xref: /linux/kernel/trace/blktrace.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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/export.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 	/*
143 	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
144 	 * message to the trace.
145 	 */
146 	if (!(bt->act_mask & BLK_TC_NOTIFY))
147 		return;
148 
149 	local_irq_save(flags);
150 	buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
151 	va_start(args, fmt);
152 	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
153 	va_end(args);
154 
155 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
156 	local_irq_restore(flags);
157 }
158 EXPORT_SYMBOL_GPL(__trace_note_message);
159 
160 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
161 			 pid_t pid)
162 {
163 	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
164 		return 1;
165 	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
166 		return 1;
167 	if (bt->pid && pid != bt->pid)
168 		return 1;
169 
170 	return 0;
171 }
172 
173 /*
174  * Data direction bit lookup
175  */
176 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
177 				 BLK_TC_ACT(BLK_TC_WRITE) };
178 
179 #define BLK_TC_RAHEAD		BLK_TC_AHEAD
180 
181 /* The ilog2() calls fall out because they're constant */
182 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
183 	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
184 
185 /*
186  * The worker for the various blk_add_trace*() types. Fills out a
187  * blk_io_trace structure and places it in a per-cpu subbuffer.
188  */
189 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
190 		     int rw, u32 what, int error, int pdu_len, void *pdu_data)
191 {
192 	struct task_struct *tsk = current;
193 	struct ring_buffer_event *event = NULL;
194 	struct ring_buffer *buffer = NULL;
195 	struct blk_io_trace *t;
196 	unsigned long flags = 0;
197 	unsigned long *sequence;
198 	pid_t pid;
199 	int cpu, pc = 0;
200 	bool blk_tracer = blk_tracer_enabled;
201 
202 	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
203 		return;
204 
205 	what |= ddir_act[rw & WRITE];
206 	what |= MASK_TC_BIT(rw, SYNC);
207 	what |= MASK_TC_BIT(rw, RAHEAD);
208 	what |= MASK_TC_BIT(rw, META);
209 	what |= MASK_TC_BIT(rw, DISCARD);
210 	what |= MASK_TC_BIT(rw, FLUSH);
211 	what |= MASK_TC_BIT(rw, FUA);
212 
213 	pid = tsk->pid;
214 	if (act_log_check(bt, what, sector, pid))
215 		return;
216 	cpu = raw_smp_processor_id();
217 
218 	if (blk_tracer) {
219 		tracing_record_cmdline(current);
220 
221 		buffer = blk_tr->buffer;
222 		pc = preempt_count();
223 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
224 						  sizeof(*t) + pdu_len,
225 						  0, pc);
226 		if (!event)
227 			return;
228 		t = ring_buffer_event_data(event);
229 		goto record_it;
230 	}
231 
232 	/*
233 	 * A word about the locking here - we disable interrupts to reserve
234 	 * some space in the relay per-cpu buffer, to prevent an irq
235 	 * from coming in and stepping on our toes.
236 	 */
237 	local_irq_save(flags);
238 
239 	if (unlikely(tsk->btrace_seq != blktrace_seq))
240 		trace_note_tsk(bt, tsk);
241 
242 	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
243 	if (t) {
244 		sequence = per_cpu_ptr(bt->sequence, cpu);
245 
246 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
247 		t->sequence = ++(*sequence);
248 		t->time = ktime_to_ns(ktime_get());
249 record_it:
250 		/*
251 		 * These two are not needed in ftrace as they are in the
252 		 * generic trace_entry, filled by tracing_generic_entry_update,
253 		 * but for the trace_event->bin() synthesizer benefit we do it
254 		 * here too.
255 		 */
256 		t->cpu = cpu;
257 		t->pid = pid;
258 
259 		t->sector = sector;
260 		t->bytes = bytes;
261 		t->action = what;
262 		t->device = bt->dev;
263 		t->error = error;
264 		t->pdu_len = pdu_len;
265 
266 		if (pdu_len)
267 			memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
268 
269 		if (blk_tracer) {
270 			trace_buffer_unlock_commit(buffer, event, 0, pc);
271 			return;
272 		}
273 	}
274 
275 	local_irq_restore(flags);
276 }
277 
278 static struct dentry *blk_tree_root;
279 static DEFINE_MUTEX(blk_tree_mutex);
280 
281 static void blk_trace_free(struct blk_trace *bt)
282 {
283 	debugfs_remove(bt->msg_file);
284 	debugfs_remove(bt->dropped_file);
285 	relay_close(bt->rchan);
286 	debugfs_remove(bt->dir);
287 	free_percpu(bt->sequence);
288 	free_percpu(bt->msg_data);
289 	kfree(bt);
290 }
291 
292 static void blk_trace_cleanup(struct blk_trace *bt)
293 {
294 	blk_trace_free(bt);
295 	if (atomic_dec_and_test(&blk_probes_ref))
296 		blk_unregister_tracepoints();
297 }
298 
299 int blk_trace_remove(struct request_queue *q)
300 {
301 	struct blk_trace *bt;
302 
303 	bt = xchg(&q->blk_trace, NULL);
304 	if (!bt)
305 		return -EINVAL;
306 
307 	if (bt->trace_state != Blktrace_running)
308 		blk_trace_cleanup(bt);
309 
310 	return 0;
311 }
312 EXPORT_SYMBOL_GPL(blk_trace_remove);
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 =		simple_open,
328 	.read =		blk_dropped_read,
329 	.llseek =	default_llseek,
330 };
331 
332 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
333 				size_t count, loff_t *ppos)
334 {
335 	char *msg;
336 	struct blk_trace *bt;
337 
338 	if (count >= BLK_TN_MAX_MSG)
339 		return -EINVAL;
340 
341 	msg = kmalloc(count + 1, GFP_KERNEL);
342 	if (msg == NULL)
343 		return -ENOMEM;
344 
345 	if (copy_from_user(msg, buffer, count)) {
346 		kfree(msg);
347 		return -EFAULT;
348 	}
349 
350 	msg[count] = '\0';
351 	bt = filp->private_data;
352 	__trace_note_message(bt, "%s", msg);
353 	kfree(msg);
354 
355 	return count;
356 }
357 
358 static const struct file_operations blk_msg_fops = {
359 	.owner =	THIS_MODULE,
360 	.open =		simple_open,
361 	.write =	blk_msg_write,
362 	.llseek =	noop_llseek,
363 };
364 
365 /*
366  * Keep track of how many times we encountered a full subbuffer, to aid
367  * the user space app in telling how many lost events there were.
