log the bio offset + size.
647  *
648  **/
649 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
650                                     u32 what)
651 {
652         struct blk_trace *bt = q->blk_trace;
653         int rw = rq->cmd_flags & 0x03;
654
655         if (likely(!bt))
656                 return;
657
658         if (blk_discard_rq(rq))
659                 rw |= (1 << BIO_RW_DISCARD);
660
661         if (blk_pc_request(rq)) {
662                 what |= BLK_TC_ACT(BLK_TC_PC);
663                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
664                                 what, rq->errors, rq->cmd_len, rq->cmd);
665         } else  {
666                 what |= BLK_TC_ACT(BLK_TC_FS);
667                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
668                                 what, rq->errors, 0, NULL);
669         }
670 }
671
672 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
673 {
674         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
675 }
676
677 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
678 {
679         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
680 }
681
682 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
683 {
684         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
685 }
686
687 static void blk_add_trace_rq_requeue(struct request_queue *q,
688                                      struct request *rq)
689 {
690         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
691 }
692
693 static void blk_add_trace_rq_complete(struct request_queue *q,
694                                       struct request *rq)
695 {
696         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
697 }
698
699 /**
700  * blk_add_trace_bio - Add a trace for a bio oriented action
701  * @q:          queue the io is for
702  * @bio:        the source bio
703  * @what:       the action
704  *
705  * Description:
706  *     Records an action against a bio. Will log the bio offset + size.
707  *
708  **/
709 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
710                                      u32 what)
711 {
712         struct blk_trace *bt = q->blk_trace;
713
714         if (likely(!bt))
715                 return;
716
717         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
718                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
719 }
720
721 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
722 {
723         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
724 }
725
726 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
727 {
728         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
729 }
730
731 static void blk_add_trace_bio_backmerge(struct request_queue *q,
732                                         struct bio *bio)
733 {
734         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
735 }
736
737 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
738                                          struct bio *bio)
739 {
740         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
741 }
742
743 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
744 {
745         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
746 }
747
748 static void blk_add_trace_getrq(struct request_queue *q,
749                                 struct bio *bio, int rw)
750 {
751         if (bio)
752                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
753         else {
754                 struct blk_trace *bt = q->blk_trace;
755
756                 if (bt)
757                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
758         }
759 }
760
761
762 static void blk_add_trace_sleeprq(struct request_queue *q,
763                                   struct bio *bio, int rw)
764 {
765         if (bio)
766                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
767         else {
768                 struct blk_trace *bt = q->blk_trace;
769
770                 if (bt)
771                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
772                                         0, 0, NULL);
773         }
774 }
775
776 static void blk_add_trace_plug(struct request_queue *q)
777 {
778         struct blk_trace *bt = q->blk_trace;
779
780         if (bt)
781                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
782 }
783
784 static void blk_add_trace_unplug_io(struct request_queue *q)
785 {
786         struct blk_trace *bt = q->blk_trace;
787
788         if (bt) {
789                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
790                 __be64 rpdu = cpu_to_be64(pdu);
791
792                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
793                                 sizeof(rpdu), &rpdu);
794         }
795 }
796
797 static void blk_add_trace_unplug_timer(struct request_queue *q)
798 {
799         struct blk_trace *bt = q->blk_trace;
800
801         if (bt) {
802                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
803                 __be64 rpdu = cpu_to_be64(pdu);
804
805                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
806                                 sizeof(rpdu), &rpdu);
807         }
808 }
809
810 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
811                                 unsigned int pdu)
812 {
813         struct blk_trace *bt = q->blk_trace;
814
815         if (bt) {
816                 __be64 rpdu = cpu_to_be64(pdu);
817
818                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
819                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
820                                 sizeof(rpdu), &rpdu);
821         }
822 }
823
824 /**
825  * blk_add_trace_remap - Add a trace for a remap operation
826  * @q:          queue the io is for
827  * @bio:        the source bio
828  * @dev:        target device
829  * @from:       source sector
830  *
831  * Description:
832  *     Device mapper or raid target sometimes need to split a bio because
833  *     it spans a stripe (or similar). Add a trace for that action.
