block: add secure discard
[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/slab.h>
25 #include <linux/debugfs.h>
26 #include <linux/smp_lock.h>
27 #include <linux/time.h>
28 #include <linux/uaccess.h>
29
30 #include <trace/events/block.h>
31
32 #include "trace_output.h"
33
34 #ifdef CONFIG_BLK_DEV_IO_TRACE
35
36 static unsigned int blktrace_seq __read_mostly = 1;
37
38 static struct trace_array *blk_tr;
39 static bool blk_tracer_enabled __read_mostly;
40
41 /* Select an alternative, minimalistic output than the original one */
42 #define TRACE_BLK_OPT_CLASSIC   0x1
43
44 static struct tracer_opt blk_tracer_opts[] = {
45         /* Default disable the minimalistic output */
46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
47         { }
48 };
49
50 static struct tracer_flags blk_tracer_flags = {
51         .val  = 0,
52         .opts = blk_tracer_opts,
53 };
54
55 /* Global reference count of probes */
56 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
57
58 static void blk_register_tracepoints(void);
59 static void blk_unregister_tracepoints(void);
60
61 /*
62  * Send out a notify message.
63  */
64 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
65                        const void *data, size_t len)
66 {
67         struct blk_io_trace *t;
68         struct ring_buffer_event *event = NULL;
69         struct ring_buffer *buffer = NULL;
70         int pc = 0;
71         int cpu = smp_processor_id();
72         bool blk_tracer = blk_tracer_enabled;
73
74         if (blk_tracer) {
75                 buffer = blk_tr->buffer;
76                 pc = preempt_count();
77                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
78                                                   sizeof(*t) + len,
79                                                   0, pc);
80                 if (!event)
81                         return;
82                 t = ring_buffer_event_data(event);
83                 goto record_it;
84         }
85
86         if (!bt->rchan)
87                 return;
88
89         t = relay_reserve(bt->rchan, sizeof(*t) + len);
90         if (t) {
91                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
92                 t->time = ktime_to_ns(ktime_get());
93 record_it:
94                 t->device = bt->dev;
95                 t->action = action;
96                 t->pid = pid;
97                 t->cpu = cpu;
98                 t->pdu_len = len;
99                 memcpy((void *) t + sizeof(*t), data, len);
100
101                 if (blk_tracer)
102                         trace_buffer_unlock_commit(buffer, event, 0, pc);
103         }
104 }
105
106 /*
107  * Send out a notify for this process, if we haven't done so since a trace
108  * started
109  */
110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
111 {
112         tsk->btrace_seq = blktrace_seq;
113         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
114 }
115
116 static void trace_note_time(struct blk_trace *bt)
117 {
118         struct timespec now;
119         unsigned long flags;
120         u32 words[2];
121
122         getnstimeofday(&now);
123         words[0] = now.tv_sec;
124         words[1] = now.tv_nsec;
125
126         local_irq_save(flags);
127         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
128         local_irq_restore(flags);
129 }
130
131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
132 {
133         int n;
134         va_list args;
135         unsigned long flags;
136         char *buf;
137
138         if (unlikely(bt->trace_state != Blktrace_running &&
139                      !blk_tracer_enabled))
140                 return;
141
142         local_irq_save(flags);
143         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
144         va_start(args, fmt);
145         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
146         va_end(args);
147
148         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
149         local_irq_restore(flags);
150 }
151 EXPORT_SYMBOL_GPL(__trace_note_message);
152
153 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
154                          pid_t pid)
155 {
156         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
157                 return 1;
158         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
159                 return 1;
160         if (bt->pid && pid != bt->pid)
161                 return 1;
162
163         return 0;
164 }
165
166 /*
167  * Data direction bit lookup
168  */
169 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
170                                  BLK_TC_ACT(BLK_TC_WRITE) };
171
172 #define BLK_TC_HARDBARRIER      BLK_TC_BARRIER
173 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
174
175 /* The ilog2() calls fall out because they're constant */
176 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
177           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
178
179 /*
180  * The worker for the various blk_add_trace*() types. Fills out a
181  * blk_io_trace structure and places it in a per-cpu subbuffer.
182  */
183 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
184                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
185 {
186         struct task_struct *tsk = current;
187         struct ring_buffer_event *event = NULL;
188         struct ring_buffer *buffer = NULL;
189         struct blk_io_trace *t;
190         unsigned long flags = 0;
191         unsigned long *sequence;
192         pid_t pid;
193         int cpu, pc = 0;
194         bool blk_tracer = blk_tracer_enabled;
195
196         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
197                 return;
198
199         what |= ddir_act[rw & WRITE];
200         what |= MASK_TC_BIT(rw, HARDBARRIER);
201         what |= MASK_TC_BIT(rw, SYNC);
202         what |= MASK_TC_BIT(rw, RAHEAD);
203         what |= MASK_TC_BIT(rw, META);
204         what |= MASK_TC_BIT(rw, DISCARD);
205
206         pid = tsk->pid;
207         if (act_log_check(bt, what, sector, pid))
208                 return;
209         cpu = raw_smp_processor_id();
210
211         if (blk_tracer) {
212                 tracing_record_cmdline(current);
213
214                 buffer = blk_tr->buffer;
215                 pc = preempt_count();
216                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
217                                                   sizeof(*t) + pdu_len,
218                                                   0, pc);
219                 if (!event)
220                         return;
221                 t = ring_buffer_event_data(event);
222                 goto record_it;
223         }
224
225         /*
226          * A word about the locking here - we disable interrupts to reserve
227          * some space in the relay per-cpu buffer, to prevent an irq
228          * from coming in and stepping on our toes.
229          */
230         local_irq_save(flags);
231
232         if (unlikely(tsk->btrace_seq != blktrace_seq))
233                 trace_note_tsk(bt, tsk);
234
235         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
236         if (t) {
237                 sequence = per_cpu_ptr(bt->sequence, cpu);
238
239                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
240                 t->sequence = ++(*sequence);
241                 t->time = ktime_to_ns(ktime_get());
242 record_it:
243                 /*
244                  * These two are not needed in ftrace as they are in the
245                  * generic trace_entry, filled by tracing_generic_entry_update,
246                  * but for the trace_event->bin() synthesizer benefit we do it
247                  * here too.