368  */
369 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
370 				     void *prev_subbuf, size_t prev_padding)
371 {
372 	struct blk_trace *bt;
373 
374 	if (!relay_buf_full(buf))
375 		return 1;
376 
377 	bt = buf->chan->private_data;
378 	atomic_inc(&bt->dropped);
379 	return 0;
380 }
381 
382 static int blk_remove_buf_file_callback(struct dentry *dentry)
383 {
384 	debugfs_remove(dentry);
385 
386 	return 0;
387 }
388 
389 static struct dentry *blk_create_buf_file_callback(const char *filename,
390 						   struct dentry *parent,
391 						   umode_t mode,
392 						   struct rchan_buf *buf,
393 						   int *is_global)
394 {
395 	return debugfs_create_file(filename, mode, parent, buf,
396 					&relay_file_operations);
397 }
398 
399 static struct rchan_callbacks blk_relay_callbacks = {
400 	.subbuf_start		= blk_subbuf_start_callback,
401 	.create_buf_file	= blk_create_buf_file_callback,
402 	.remove_buf_file	= blk_remove_buf_file_callback,
403 };
404 
405 static void blk_trace_setup_lba(struct blk_trace *bt,
406 				struct block_device *bdev)
407 {
408 	struct hd_struct *part = NULL;
409 
410 	if (bdev)
411 		part = bdev->bd_part;
412 
413 	if (part) {
414 		bt->start_lba = part->start_sect;
415 		bt->end_lba = part->start_sect + part->nr_sects;
416 	} else {
417 		bt->start_lba = 0;
418 		bt->end_lba = -1ULL;
419 	}
420 }
421 
422 /*
423  * Setup everything required to start tracing
424  */
425 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
426 		       struct block_device *bdev,
427 		       struct blk_user_trace_setup *buts)
428 {
429 	struct blk_trace *old_bt, *bt = NULL;
430 	struct dentry *dir = NULL;
431 	int ret, i;
432 
433 	if (!buts->buf_size || !buts->buf_nr)
434 		return -EINVAL;
435 
436 	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
437 	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
438 
439 	/*
440 	 * some device names have larger paths - convert the slashes
441 	 * to underscores for this to work as expected
442 	 */
443 	for (i = 0; i < strlen(buts->name); i++)
444 		if (buts->name[i] == '/')
445 			buts->name[i] = '_';
446 
447 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
448 	if (!bt)
449 		return -ENOMEM;
450 
451 	ret = -ENOMEM;
452 	bt->sequence = alloc_percpu(unsigned long);
453 	if (!bt->sequence)
454 		goto err;
455 
456 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
457 	if (!bt->msg_data)
458 		goto err;
459 
460 	ret = -ENOENT;
461 
462 	mutex_lock(&blk_tree_mutex);
463 	if (!blk_tree_root) {
464 		blk_tree_root = debugfs_create_dir("block", NULL);
465 		if (!blk_tree_root) {
466 			mutex_unlock(&blk_tree_mutex);
467 			goto err;
468 		}
469 	}
470 	mutex_unlock(&blk_tree_mutex);
471 
472 	dir = debugfs_create_dir(buts->name, blk_tree_root);
473 
474 	if (!dir)
475 		goto err;
476 
477 	bt->dir = dir;
478 	bt->dev = dev;
479 	atomic_set(&bt->dropped, 0);
480 
481 	ret = -EIO;
482 	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
483 					       &blk_dropped_fops);
484 	if (!bt->dropped_file)
485 		goto err;
486 
487 	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
488 	if (!bt->msg_file)
489 		goto err;
490 
491 	bt->rchan = relay_open("trace", dir, buts->buf_size,
492 				buts->buf_nr, &blk_relay_callbacks, bt);
493 	if (!bt->rchan)
494 		goto err;
495 
496 	bt->act_mask = buts->act_mask;
497 	if (!bt->act_mask)
498 		bt->act_mask = (u16) -1;
499 
500 	blk_trace_setup_lba(bt, bdev);
501 
502 	/* overwrite with user settings */
503 	if (buts->start_lba)
504 		bt->start_lba = buts->start_lba;
505 	if (buts->end_lba)
506 		bt->end_lba = buts->end_lba;
507 
508 	bt->pid = buts->pid;
509 	bt->trace_state = Blktrace_setup;
510 
511 	ret = -EBUSY;
512 	old_bt = xchg(&q->blk_trace, bt);
513 	if (old_bt) {
514 		(void) xchg(&q->blk_trace, old_bt);
515 		goto err;
516 	}
517 
518 	if (atomic_inc_return(&blk_probes_ref) == 1)
519 		blk_register_tracepoints();
520 
521 	return 0;
522 err:
523 	blk_trace_free(bt);
524 	return ret;
525 }
526 
527 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
528 		    struct block_device *bdev,
529 		    char __user *arg)
530 {
531 	struct blk_user_trace_setup buts;
532 	int ret;
533 
534 	ret = copy_from_user(&buts, arg, sizeof(buts));
535 	if (ret)
536 		return -EFAULT;
537 
538 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
539 	if (ret)
540 		return ret;
541 
542 	if (copy_to_user(arg, &buts, sizeof(buts))) {
543 		blk_trace_remove(q);
544 		return -EFAULT;
545 	}
546 	return 0;
547 }
548 EXPORT_SYMBOL_GPL(blk_trace_setup);
549 
550 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
551 static int compat_blk_trace_setup(struct request_queue *q, char *name,
552 				  dev_t dev, struct block_device *bdev,
553 				  char __user *arg)
554 {
555 	struct blk_user_trace_setup buts;
556 	struct compat_blk_user_trace_setup cbuts;
557 	int ret;
558 
559 	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
560 		return -EFAULT;