834  *
835  **/
836 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
837                                        dev_t dev, sector_t from)
838 {
839         struct blk_trace *bt = q->blk_trace;
840         struct blk_io_trace_remap r;
841
842         if (likely(!bt))
843                 return;
844
845         r.device_from = cpu_to_be32(dev);
846         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
847         r.sector_from = cpu_to_be64(from);
848
849         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
850                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
851                         sizeof(r), &r);
852 }
853
854 /**
855  * blk_add_driver_data - Add binary message with driver-specific data
856  * @q:          queue the io is for
857  * @rq:         io request
858  * @data:       driver-specific data
859  * @len:        length of driver-specific data
860  *
861  * Description:
862  *     Some drivers might want to write driver-specific data per request.
863  *
864  **/
865 void blk_add_driver_data(struct request_queue *q,
866                          struct request *rq,
867                          void *data, size_t len)
868 {
869         struct blk_trace *bt = q->blk_trace;
870
871         if (likely(!bt))
872                 return;
873
874         if (blk_pc_request(rq))
875                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
876                                 BLK_TA_DRV_DATA, rq->errors, len, data);
877         else
878                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
879                                 BLK_TA_DRV_DATA, rq->errors, len, data);
880 }
881 EXPORT_SYMBOL_GPL(blk_add_driver_data);
882
883 static void blk_register_tracepoints(void)
884 {
885         int ret;
886
887         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
888         WARN_ON(ret);
889         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
890         WARN_ON(ret);
891         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
892         WARN_ON(ret);
893         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
894         WARN_ON(ret);
895         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
896         WARN_ON(ret);
897         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
898         WARN_ON(ret);
899         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
900         WARN_ON(ret);
901         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
902         WARN_ON(ret);
903         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
904         WARN_ON(ret);
905         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
906         WARN_ON(ret);
907         ret = register_trace_block_getrq(blk_add_trace_getrq);
908         WARN_ON(ret);
909         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
910         WARN_ON(ret);
911         ret = register_trace_block_plug(blk_add_trace_plug);
912         WARN_ON(ret);
913         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
914         WARN_ON(ret);
915         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
916         WARN_ON(ret);
917         ret = register_trace_block_split(blk_add_trace_split);
918         WARN_ON(ret);
919         ret = register_trace_block_remap(blk_add_trace_remap);
920         WARN_ON(ret);
921 }
922
923 static void blk_unregister_tracepoints(void)
924 {
925         unregister_trace_block_remap(blk_add_trace_remap);
926         unregister_trace_block_split(blk_add_trace_split);
927         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
928         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
929         unregister_trace_block_plug(blk_add_trace_plug);
930         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
931         unregister_trace_block_getrq(blk_add_trace_getrq);
932         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
933         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
934         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
935         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
936         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
937         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
938         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
939         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
940         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
941         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
942
943         tracepoint_synchronize_unregister();
944 }
945
946 /*
947  * struct blk_io_tracer formatting routines
948  */
949
950 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
951 {
952         int i = 0;
953         int tc = t->action >> BLK_TC_SHIFT;
954
955         if (t->action == BLK_TN_MESSAGE) {
956                 rwbs[i++] = 'N';
957                 goto out;
958         }
959
960         if (tc & BLK_TC_DISCARD)
961                 rwbs[i++] = 'D';
962         else if (tc & BLK_TC_WRITE)
963                 rwbs[i++] = 'W';
964         else if (t->bytes)
965                 rwbs[i++] = 'R';
966         else
967                 rwbs[i++] = 'N';
968
969         if (tc & BLK_TC_AHEAD)
970                 rwbs[i++] = 'A';
971         if (tc & BLK_TC_BARRIER)
972                 rwbs[i++] = 'B';
973         if (tc & BLK_TC_SYNC)
974                 rwbs[i++] = 'S';
975         if (tc & BLK_TC_META)
976                 rwbs[i++] = 'M';
977 out:
978         rwbs[i] = '\0';
979 }
980
981 static inline
982 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
983 {
984         return (const struct blk_io_trace *)ent;
985 }
986
