blktrace: report EBUSY correctly
[linux-2.6.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/time.h>
26 #include <trace/block.h>
27 #include <linux/uaccess.h>
28 #include "trace_output.h"
29
30 static unsigned int blktrace_seq __read_mostly = 1;
31
32 static struct trace_array *blk_tr;
33 static bool blk_tracer_enabled __read_mostly;
34
35 /* Select an alternative, minimalistic output than the original one */
36 #define TRACE_BLK_OPT_CLASSIC   0x1
37
38 static struct tracer_opt blk_tracer_opts[] = {
39         /* Default disable the minimalistic output */
40         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
41         { }
42 };
43
44 static struct tracer_flags blk_tracer_flags = {
45         .val  = 0,
46         .opts = blk_tracer_opts,
47 };
48
49 /* Global reference count of probes */
50 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
51
52 static void blk_register_tracepoints(void);
53 static void blk_unregister_tracepoints(void);
54
55 /*
56  * Send out a notify message.
57  */
58 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
59                        const void *data, size_t len)
60 {
61         struct blk_io_trace *t;
62
63         if (!bt->rchan)
64                 return;
65
66         t = relay_reserve(bt->rchan, sizeof(*t) + len);
67         if (t) {
68                 const int cpu = smp_processor_id();
69
70                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
71                 t->time = ktime_to_ns(ktime_get());
72                 t->device = bt->dev;
73                 t->action = action;
74                 t->pid = pid;
75                 t->cpu = cpu;
76                 t->pdu_len = len;
77                 memcpy((void *) t + sizeof(*t), data, len);
78         }
79 }
80
81 /*
82  * Send out a notify for this process, if we haven't done so since a trace
83  * started
84  */
85 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
86 {
87         tsk->btrace_seq = blktrace_seq;
88         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
89 }
90
91 static void trace_note_time(struct blk_trace *bt)
92 {
93         struct timespec now;
94         unsigned long flags;
95         u32 words[2];
96
97         getnstimeofday(&now);
98         words[0] = now.tv_sec;
99         words[1] = now.tv_nsec;
100
101         local_irq_save(flags);
102         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
103         local_irq_restore(flags);
104 }
105
106 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
107 {
108         int n;
109         va_list args;
110         unsigned long flags;
111         char *buf;
112
113         if (blk_tr) {
114                 va_start(args, fmt);
115                 ftrace_vprintk(fmt, args);
116                 va_end(args);
117                 return;
118         }
119
120         if (!bt->msg_data)
121                 return;
122
123         local_irq_save(flags);
124         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
125         va_start(args, fmt);
126         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
127         va_end(args);
128
129         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
130         local_irq_restore(flags);
131 }
132 EXPORT_SYMBOL_GPL(__trace_note_message);
133
134 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
135                          pid_t pid)
136 {
137         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
138                 return 1;
139         if (sector < bt->start_lba || sector > bt->end_lba)
140                 return 1;
141         if (bt->pid && pid != bt->pid)
142                 return 1;
143
144         return 0;
145 }
146
147 /*
148  * Data direction bit lookup
149  */
150 static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ),
151                                          BLK_TC_ACT(BLK_TC_WRITE) };
152
153 /* The ilog2() calls fall out because they're constant */
154 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
155           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
156
157 /*
158  * The worker for the various blk_add_trace*() types. Fills out a
159  * blk_io_trace structure and places it in a per-cpu subbuffer.
160  */
161 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
162                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
163 {
164         struct task_struct *tsk = current;
165         struct ring_buffer_event *event = NULL;
166         struct blk_io_trace *t;
167         unsigned long flags = 0;
168         unsigned long *sequence;
169         pid_t pid;
170         int cpu, pc = 0;
171
172         if (unlikely(bt->trace_state != Blktrace_running ||
173                      !blk_tracer_enabled))
174                 return;
175
176         what |= ddir_act[rw & WRITE];
177         what |= MASK_TC_BIT(rw, BARRIER);
178         what |= MASK_TC_BIT(rw, SYNCIO);
179         what |= MASK_TC_BIT(rw, AHEAD);
180         what |= MASK_TC_BIT(rw, META);
181         what |= MASK_TC_BIT(rw, DISCARD);
182
183         pid = tsk->pid;
184         if (unlikely(act_log_check(bt, what, sector, pid)))
185                 return;
186         cpu = raw_smp_processor_id();
187
188         if (blk_tr) {
189                 tracing_record_cmdline(current);
190
191                 pc = preempt_count();
192                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
193                                                   sizeof(*t) + pdu_len,
194                                                   0, pc);
195                 if (!event)
196                         return;
197                 t = ring_buffer_event_data(event);
198                 goto record_it;
199         }
200
201         /*
202          * A word about the locking here - we disable interrupts to reserve
203          * some space in the relay per-cpu buffer, to prevent an irq
204          * from coming in and stepping on our toes.
205          */
206         local_irq_save(flags);
207
208         if (unlikely(tsk->btrace_seq != blktrace_seq))
209                 trace_note_tsk(bt, tsk);
210
211         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
212         if (t) {
213                 sequence = per_cpu_ptr(bt->sequence, cpu);
214
215                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
216                 t->sequence = ++(*sequence);
217                 t->time = ktime_to_ns(ktime_get());
218 record_it:
219                 /*
220                  * These two are not needed in ftrace as they are in the
221                  * generic trace_entry, filled by tracing_generic_entry_update,
222                  * but for the trace_event->bin() synthesizer benefit we do it
223                  * here too.