561 
562 	buts = (struct blk_user_trace_setup) {
563 		.act_mask = cbuts.act_mask,
564 		.buf_size = cbuts.buf_size,
565 		.buf_nr = cbuts.buf_nr,
566 		.start_lba = cbuts.start_lba,
567 		.end_lba = cbuts.end_lba,
568 		.pid = cbuts.pid,
569 	};
570 	memcpy(&buts.name, &cbuts.name, 32);
571 
572 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
573 	if (ret)
574 		return ret;
575 
576 	if (copy_to_user(arg, &buts.name, 32)) {
577 		blk_trace_remove(q);
578 		return -EFAULT;
579 	}
580 
581 	return 0;
582 }
583 #endif
584 
585 int blk_trace_startstop(struct request_queue *q, int start)
586 {
587 	int ret;
588 	struct blk_trace *bt = q->blk_trace;
589 
590 	if (bt == NULL)
591 		return -EINVAL;
592 
593 	/*
594 	 * For starting a trace, we can transition from a setup or stopped
595 	 * trace. For stopping a trace, the state must be running
596 	 */
597 	ret = -EINVAL;
598 	if (start) {
599 		if (bt->trace_state == Blktrace_setup ||
600 		    bt->trace_state == Blktrace_stopped) {
601 			blktrace_seq++;
602 			smp_mb();
603 			bt->trace_state = Blktrace_running;
604 
605 			trace_note_time(bt);
606 			ret = 0;
607 		}
608 	} else {
609 		if (bt->trace_state == Blktrace_running) {
610 			bt->trace_state = Blktrace_stopped;
611 			relay_flush(bt->rchan);
612 			ret = 0;
613 		}
614 	}
615 
616 	return ret;
617 }
618 EXPORT_SYMBOL_GPL(blk_trace_startstop);
619 
620 /**
621  * blk_trace_ioctl: - handle the ioctls associated with tracing
622  * @bdev:	the block device
623  * @cmd:	the ioctl cmd
624  * @arg:	the argument data, if any
625  *
626  **/
627 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
628 {
629 	struct request_queue *q;
630 	int ret, start = 0;
631 	char b[BDEVNAME_SIZE];
632 
633 	q = bdev_get_queue(bdev);
634 	if (!q)
635 		return -ENXIO;
636 
637 	mutex_lock(&bdev->bd_mutex);
638 
639 	switch (cmd) {
640 	case BLKTRACESETUP:
641 		bdevname(bdev, b);
642 		ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
643 		break;
644 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
645 	case BLKTRACESETUP32:
646 		bdevname(bdev, b);
647 		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
648 		break;
649 #endif
650 	case BLKTRACESTART:
651 		start = 1;
652 	case BLKTRACESTOP:
653 		ret = blk_trace_startstop(q, start);
654 		break;
655 	case BLKTRACETEARDOWN:
656 		ret = blk_trace_remove(q);
657 		break;
658 	default:
659 		ret = -ENOTTY;
660 		break;
661 	}
662 
663 	mutex_unlock(&bdev->bd_mutex);
664 	return ret;
665 }
666 
667 /**
668  * blk_trace_shutdown: - stop and cleanup trace structures
669  * @q:    the request queue associated with the device
670  *
671  **/
672 void blk_trace_shutdown(struct request_queue *q)
673 {
674 	if (q->blk_trace) {
675 		blk_trace_startstop(q, 0);
676 		blk_trace_remove(q);
677 	}
678 }
679 
680 /*
681  * blktrace probes
682  */
683 
684 /**
685  * blk_add_trace_rq - Add a trace for a request oriented action
686  * @q:		queue the io is for
687  * @rq:		the source request
688  * @what:	the action
689  *
690  * Description:
691  *     Records an action against a request. Will log the bio offset + size.
692  *
693  **/
694 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
695 			     u32 what)
696 {
697 	struct blk_trace *bt = q->blk_trace;
698 
699 	if (likely(!bt))
700 		return;
701 
702 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
703 		what |= BLK_TC_ACT(BLK_TC_PC);
704 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), rq->cmd_flags,
705 				what, rq->errors, rq->cmd_len, rq->cmd);
706 	} else  {
707 		what |= BLK_TC_ACT(BLK_TC_FS);
708 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
709 				rq->cmd_flags, what, rq->errors, 0, NULL);
710 	}
711 }
712 
713 static void blk_add_trace_rq_abort(void *ignore,
714 				   struct request_queue *q, struct request *rq)
715 {
716 	blk_add_trace_rq(q, rq, BLK_TA_ABORT);
717 }
718 
719 static void blk_add_trace_rq_insert(void *ignore,
720 				    struct request_queue *q, struct request *rq)
721 {
722 	blk_add_trace_rq(q, rq, BLK_TA_INSERT);
723 }
724 
725 static void blk_add_trace_rq_issue(void *ignore,
726 				   struct request_queue *q, struct request *rq)
727 {
728 	blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
729 }
730 
731 static void blk_add_trace_rq_requeue(void *ignore,
732 				     struct request_queue *q,
733 				     struct request *rq)
734 {
735 	blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
736 }
737 
738 static void blk_add_trace_rq_complete(void *ignore,
739 				      struct request_queue *q,
740 				      struct request *rq)
741 {
742 	blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
743 }
744 
745 /**
746  * blk_add_trace_bio - Add a trace for a bio oriented action
747  * @q:		queue the io is for
748  * @bio:	the source bio
749  * @what:	the action
750  * @error:	error, if any
751  *
752  * Description:
753  *     Records an action against a bio. Will log the bio offset + size.