987 static inline const void *pdu_start(const struct trace_entry *ent)
988 {
989         return te_blk_io_trace(ent) + 1;
990 }
991
992 static inline u32 t_action(const struct trace_entry *ent)
993 {
994         return te_blk_io_trace(ent)->action;
995 }
996
997 static inline u32 t_bytes(const struct trace_entry *ent)
998 {
999         return te_blk_io_trace(ent)->bytes;
1000 }
1001
1002 static inline u32 t_sec(const struct trace_entry *ent)
1003 {
1004         return te_blk_io_trace(ent)->bytes >> 9;
1005 }
1006
1007 static inline unsigned long long t_sector(const struct trace_entry *ent)
1008 {
1009         return te_blk_io_trace(ent)->sector;
1010 }
1011
1012 static inline __u16 t_error(const struct trace_entry *ent)
1013 {
1014         return te_blk_io_trace(ent)->error;
1015 }
1016
1017 static __u64 get_pdu_int(const struct trace_entry *ent)
1018 {
1019         const __u64 *val = pdu_start(ent);
1020         return be64_to_cpu(*val);
1021 }
1022
1023 static void get_pdu_remap(const struct trace_entry *ent,
1024                           struct blk_io_trace_remap *r)
1025 {
1026         const struct blk_io_trace_remap *__r = pdu_start(ent);
1027         __u64 sector_from = __r->sector_from;
1028
1029         r->device_from = be32_to_cpu(__r->device_from);
1030         r->device_to   = be32_to_cpu(__r->device_to);
1031         r->sector_from = be64_to_cpu(sector_from);
1032 }
1033
1034 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1035
1036 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1037 {
1038         char rwbs[6];
1039         unsigned long long ts  = iter->ts;
1040         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1041         unsigned secs          = (unsigned long)ts;
1042         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1043
1044         fill_rwbs(rwbs, t);
1045
1046         return trace_seq_printf(&iter->seq,
1047                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1048                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1049                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1050 }
1051
1052 static int blk_log_action(struct trace_iterator *iter, const char *act)
1053 {
1054         char rwbs[6];
1055         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1056
1057         fill_rwbs(rwbs, t);
1058         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1059                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1060 }
1061
1062 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1063 {
1064         const unsigned char *pdu_buf;
1065         int pdu_len;
1066         int i, end, ret;
1067
1068         pdu_buf = pdu_start(ent);
1069         pdu_len = te_blk_io_trace(ent)->pdu_len;
1070
1071         if (!pdu_len)
1072                 return 1;
1073
1074         /* find the last zero that needs to be printed */
1075         for (end = pdu_len - 1; end >= 0; end--)
1076                 if (pdu_buf[end])
1077                         break;
1078         end++;
1079
1080         if (!trace_seq_putc(s, '('))
1081                 return 0;
1082
1083         for (i = 0; i < pdu_len; i++) {
1084
1085                 ret = trace_seq_printf(s, "%s%02x",
1086                                        i == 0 ? "" : " ", pdu_buf[i]);
1087                 if (!ret)
1088                         return ret;
1089
1090                 /*
1091                  * stop when the rest is just zeroes and indicate so
1092                  * with a ".." appended
1093                  */
1094                 if (i == end && end != pdu_len - 1)
1095                         return trace_seq_puts(s, " ..) ");
1096         }
1097
1098         return trace_seq_puts(s, ") ");
1099 }
1100
1101 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1102 {
1103         char cmd[TASK_COMM_LEN];
1104
1105         trace_find_cmdline(ent->pid, cmd);
1106
1107         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1108                 int ret;
1109
1110                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1111                 if (!ret)
1112                         return 0;
1113                 ret = blk_log_dump_pdu(s, ent);
1114                 if (!ret)
1115                         return 0;
1116                 return trace_seq_printf(s, "[%s]\n", cmd);
1117         } else {
1118                 if (t_sec(ent))
1119                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1120                                                 t_sector(ent), t_sec(ent), cmd);
1121                 return trace_seq_printf(s, "[%s]\n", cmd);
1122         }
1123 }
1124
1125 static int blk_log_with_error(struct trace_seq *s,
1126                               const struct trace_entry *ent)
1127 {
1128         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1129                 int ret;
1130
1131                 ret = blk_log_dump_pdu(s, ent);
1132                 if (ret)
1133                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1134                 return 0;
1135         } else {
1136                 if (t_sec(ent))
1137                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1138                                                 t_sector(ent),
1139                                                 t_sec(ent), t_error(ent));
1140                 return trace_seq_printf(s, "%llu [%d]\n",
1141                                         t_sector(ent), t_error(ent));
1142         }
1143 }
1144
1145 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1146 {
1147         struct blk_io_trace_remap r = { .device_from = 0, };
1148
1149         get_pdu_remap(ent, &r);