224                  */
225                 t->cpu = cpu;
226                 t->pid = pid;
227
228                 t->sector = sector;
229                 t->bytes = bytes;
230                 t->action = what;
231                 t->device = bt->dev;
232                 t->error = error;
233                 t->pdu_len = pdu_len;
234
235                 if (pdu_len)
236                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
237
238                 if (blk_tr) {
239                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
240                         return;
241                 }
242         }
243
244         local_irq_restore(flags);
245 }
246
247 static struct dentry *blk_tree_root;
248 static DEFINE_MUTEX(blk_tree_mutex);
249
250 static void blk_trace_cleanup(struct blk_trace *bt)
251 {
252         debugfs_remove(bt->msg_file);
253         debugfs_remove(bt->dropped_file);
254         relay_close(bt->rchan);
255         free_percpu(bt->sequence);
256         free_percpu(bt->msg_data);
257         kfree(bt);
258         if (atomic_dec_and_test(&blk_probes_ref))
259                 blk_unregister_tracepoints();
260 }
261
262 int blk_trace_remove(struct request_queue *q)
263 {
264         struct blk_trace *bt;
265
266         bt = xchg(&q->blk_trace, NULL);
267         if (!bt)
268                 return -EINVAL;
269
270         if (bt->trace_state == Blktrace_setup ||
271             bt->trace_state == Blktrace_stopped)
272                 blk_trace_cleanup(bt);
273
274         return 0;
275 }
276 EXPORT_SYMBOL_GPL(blk_trace_remove);
277
278 static int blk_dropped_open(struct inode *inode, struct file *filp)
279 {
280         filp->private_data = inode->i_private;
281
282         return 0;
283 }
284
285 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
286                                 size_t count, loff_t *ppos)
287 {
288         struct blk_trace *bt = filp->private_data;
289         char buf[16];
290
291         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
292
293         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
294 }
295
296 static const struct file_operations blk_dropped_fops = {
297         .owner =        THIS_MODULE,
298         .open =         blk_dropped_open,
299         .read =         blk_dropped_read,
300 };
301
302 static int blk_msg_open(struct inode *inode, struct file *filp)
303 {
304         filp->private_data = inode->i_private;
305
306         return 0;
307 }
308
309 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
310                                 size_t count, loff_t *ppos)
311 {
312         char *msg;
313         struct blk_trace *bt;
314
315         if (count > BLK_TN_MAX_MSG)
316                 return -EINVAL;
317
318         msg = kmalloc(count, GFP_KERNEL);
319         if (msg == NULL)
320                 return -ENOMEM;
321
322         if (copy_from_user(msg, buffer, count)) {
323                 kfree(msg);
324                 return -EFAULT;
325         }
326
327         bt = filp->private_data;
328         __trace_note_message(bt, "%s", msg);
329         kfree(msg);
330
331         return count;
332 }
333
334 static const struct file_operations blk_msg_fops = {
335         .owner =        THIS_MODULE,
336         .open =         blk_msg_open,
337         .write =        blk_msg_write,
338 };
339
340 /*
341  * Keep track of how many times we encountered a full subbuffer, to aid
342  * the user space app in telling how many lost events there were.
343  */
344 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
345                                      void *prev_subbuf, size_t prev_padding)
346 {
347         struct blk_trace *bt;
348
349         if (!relay_buf_full(buf))
350                 return 1;
351
352         bt = buf->chan->private_data;
353         atomic_inc(&bt->dropped);
354         return 0;
355 }
356
357 static int blk_remove_buf_file_callback(struct dentry *dentry)
358 {
359         struct dentry *parent = dentry->d_parent;
360         debugfs_remove(dentry);
361
362         /*
363         * this will fail for all but the last file, but that is ok. what we
364         * care about is the top level buts->name directory going away, when
365         * the last trace file is gone. Then we don't have to rmdir() that
366         * manually on trace stop, so it nicely solves the issue with
367         * force killing of running traces.