248                  */
249                 t->cpu = cpu;
250                 t->pid = pid;
251
252                 t->sector = sector;
253                 t->bytes = bytes;
254                 t->action = what;
255                 t->device = bt->dev;
256                 t->error = error;
257                 t->pdu_len = pdu_len;
258
259                 if (pdu_len)
260                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
261
262                 if (blk_tracer) {
263                         trace_buffer_unlock_commit(buffer, event, 0, pc);
264                         return;
265                 }
266         }
267
268         local_irq_restore(flags);
269 }
270
271 static struct dentry *blk_tree_root;
272 static DEFINE_MUTEX(blk_tree_mutex);
273
274 static void blk_trace_free(struct blk_trace *bt)
275 {
276         debugfs_remove(bt->msg_file);
277         debugfs_remove(bt->dropped_file);
278         relay_close(bt->rchan);
279         debugfs_remove(bt->dir);
280         free_percpu(bt->sequence);
281         free_percpu(bt->msg_data);
282         kfree(bt);
283 }
284
285 static void blk_trace_cleanup(struct blk_trace *bt)
286 {
287         blk_trace_free(bt);
288         if (atomic_dec_and_test(&blk_probes_ref))
289                 blk_unregister_tracepoints();
290 }
291
292 int blk_trace_remove(struct request_queue *q)
293 {
294         struct blk_trace *bt;
295
296         bt = xchg(&q->blk_trace, NULL);
297         if (!bt)
298                 return -EINVAL;
299
300         if (bt->trace_state != Blktrace_running)
301                 blk_trace_cleanup(bt);
302
303         return 0;
304 }
305 EXPORT_SYMBOL_GPL(blk_trace_remove);
306
307 static int blk_dropped_open(struct inode *inode, struct file *filp)
308 {
309         filp->private_data = inode->i_private;
310
311         return 0;
312 }
313
314 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
315                                 size_t count, loff_t *ppos)
316 {
317         struct blk_trace *bt = filp->private_data;
318         char buf[16];
319
320         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
321
322         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
323 }
324
325 static const struct file_operations blk_dropped_fops = {
326         .owner =        THIS_MODULE,
327         .open =         blk_dropped_open,
328         .read =         blk_dropped_read,
329 };
330
331 static int blk_msg_open(struct inode *inode, struct file *filp)
332 {
333         filp->private_data = inode->i_private;
334
335         return 0;
336 }
337
338 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
339                                 size_t count, loff_t *ppos)
340 {
341         char *msg;
342         struct blk_trace *bt;
343
344         if (count >= BLK_TN_MAX_MSG)
345                 return -EINVAL;
346
347         msg = kmalloc(count + 1, GFP_KERNEL);
348         if (msg == NULL)
349                 return -ENOMEM;
350
351         if (copy_from_user(msg, buffer, count)) {
352                 kfree(msg);
353                 return -EFAULT;
354         }
355
356         msg[count] = '\0';
357         bt = filp->private_data;
358         __trace_note_message(bt, "%s", msg);
359         kfree(msg);
360
361         return count;
362 }
363
364 static const struct file_operations blk_msg_fops = {
365         .owner =        THIS_MODULE,
366         .open =         blk_msg_open,
367         .write =        blk_msg_write,
368 };
369
370 /*
371  * Keep track of how many times we encountered a full subbuffer, to aid
372  * the user space app in telling how many lost events there were.
373  */
374 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
375                                      void *prev_subbuf, size_t prev_padding)
376 {
377         struct blk_trace *bt;
378
379         if (!relay_buf_full(buf))
380                 return 1;
381
382         bt = buf->chan->private_data;
383         atomic_inc(&bt->dropped);
384         return 0;
385 }
386
387 static int blk_remove_buf_file_callback(struct dentry *dentry)
388 {
389         debugfs_remove(dentry);
390
391         return 0;
392 }
393
394 static struct dentry *blk_create_buf_file_callback(const char *filename,
395                                                    struct dentry *parent,
396                                                    int mode,
397                                                    struct rchan_buf *buf,
398                                                    int *is_global)
399 {
400         return debugfs_create_file(filename, mode, parent, buf,
401                                         &relay_file_operations);
402 }
403
404 static struct rchan_callbacks blk_relay_callbacks = {
405         .subbuf_start           = blk_subbuf_start_callback,
406         .create_buf_file        = blk_create_buf_file_callback,
407         .remove_buf_file        = blk_remove_buf_file_callback,
408 };
409
410 static void blk_trace_setup_lba(struct blk_trace *bt,
411                                 struct block_device *bdev)
412 {
413         struct hd_struct *part = NULL;
414
415         if (bdev)
416                 part = bdev->bd_part;
417
418         if (part) {
419                 bt->start_lba = part->start_sect;
420                 bt->end_lba = part->start_sect + part->nr_sects;
421         } else {
422                 bt->start_lba = 0;
423                 bt->end_lba = -1ULL;
424         }
425 }
426
427 /*
428  * Setup everything required to start tracing
429  */
430 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
431                        struct block_device *bdev,
432                        struct blk_user_trace_setup *buts)
433 {
434         struct blk_trace *old_bt, *bt = NULL;
435         struct dentry *dir = NULL;
436         int ret, i;
437
438         if (!buts->buf_size || !buts->buf_nr)
439                 return -EINVAL;
440
441         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
442         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
443
444         /*
445          * some device names have larger paths - convert the slashes
446          * to underscores for this to work as expected
447          */
448         for (i = 0; i < strlen(buts->name); i++)
449                 if (buts->name[i] == '/')
450                         buts->name[i] = '_';
451
452         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
453         if (!bt)
454                 return -ENOMEM;
455
456         ret = -ENOMEM;
457         bt->sequence = alloc_percpu(unsigned long);
458         if (!bt->sequence)
459                 goto err;
460
461         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
462         if (!bt->msg_data)
463                 goto err;
464
465         ret = -ENOENT;
466
467         mutex_lock(&blk_tree_mutex);
468         if (!blk_tree_root) {
469                 blk_tree_root = debugfs_create_dir("block", NULL);
470                 if (!blk_tree_root) {
471                         mutex_unlock(&blk_tree_mutex);