1150         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1151                                 t_sector(ent), t_sec(ent),
1152                                 MAJOR(r.device_from), MINOR(r.device_from),
1153                                 (unsigned long long)r.sector_from);
1154 }
1155
1156 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1157 {
1158         char cmd[TASK_COMM_LEN];
1159
1160         trace_find_cmdline(ent->pid, cmd);
1161
1162         return trace_seq_printf(s, "[%s]\n", cmd);
1163 }
1164
1165 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1166 {
1167         char cmd[TASK_COMM_LEN];
1168
1169         trace_find_cmdline(ent->pid, cmd);
1170
1171         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1172 }
1173
1174 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1175 {
1176         char cmd[TASK_COMM_LEN];
1177
1178         trace_find_cmdline(ent->pid, cmd);
1179
1180         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1181                                 get_pdu_int(ent), cmd);
1182 }
1183
1184 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1185 {
1186         int ret;
1187         const struct blk_io_trace *t = te_blk_io_trace(ent);
1188
1189         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1190         if (ret)
1191                 return trace_seq_putc(s, '\n');
1192         return ret;
1193 }
1194
1195 /*
1196  * struct tracer operations
1197  */
1198
1199 static void blk_tracer_print_header(struct seq_file *m)
1200 {
1201         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1202                 return;
1203         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1204                     "#  |     |     |           |   |   |\n");
1205 }
1206
1207 static void blk_tracer_start(struct trace_array *tr)
1208 {
1209         blk_tracer_enabled = true;
1210 }
1211
1212 static int blk_tracer_init(struct trace_array *tr)
1213 {
1214         blk_tr = tr;
1215         blk_tracer_start(tr);
1216         return 0;
1217 }
1218
1219 static void blk_tracer_stop(struct trace_array *tr)
1220 {
1221         blk_tracer_enabled = false;
1222 }
1223
1224 static void blk_tracer_reset(struct trace_array *tr)
1225 {
1226         blk_tracer_stop(tr);
1227 }
1228
1229 static const struct {
1230         const char *act[2];
1231         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1232 } what2act[] = {
1233         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1234         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1235         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1236         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1237         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1238         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1239         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1240         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1241         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1242         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1243         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1244         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1245         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1246         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1247         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1248 };
1249
1250 static enum print_line_t print_one_line(struct trace_iterator *iter,
1251                                         bool classic)
1252 {
1253         struct trace_seq *s = &iter->seq;
1254         const struct blk_io_trace *t;
1255         u16 what;
1256         int ret;
1257         bool long_act;
1258         blk_log_action_t *log_action;
1259
1260         t          = te_blk_io_trace(iter->ent);
1261         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1262         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1263         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1264
1265         if (t->action == BLK_TN_MESSAGE) {
1266                 ret = log_action(iter, long_act ? "message" : "m");
1267                 if (ret)
1268                         ret = blk_log_msg(s, iter->ent);
1269                 goto out;
1270         }
1271
1272         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1273                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1274         else {
1275                 ret = log_action(iter, what2act[what].act[long_act]);
1276                 if (ret)
1277                         ret = what2act[what].print(s, iter->ent);
1278         }
1279 out:
1280         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1281 }
1282
1283 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1284                                                int flags)
1285 {
1286         return print_one_line(iter, false);
1287 }
1288
1289 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1290 {
1291         struct trace_seq *s = &iter->seq;
1292         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1293         const int offset = offsetof(struct blk_io_trace, sector);
1294         struct blk_io_trace old = {
1295                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1296                 .time     = iter->ts,
1297         };
1298
1299         if (!trace_seq_putmem(s, &old, offset))
1300                 return 0;
1301         return trace_seq_putmem(s, &t->sector,
1302                                 sizeof(old) - offset + t->pdu_len);
1303 }
1304
1305 static enum print_line_t
1306 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1307 {
1308         return blk_trace_synthesize_old_trace(iter) ?