754  *
755  **/
756 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
757 			      u32 what, int error)
758 {
759 	struct blk_trace *bt = q->blk_trace;
760 
761 	if (likely(!bt))
762 		return;
763 
764 	if (!error && !bio_flagged(bio, BIO_UPTODATE))
765 		error = EIO;
766 
767 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
768 			error, 0, NULL);
769 }
770 
771 static void blk_add_trace_bio_bounce(void *ignore,
772 				     struct request_queue *q, struct bio *bio)
773 {
774 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
775 }
776 
777 static void blk_add_trace_bio_complete(void *ignore,
778 				       struct request_queue *q, struct bio *bio,
779 				       int error)
780 {
781 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
782 }
783 
784 static void blk_add_trace_bio_backmerge(void *ignore,
785 					struct request_queue *q,
786 					struct bio *bio)
787 {
788 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
789 }
790 
791 static void blk_add_trace_bio_frontmerge(void *ignore,
792 					 struct request_queue *q,
793 					 struct bio *bio)
794 {
795 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
796 }
797 
798 static void blk_add_trace_bio_queue(void *ignore,
799 				    struct request_queue *q, struct bio *bio)
800 {
801 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
802 }
803 
804 static void blk_add_trace_getrq(void *ignore,
805 				struct request_queue *q,
806 				struct bio *bio, int rw)
807 {
808 	if (bio)
809 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
810 	else {
811 		struct blk_trace *bt = q->blk_trace;
812 
813 		if (bt)
814 			__blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
815 	}
816 }
817 
818 
819 static void blk_add_trace_sleeprq(void *ignore,
820 				  struct request_queue *q,
821 				  struct bio *bio, int rw)
822 {
823 	if (bio)
824 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
825 	else {
826 		struct blk_trace *bt = q->blk_trace;
827 
828 		if (bt)
829 			__blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
830 					0, 0, NULL);
831 	}
832 }
833 
834 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
835 {
836 	struct blk_trace *bt = q->blk_trace;
837 
838 	if (bt)
839 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
840 }
841 
842 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
843 				    unsigned int depth, bool explicit)
844 {
845 	struct blk_trace *bt = q->blk_trace;
846 
847 	if (bt) {
848 		__be64 rpdu = cpu_to_be64(depth);
849 		u32 what;
850 
851 		if (explicit)
852 			what = BLK_TA_UNPLUG_IO;
853 		else
854 			what = BLK_TA_UNPLUG_TIMER;
855 
856 		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
857 	}
858 }
859 
860 static void blk_add_trace_split(void *ignore,
861 				struct request_queue *q, struct bio *bio,
862 				unsigned int pdu)
863 {
864 	struct blk_trace *bt = q->blk_trace;
865 
866 	if (bt) {
867 		__be64 rpdu = cpu_to_be64(pdu);
868 
869 		__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
870 				BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
871 				sizeof(rpdu), &rpdu);
872 	}
873 }
874 
875 /**
876  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
877  * @ignore:	trace callback data parameter (not used)
878  * @q:		queue the io is for
879  * @bio:	the source bio
880  * @dev:	target device
881  * @from:	source sector
882  *
883  * Description:
884  *     Device mapper or raid target sometimes need to split a bio because
885  *     it spans a stripe (or similar). Add a trace for that action.
886  *
887  **/
888 static void blk_add_trace_bio_remap(void *ignore,
889 				    struct request_queue *q, struct bio *bio,
890 				    dev_t dev, sector_t from)
891 {
892 	struct blk_trace *bt = q->blk_trace;
893 	struct blk_io_trace_remap r;
894 
895 	if (likely(!bt))
896 		return;
897 
898 	r.device_from = cpu_to_be32(dev);
899 	r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
900 	r.sector_from = cpu_to_be64(from);
901 
902 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
903 			BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
904 			sizeof(r), &r);
905 }
906 
907 /**
908  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
909  * @ignore:	trace callback data parameter (not used)
910  * @q:		queue the io is for
911  * @rq:		the source request
912  * @dev:	target device
913  * @from:	source sector
914  *
915  * Description:
916  *     Device mapper remaps request to other devices.
917  *     Add a trace for that action.
918  *
919  **/
920 static void blk_add_trace_rq_remap(void *ignore,
921 				   struct request_queue *q,
922 				   struct request *rq, dev_t dev,
923 				   sector_t from)
924 {
925 	struct blk_trace *bt = q->blk_trace;
926 	struct blk_io_trace_remap r;
927 
928 	if (likely(!bt))
929 		return;
930 
931 	r.device_from = cpu_to_be32(dev);
932 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
933 	r.sector_from = cpu_to_be64(from);
934 
935 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
936 			rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
937 			sizeof(r), &r);
938 }
939 
940 /**
941  * blk_add_driver_data - Add binary message with driver-specific data
942  * @q:		queue the io is for
943  * @rq:		io request
944  * @data:	driver-specific data
945  * @len:	length of driver-specific data
946  *
947  * Description:
948  *     Some drivers might want to write driver-specific data per request.