472                         goto err;
473                 }
474         }
475         mutex_unlock(&blk_tree_mutex);
476
477         dir = debugfs_create_dir(buts->name, blk_tree_root);
478
479         if (!dir)
480                 goto err;
481
482         bt->dir = dir;
483         bt->dev = dev;
484         atomic_set(&bt->dropped, 0);
485
486         ret = -EIO;
487         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
488                                                &blk_dropped_fops);
489         if (!bt->dropped_file)
490                 goto err;
491
492         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
493         if (!bt->msg_file)
494                 goto err;
495
496         bt->rchan = relay_open("trace", dir, buts->buf_size,
497                                 buts->buf_nr, &blk_relay_callbacks, bt);
498         if (!bt->rchan)
499                 goto err;
500
501         bt->act_mask = buts->act_mask;
502         if (!bt->act_mask)
503                 bt->act_mask = (u16) -1;
504
505         blk_trace_setup_lba(bt, bdev);
506
507         /* overwrite with user settings */
508         if (buts->start_lba)
509                 bt->start_lba = buts->start_lba;
510         if (buts->end_lba)
511                 bt->end_lba = buts->end_lba;
512
513         bt->pid = buts->pid;
514         bt->trace_state = Blktrace_setup;
515
516         ret = -EBUSY;
517         old_bt = xchg(&q->blk_trace, bt);
518         if (old_bt) {
519                 (void) xchg(&q->blk_trace, old_bt);
520                 goto err;
521         }
522
523         if (atomic_inc_return(&blk_probes_ref) == 1)
524                 blk_register_tracepoints();
525
526         return 0;
527 err:
528         blk_trace_free(bt);
529         return ret;
530 }
531
532 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
533                     struct block_device *bdev,
534                     char __user *arg)
535 {
536         struct blk_user_trace_setup buts;
537         int ret;
538
539         ret = copy_from_user(&buts, arg, sizeof(buts));
540         if (ret)
541                 return -EFAULT;
542
543         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
544         if (ret)
545                 return ret;
546
547         if (copy_to_user(arg, &buts, sizeof(buts))) {
548                 blk_trace_remove(q);
549                 return -EFAULT;
550         }
551         return 0;
552 }
553 EXPORT_SYMBOL_GPL(blk_trace_setup);
554
555 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
556 static int compat_blk_trace_setup(struct request_queue *q, char *name,
557                                   dev_t dev, struct block_device *bdev,
558                                   char __user *arg)
559 {
560         struct blk_user_trace_setup buts;
561         struct compat_blk_user_trace_setup cbuts;
562         int ret;
563
564         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
565                 return -EFAULT;
566
567         buts = (struct blk_user_trace_setup) {
568                 .act_mask = cbuts.act_mask,
569                 .buf_size = cbuts.buf_size,
570                 .buf_nr = cbuts.buf_nr,
571                 .start_lba = cbuts.start_lba,
572                 .end_lba = cbuts.end_lba,
573                 .pid = cbuts.pid,
574         };
575         memcpy(&buts.name, &cbuts.name, 32);
576
577         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
578         if (ret)
579                 return ret;
580
581         if (copy_to_user(arg, &buts.name, 32)) {
582                 blk_trace_remove(q);
583                 return -EFAULT;
584         }
585
586         return 0;
587 }
588 #endif
589
590 int blk_trace_startstop(struct request_queue *q, int start)
591 {
592         int ret;
593         struct blk_trace *bt = q->blk_trace;
594
595         if (bt == NULL)
596                 return -EINVAL;
597
598         /*
599          * For starting a trace, we can transition from a setup or stopped
600          * trace. For stopping a trace, the state must be running
601          */
602         ret = -EINVAL;
603         if (start) {
604                 if (bt->trace_state == Blktrace_setup ||
605                     bt->trace_state == Blktrace_stopped) {
606                         blktrace_seq++;
607                         smp_mb();
608                         bt->trace_state = Blktrace_running;
609
610                         trace_note_time(bt);
611                         ret = 0;
612                 }
613         } else {
614                 if (bt->trace_state == Blktrace_running) {
615                         bt->trace_state = Blktrace_stopped;
616                         relay_flush(bt->rchan);
617                         ret = 0;
618                 }
619         }
620
621         return ret;
622 }
623 EXPORT_SYMBOL_GPL(blk_trace_startstop);
624
625 /**
626  * blk_trace_ioctl: - handle the ioctls associated with tracing
627  * @bdev:       the block device
628  * @cmd:        the ioctl cmd
629  * @arg:        the argument data, if any
630  *
631  **/
632 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
633 {
634         struct request_queue *q;
635         int ret, start = 0;
636         char b[BDEVNAME_SIZE];
637
638         q = bdev_get_queue(bdev);
639         if (!q)
640                 return -ENXIO;
641
642         lock_kernel();
643         mutex_lock(&bdev->bd_mutex);
644
645         switch (cmd) {
646         case BLKTRACESETUP:
647                 bdevname(bdev, b);
648                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
649                 break;
650 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
651         case BLKTRACESETUP32:
652                 bdevname(bdev, b);
653                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
654                 break;
655 #endif
656         case BLKTRACESTART:
657                 start = 1;
658         case BLKTRACESTOP:
659                 ret = blk_trace_startstop(q, start);
660                 break;
661         case BLKTRACETEARDOWN:
662                 ret = blk_trace_remove(q);
663                 break;
664         default:
665                 ret = -ENOTTY;
666                 break;
667         }
668
669         mutex_unlock(&bdev->bd_mutex);
670         unlock_kernel();
671         return ret;
672 }
673
674 /**
675  * blk_trace_shutdown: - stop and cleanup trace structures
676  * @q:    the request queue associated with the device
677  *
678  **/
679 void blk_trace_shutdown(struct request_queue *q)
680 {
681         if (q->blk_trace) {
682                 blk_trace_startstop(q, 0);
683                 blk_trace_remove(q);
684         }
685 }
686
687 /*
688  * blktrace probes
689  */
690
691 /**
692  * blk_add_trace_rq - Add a trace for a request oriented action
693  * @q:          queue the io is for
694  * @rq:         the source request
695  * @what:       the action
696  *
697  * Description:
698  *     Records an action against a request. Will log the bio offset + size.