368         */
369
370         debugfs_remove(parent);
371         return 0;
372 }
373
374 static struct dentry *blk_create_buf_file_callback(const char *filename,
375                                                    struct dentry *parent,
376                                                    int mode,
377                                                    struct rchan_buf *buf,
378                                                    int *is_global)
379 {
380         return debugfs_create_file(filename, mode, parent, buf,
381                                         &relay_file_operations);
382 }
383
384 static struct rchan_callbacks blk_relay_callbacks = {
385         .subbuf_start           = blk_subbuf_start_callback,
386         .create_buf_file        = blk_create_buf_file_callback,
387         .remove_buf_file        = blk_remove_buf_file_callback,
388 };
389
390 /*
391  * Setup everything required to start tracing
392  */
393 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
394                         struct blk_user_trace_setup *buts)
395 {
396         struct blk_trace *old_bt, *bt = NULL;
397         struct dentry *dir = NULL;
398         int ret, i;
399
400         if (!buts->buf_size || !buts->buf_nr)
401                 return -EINVAL;
402
403         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
404         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
405
406         /*
407          * some device names have larger paths - convert the slashes
408          * to underscores for this to work as expected
409          */
410         for (i = 0; i < strlen(buts->name); i++)
411                 if (buts->name[i] == '/')
412                         buts->name[i] = '_';
413
414         ret = -ENOMEM;
415         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
416         if (!bt)
417                 goto err;
418
419         bt->sequence = alloc_percpu(unsigned long);
420         if (!bt->sequence)
421                 goto err;
422
423         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
424         if (!bt->msg_data)
425                 goto err;
426
427         ret = -ENOENT;
428
429         if (!blk_tree_root) {
430                 blk_tree_root = debugfs_create_dir("block", NULL);
431                 if (!blk_tree_root)
432                         goto err;
433         }
434
435         dir = debugfs_create_dir(buts->name, blk_tree_root);
436
437         if (!dir)
438                 goto err;
439
440         bt->dir = dir;
441         bt->dev = dev;
442         atomic_set(&bt->dropped, 0);
443
444         ret = -EIO;
445         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
446                                                &blk_dropped_fops);
447         if (!bt->dropped_file)
448                 goto err;
449
450         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
451         if (!bt->msg_file)
452                 goto err;
453
454         bt->rchan = relay_open("trace", dir, buts->buf_size,
455                                 buts->buf_nr, &blk_relay_callbacks, bt);
456         if (!bt->rchan)
457                 goto err;
458
459         bt->act_mask = buts->act_mask;
460         if (!bt->act_mask)
461                 bt->act_mask = (u16) -1;
462
463         bt->start_lba = buts->start_lba;
464         bt->end_lba = buts->end_lba;
465         if (!bt->end_lba)
466                 bt->end_lba = -1ULL;
467
468         bt->pid = buts->pid;
469         bt->trace_state = Blktrace_setup;
470
471         ret = -EBUSY;
472         old_bt = xchg(&q->blk_trace, bt);
473         if (old_bt) {
474                 (void) xchg(&q->blk_trace, old_bt);
475                 goto err;
476         }
477
478         if (atomic_add_return(1, &blk_probes_ref) == 1)
479                 blk_register_tracepoints();
480
481         return 0;
482 err:
483         if (bt) {
484                 if (bt->msg_file)
485                         debugfs_remove(bt->msg_file);
486                 if (bt->dropped_file)
487                         debugfs_remove(bt->dropped_file);
488                 free_percpu(bt->sequence);
489                 free_percpu(bt->msg_data);
490                 if (bt->rchan)
491                         relay_close(bt->rchan);
492                 kfree(bt);
493         }
494         return ret;
495 }
496
497 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
498                     char __user *arg)
499 {
500         struct blk_user_trace_setup buts;
501         int ret;
502
503         ret = copy_from_user(&buts, arg, sizeof(buts));
504         if (ret)
505                 return -EFAULT;
506
507         ret = do_blk_trace_setup(q, name, dev, &buts);
508         if (ret)
509                 return ret;
510
511         if (copy_to_user(arg, &buts, sizeof(buts)))
512                 return -EFAULT;
513
514         return 0;
515 }
516 EXPORT_SYMBOL_GPL(blk_trace_setup);
517
518 int blk_trace_startstop(struct request_queue *q, int start)
519 {
520         int ret;
521         struct blk_trace *bt = q->blk_trace;
522
523         if (bt == NULL)
524                 return -EINVAL;
525
526         /*
527          * For starting a trace, we can transition from a setup or stopped
528          * trace. For stopping a trace, the state must be running
529          */
530         ret = -EINVAL;
531         if (start) {
532                 if (bt->trace_state == Blktrace_setup ||
533                     bt->trace_state == Blktrace_stopped) {
534                         blktrace_seq++;
535                         smp_mb();
536                         bt->trace_state = Blktrace_running;
537
538                         trace_note_time(bt);
539                         ret = 0;
540                 }
541         } else {
542                 if (bt->trace_state == Blktrace_running) {
543                         bt->trace_state = Blktrace_stopped;
544                         relay_flush(bt->rchan);
545                         ret = 0;
546                 }
547         }
548
549         return ret;
550 }
551 EXPORT_SYMBOL_GPL(blk_trace_startstop);
552
553 /**
554  * blk_trace_ioctl: - handle the ioctls associated with tracing
555  * @bdev:       the block device
556  * @cmd:        the ioctl cmd
557  * @arg:        the argument data, if any
558  *
559  **/
560 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
561 {
562         struct request_queue *q;
563         int ret, start = 0;
564         char b[BDEVNAME_SIZE];
565
566         q = bdev_get_queue(bdev);
567         if (!q)
568                 return -ENXIO;
569
570         mutex_lock(&bdev->bd_mutex);
571
572         switch (cmd) {
573         case BLKTRACESETUP:
574                 bdevname(bdev, b);
575                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
576                 break;
577         case BLKTRACESTART:
578                 start = 1;
579         case BLKTRACESTOP:
580                 ret = blk_trace_startstop(q, start);
581                 break;
582         case BLKTRACETEARDOWN:
583                 ret = blk_trace_remove(q);
584                 break;
585         default:
586                 ret = -ENOTTY;
587                 break;
588         }
589
590         mutex_unlock(&bdev->bd_mutex);
591         return ret;
592 }
593
594 /**
595  * blk_trace_shutdown: - stop and cleanup trace structures
596  * @q:    the request queue associated with the device
597  *
598  **/
599 void blk_trace_shutdown(struct request_queue *q)
600 {
601         if (q->blk_trace) {
602                 blk_trace_startstop(q, 0);
603                 blk_trace_remove(q);
604         }
605 }
606
607 /*
608  * blktrace probes
609  */
610
611 /**
612  * blk_add_trace_rq - Add a trace for a request oriented action
613  * @q:          queue the io is for
614  * @rq:         the source request
615  * @what:       the action
616  *
617  * Description:
618  *     Records an action against a request. Will log the bio offset + size.