949  *
950  **/
951 void blk_add_driver_data(struct request_queue *q,
952 			 struct request *rq,
953 			 void *data, size_t len)
954 {
955 	struct blk_trace *bt = q->blk_trace;
956 
957 	if (likely(!bt))
958 		return;
959 
960 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
961 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
962 				BLK_TA_DRV_DATA, rq->errors, len, data);
963 	else
964 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
965 				BLK_TA_DRV_DATA, rq->errors, len, data);
966 }
967 EXPORT_SYMBOL_GPL(blk_add_driver_data);
968 
969 static void blk_register_tracepoints(void)
970 {
971 	int ret;
972 
973 	ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
974 	WARN_ON(ret);
975 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
976 	WARN_ON(ret);
977 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
978 	WARN_ON(ret);
979 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
980 	WARN_ON(ret);
981 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
982 	WARN_ON(ret);
983 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
984 	WARN_ON(ret);
985 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
986 	WARN_ON(ret);
987 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
988 	WARN_ON(ret);
989 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
990 	WARN_ON(ret);
991 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
992 	WARN_ON(ret);
993 	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
994 	WARN_ON(ret);
995 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
996 	WARN_ON(ret);
997 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
998 	WARN_ON(ret);
999 	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1000 	WARN_ON(ret);
1001 	ret = register_trace_block_split(blk_add_trace_split, NULL);
1002 	WARN_ON(ret);
1003 	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1004 	WARN_ON(ret);
1005 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1006 	WARN_ON(ret);
1007 }
1008 
1009 static void blk_unregister_tracepoints(void)
1010 {
1011 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1012 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1013 	unregister_trace_block_split(blk_add_trace_split, NULL);
1014 	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1015 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1016 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1017 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1018 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1019 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1020 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1021 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1022 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1023 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1024 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1025 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1026 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1027 	unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1028 
1029 	tracepoint_synchronize_unregister();
1030 }
1031 
1032 /*
1033  * struct blk_io_tracer formatting routines
1034  */
1035 
1036 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1037 {
1038 	int i = 0;
1039 	int tc = t->action >> BLK_TC_SHIFT;
1040 
1041 	if (t->action == BLK_TN_MESSAGE) {
1042 		rwbs[i++] = 'N';
1043 		goto out;
1044 	}
1045 
1046 	if (tc & BLK_TC_FLUSH)
1047 		rwbs[i++] = 'F';
1048 
1049 	if (tc & BLK_TC_DISCARD)
1050 		rwbs[i++] = 'D';
1051 	else if (tc & BLK_TC_WRITE)
1052 		rwbs[i++] = 'W';
1053 	else if (t->bytes)
1054 		rwbs[i++] = 'R';
1055 	else
1056 		rwbs[i++] = 'N';
1057 
1058 	if (tc & BLK_TC_FUA)
1059 		rwbs[i++] = 'F';
1060 	if (tc & BLK_TC_AHEAD)
1061 		rwbs[i++] = 'A';
1062 	if (tc & BLK_TC_SYNC)
1063 		rwbs[i++] = 'S';
1064 	if (tc & BLK_TC_META)
1065 		rwbs[i++] = 'M';
1066 out:
1067 	rwbs[i] = '\0';
1068 }
1069 
1070 static inline
1071 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1072 {
1073 	return (const struct blk_io_trace *)ent;
1074 }
1075 
1076 static inline const void *pdu_start(const struct trace_entry *ent)
1077 {
1078 	return te_blk_io_trace(ent) + 1;
1079 }
1080 
1081 static inline u32 t_action(const struct trace_entry *ent)
1082 {
1083 	return te_blk_io_trace(ent)->action;
1084 }
1085 
1086 static inline u32 t_bytes(const struct trace_entry *ent)
1087 {
1088 	return te_blk_io_trace(ent)->bytes;
1089 }
1090 
1091 static inline u32 t_sec(const struct trace_entry *ent)
1092 {
1093 	return te_blk_io_trace(ent)->bytes >> 9;
1094 }
1095 
1096 static inline unsigned long long t_sector(const struct trace_entry *ent)
1097 {
1098 	return te_blk_io_trace(ent)->sector;
1099 }
1100 
1101 static inline __u16 t_error(const struct trace_entry *ent)
1102 {
1103 	return te_blk_io_trace(ent)->error;
1104 }
1105 
1106 static __u64 get_pdu_int(const struct trace_entry *ent)
1107 {
1108 	const __u64 *val = pdu_start(ent);
1109 	return be64_to_cpu(*val);
1110 }
1111 
1112 static void get_pdu_remap(const struct trace_entry *ent,
1113 			  struct blk_io_trace_remap *r)
1114 {
1115 	const struct blk_io_trace_remap *__r = pdu_start(ent);
1116 	__u64 sector_from = __r->sector_from;
1117 
1118 	r->device_from = be32_to_cpu(__r->device_from);
1119 	r->device_to   = be32_to_cpu(__r->device_to);
1120 	r->sector_from = be64_to_cpu(sector_from);
1121 }
1122 
1123 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1124 
1125 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1126 {
1127 	char rwbs[RWBS_LEN];
1128 	unsigned long long ts  = iter->ts;
1129 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1130 	unsigned secs	       = (unsigned long)ts;
1131 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1132 
1133 	fill_rwbs(rwbs, t);
1134 
1135 	return trace_seq_printf(&iter->seq,
1136 				"%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1137 				MAJOR(t->device), MINOR(t->device), iter->cpu,
1138 				secs, nsec_rem, iter->ent->pid, act, rwbs);
1139 }
1140 
1141 static int blk_log_action(struct trace_iterator *iter, const char *act)
1142 {
1143 	char rwbs[RWBS_LEN];
1144 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1145 
1146 	fill_rwbs(rwbs, t);
1147 	return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1148 				MAJOR(t->device), MINOR(t->device), act, rwbs);
1149 }
1150 
1151 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1152 {
1153 	const unsigned char *pdu_buf;
1154 	int pdu_len;
1155 	int i, end, ret;
1156 
1157 	pdu_buf = pdu_start(ent);
1158 	pdu_len = te_blk_io_trace(ent)->pdu_len;
1159 
1160 	if (!pdu_len)
1161 		return 1;
1162 
1163 	/* find the last zero that needs to be printed */
1164 	for (end = pdu_len - 1; end >= 0; end--)
1165 		if (pdu_buf[end])
1166 			break;
1167 	end++;
1168 
1169 	if (!trace_seq_putc(s, '('))
1170 		return 0;