699  *
700  **/
701 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
702                                     u32 what)
703 {
704         struct blk_trace *bt = q->blk_trace;
705         int rw = rq->cmd_flags & 0x03;
706
707         if (likely(!bt))
708                 return;
709
710         if (rq->cmd_flags & REQ_DISCARD)
711                 rw |= REQ_DISCARD;
712
713         if (rq->cmd_flags & REQ_SECURE)
714                 rw |= REQ_SECURE;
715
716         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
717                 what |= BLK_TC_ACT(BLK_TC_PC);
718                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
719                                 what, rq->errors, rq->cmd_len, rq->cmd);
720         } else  {
721                 what |= BLK_TC_ACT(BLK_TC_FS);
722                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
723                                 what, rq->errors, 0, NULL);
724         }
725 }
726
727 static void blk_add_trace_rq_abort(void *ignore,
728                                    struct request_queue *q, struct request *rq)
729 {
730         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
731 }
732
733 static void blk_add_trace_rq_insert(void *ignore,
734                                     struct request_queue *q, struct request *rq)
735 {
736         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
737 }
738
739 static void blk_add_trace_rq_issue(void *ignore,
740                                    struct request_queue *q, struct request *rq)
741 {
742         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
743 }
744
745 static void blk_add_trace_rq_requeue(void *ignore,
746                                      struct request_queue *q,
747                                      struct request *rq)
748 {
749         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
750 }
751
752 static void blk_add_trace_rq_complete(void *ignore,
753                                       struct request_queue *q,
754                                       struct request *rq)
755 {
756         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
757 }
758
759 /**
760  * blk_add_trace_bio - Add a trace for a bio oriented action
761  * @q:          queue the io is for
762  * @bio:        the source bio
763  * @what:       the action
764  *
765  * Description:
766  *     Records an action against a bio. Will log the bio offset + size.
767  *
768  **/
769 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
770                                      u32 what)
771 {
772         struct blk_trace *bt = q->blk_trace;
773
774         if (likely(!bt))
775                 return;
776
777         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
778                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
779 }
780
781 static void blk_add_trace_bio_bounce(void *ignore,
782                                      struct request_queue *q, struct bio *bio)
783 {
784         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
785 }
786
787 static void blk_add_trace_bio_complete(void *ignore,
788                                        struct request_queue *q, struct bio *bio)
789 {
790         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
791 }
792
793 static void blk_add_trace_bio_backmerge(void *ignore,
794                                         struct request_queue *q,
795                                         struct bio *bio)
796 {
797         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
798 }
799
800 static void blk_add_trace_bio_frontmerge(void *ignore,
801                                          struct request_queue *q,
802                                          struct bio *bio)
803 {
804         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
805 }
806
807 static void blk_add_trace_bio_queue(void *ignore,
808                                     struct request_queue *q, struct bio *bio)
809 {
810         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
811 }
812
813 static void blk_add_trace_getrq(void *ignore,
814                                 struct request_queue *q,
815                                 struct bio *bio, int rw)
816 {
817         if (bio)
818                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
819         else {
820                 struct blk_trace *bt = q->blk_trace;
821
822                 if (bt)
823                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
824         }
825 }
826
827
828 static void blk_add_trace_sleeprq(void *ignore,
829                                   struct request_queue *q,
830                                   struct bio *bio, int rw)
831 {
832         if (bio)
833                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
834         else {
835                 struct blk_trace *bt = q->blk_trace;
836
837                 if (bt)
838                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
839                                         0, 0, NULL);
840         }
841 }
842
843 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
844 {
845         struct blk_trace *bt = q->blk_trace;
846
847         if (bt)
848                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
849 }
850
851 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
852 {
853         struct blk_trace *bt = q->blk_trace;
854
855         if (bt) {
856                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
857                 __be64 rpdu = cpu_to_be64(pdu);
858
859                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
860                                 sizeof(rpdu), &rpdu);
861         }
862 }
863
864 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q)
865 {
866         struct blk_trace *bt = q->blk_trace;
867
868         if (bt) {
869                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
870                 __be64 rpdu = cpu_to_be64(pdu);
871
872                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
873                                 sizeof(rpdu), &rpdu);
874         }
875 }
876
877 static void blk_add_trace_split(void *ignore,
878                                 struct request_queue *q, struct bio *bio,
879                                 unsigned int pdu)
880 {
881         struct blk_trace *bt = q->blk_trace;
882
883         if (bt) {
884                 __be64 rpdu = cpu_to_be64(pdu);
885
886                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
887                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
888                                 sizeof(rpdu), &rpdu);
889         }
890 }
891
892 /**
893  * blk_add_trace_remap - Add a trace for a remap operation
894  * @ignore:     trace callback data parameter (not used)
895  * @q:          queue the io is for
896  * @bio:        the source bio
897  * @dev:        target device
898  * @from:       source sector
899  *
900  * Description:
901  *     Device mapper or raid target sometimes need to split a bio because
902  *     it spans a stripe (or similar). Add a trace for that action.
903  *
904  **/
905 static void blk_add_trace_remap(void *ignore,
906                                 struct request_queue *q, struct bio *bio,
907                                 dev_t dev, sector_t from)
908 {
909         struct blk_trace *bt = q->blk_trace;
910         struct blk_io_trace_remap r;
911
912         if (likely(!bt))
913                 return;
914
915         r.device_from = cpu_to_be32(dev);
916         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
917         r.sector_from = cpu_to_be64(from);
918
919         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
920                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
921                         sizeof(r), &r);
922 }
923
924 /**
925  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
926  * @ignore:     trace callback data parameter (not used)
927  * @q:          queue the io is for
928  * @rq:         the source request
929  * @dev:        target device
930  * @from:       source sector
931  *
932  * Description:
933  *     Device mapper remaps request to other devices.
934  *     Add a trace for that action.
935  *
936  **/
937 static void blk_add_trace_rq_remap(void *ignore,
938                                    struct request_queue *q,
939                                    struct request *rq, dev_t dev,
940                                    sector_t from)
941 {
942         struct blk_trace *bt = q->blk_trace;
943         struct blk_io_trace_remap r;
944
945         if (likely(!bt))
946                 return;
947
948         r.device_from = cpu_to_be32(dev);
949         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
950         r.sector_from = cpu_to_be64(from);
951
952         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
953                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
954                         sizeof(r), &r);
955 }
956
957 /**
958  * blk_add_driver_data - Add binary message with driver-specific data
959  * @q:          queue the io is for
960  * @rq:         io request
961  * @data:       driver-specific data
962  * @len:        length of driver-specific data
963  *
964  * Description:
965  *     Some drivers might want to write driver-specific data per request.