619  *
620  **/
621 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
622                                     u32 what)
623 {
624         struct blk_trace *bt = q->blk_trace;
625         int rw = rq->cmd_flags & 0x03;
626
627         if (likely(!bt))
628                 return;
629
630         if (blk_discard_rq(rq))
631                 rw |= (1 << BIO_RW_DISCARD);
632
633         if (blk_pc_request(rq)) {
634                 what |= BLK_TC_ACT(BLK_TC_PC);
635                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
636                                 sizeof(rq->cmd), rq->cmd);
637         } else  {
638                 what |= BLK_TC_ACT(BLK_TC_FS);
639                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
640                                 rw, what, rq->errors, 0, NULL);
641         }
642 }
643
644 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
645 {
646         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
647 }
648
649 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
650 {
651         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
652 }
653
654 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
655 {
656         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
657 }
658
659 static void blk_add_trace_rq_requeue(struct request_queue *q,
660                                      struct request *rq)
661 {
662         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
663 }
664
665 static void blk_add_trace_rq_complete(struct request_queue *q,
666                                       struct request *rq)
667 {
668         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
669 }
670
671 /**
672  * blk_add_trace_bio - Add a trace for a bio oriented action
673  * @q:          queue the io is for
674  * @bio:        the source bio
675  * @what:       the action
676  *
677  * Description:
678  *     Records an action against a bio. Will log the bio offset + size.
679  *
680  **/
681 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
682                                      u32 what)
683 {
684         struct blk_trace *bt = q->blk_trace;
685
686         if (likely(!bt))
687                 return;
688
689         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
690                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
691 }
692
693 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
694 {
695         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
696 }
697
698 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
699 {
700         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
701 }
702
703 static void blk_add_trace_bio_backmerge(struct request_queue *q,
704                                         struct bio *bio)
705 {
706         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
707 }
708
709 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
710                                          struct bio *bio)
711 {
712         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
713 }
714
715 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
716 {
717         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
718 }
719
720 static void blk_add_trace_getrq(struct request_queue *q,
721                                 struct bio *bio, int rw)
722 {
723         if (bio)
724                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
725         else {
726                 struct blk_trace *bt = q->blk_trace;
727
728                 if (bt)
729                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
730         }
731 }
732
733
734 static void blk_add_trace_sleeprq(struct request_queue *q,
735                                   struct bio *bio, int rw)
736 {
737         if (bio)
738                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
739         else {
740                 struct blk_trace *bt = q->blk_trace;
741
742                 if (bt)
743                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
744                                         0, 0, NULL);
745         }
746 }
747
748 static void blk_add_trace_plug(struct request_queue *q)
749 {
750         struct blk_trace *bt = q->blk_trace;
751
752         if (bt)
753                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
754 }
755
756 static void blk_add_trace_unplug_io(struct request_queue *q)
757 {
758         struct blk_trace *bt = q->blk_trace;
759
760         if (bt) {
761                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
762                 __be64 rpdu = cpu_to_be64(pdu);
763
764                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
765                                 sizeof(rpdu), &rpdu);
766         }
767 }
768
769 static void blk_add_trace_unplug_timer(struct request_queue *q)
770 {
771         struct blk_trace *bt = q->blk_trace;
772
773         if (bt) {
774                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
775                 __be64 rpdu = cpu_to_be64(pdu);
776
777                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
778                                 sizeof(rpdu), &rpdu);
779         }
780 }
781
782 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
783                                 unsigned int pdu)
784 {
785         struct blk_trace *bt = q->blk_trace;
786
787         if (bt) {
788                 __be64 rpdu = cpu_to_be64(pdu);
789
790                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
791                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
792                                 sizeof(rpdu), &rpdu);
793         }
794 }
795
796 /**
797  * blk_add_trace_remap - Add a trace for a remap operation
798  * @q:          queue the io is for
799  * @bio:        the source bio
800  * @dev:        target device
801  * @from:       source sector
802  * @to:         target sector
803  *
804  * Description:
805  *     Device mapper or raid target sometimes need to split a bio because
806  *     it spans a stripe (or similar). Add a trace for that action.
807  *
808  **/
809 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
810                                        dev_t dev, sector_t from, sector_t to)
811 {
812         struct blk_trace *bt = q->blk_trace;
813         struct blk_io_trace_remap r;
814
815         if (likely(!bt))
816                 return;
817
818         r.device = cpu_to_be32(dev);
819         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
820         r.sector = cpu_to_be64(to);
821
822         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
823                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
824 }
825
826 /**
827  * blk_add_driver_data - Add binary message with driver-specific data
828  * @q:          queue the io is for
829  * @rq:         io request
830  * @data:       driver-specific data
831  * @len:        length of driver-specific data
832  *
833  * Description:
834  *     Some drivers might want to write driver-specific data per request.