1171 
1172 	for (i = 0; i < pdu_len; i++) {
1173 
1174 		ret = trace_seq_printf(s, "%s%02x",
1175 				       i == 0 ? "" : " ", pdu_buf[i]);
1176 		if (!ret)
1177 			return ret;
1178 
1179 		/*
1180 		 * stop when the rest is just zeroes and indicate so
1181 		 * with a ".." appended
1182 		 */
1183 		if (i == end && end != pdu_len - 1)
1184 			return trace_seq_puts(s, " ..) ");
1185 	}
1186 
1187 	return trace_seq_puts(s, ") ");
1188 }
1189 
1190 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1191 {
1192 	char cmd[TASK_COMM_LEN];
1193 
1194 	trace_find_cmdline(ent->pid, cmd);
1195 
1196 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1197 		int ret;
1198 
1199 		ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1200 		if (!ret)
1201 			return 0;
1202 		ret = blk_log_dump_pdu(s, ent);
1203 		if (!ret)
1204 			return 0;
1205 		return trace_seq_printf(s, "[%s]\n", cmd);
1206 	} else {
1207 		if (t_sec(ent))
1208 			return trace_seq_printf(s, "%llu + %u [%s]\n",
1209 						t_sector(ent), t_sec(ent), cmd);
1210 		return trace_seq_printf(s, "[%s]\n", cmd);
1211 	}
1212 }
1213 
1214 static int blk_log_with_error(struct trace_seq *s,
1215 			      const struct trace_entry *ent)
1216 {
1217 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1218 		int ret;
1219 
1220 		ret = blk_log_dump_pdu(s, ent);
1221 		if (ret)
1222 			return trace_seq_printf(s, "[%d]\n", t_error(ent));
1223 		return 0;
1224 	} else {
1225 		if (t_sec(ent))
1226 			return trace_seq_printf(s, "%llu + %u [%d]\n",
1227 						t_sector(ent),
1228 						t_sec(ent), t_error(ent));
1229 		return trace_seq_printf(s, "%llu [%d]\n",
1230 					t_sector(ent), t_error(ent));
1231 	}
1232 }
1233 
1234 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1235 {
1236 	struct blk_io_trace_remap r = { .device_from = 0, };
1237 
1238 	get_pdu_remap(ent, &r);
1239 	return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1240 				t_sector(ent), t_sec(ent),
1241 				MAJOR(r.device_from), MINOR(r.device_from),
1242 				(unsigned long long)r.sector_from);
1243 }
1244 
1245 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1246 {
1247 	char cmd[TASK_COMM_LEN];
1248 
1249 	trace_find_cmdline(ent->pid, cmd);
1250 
1251 	return trace_seq_printf(s, "[%s]\n", cmd);
1252 }
1253 
1254 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1255 {
1256 	char cmd[TASK_COMM_LEN];
1257 
1258 	trace_find_cmdline(ent->pid, cmd);
1259 
1260 	return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1261 }
1262 
1263 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1264 {
1265 	char cmd[TASK_COMM_LEN];
1266 
1267 	trace_find_cmdline(ent->pid, cmd);
1268 
1269 	return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1270 				get_pdu_int(ent), cmd);
1271 }
1272 
1273 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1274 {
1275 	int ret;
1276 	const struct blk_io_trace *t = te_blk_io_trace(ent);
1277 
1278 	ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1279 	if (ret)
1280 		return trace_seq_putc(s, '\n');
1281 	return ret;
1282 }
1283 
1284 /*
1285  * struct tracer operations
1286  */
1287 
1288 static void blk_tracer_print_header(struct seq_file *m)
1289 {
1290 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1291 		return;
1292 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1293 		    "#  |     |     |           |   |   |\n");
1294 }
1295 
1296 static void blk_tracer_start(struct trace_array *tr)
1297 {
1298 	blk_tracer_enabled = true;
1299 }
1300 
1301 static int blk_tracer_init(struct trace_array *tr)
1302 {
1303 	blk_tr = tr;
1304 	blk_tracer_start(tr);
1305 	return 0;
1306 }
1307 
1308 static void blk_tracer_stop(struct trace_array *tr)
1309 {
1310 	blk_tracer_enabled = false;
1311 }
1312 
1313 static void blk_tracer_reset(struct trace_array *tr)
1314 {
1315 	blk_tracer_stop(tr);
1316 }
1317 
1318 static const struct {
1319 	const char *act[2];
1320 	int	   (*print)(struct trace_seq *s, const struct trace_entry *ent);
1321 } what2act[] = {
1322 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1323 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1324 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1325 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1326 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1327 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1328 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1329 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1330 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1331 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1332 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1333 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1334 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1335 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1336 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1337 };
1338 
1339 static enum print_line_t print_one_line(struct trace_iterator *iter,
1340 					bool classic)
1341 {
1342 	struct trace_seq *s = &iter->seq;
1343 	const struct blk_io_trace *t;
1344 	u16 what;
1345 	int ret;
1346 	bool long_act;
1347 	blk_log_action_t *log_action;
1348 
1349 	t	   = te_blk_io_trace(iter->ent);
1350 	what	   = t->action & ((1 << BLK_TC_SHIFT) - 1);
1351 	long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1352 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1353 
1354 	if (t->action == BLK_TN_MESSAGE) {
1355 		ret = log_action(iter, long_act ? "message" : "m");
1356 		if (ret)
1357 			ret = blk_log_msg(s, iter->ent);
1358 		goto out;
1359 	}
1360 
1361 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1362 		ret = trace_seq_printf(s, "Unknown action %x\n", what);
1363 	else {
1364 		ret = log_action(iter, what2act[what].act[long_act]);
1365 		if (ret)
1366 			ret = what2act[what].print(s, iter->ent);
1367 	}
1368 out:
1369 	return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1370 }
1371 
1372 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1373 					       int flags, struct trace_event *event)
1374 {
1375 	return print_one_line(iter, false);
1376 }
1377 
1378 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1379 {
1380 	struct trace_seq *s = &iter->seq;
1381 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1382 	const int offset = offsetof(struct blk_io_trace, sector);
1383 	struct blk_io_trace old = {
1384 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1385 		.time     = iter->ts,
1386 	};
1387 
1388 	if (!trace_seq_putmem(s, &old, offset))
1389 		return 0;
1390 	return trace_seq_putmem(s, &t->sector,
1391 				sizeof(old) - offset + t->pdu_len);
1392 }
1393 
1394 static enum print_line_t
1395 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1396 			     struct trace_event *event)
1397 {
1398 	return blk_trace_synthesize_old_trace(iter) ?