966  *
967  **/
968 void blk_add_driver_data(struct request_queue *q,
969                          struct request *rq,
970                          void *data, size_t len)
971 {
972         struct blk_trace *bt = q->blk_trace;
973
974         if (likely(!bt))
975                 return;
976
977         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
978                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
979                                 BLK_TA_DRV_DATA, rq->errors, len, data);
980         else
981                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
982                                 BLK_TA_DRV_DATA, rq->errors, len, data);
983 }
984 EXPORT_SYMBOL_GPL(blk_add_driver_data);
985
986 static void blk_register_tracepoints(void)
987 {
988         int ret;
989
990         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
991         WARN_ON(ret);
992         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
993         WARN_ON(ret);
994         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
995         WARN_ON(ret);
996         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
997         WARN_ON(ret);
998         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
999         WARN_ON(ret);
1000         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1001         WARN_ON(ret);
1002         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1003         WARN_ON(ret);
1004         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1005         WARN_ON(ret);
1006         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1007         WARN_ON(ret);
1008         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1009         WARN_ON(ret);
1010         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1011         WARN_ON(ret);
1012         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1013         WARN_ON(ret);
1014         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1015         WARN_ON(ret);
1016         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1017         WARN_ON(ret);
1018         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1019         WARN_ON(ret);
1020         ret = register_trace_block_split(blk_add_trace_split, NULL);
1021         WARN_ON(ret);
1022         ret = register_trace_block_remap(blk_add_trace_remap, NULL);
1023         WARN_ON(ret);
1024         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1025         WARN_ON(ret);
1026 }
1027
1028 static void blk_unregister_tracepoints(void)
1029 {
1030         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1031         unregister_trace_block_remap(blk_add_trace_remap, NULL);
1032         unregister_trace_block_split(blk_add_trace_split, NULL);
1033         unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1034         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1035         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1036         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1037         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1038         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1039         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1040         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1041         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1042         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1043         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1044         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1045         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1046         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1047         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1048
1049         tracepoint_synchronize_unregister();
1050 }
1051
1052 /*
1053  * struct blk_io_tracer formatting routines
1054  */
1055
1056 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1057 {
1058         int i = 0;
1059         int tc = t->action >> BLK_TC_SHIFT;
1060
1061         if (t->action == BLK_TN_MESSAGE) {
1062                 rwbs[i++] = 'N';
1063                 goto out;
1064         }
1065
1066         if (tc & BLK_TC_DISCARD)
1067                 rwbs[i++] = 'D';
1068         else if (tc & BLK_TC_WRITE)
1069                 rwbs[i++] = 'W';
1070         else if (t->bytes)
1071                 rwbs[i++] = 'R';
1072         else
1073                 rwbs[i++] = 'N';
1074
1075         if (tc & BLK_TC_AHEAD)
1076                 rwbs[i++] = 'A';
1077         if (tc & BLK_TC_BARRIER)
1078                 rwbs[i++] = 'B';
1079         if (tc & BLK_TC_SYNC)
1080                 rwbs[i++] = 'S';
1081         if (tc & BLK_TC_META)
1082                 rwbs[i++] = 'M';
1083 out:
1084         rwbs[i] = '\0';
1085 }
1086
1087 static inline
1088 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1089 {
1090         return (const struct blk_io_trace *)ent;
1091 }
1092
1093 static inline const void *pdu_start(const struct trace_entry *ent)
1094 {
1095         return te_blk_io_trace(ent) + 1;
1096 }
1097
1098 static inline u32 t_action(const struct trace_entry *ent)
1099 {
1100         return te_blk_io_trace(ent)->action;
1101 }
1102
1103 static inline u32 t_bytes(const struct trace_entry *ent)
1104 {
1105         return te_blk_io_trace(ent)->bytes;
1106 }
1107
1108 static inline u32 t_sec(const struct trace_entry *ent)
1109 {
1110         return te_blk_io_trace(ent)->bytes >> 9;
1111 }
1112
1113 static inline unsigned long long t_sector(const struct trace_entry *ent)
1114 {
1115         return te_blk_io_trace(ent)->sector;
1116 }
1117
1118 static inline __u16 t_error(const struct trace_entry *ent)
1119 {
1120         return te_blk_io_trace(ent)->error;
1121 }
1122
1123 static __u64 get_pdu_int(const struct trace_entry *ent)
1124 {
1125         const __u64 *val = pdu_start(ent);
1126         return be64_to_cpu(*val);
1127 }
1128
1129 static void get_pdu_remap(const struct trace_entry *ent,
1130                           struct blk_io_trace_remap *r)
1131 {
1132         const struct blk_io_trace_remap *__r = pdu_start(ent);
1133         __u64 sector_from = __r->sector_from;
1134
1135         r->device_from = be32_to_cpu(__r->device_from);
1136         r->device_to   = be32_to_cpu(__r->device_to);
1137         r->sector_from = be64_to_cpu(sector_from);
1138 }
1139
1140 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1141
1142 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1143 {
1144         char rwbs[6];
1145         unsigned long long ts  = iter->ts;
1146         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1147         unsigned secs          = (unsigned long)ts;
1148         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1149
1150         fill_rwbs(rwbs, t);
1151
1152         return trace_seq_printf(&iter->seq,
1153                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1154                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1155                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1156 }
1157
1158 static int blk_log_action(struct trace_iterator *iter, const char *act)
1159 {
1160         char rwbs[6];
1161         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1162
1163         fill_rwbs(rwbs, t);
1164         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1165                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1166 }
1167
1168 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1169 {
1170         const unsigned char *pdu_buf;
1171         int pdu_len;
1172         int i, end, ret;
1173
1174         pdu_buf = pdu_start(ent);
1175         pdu_len = te_blk_io_trace(ent)->pdu_len;
1176
1177         if (!pdu_len)
1178                 return 1;
1179
1180         /* find the last zero that needs to be printed */
1181         for (end = pdu_len - 1; end >= 0; end--)
1182                 if (pdu_buf[end])
1183                         break;
1184         end++;
1185
1186         if (!trace_seq_putc(s, '('))
1187                 return 0;
1188
1189         for (i = 0; i < pdu_len; i++) {
1190
1191                 ret = trace_seq_printf(s, "%s%02x",