835  *
836  **/
837 void blk_add_driver_data(struct request_queue *q,
838                          struct request *rq,
839                          void *data, size_t len)
840 {
841         struct blk_trace *bt = q->blk_trace;
842
843         if (likely(!bt))
844                 return;
845
846         if (blk_pc_request(rq))
847                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
848                                 rq->errors, len, data);
849         else
850                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
851                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
852 }
853 EXPORT_SYMBOL_GPL(blk_add_driver_data);
854
855 static void blk_register_tracepoints(void)
856 {
857         int ret;
858
859         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
860         WARN_ON(ret);
861         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
862         WARN_ON(ret);
863         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
864         WARN_ON(ret);
865         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
866         WARN_ON(ret);
867         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
868         WARN_ON(ret);
869         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
870         WARN_ON(ret);
871         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
872         WARN_ON(ret);
873         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
874         WARN_ON(ret);
875         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
876         WARN_ON(ret);
877         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
878         WARN_ON(ret);
879         ret = register_trace_block_getrq(blk_add_trace_getrq);
880         WARN_ON(ret);
881         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
882         WARN_ON(ret);
883         ret = register_trace_block_plug(blk_add_trace_plug);
884         WARN_ON(ret);
885         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
886         WARN_ON(ret);
887         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
888         WARN_ON(ret);
889         ret = register_trace_block_split(blk_add_trace_split);
890         WARN_ON(ret);
891         ret = register_trace_block_remap(blk_add_trace_remap);
892         WARN_ON(ret);
893 }
894
895 static void blk_unregister_tracepoints(void)
896 {
897         unregister_trace_block_remap(blk_add_trace_remap);
898         unregister_trace_block_split(blk_add_trace_split);
899         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
900         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
901         unregister_trace_block_plug(blk_add_trace_plug);
902         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
903         unregister_trace_block_getrq(blk_add_trace_getrq);
904         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
905         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
906         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
907         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
908         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
909         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
910         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
911         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
912         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
913         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
914
915         tracepoint_synchronize_unregister();
916 }
917
918 /*
919  * struct blk_io_tracer formatting routines
920  */
921
922 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
923 {
924         int i = 0;
925
926         if (t->action & BLK_TC_DISCARD)
927                 rwbs[i++] = 'D';
928         else if (t->action & BLK_TC_WRITE)
929                 rwbs[i++] = 'W';
930         else if (t->bytes)
931                 rwbs[i++] = 'R';
932         else
933                 rwbs[i++] = 'N';
934
935         if (t->action & BLK_TC_AHEAD)
936                 rwbs[i++] = 'A';
937         if (t->action & BLK_TC_BARRIER)
938                 rwbs[i++] = 'B';
939         if (t->action & BLK_TC_SYNC)
940                 rwbs[i++] = 'S';
941         if (t->action & BLK_TC_META)
942                 rwbs[i++] = 'M';
943
944         rwbs[i] = '\0';
945 }
946
947 static inline
948 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
949 {
950         return (const struct blk_io_trace *)ent;
951 }
952
953 static inline const void *pdu_start(const struct trace_entry *ent)
954 {
955         return te_blk_io_trace(ent) + 1;
956 }
957
958 static inline u32 t_sec(const struct trace_entry *ent)
959 {
960         return te_blk_io_trace(ent)->bytes >> 9;
961 }
962
963 static inline unsigned long long t_sector(const struct trace_entry *ent)
964 {
965         return te_blk_io_trace(ent)->sector;
966 }
967
968 static inline __u16 t_error(const struct trace_entry *ent)
969 {
970         return te_blk_io_trace(ent)->sector;
971 }
972
973 static __u64 get_pdu_int(const struct trace_entry *ent)
974 {
975         const __u64 *val = pdu_start(ent);
976         return be64_to_cpu(*val);
977 }
978
979 static void get_pdu_remap(const struct trace_entry *ent,
980                           struct blk_io_trace_remap *r)
981 {
982         const struct blk_io_trace_remap *__r = pdu_start(ent);
983         __u64 sector = __r->sector;
984
985         r->device = be32_to_cpu(__r->device);
986         r->device_from = be32_to_cpu(__r->device_from);
987         r->sector = be64_to_cpu(sector);
988 }
989
990 static int blk_log_action_iter(struct trace_iterator *iter, const char *act)
991 {
992         char rwbs[6];
993         unsigned long long ts  = ns2usecs(iter->ts);
994         unsigned long usec_rem = do_div(ts, USEC_PER_SEC);
995         unsigned secs          = (unsigned long)ts;
996         const struct trace_entry *ent = iter->ent;
997         const struct blk_io_trace *t = (const struct blk_io_trace *)ent;
998
999         fill_rwbs(rwbs, t);
1000
1001         return trace_seq_printf(&iter->seq,
1002                                 "%3d,%-3d %2d %5d.%06lu %5u %2s %3s ",
1003                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1004                                 secs, usec_rem, ent->pid, act, rwbs);
1005 }
1006
1007 static int blk_log_action_seq(struct trace_seq *s, const struct blk_io_trace *t,
1008                               const char *act)