1399 			TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1400 }
1401 
1402 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1403 {
1404 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1405 		return TRACE_TYPE_UNHANDLED;
1406 
1407 	return print_one_line(iter, true);
1408 }
1409 
1410 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1411 {
1412 	/* don't output context-info for blk_classic output */
1413 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1414 		if (set)
1415 			trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1416 		else
1417 			trace_flags |= TRACE_ITER_CONTEXT_INFO;
1418 	}
1419 	return 0;
1420 }
1421 
1422 static struct tracer blk_tracer __read_mostly = {
1423 	.name		= "blk",
1424 	.init		= blk_tracer_init,
1425 	.reset		= blk_tracer_reset,
1426 	.start		= blk_tracer_start,
1427 	.stop		= blk_tracer_stop,
1428 	.print_header	= blk_tracer_print_header,
1429 	.print_line	= blk_tracer_print_line,
1430 	.flags		= &blk_tracer_flags,
1431 	.set_flag	= blk_tracer_set_flag,
1432 };
1433 
1434 static struct trace_event_functions trace_blk_event_funcs = {
1435 	.trace		= blk_trace_event_print,
1436 	.binary		= blk_trace_event_print_binary,
1437 };
1438 
1439 static struct trace_event trace_blk_event = {
1440 	.type		= TRACE_BLK,
1441 	.funcs		= &trace_blk_event_funcs,
1442 };
1443 
1444 static int __init init_blk_tracer(void)
1445 {
1446 	if (!register_ftrace_event(&trace_blk_event)) {
1447 		pr_warning("Warning: could not register block events\n");
1448 		return 1;
1449 	}
1450 
1451 	if (register_tracer(&blk_tracer) != 0) {
1452 		pr_warning("Warning: could not register the block tracer\n");
1453 		unregister_ftrace_event(&trace_blk_event);
1454 		return 1;
1455 	}
1456 
1457 	return 0;
1458 }
1459 
1460 device_initcall(init_blk_tracer);
1461 
1462 static int blk_trace_remove_queue(struct request_queue *q)
1463 {
1464 	struct blk_trace *bt;
1465 
1466 	bt = xchg(&q->blk_trace, NULL);
1467 	if (bt == NULL)
1468 		return -EINVAL;
1469 
1470 	if (atomic_dec_and_test(&blk_probes_ref))
1471 		blk_unregister_tracepoints();
1472 
1473 	blk_trace_free(bt);
1474 	return 0;
1475 }
1476 
1477 /*
1478  * Setup everything required to start tracing
1479  */
1480 static int blk_trace_setup_queue(struct request_queue *q,
1481 				 struct block_device *bdev)
1482 {
1483 	struct blk_trace *old_bt, *bt = NULL;
1484 	int ret = -ENOMEM;
1485 
1486 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1487 	if (!bt)
1488 		return -ENOMEM;
1489 
1490 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1491 	if (!bt->msg_data)
1492 		goto free_bt;
1493 
1494 	bt->dev = bdev->bd_dev;
1495 	bt->act_mask = (u16)-1;
1496 
1497 	blk_trace_setup_lba(bt, bdev);
1498 
1499 	old_bt = xchg(&q->blk_trace, bt);
1500 	if (old_bt != NULL) {
1501 		(void)xchg(&q->blk_trace, old_bt);
1502 		ret = -EBUSY;
1503 		goto free_bt;
1504 	}
1505 
1506 	if (atomic_inc_return(&blk_probes_ref) == 1)
1507 		blk_register_tracepoints();
1508 	return 0;
1509 
1510 free_bt:
1511 	blk_trace_free(bt);
1512 	return ret;
1513 }
1514 
1515 /*
1516  * sysfs interface to enable and configure tracing
1517  */
1518 
1519 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1520 					 struct device_attribute *attr,
1521 					 char *buf);
1522 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1523 					  struct device_attribute *attr,
1524 					  const char *buf, size_t count);
1525 #define BLK_TRACE_DEVICE_ATTR(_name) \
1526 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1527 		    sysfs_blk_trace_attr_show, \
1528 		    sysfs_blk_trace_attr_store)
1529 
1530 static BLK_TRACE_DEVICE_ATTR(enable);
1531 static BLK_TRACE_DEVICE_ATTR(act_mask);
1532 static BLK_TRACE_DEVICE_ATTR(pid);
1533 static BLK_TRACE_DEVICE_ATTR(start_lba);
1534 static BLK_TRACE_DEVICE_ATTR(end_lba);
1535 
1536 static struct attribute *blk_trace_attrs[] = {
1537 	&dev_attr_enable.attr,
1538 	&dev_attr_act_mask.attr,
1539 	&dev_attr_pid.attr,
1540 	&dev_attr_start_lba.attr,
1541 	&dev_attr_end_lba.attr,
1542 	NULL
1543 };
1544 
1545 struct attribute_group blk_trace_attr_group = {
1546 	.name  = "trace",
1547 	.attrs = blk_trace_attrs,
1548 };
1549 
1550 static const struct {
1551 	int mask;
1552 	const char *str;
1553 } mask_maps[] = {
1554 	{ BLK_TC_READ,		"read"		},
1555 	{ BLK_TC_WRITE,		"write"		},
1556 	{ BLK_TC_FLUSH,		"flush"		},
1557 	{ BLK_TC_SYNC,		"sync"		},
1558 	{ BLK_TC_QUEUE,		"queue"		},
1559 	{ BLK_TC_REQUEUE,	"requeue"	},
1560 	{ BLK_TC_ISSUE,		"issue"		},
1561 	{ BLK_TC_COMPLETE,	"complete"	},
1562 	{ BLK_TC_FS,		"fs"		},
1563 	{ BLK_TC_PC,		"pc"		},
1564 	{ BLK_TC_AHEAD,		"ahead"		},
1565 	{ BLK_TC_META,		"meta"		},
1566 	{ BLK_TC_DISCARD,	"discard"	},
1567 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1568 	{ BLK_TC_FUA,		"fua"		},
1569 };
1570 
1571 static int blk_trace_str2mask(const char *str)
1572 {
1573 	int i;
1574 	int mask = 0;
1575 	char *buf, *s, *token;
1576 
1577 	buf = kstrdup(str, GFP_KERNEL);
1578 	if (buf == NULL)
1579 		return -ENOMEM;
1580 	s = strstrip(buf);
1581 
1582 	while (1) {
1583 		token = strsep(&s, ",");
1584 		if (token == NULL)
1585 			break;
1586 
1587 		if (*token == '\0')
1588 			continue;
1589 
1590 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1591 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1592 				mask |= mask_maps[i].mask;