1192                                        i == 0 ? "" : " ", pdu_buf[i]);
1193                 if (!ret)
1194                         return ret;
1195
1196                 /*
1197                  * stop when the rest is just zeroes and indicate so
1198                  * with a ".." appended
1199                  */
1200                 if (i == end && end != pdu_len - 1)
1201                         return trace_seq_puts(s, " ..) ");
1202         }
1203
1204         return trace_seq_puts(s, ") ");
1205 }
1206
1207 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1208 {
1209         char cmd[TASK_COMM_LEN];
1210
1211         trace_find_cmdline(ent->pid, cmd);
1212
1213         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1214                 int ret;
1215
1216                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1217                 if (!ret)
1218                         return 0;
1219                 ret = blk_log_dump_pdu(s, ent);
1220                 if (!ret)
1221                         return 0;
1222                 return trace_seq_printf(s, "[%s]\n", cmd);
1223         } else {
1224                 if (t_sec(ent))
1225                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1226                                                 t_sector(ent), t_sec(ent), cmd);
1227                 return trace_seq_printf(s, "[%s]\n", cmd);
1228         }
1229 }
1230
1231 static int blk_log_with_error(struct trace_seq *s,
1232                               const struct trace_entry *ent)
1233 {
1234         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1235                 int ret;
1236
1237                 ret = blk_log_dump_pdu(s, ent);
1238                 if (ret)
1239                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1240                 return 0;
1241         } else {
1242                 if (t_sec(ent))
1243                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1244                                                 t_sector(ent),
1245                                                 t_sec(ent), t_error(ent));
1246                 return trace_seq_printf(s, "%llu [%d]\n",
1247                                         t_sector(ent), t_error(ent));
1248         }
1249 }
1250
1251 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1252 {
1253         struct blk_io_trace_remap r = { .device_from = 0, };
1254
1255         get_pdu_remap(ent, &r);
1256         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1257                                 t_sector(ent), t_sec(ent),
1258                                 MAJOR(r.device_from), MINOR(r.device_from),
1259                                 (unsigned long long)r.sector_from);
1260 }
1261
1262 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1263 {
1264         char cmd[TASK_COMM_LEN];
1265
1266         trace_find_cmdline(ent->pid, cmd);
1267
1268         return trace_seq_printf(s, "[%s]\n", cmd);
1269 }
1270
1271 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1272 {
1273         char cmd[TASK_COMM_LEN];
1274
1275         trace_find_cmdline(ent->pid, cmd);
1276
1277         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1278 }
1279
1280 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1281 {
1282         char cmd[TASK_COMM_LEN];
1283
1284         trace_find_cmdline(ent->pid, cmd);
1285
1286         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1287                                 get_pdu_int(ent), cmd);
1288 }
1289
1290 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1291 {
1292         int ret;
1293         const struct blk_io_trace *t = te_blk_io_trace(ent);
1294
1295         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1296         if (ret)
1297                 return trace_seq_putc(s, '\n');
1298         return ret;
1299 }
1300
1301 /*
1302  * struct tracer operations
1303  */
1304
1305 static void blk_tracer_print_header(struct seq_file *m)
1306 {
1307         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1308                 return;
1309         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1310                     "#  |     |     |           |   |   |\n");
1311 }
1312
1313 static void blk_tracer_start(struct trace_array *tr)
1314 {
1315         blk_tracer_enabled = true;
1316 }
1317
1318 static int blk_tracer_init(struct trace_array *tr)
1319 {
1320         blk_tr = tr;
1321         blk_tracer_start(tr);
1322         return 0;
1323 }
1324
1325 static void blk_tracer_stop(struct trace_array *tr)
1326 {
1327         blk_tracer_enabled = false;
1328 }
1329
1330 static void blk_tracer_reset(struct trace_array *tr)
1331 {
1332         blk_tracer_stop(tr);
1333 }
1334
1335 static const struct {
1336         const char *act[2];
1337         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1338 } what2act[] = {
1339         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1340         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1341         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1342         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1343         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1344         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1345         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1346         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1347         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1348         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1349         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1350         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1351         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1352         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1353         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1354 };
1355
1356 static enum print_line_t print_one_line(struct trace_iterator *iter,
1357                                         bool classic)
1358 {
1359         struct trace_seq *s = &iter->seq;
1360         const struct blk_io_trace *t;
1361         u16 what;
1362         int ret;
1363         bool long_act;
1364         blk_log_action_t *log_action;
1365
1366         t          = te_blk_io_trace(iter->ent);
1367         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1368         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1369         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1370
1371         if (t->action == BLK_TN_MESSAGE) {
1372                 ret = log_action(iter, long_act ? "message" : "m");
1373                 if (ret)
1374                         ret = blk_log_msg(s, iter->ent);
1375                 goto out;
1376         }
1377
1378         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1379                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1380         else {
1381                 ret = log_action(iter, what2act[what].act[long_act]);
1382                 if (ret)
1383                         ret = what2act[what].print(s, iter->ent);
1384         }
1385 out:
1386         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1387 }
1388
1389 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1390                                                int flags, struct trace_event *event)
1391 {
1392         return print_one_line(iter, false);
1393 }
1394
1395 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1396 {
1397         struct trace_seq *s = &iter->seq;
1398         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1399         const int offset = offsetof(struct blk_io_trace, sector);
1400         struct blk_io_trace old = {
1401                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1402                 .time     = iter->ts,
1403         };
1404
1405         if (!trace_seq_putmem(s, &old, offset))
1406                 return 0;
1407         return trace_seq_putmem(s, &t->sector,
1408                                 sizeof(old) - offset + t->pdu_len);
1409 }
1410
1411 static enum print_line_t
1412 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1413                              struct trace_event *event)
1414 {
1415         return blk_trace_synthesize_old_trace(iter) ?