1309                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1310 }
1311
1312 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1313 {
1314         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1315                 return TRACE_TYPE_UNHANDLED;
1316
1317         return print_one_line(iter, true);
1318 }
1319
1320 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1321 {
1322         /* don't output context-info for blk_classic output */
1323         if (bit == TRACE_BLK_OPT_CLASSIC) {
1324                 if (set)
1325                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1326                 else
1327                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1328         }
1329         return 0;
1330 }
1331
1332 static struct tracer blk_tracer __read_mostly = {
1333         .name           = "blk",
1334         .init           = blk_tracer_init,
1335         .reset          = blk_tracer_reset,
1336         .start          = blk_tracer_start,
1337         .stop           = blk_tracer_stop,
1338         .print_header   = blk_tracer_print_header,
1339         .print_line     = blk_tracer_print_line,
1340         .flags          = &blk_tracer_flags,
1341         .set_flag       = blk_tracer_set_flag,
1342 };
1343
1344 static struct trace_event trace_blk_event = {
1345         .type           = TRACE_BLK,
1346         .trace          = blk_trace_event_print,
1347         .binary         = blk_trace_event_print_binary,
1348 };
1349
1350 static int __init init_blk_tracer(void)
1351 {
1352         if (!register_ftrace_event(&trace_blk_event)) {
1353                 pr_warning("Warning: could not register block events\n");
1354                 return 1;
1355         }
1356
1357         if (register_tracer(&blk_tracer) != 0) {
1358                 pr_warning("Warning: could not register the block tracer\n");
1359                 unregister_ftrace_event(&trace_blk_event);
1360                 return 1;
1361         }
1362
1363         return 0;
1364 }
1365
1366 device_initcall(init_blk_tracer);
1367
1368 static int blk_trace_remove_queue(struct request_queue *q)
1369 {
1370         struct blk_trace *bt;
1371
1372         bt = xchg(&q->blk_trace, NULL);
1373         if (bt == NULL)
1374                 return -EINVAL;
1375
1376         if (atomic_dec_and_test(&blk_probes_ref))
1377                 blk_unregister_tracepoints();
1378
1379         blk_trace_free(bt);
1380         return 0;
1381 }
1382
1383 /*
1384  * Setup everything required to start tracing
1385  */
1386 static int blk_trace_setup_queue(struct request_queue *q,
1387                                  struct block_device *bdev)
1388 {
1389         struct blk_trace *old_bt, *bt = NULL;
1390         int ret = -ENOMEM;
1391
1392         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1393         if (!bt)
1394                 return -ENOMEM;
1395
1396         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1397         if (!bt->msg_data)
1398                 goto free_bt;
1399
1400         bt->dev = bdev->bd_dev;
1401         bt->act_mask = (u16)-1;
1402
1403         blk_trace_setup_lba(bt, bdev);
1404
1405         old_bt = xchg(&q->blk_trace, bt);
1406         if (old_bt != NULL) {
1407                 (void)xchg(&q->blk_trace, old_bt);
1408                 ret = -EBUSY;
1409                 goto free_bt;
1410         }
1411
1412         if (atomic_inc_return(&blk_probes_ref) == 1)
1413                 blk_register_tracepoints();
1414         return 0;
1415
1416 free_bt:
1417         blk_trace_free(bt);
1418         return ret;
1419 }
1420
1421 /*
1422  * sysfs interface to enable and configure tracing
1423  */
1424
1425 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1426                                          struct device_attribute *attr,
1427                                          char *buf);
1428 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1429                                           struct device_attribute *attr,
1430                                           const char *buf, size_t count);
1431 #define BLK_TRACE_DEVICE_ATTR(_name) \
1432         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1433                     sysfs_blk_trace_attr_show, \
1434                     sysfs_blk_trace_attr_store)
1435
1436 static BLK_TRACE_DEVICE_ATTR(enable);
1437 static BLK_TRACE_DEVICE_ATTR(act_mask);
1438 static BLK_TRACE_DEVICE_ATTR(pid);
1439 static BLK_TRACE_DEVICE_ATTR(start_lba);
1440 static BLK_TRACE_DEVICE_ATTR(end_lba);
1441
1442 static struct attribute *blk_trace_attrs[] = {
1443         &dev_attr_enable.attr,