1593 				break;
1594 			}
1595 		}
1596 		if (i == ARRAY_SIZE(mask_maps)) {
1597 			mask = -EINVAL;
1598 			break;
1599 		}
1600 	}
1601 	kfree(buf);
1602 
1603 	return mask;
1604 }
1605 
1606 static ssize_t blk_trace_mask2str(char *buf, int mask)
1607 {
1608 	int i;
1609 	char *p = buf;
1610 
1611 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1612 		if (mask & mask_maps[i].mask) {
1613 			p += sprintf(p, "%s%s",
1614 				    (p == buf) ? "" : ",", mask_maps[i].str);
1615 		}
1616 	}
1617 	*p++ = '\n';
1618 
1619 	return p - buf;
1620 }
1621 
1622 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1623 {
1624 	if (bdev->bd_disk == NULL)
1625 		return NULL;
1626 
1627 	return bdev_get_queue(bdev);
1628 }
1629 
1630 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1631 					 struct device_attribute *attr,
1632 					 char *buf)
1633 {
1634 	struct hd_struct *p = dev_to_part(dev);
1635 	struct request_queue *q;
1636 	struct block_device *bdev;
1637 	ssize_t ret = -ENXIO;
1638 
1639 	bdev = bdget(part_devt(p));
1640 	if (bdev == NULL)
1641 		goto out;
1642 
1643 	q = blk_trace_get_queue(bdev);
1644 	if (q == NULL)
1645 		goto out_bdput;
1646 
1647 	mutex_lock(&bdev->bd_mutex);
1648 
1649 	if (attr == &dev_attr_enable) {
1650 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1651 		goto out_unlock_bdev;
1652 	}
1653 
1654 	if (q->blk_trace == NULL)
1655 		ret = sprintf(buf, "disabled\n");
1656 	else if (attr == &dev_attr_act_mask)
1657 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1658 	else if (attr == &dev_attr_pid)
1659 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1660 	else if (attr == &dev_attr_start_lba)
1661 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1662 	else if (attr == &dev_attr_end_lba)
1663 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1664 
1665 out_unlock_bdev:
1666 	mutex_unlock(&bdev->bd_mutex);
1667 out_bdput:
1668 	bdput(bdev);
1669 out:
1670 	return ret;
1671 }
1672 
1673 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1674 					  struct device_attribute *attr,
1675 					  const char *buf, size_t count)
1676 {
1677 	struct block_device *bdev;
1678 	struct request_queue *q;
1679 	struct hd_struct *p;
1680 	u64 value;
1681 	ssize_t ret = -EINVAL;
1682 
1683 	if (count == 0)
1684 		goto out;
1685 
1686 	if (attr == &dev_attr_act_mask) {
1687 		if (sscanf(buf, "%llx", &value) != 1) {
1688 			/* Assume it is a list of trace category names */
1689 			ret = blk_trace_str2mask(buf);
1690 			if (ret < 0)
1691 				goto out;
1692 			value = ret;
1693 		}
1694 	} else if (sscanf(buf, "%llu", &value) != 1)
1695 		goto out;
1696 
1697 	ret = -ENXIO;
1698 
1699 	p = dev_to_part(dev);
1700 	bdev = bdget(part_devt(p));
1701 	if (bdev == NULL)
1702 		goto out;
1703 
1704 	q = blk_trace_get_queue(bdev);
1705 	if (q == NULL)
1706 		goto out_bdput;
1707 
1708 	mutex_lock(&bdev->bd_mutex);
1709 
1710 	if (attr == &dev_attr_enable) {
1711 		if (value)
1712 			ret = blk_trace_setup_queue(q, bdev);
1713 		else
1714 			ret = blk_trace_remove_queue(q);
1715 		goto out_unlock_bdev;
1716 	}
1717 
1718 	ret = 0;
1719 	if (q->blk_trace == NULL)
1720 		ret = blk_trace_setup_queue(q, bdev);
1721 
1722 	if (ret == 0) {
1723 		if (attr == &dev_attr_act_mask)
1724 			q->blk_trace->act_mask = value;
1725 		else if (attr == &dev_attr_pid)
1726 			q->blk_trace->pid = value;
1727 		else if (attr == &dev_attr_start_lba)
1728 			q->blk_trace->start_lba = value;
1729 		else if (attr == &dev_attr_end_lba)
1730 			q->blk_trace->end_lba = value;
1731 	}
1732 
1733 out_unlock_bdev:
1734 	mutex_unlock(&bdev->bd_mutex);
1735 out_bdput:
1736 	bdput(bdev);
1737 out:
1738 	return ret ? ret : count;
1739 }
1740 
1741 int blk_trace_init_sysfs(struct device *dev)
1742 {
1743 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1744 }
1745 
1746 void blk_trace_remove_sysfs(struct device *dev)
1747 {
1748 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1749 }
1750 
1751 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1752 
1753 #ifdef CONFIG_EVENT_TRACING
1754 
1755 void blk_dump_cmd(char *buf, struct request *rq)
1756 {
1757 	int i, end;
1758 	int len = rq->cmd_len;
1759 	unsigned char *cmd = rq->cmd;
1760 
1761 	if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1762 		buf[0] = '\0';
1763 		return;
1764 	}
1765 
1766 	for (end = len - 1; end >= 0; end--)
1767 		if (cmd[end])
1768 			break;
1769 	end++;
1770 
1771 	for (i = 0; i < len; i++) {
1772 		buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1773 		if (i == end && end != len - 1) {
1774 			sprintf(buf, " ..");
1775 			break;
1776 		}
1777 	}
1778 }
1779 
1780 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1781 {
1782 	int i = 0;
1783 
1784 	if (rw & REQ_FLUSH)
1785 		rwbs[i++] = 'F';
1786 
1787 	if (rw & WRITE)
1788 		rwbs[i++] = 'W';
1789 	else if (rw & REQ_DISCARD)
1790 		rwbs[i++] = 'D';
1791 	else if (bytes)
1792 		rwbs[i++] = 'R';
1793 	else
1794 		rwbs[i++] = 'N';
1795 
1796 	if (rw & REQ_FUA)
1797 		rwbs[i++] = 'F';
1798 	if (rw & REQ_RAHEAD)
1799 		rwbs[i++] = 'A';
1800 	if (rw & REQ_SYNC)
1801 		rwbs[i++] = 'S';
1802 	if (rw & REQ_META)
1803 		rwbs[i++] = 'M';
1804 	if (rw & REQ_SECURE)
1805 		rwbs[i++] = 'E';
1806 
1807 	rwbs[i] = '\0';
1808 }
1809 
1810 #endif /* CONFIG_EVENT_TRACING */
1811 
1812