1009 {
1010         char rwbs[6];
1011         fill_rwbs(rwbs, t);
1012         return trace_seq_printf(s, "%3d,%-3d %2s %3s ",
1013                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1014 }
1015
1016 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1017 {
1018         char cmd[TASK_COMM_LEN];
1019
1020         trace_find_cmdline(ent->pid, cmd);
1021
1022         if (t_sec(ent))
1023                 return trace_seq_printf(s, "%llu + %u [%s]\n",
1024                                         t_sector(ent), t_sec(ent), cmd);
1025         return trace_seq_printf(s, "[%s]\n", cmd);
1026 }
1027
1028 static int blk_log_with_error(struct trace_seq *s,
1029                               const struct trace_entry *ent)
1030 {
1031         if (t_sec(ent))
1032                 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1033                                         t_sec(ent), t_error(ent));
1034         return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1035 }
1036
1037 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1038 {
1039         struct blk_io_trace_remap r = { .device = 0, };
1040
1041         get_pdu_remap(ent, &r);
1042         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1043                                t_sector(ent),
1044                                t_sec(ent), MAJOR(r.device), MINOR(r.device),
1045                                (unsigned long long)r.sector);
1046 }
1047
1048 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1049 {
1050         char cmd[TASK_COMM_LEN];
1051
1052         trace_find_cmdline(ent->pid, cmd);
1053
1054         return trace_seq_printf(s, "[%s]\n", cmd);
1055 }
1056
1057 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1058 {
1059         char cmd[TASK_COMM_LEN];
1060
1061         trace_find_cmdline(ent->pid, cmd);
1062
1063         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1064 }
1065
1066 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1067 {
1068         char cmd[TASK_COMM_LEN];
1069
1070         trace_find_cmdline(ent->pid, cmd);
1071
1072         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1073                                 get_pdu_int(ent), cmd);
1074 }
1075
1076 /*
1077  * struct tracer operations
1078  */
1079
1080 static void blk_tracer_print_header(struct seq_file *m)
1081 {
1082         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1083                 return;
1084         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1085                     "#  |     |     |           |   |   |\n");
1086 }
1087
1088 static void blk_tracer_start(struct trace_array *tr)
1089 {
1090         if (atomic_add_return(1, &blk_probes_ref) == 1)
1091                 blk_register_tracepoints();
1092         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1093 }
1094
1095 static int blk_tracer_init(struct trace_array *tr)
1096 {
1097         blk_tr = tr;
1098         blk_tracer_start(tr);
1099         blk_tracer_enabled = true;
1100         return 0;
1101 }
1102
1103 static void blk_tracer_stop(struct trace_array *tr)
1104 {
1105         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1106         if (atomic_dec_and_test(&blk_probes_ref))
1107                 blk_unregister_tracepoints();
1108 }
1109
1110 static void blk_tracer_reset(struct trace_array *tr)
1111 {
1112         if (!atomic_read(&blk_probes_ref))
1113                 return;
1114
1115         blk_tracer_enabled = false;
1116         blk_tracer_stop(tr);
1117 }
1118
1119 static struct {
1120         const char *act[2];
1121         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1122 } what2act[] __read_mostly = {
1123         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1124         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1125         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1126         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1127         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1128         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1129         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1130         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1131         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1132         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1133         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1134         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1135         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1136         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1137         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1138 };
1139
1140 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1141                                                int flags)
1142 {
1143         struct trace_seq *s = &iter->seq;
1144         const struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1145         const u16 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1146         int ret;
1147
1148         if (!trace_print_context(iter))
1149                 return TRACE_TYPE_PARTIAL_LINE;
1150
1151         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1152                 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1153         else {
1154                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1155                 ret = blk_log_action_seq(s, t, what2act[what].act[long_act]);
1156                 if (ret)
1157                         ret = what2act[what].print(s, iter->ent);
1158         }
1159
1160         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1161 }
1162
1163 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1164 {
1165         struct trace_seq *s = &iter->seq;
1166         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1167         const int offset = offsetof(struct blk_io_trace, sector);
1168         struct blk_io_trace old = {
1169                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1170                 .time     = ns2usecs(iter->ts),
1171         };
1172
1173         if (!trace_seq_putmem(s, &old, offset))
1174                 return 0;
1175         return trace_seq_putmem(s, &t->sector,
1176                                 sizeof(old) - offset + t->pdu_len);
1177 }
1178
1179 static enum print_line_t
1180 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1181 {
1182         return blk_trace_synthesize_old_trace(iter) ?