1416                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1417 }
1418
1419 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1420 {
1421         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1422                 return TRACE_TYPE_UNHANDLED;
1423
1424         return print_one_line(iter, true);
1425 }
1426
1427 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1428 {
1429         /* don't output context-info for blk_classic output */
1430         if (bit == TRACE_BLK_OPT_CLASSIC) {
1431                 if (set)
1432                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1433                 else
1434                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1435         }
1436         return 0;
1437 }
1438
1439 static struct tracer blk_tracer __read_mostly = {
1440         .name           = "blk",
1441         .init           = blk_tracer_init,
1442         .reset          = blk_tracer_reset,
1443         .start          = blk_tracer_start,
1444         .stop           = blk_tracer_stop,
1445         .print_header   = blk_tracer_print_header,
1446         .print_line     = blk_tracer_print_line,
1447         .flags          = &blk_tracer_flags,
1448         .set_flag       = blk_tracer_set_flag,
1449 };
1450
1451 static struct trace_event_functions trace_blk_event_funcs = {
1452         .trace          = blk_trace_event_print,
1453         .binary         = blk_trace_event_print_binary,
1454 };
1455
1456 static struct trace_event trace_blk_event = {
1457         .type           = TRACE_BLK,
1458         .funcs          = &trace_blk_event_funcs,
1459 };
1460
1461 static int __init init_blk_tracer(void)
1462 {
1463         if (!register_ftrace_event(&trace_blk_event)) {
1464                 pr_warning("Warning: could not register block events\n");
1465                 return 1;
1466         }
1467
1468         if (register_tracer(&blk_tracer) != 0) {
1469                 pr_warning("Warning: could not register the block tracer\n");
1470                 unregister_ftrace_event(&trace_blk_event);
1471                 return 1;
1472         }
1473
1474         return 0;
1475 }
1476
1477 device_initcall(init_blk_tracer);
1478
1479 static int blk_trace_remove_queue(struct request_queue *q)
1480 {
1481         struct blk_trace *bt;
1482
1483         bt = xchg(&q->blk_trace, NULL);
1484         if (bt == NULL)
1485                 return -EINVAL;
1486
1487         if (atomic_dec_and_test(&blk_probes_ref))
1488                 blk_unregister_tracepoints();
1489
1490         blk_trace_free(bt);
1491         return 0;
1492 }
1493
1494 /*
1495  * Setup everything required to start tracing
1496  */
1497 static int blk_trace_setup_queue(struct request_queue *q,
1498                                  struct block_device *bdev)
1499 {
1500         struct blk_trace *old_bt, *bt = NULL;
1501         int ret = -ENOMEM;
1502
1503         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1504         if (!bt)
1505                 return -ENOMEM;
1506
1507         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1508         if (!bt->msg_data)
1509                 goto free_bt;
1510
1511         bt->dev = bdev->bd_dev;
1512         bt->act_mask = (u16)-1;
1513
1514         blk_trace_setup_lba(bt, bdev);
1515
1516         old_bt = xchg(&q->blk_trace, bt);
1517         if (old_bt != NULL) {
1518                 (void)xchg(&q->blk_trace, old_bt);
1519                 ret = -EBUSY;
1520                 goto free_bt;
1521         }
1522
1523         if (atomic_inc_return(&blk_probes_ref) == 1)
1524                 blk_register_tracepoints();
1525         return 0;
1526
1527 free_bt:
1528         blk_trace_free(bt);
1529         return ret;
1530 }
1531
1532 /*
1533  * sysfs interface to enable and configure tracing
1534  */
1535
1536 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1537                                          struct device_attribute *attr,
1538                                          char *buf);
1539 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1540                                           struct device_attribute *attr,
1541                                           const char *buf, size_t count);
1542 #define BLK_TRACE_DEVICE_ATTR(_name) \
1543         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1544                     sysfs_blk_trace_attr_show, \
1545                     sysfs_blk_trace_attr_store)
1546
1547 static BLK_TRACE_DEVICE_ATTR(enable);
1548 static BLK_TRACE_DEVICE_ATTR(act_mask);
1549 static BLK_TRACE_DEVICE_ATTR(pid);
1550 static BLK_TRACE_DEVICE_ATTR(start_lba);
1551 static BLK_TRACE_DEVICE_ATTR(end_lba);
1552
1553 static struct attribute *blk_trace_attrs[] = {
1554         &dev_attr_enable.attr,
1555         &dev_attr_act_mask.attr,
1556         &dev_attr_pid.attr,
1557         &dev_attr_start_lba.attr,
1558         &dev_attr_end_lba.attr,
1559         NULL
1560 };
1561
1562 struct attribute_group blk_trace_attr_group = {
1563         .name  = "trace",
1564         .attrs = blk_trace_attrs,
1565 };
1566
1567 static const struct {
1568         int mask;
1569         const char *str;
1570 } mask_maps[] = {
1571         { BLK_TC_READ,          "read"          },
1572         { BLK_TC_WRITE,         "write"         },
1573         { BLK_TC_BARRIER,       "barrier"       },
1574         { BLK_TC_SYNC,          "sync"          },
1575         { BLK_TC_QUEUE,         "queue"         },
1576         { BLK_TC_REQUEUE,       "requeue"       },
1577         { BLK_TC_ISSUE,         "issue"         },
1578         { BLK_TC_COMPLETE,      "complete"      },
1579         { BLK_TC_FS,            "fs"            },
1580         { BLK_TC_PC,            "pc"            },
1581         { BLK_TC_AHEAD,         "ahead"         },
1582         { BLK_TC_META,          "meta"          },
1583         { BLK_TC_DISCARD,       "discard"       },
1584         { BLK_TC_DRV_DATA,      "drv_data"      },
1585 };
1586
1587 static int blk_trace_str2mask(const char *str)
1588 {
1589         int i;
1590         int mask = 0;
1591         char *buf, *s, *token;
1592
1593         buf = kstrdup(str, GFP_KERNEL);
1594         if (buf == NULL)
1595                 return -ENOMEM;
1596         s = strstrip(buf);
1597
1598         while (1) {
1599                 token = strsep(&s, ",");
1600                 if (token == NULL)
1601                         break;
1602
1603                 if (*token == '\0')
1604                         continue;
1605
1606                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1607                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1608                                 mask |= mask_maps[i].mask;
1609                                 break;
1610                         }
1611                 }
1612                 if (i == ARRAY_SIZE(mask_maps)) {
1613                         mask = -EINVAL;
1614                         break;
1615                 }
1616         }
1617         kfree(buf);
1618
1619         return mask;
1620 }
1621
1622 static ssize_t blk_trace_mask2str(char *buf, int mask)
1623 {