1444         &dev_attr_act_mask.attr,
1445         &dev_attr_pid.attr,
1446         &dev_attr_start_lba.attr,
1447         &dev_attr_end_lba.attr,
1448         NULL
1449 };
1450
1451 struct attribute_group blk_trace_attr_group = {
1452         .name  = "trace",
1453         .attrs = blk_trace_attrs,
1454 };
1455
1456 static const struct {
1457         int mask;
1458         const char *str;
1459 } mask_maps[] = {
1460         { BLK_TC_READ,          "read"          },
1461         { BLK_TC_WRITE,         "write"         },
1462         { BLK_TC_BARRIER,       "barrier"       },
1463         { BLK_TC_SYNC,          "sync"          },
1464         { BLK_TC_QUEUE,         "queue"         },
1465         { BLK_TC_REQUEUE,       "requeue"       },
1466         { BLK_TC_ISSUE,         "issue"         },
1467         { BLK_TC_COMPLETE,      "complete"      },
1468         { BLK_TC_FS,            "fs"            },
1469         { BLK_TC_PC,            "pc"            },
1470         { BLK_TC_AHEAD,         "ahead"         },
1471         { BLK_TC_META,          "meta"          },
1472         { BLK_TC_DISCARD,       "discard"       },
1473         { BLK_TC_DRV_DATA,      "drv_data"      },
1474 };
1475
1476 static int blk_trace_str2mask(const char *str)
1477 {
1478         int i;
1479         int mask = 0;
1480         char *buf, *s, *token;
1481
1482         buf = kstrdup(str, GFP_KERNEL);
1483         if (buf == NULL)
1484                 return -ENOMEM;
1485         s = strstrip(buf);
1486
1487         while (1) {
1488                 token = strsep(&s, ",");
1489                 if (token == NULL)
1490                         break;
1491
1492                 if (*token == '\0')
1493                         continue;
1494
1495                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1496                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1497                                 mask |= mask_maps[i].mask;
1498                                 break;
1499                         }
1500                 }
1501                 if (i == ARRAY_SIZE(mask_maps)) {
1502                         mask = -EINVAL;
1503                         break;
1504                 }
1505         }
1506         kfree(buf);
1507
1508         return mask;
1509 }
1510
1511 static ssize_t blk_trace_mask2str(char *buf, int mask)
1512 {
1513         int i;
1514         char *p = buf;
1515
1516         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1517                 if (mask & mask_maps[i].mask) {
1518                         p += sprintf(p, "%s%s",
1519                                     (p == buf) ? "" : ",", mask_maps[i].str);
1520                 }
1521         }
1522         *p++ = '\n';
1523
1524         return p - buf;
1525 }
1526
1527 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1528 {
1529         if (bdev->bd_disk == NULL)
1530                 return NULL;
1531
1532         return bdev_get_queue(bdev);
1533 }
1534
1535 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1536                                          struct device_attribute *attr,
1537                                          char *buf)
1538 {
1539         struct hd_struct *p = dev_to_part(dev);
1540         struct request_queue *q;
1541         struct block_device *bdev;
1542         ssize_t ret = -ENXIO;
1543
1544         lock_kernel();
1545         bdev = bdget(part_devt(p));
1546         if (bdev == NULL)
1547                 goto out_unlock_kernel;
1548
1549         q = blk_trace_get_queue(bdev);
1550         if (q == NULL)
1551                 goto out_bdput;
1552
1553         mutex_lock(&bdev->bd_mutex);
1554
1555         if (attr == &dev_attr_enable) {
1556                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1557                 goto out_unlock_bdev;
1558         }
1559
1560         if (q->blk_trace == NULL)
1561                 ret = sprintf(buf, "disabled\n");
1562         else if (attr == &dev_attr_act_mask)
1563                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1564         else if (attr == &dev_attr_pid)
1565                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1566         else if (attr == &dev_attr_start_lba)
1567                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1568         else if (attr == &dev_attr_end_lba)
1569                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1570
1571 out_unlock_bdev:
1572         mutex_unlock(&bdev->bd_mutex);
1573 out_bdput:
1574         bdput(bdev);
1575 out_unlock_kernel:
1576         unlock_kernel();