1183                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1184 }
1185
1186 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1187 {
1188         const struct blk_io_trace *t;
1189         u16 what;
1190         int ret;
1191
1192         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1193                 return TRACE_TYPE_UNHANDLED;
1194
1195         t = (const struct blk_io_trace *)iter->ent;
1196         what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1197
1198         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1199                 ret = trace_seq_printf(&iter->seq, "Bad pc action %x\n", what);
1200         else {
1201                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1202                 ret = blk_log_action_iter(iter, what2act[what].act[long_act]);
1203                 if (ret)
1204                         ret = what2act[what].print(&iter->seq, iter->ent);
1205         }
1206
1207         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1208 }
1209
1210 static struct tracer blk_tracer __read_mostly = {
1211         .name           = "blk",
1212         .init           = blk_tracer_init,
1213         .reset          = blk_tracer_reset,
1214         .start          = blk_tracer_start,
1215         .stop           = blk_tracer_stop,
1216         .print_header   = blk_tracer_print_header,
1217         .print_line     = blk_tracer_print_line,
1218         .flags          = &blk_tracer_flags,
1219 };
1220
1221 static struct trace_event trace_blk_event = {
1222         .type           = TRACE_BLK,
1223         .trace          = blk_trace_event_print,
1224         .binary         = blk_trace_event_print_binary,
1225 };
1226
1227 static int __init init_blk_tracer(void)
1228 {
1229         if (!register_ftrace_event(&trace_blk_event)) {
1230                 pr_warning("Warning: could not register block events\n");
1231                 return 1;
1232         }
1233
1234         if (register_tracer(&blk_tracer) != 0) {
1235                 pr_warning("Warning: could not register the block tracer\n");
1236                 unregister_ftrace_event(&trace_blk_event);
1237                 return 1;
1238         }
1239
1240         return 0;
1241 }
1242
1243 device_initcall(init_blk_tracer);
1244
1245 static int blk_trace_remove_queue(struct request_queue *q)
1246 {
1247         struct blk_trace *bt;
1248
1249         bt = xchg(&q->blk_trace, NULL);
1250         if (bt == NULL)
1251                 return -EINVAL;
1252
1253         kfree(bt);
1254         return 0;
1255 }
1256
1257 /*
1258  * Setup everything required to start tracing
1259  */
1260 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1261 {
1262         struct blk_trace *old_bt, *bt = NULL;
1263
1264         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1265         if (!bt)
1266                 return -ENOMEM;
1267
1268         bt->dev = dev;
1269         bt->act_mask = (u16)-1;
1270         bt->end_lba = -1ULL;
1271         bt->trace_state = Blktrace_running;
1272
1273         old_bt = xchg(&q->blk_trace, bt);
1274         if (old_bt != NULL) {
1275                 (void)xchg(&q->blk_trace, old_bt);
1276                 kfree(bt);
1277                 return -EBUSY;
1278         }
1279
1280         return 0;
1281 }
1282
1283 /*
1284  * sysfs interface to enable and configure tracing
1285  */
1286
1287 static ssize_t sysfs_blk_trace_enable_show(struct device *dev,
1288                                            struct device_attribute *attr,
1289                                            char *buf)
1290 {
1291         struct hd_struct *p = dev_to_part(dev);
1292         struct block_device *bdev;
1293         ssize_t ret = -ENXIO;
1294
1295         lock_kernel();
1296         bdev = bdget(part_devt(p));
1297         if (bdev != NULL) {
1298                 struct request_queue *q = bdev_get_queue(bdev);
1299
1300                 if (q != NULL) {
1301                         mutex_lock(&bdev->bd_mutex);
1302                         ret = sprintf(buf, "%u\n", !!q->blk_trace);
1303                         mutex_unlock(&bdev->bd_mutex);
1304                 }
1305
1306                 bdput(bdev);
1307         }
1308
1309         unlock_kernel();
1310         return ret;
1311 }
1312
1313 static ssize_t sysfs_blk_trace_enable_store(struct device *dev,
1314                                             struct device_attribute *attr,
1315                                             const char *buf, size_t count)
1316 {
1317         struct block_device *bdev;
1318         struct request_queue *q;
1319         struct hd_struct *p;
1320         int value;
1321         ssize_t ret = -ENXIO;
1322
1323         if (count == 0 || sscanf(buf, "%d", &value) != 1)
1324                 goto out;
1325
1326         lock_kernel();
1327         p = dev_to_part(dev);
1328         bdev = bdget(part_devt(p));
1329         if (bdev == NULL)
1330                 goto out_unlock_kernel;
1331
1332         q = bdev_get_queue(bdev);
1333         if (q == NULL)
1334                 goto out_bdput;
1335
1336         mutex_lock(&bdev->bd_mutex);
1337         if (value)
1338                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1339         else
1340                 ret = blk_trace_remove_queue(q);
1341         mutex_unlock(&bdev->bd_mutex);
1342
1343         if (ret == 0)
1344                 ret = count;
1345 out_bdput:
1346         bdput(bdev);
1347 out_unlock_kernel:
1348         unlock_kernel();
1349 out:
1350         return ret;
1351 }
1352
1353 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1354                                          struct device_attribute *attr,
1355                                          char *buf);
1356 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1357                                           struct device_attribute *attr,
1358                                           const char *buf, size_t count);
1359 #define BLK_TRACE_DEVICE_ATTR(_name) \
1360         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1361                     sysfs_blk_trace_attr_show, \