1624         int i;
1625         char *p = buf;
1626
1627         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1628                 if (mask & mask_maps[i].mask) {
1629                         p += sprintf(p, "%s%s",
1630                                     (p == buf) ? "" : ",", mask_maps[i].str);
1631                 }
1632         }
1633         *p++ = '\n';
1634
1635         return p - buf;
1636 }
1637
1638 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1639 {
1640         if (bdev->bd_disk == NULL)
1641                 return NULL;
1642
1643         return bdev_get_queue(bdev);
1644 }
1645
1646 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1647                                          struct device_attribute *attr,
1648                                          char *buf)
1649 {
1650         struct hd_struct *p = dev_to_part(dev);
1651         struct request_queue *q;
1652         struct block_device *bdev;
1653         ssize_t ret = -ENXIO;
1654
1655         lock_kernel();
1656         bdev = bdget(part_devt(p));
1657         if (bdev == NULL)
1658                 goto out_unlock_kernel;
1659
1660         q = blk_trace_get_queue(bdev);
1661         if (q == NULL)
1662                 goto out_bdput;
1663
1664         mutex_lock(&bdev->bd_mutex);
1665
1666         if (attr == &dev_attr_enable) {
1667                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1668                 goto out_unlock_bdev;
1669         }
1670
1671         if (q->blk_trace == NULL)
1672                 ret = sprintf(buf, "disabled\n");
1673         else if (attr == &dev_attr_act_mask)
1674                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1675         else if (attr == &dev_attr_pid)
1676                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1677         else if (attr == &dev_attr_start_lba)
1678                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1679         else if (attr == &dev_attr_end_lba)
1680                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1681
1682 out_unlock_bdev:
1683         mutex_unlock(&bdev->bd_mutex);
1684 out_bdput:
1685         bdput(bdev);
1686 out_unlock_kernel:
1687         unlock_kernel();
1688         return ret;
1689 }
1690
1691 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1692                                           struct device_attribute *attr,
1693                                           const char *buf, size_t count)
1694 {
1695         struct block_device *bdev;
1696         struct request_queue *q;
1697         struct hd_struct *p;
1698         u64 value;
1699         ssize_t ret = -EINVAL;
1700
1701         if (count == 0)
1702                 goto out;
1703
1704         if (attr == &dev_attr_act_mask) {
1705                 if (sscanf(buf, "%llx", &value) != 1) {
1706                         /* Assume it is a list of trace category names */
1707                         ret = blk_trace_str2mask(buf);
1708                         if (ret < 0)
1709                                 goto out;
1710                         value = ret;
1711                 }
1712         } else if (sscanf(buf, "%llu", &value) != 1)
1713                 goto out;
1714
1715         ret = -ENXIO;
1716
1717         lock_kernel();
1718         p = dev_to_part(dev);
1719         bdev = bdget(part_devt(p));
1720         if (bdev == NULL)
1721                 goto out_unlock_kernel;
1722
1723         q = blk_trace_get_queue(bdev);
1724         if (q == NULL)
1725                 goto out_bdput;
1726
1727         mutex_lock(&bdev->bd_mutex);
1728
1729         if (attr == &dev_attr_enable) {
1730                 if (value)
1731                         ret = blk_trace_setup_queue(q, bdev);
1732                 else
1733                         ret = blk_trace_remove_queue(q);
1734                 goto out_unlock_bdev;
1735         }
1736
1737         ret = 0;
1738         if (q->blk_trace == NULL)
1739                 ret = blk_trace_setup_queue(q, bdev);
1740
1741         if (ret == 0) {
1742                 if (attr == &dev_attr_act_mask)
1743                         q->blk_trace->act_mask = value;
1744                 else if (attr == &dev_attr_pid)
1745                         q->blk_trace->pid = value;
1746                 else if (attr == &dev_attr_start_lba)
1747                         q->blk_trace->start_lba = value;
1748                 else if (attr == &dev_attr_end_lba)
1749                         q->blk_trace->end_lba = value;
1750         }
1751
1752 out_unlock_bdev:
1753         mutex_unlock(&bdev->bd_mutex);
1754 out_bdput:
1755         bdput(bdev);
1756 out_unlock_kernel:
1757         unlock_kernel();
1758 out:
1759         return ret ? ret : count;
1760 }
1761
1762 int blk_trace_init_sysfs(struct device *dev)
1763 {
1764         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1765 }
1766
1767 void blk_trace_remove_sysfs(struct device *dev)
1768 {
1769         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1770 }
1771
1772 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1773
1774 #ifdef CONFIG_EVENT_TRACING
1775
1776 void blk_dump_cmd(char *buf, struct request *rq)
1777 {
1778         int i, end;
1779         int len = rq->cmd_len;
1780         unsigned char *cmd = rq->cmd;
1781
1782         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1783                 buf[0] = '\0';
1784                 return;
1785         }
1786
1787         for (end = len - 1; end >= 0; end--)
1788                 if (cmd[end])
1789                         break;
1790         end++;
1791
1792         for (i = 0; i < len; i++) {
1793                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1794                 if (i == end && end != len - 1) {
1795                         sprintf(buf, " ..");
1796                         break;
1797                 }
1798         }
1799 }
1800
1801 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1802 {
1803         int i = 0;
1804
1805         if (rw & WRITE)
1806                 rwbs[i++] = 'W';
1807         else if (rw & REQ_DISCARD)
1808                 rwbs[i++] = 'D';
1809         else if (bytes)
1810                 rwbs[i++] = 'R';
1811         else
1812                 rwbs[i++] = 'N';
1813
1814         if (rw & REQ_RAHEAD)
1815                 rwbs[i++] = 'A';
1816         if (rw & REQ_HARDBARRIER)
1817                 rwbs[i++] = 'B';
1818         if (rw & REQ_SYNC)
1819                 rwbs[i++] = 'S';
1820         if (rw & REQ_META)
1821                 rwbs[i++] = 'M';
1822         if (rw & REQ_SECURE)
1823                 rwbs[i++] = 'E';
1824
1825         rwbs[i] = '\0';
1826 }
1827
1828 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1829 {
1830         int rw = rq->cmd_flags & 0x03;
1831         int bytes;
1832
1833         if (rq->cmd_flags & REQ_DISCARD)
1834                 rw |= REQ_DISCARD;
1835
1836         if (rq->cmd_flags & REQ_SECURE)
1837                 rw |= REQ_SECURE;
1838
1839         bytes = blk_rq_bytes(rq);
1840
1841         blk_fill_rwbs(rwbs, rw, bytes);
1842 }
1843
1844 #endif /* CONFIG_EVENT_TRACING */
1845