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