1362                     sysfs_blk_trace_attr_store)
1363
1364 static DEVICE_ATTR(enable, S_IRUGO | S_IWUSR,
1365                    sysfs_blk_trace_enable_show, sysfs_blk_trace_enable_store);
1366 static BLK_TRACE_DEVICE_ATTR(act_mask);
1367 static BLK_TRACE_DEVICE_ATTR(pid);
1368 static BLK_TRACE_DEVICE_ATTR(start_lba);
1369 static BLK_TRACE_DEVICE_ATTR(end_lba);
1370
1371 static struct attribute *blk_trace_attrs[] = {
1372         &dev_attr_enable.attr,
1373         &dev_attr_act_mask.attr,
1374         &dev_attr_pid.attr,
1375         &dev_attr_start_lba.attr,
1376         &dev_attr_end_lba.attr,
1377         NULL
1378 };
1379
1380 struct attribute_group blk_trace_attr_group = {
1381         .name  = "trace",
1382         .attrs = blk_trace_attrs,
1383 };
1384
1385 static int blk_str2act_mask(const char *str)
1386 {
1387         int mask = 0;
1388         char *copy = kstrdup(str, GFP_KERNEL), *s;
1389
1390         if (copy == NULL)
1391                 return -ENOMEM;
1392
1393         s = strstrip(copy);
1394
1395         while (1) {
1396                 char *sep = strchr(s, ',');
1397
1398                 if (sep != NULL)
1399                         *sep = '\0';
1400
1401                 if (strcasecmp(s, "barrier") == 0)
1402                         mask |= BLK_TC_BARRIER;
1403                 else if (strcasecmp(s, "complete") == 0)
1404                         mask |= BLK_TC_COMPLETE;
1405                 else if (strcasecmp(s, "fs") == 0)
1406                         mask |= BLK_TC_FS;
1407                 else if (strcasecmp(s, "issue") == 0)
1408                         mask |= BLK_TC_ISSUE;
1409                 else if (strcasecmp(s, "pc") == 0)
1410                         mask |= BLK_TC_PC;
1411                 else if (strcasecmp(s, "queue") == 0)
1412                         mask |= BLK_TC_QUEUE;
1413                 else if (strcasecmp(s, "read") == 0)
1414                         mask |= BLK_TC_READ;
1415                 else if (strcasecmp(s, "requeue") == 0)
1416                         mask |= BLK_TC_REQUEUE;
1417                 else if (strcasecmp(s, "sync") == 0)
1418                         mask |= BLK_TC_SYNC;
1419                 else if (strcasecmp(s, "write") == 0)
1420                         mask |= BLK_TC_WRITE;
1421
1422                 if (sep == NULL)
1423                         break;
1424
1425                 s = sep + 1;
1426         }
1427         kfree(copy);
1428
1429         return mask;
1430 }
1431
1432 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1433                                          struct device_attribute *attr,
1434                                          char *buf)
1435 {
1436         struct hd_struct *p = dev_to_part(dev);
1437         struct request_queue *q;
1438         struct block_device *bdev;
1439         ssize_t ret = -ENXIO;
1440
1441         lock_kernel();
1442         bdev = bdget(part_devt(p));
1443         if (bdev == NULL)
1444                 goto out_unlock_kernel;
1445
1446         q = bdev_get_queue(bdev);
1447         if (q == NULL)
1448                 goto out_bdput;
1449         mutex_lock(&bdev->bd_mutex);
1450         if (q->blk_trace == NULL)
1451                 ret = sprintf(buf, "disabled\n");
1452         else if (attr == &dev_attr_act_mask)
1453                 ret = sprintf(buf, "%#x\n", q->blk_trace->act_mask);
1454         else if (attr == &dev_attr_pid)
1455                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1456         else if (attr == &dev_attr_start_lba)
1457                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1458         else if (attr == &dev_attr_end_lba)
1459                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1460         mutex_unlock(&bdev->bd_mutex);
1461 out_bdput:
1462         bdput(bdev);
1463 out_unlock_kernel:
1464         unlock_kernel();
1465         return ret;
1466 }
1467
1468 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1469                                           struct device_attribute *attr,
1470                                           const char *buf, size_t count)
1471 {
1472         struct block_device *bdev;
1473         struct request_queue *q;
1474         struct hd_struct *p;
1475         u64 value;
1476         ssize_t ret = -ENXIO;
1477
1478         if (count == 0)
1479                 goto out;
1480
1481         if (attr == &dev_attr_act_mask) {
1482                 if (sscanf(buf, "%llx", &value) != 1) {
1483                         /* Assume it is a list of trace category names */
1484                         value = blk_str2act_mask(buf);
1485                         if (value < 0)
1486                                 goto out;
1487                 }
1488         } else if (sscanf(buf, "%llu", &value) != 1)
1489                 goto out;
1490
1491         lock_kernel();
1492         p = dev_to_part(dev);
1493         bdev = bdget(part_devt(p));
1494         if (bdev == NULL)
1495                 goto out_unlock_kernel;
1496
1497         q = bdev_get_queue(bdev);
1498         if (q == NULL)
1499                 goto out_bdput;
1500
1501         mutex_lock(&bdev->bd_mutex);
1502         ret = 0;
1503         if (q->blk_trace == NULL)
1504                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1505
1506         if (ret == 0) {
1507                 if (attr == &dev_attr_act_mask)
1508                         q->blk_trace->act_mask = value;
1509                 else if (attr == &dev_attr_pid)
1510                         q->blk_trace->pid = value;
1511                 else if (attr == &dev_attr_start_lba)
1512                         q->blk_trace->start_lba = value;
1513                 else if (attr == &dev_attr_end_lba)
1514                         q->blk_trace->end_lba = value;
1515                 ret = count;
1516         }
1517         mutex_unlock(&bdev->bd_mutex);
1518 out_bdput:
1519         bdput(bdev);
1520 out_unlock_kernel:
1521         unlock_kernel();
1522 out:
1523         return ret;
1524 }