1577         return ret;
1578 }
1579
1580 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1581                                           struct device_attribute *attr,
1582                                           const char *buf, size_t count)
1583 {
1584         struct block_device *bdev;
1585         struct request_queue *q;
1586         struct hd_struct *p;
1587         u64 value;
1588         ssize_t ret = -EINVAL;
1589
1590         if (count == 0)
1591                 goto out;
1592
1593         if (attr == &dev_attr_act_mask) {
1594                 if (sscanf(buf, "%llx", &value) != 1) {
1595                         /* Assume it is a list of trace category names */
1596                         ret = blk_trace_str2mask(buf);
1597                         if (ret < 0)
1598                                 goto out;
1599                         value = ret;
1600                 }
1601         } else if (sscanf(buf, "%llu", &value) != 1)
1602                 goto out;
1603
1604         ret = -ENXIO;
1605
1606         lock_kernel();
1607         p = dev_to_part(dev);
1608         bdev = bdget(part_devt(p));
1609         if (bdev == NULL)
1610                 goto out_unlock_kernel;
1611
1612         q = blk_trace_get_queue(bdev);
1613         if (q == NULL)
1614                 goto out_bdput;
1615
1616         mutex_lock(&bdev->bd_mutex);
1617
1618         if (attr == &dev_attr_enable) {
1619                 if (value)
1620                         ret = blk_trace_setup_queue(q, bdev);
1621                 else
1622                         ret = blk_trace_remove_queue(q);
1623                 goto out_unlock_bdev;
1624         }
1625
1626         ret = 0;
1627         if (q->blk_trace == NULL)
1628                 ret = blk_trace_setup_queue(q, bdev);
1629
1630         if (ret == 0) {
1631                 if (attr == &dev_attr_act_mask)
1632                         q->blk_trace->act_mask = value;
1633                 else if (attr == &dev_attr_pid)
1634                         q->blk_trace->pid = value;
1635                 else if (attr == &dev_attr_start_lba)
1636                         q->blk_trace->start_lba = value;
1637                 else if (attr == &dev_attr_end_lba)
1638                         q->blk_trace->end_lba = value;
1639         }
1640
1641 out_unlock_bdev:
1642         mutex_unlock(&bdev->bd_mutex);
1643 out_bdput:
1644         bdput(bdev);
1645 out_unlock_kernel:
1646         unlock_kernel();
1647 out:
1648         return ret ? ret : count;
1649 }
1650
1651 int blk_trace_init_sysfs(struct device *dev)
1652 {
1653         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1654 }
1655
1656 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1657
1658 #ifdef CONFIG_EVENT_TRACING
1659
1660 void blk_dump_cmd(char *buf, struct request *rq)
1661 {
1662         int i, end;
1663         int len = rq->cmd_len;
1664         unsigned char *cmd = rq->cmd;
1665
1666         if (!blk_pc_request(rq)) {
1667                 buf[0] = '\0';
1668                 return;
1669         }
1670
1671         for (end = len - 1; end >= 0; end--)
1672                 if (cmd[end])
1673                         break;
1674         end++;
1675
1676         for (i = 0; i < len; i++) {
1677                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1678                 if (i == end && end != len - 1) {
1679                         sprintf(buf, " ..");
1680                         break;
1681                 }
1682         }
1683 }
1684
1685 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1686 {
1687         int i = 0;
1688
1689         if (rw & WRITE)
1690                 rwbs[i++] = 'W';
1691         else if (rw & 1 << BIO_RW_DISCARD)
1692                 rwbs[i++] = 'D';
1693         else if (bytes)
1694                 rwbs[i++] = 'R';
1695         else
1696                 rwbs[i++] = 'N';
1697
1698         if (rw & 1 << BIO_RW_AHEAD)
1699                 rwbs[i++] = 'A';
1700         if (rw & 1 << BIO_RW_BARRIER)
1701                 rwbs[i++] = 'B';
1702         if (rw & 1 << BIO_RW_SYNCIO)
1703                 rwbs[i++] = 'S';
1704         if (rw & 1 << BIO_RW_META)
1705                 rwbs[i++] = 'M';
1706
1707         rwbs[i] = '\0';
1708 }
1709
1710 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1711 {
1712         int rw = rq->cmd_flags & 0x03;
1713         int bytes;
1714
1715         if (blk_discard_rq(rq))
1716                 rw |= (1 << BIO_RW_DISCARD);
1717
1718         bytes = blk_rq_bytes(rq);
1719
1720         blk_fill_rwbs(rwbs, rw, bytes);
1721 }
1722
1723 #endif /* CONFIG_EVENT_TRACING */
1724