xref: /openbmc/linux/kernel/trace/blktrace.c (revision 9d56dd3b083a3bec56e9da35ce07baca81030b03)
1 /*
2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
16  *
17  */
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/debugfs.h>
25 #include <linux/smp_lock.h>
26 #include <linux/time.h>
27 #include <linux/uaccess.h>
28 
29 #include <trace/events/block.h>
30 
31 #include "trace_output.h"
32 
33 #ifdef CONFIG_BLK_DEV_IO_TRACE
34 
35 static unsigned int blktrace_seq __read_mostly = 1;
36 
37 static struct trace_array *blk_tr;
38 static bool blk_tracer_enabled __read_mostly;
39 
40 /* Select an alternative, minimalistic output than the original one */
41 #define TRACE_BLK_OPT_CLASSIC	0x1
42 
43 static struct tracer_opt blk_tracer_opts[] = {
44 	/* Default disable the minimalistic output */
45 	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
46 	{ }
47 };
48 
49 static struct tracer_flags blk_tracer_flags = {
50 	.val  = 0,
51 	.opts = blk_tracer_opts,
52 };
53 
54 /* Global reference count of probes */
55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
56 
57 static void blk_register_tracepoints(void);
58 static void blk_unregister_tracepoints(void);
59 
60 /*
61  * Send out a notify message.
62  */
63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
64 		       const void *data, size_t len)
65 {
66 	struct blk_io_trace *t;
67 	struct ring_buffer_event *event = NULL;
68 	struct ring_buffer *buffer = NULL;
69 	int pc = 0;
70 	int cpu = smp_processor_id();
71 	bool blk_tracer = blk_tracer_enabled;
72 
73 	if (blk_tracer) {
74 		buffer = blk_tr->buffer;
75 		pc = preempt_count();
76 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
77 						  sizeof(*t) + len,
78 						  0, pc);
79 		if (!event)
80 			return;
81 		t = ring_buffer_event_data(event);
82 		goto record_it;
83 	}
84 
85 	if (!bt->rchan)
86 		return;
87 
88 	t = relay_reserve(bt->rchan, sizeof(*t) + len);
89 	if (t) {
90 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
91 		t->time = ktime_to_ns(ktime_get());
92 record_it:
93 		t->device = bt->dev;
94 		t->action = action;
95 		t->pid = pid;
96 		t->cpu = cpu;
97 		t->pdu_len = len;
98 		memcpy((void *) t + sizeof(*t), data, len);
99 
100 		if (blk_tracer)
101 			trace_buffer_unlock_commit(buffer, event, 0, pc);
102 	}
103 }
104 
105 /*
106  * Send out a notify for this process, if we haven't done so since a trace
107  * started
108  */
109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
110 {
111 	tsk->btrace_seq = blktrace_seq;
112 	trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
113 }
114 
115 static void trace_note_time(struct blk_trace *bt)
116 {
117 	struct timespec now;
118 	unsigned long flags;
119 	u32 words[2];
120 
121 	getnstimeofday(&now);
122 	words[0] = now.tv_sec;
123 	words[1] = now.tv_nsec;
124 
125 	local_irq_save(flags);
126 	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
127 	local_irq_restore(flags);
128 }
129 
130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
131 {
132 	int n;
133 	va_list args;
134 	unsigned long flags;
135 	char *buf;
136 
137 	if (unlikely(bt->trace_state != Blktrace_running &&
138 		     !blk_tracer_enabled))
139 		return;
140 
141 	local_irq_save(flags);
142 	buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
143 	va_start(args, fmt);
144 	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
145 	va_end(args);
146 
147 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
148 	local_irq_restore(flags);
149 }
150 EXPORT_SYMBOL_GPL(__trace_note_message);
151 
152 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
153 			 pid_t pid)
154 {
155 	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
156 		return 1;
157 	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
158 		return 1;
159 	if (bt->pid && pid != bt->pid)
160 		return 1;
161 
162 	return 0;
163 }
164 
165 /*
166  * Data direction bit lookup
167  */
168 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
169 				 BLK_TC_ACT(BLK_TC_WRITE) };
170 
171 /* The ilog2() calls fall out because they're constant */
172 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
173 	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
174 
175 /*
176  * The worker for the various blk_add_trace*() types. Fills out a
177  * blk_io_trace structure and places it in a per-cpu subbuffer.
178  */
179 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
180 		     int rw, u32 what, int error, int pdu_len, void *pdu_data)
181 {
182 	struct task_struct *tsk = current;
183 	struct ring_buffer_event *event = NULL;
184 	struct ring_buffer *buffer = NULL;
185 	struct blk_io_trace *t;
186 	unsigned long flags = 0;
187 	unsigned long *sequence;
188 	pid_t pid;
189 	int cpu, pc = 0;
190 	bool blk_tracer = blk_tracer_enabled;
191 
192 	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
193 		return;
194 
195 	what |= ddir_act[rw & WRITE];
196 	what |= MASK_TC_BIT(rw, BARRIER);
197 	what |= MASK_TC_BIT(rw, SYNCIO);
198 	what |= MASK_TC_BIT(rw, AHEAD);
199 	what |= MASK_TC_BIT(rw, META);
200 	what |= MASK_TC_BIT(rw, DISCARD);
201 
202 	pid = tsk->pid;
203 	if (act_log_check(bt, what, sector, pid))
204 		return;
205 	cpu = raw_smp_processor_id();
206 
207 	if (blk_tracer) {
208 		tracing_record_cmdline(current);
209 
210 		buffer = blk_tr->buffer;
211 		pc = preempt_count();
212 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
213 						  sizeof(*t) + pdu_len,
214 						  0, pc);
215 		if (!event)
216 			return;
217 		t = ring_buffer_event_data(event);
218 		goto record_it;
219 	}
220 
221 	/*
222 	 * A word about the locking here - we disable interrupts to reserve
223 	 * some space in the relay per-cpu buffer, to prevent an irq
224 	 * from coming in and stepping on our toes.
225 	 */
226 	local_irq_save(flags);
227 
228 	if (unlikely(tsk->btrace_seq != blktrace_seq))
229 		trace_note_tsk(bt, tsk);
230 
231 	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
232 	if (t) {
233 		sequence = per_cpu_ptr(bt->sequence, cpu);
234 
235 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
236 		t->sequence = ++(*sequence);
237 		t->time = ktime_to_ns(ktime_get());
238 record_it:
239 		/*
240 		 * These two are not needed in ftrace as they are in the
241 		 * generic trace_entry, filled by tracing_generic_entry_update,
242 		 * but for the trace_event->bin() synthesizer benefit we do it
243 		 * here too.
244 		 */
245 		t->cpu = cpu;
246 		t->pid = pid;
247 
248 		t->sector = sector;
249 		t->bytes = bytes;
250 		t->action = what;
251 		t->device = bt->dev;
252 		t->error = error;
253 		t->pdu_len = pdu_len;
254 
255 		if (pdu_len)
256 			memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
257 
258 		if (blk_tracer) {
259 			trace_buffer_unlock_commit(buffer, event, 0, pc);
260 			return;
261 		}
262 	}
263 
264 	local_irq_restore(flags);
265 }
266 
267 static struct dentry *blk_tree_root;
268 static DEFINE_MUTEX(blk_tree_mutex);
269 
270 static void blk_trace_free(struct blk_trace *bt)
271 {
272 	debugfs_remove(bt->msg_file);
273 	debugfs_remove(bt->dropped_file);
274 	relay_close(bt->rchan);
275 	debugfs_remove(bt->dir);
276 	free_percpu(bt->sequence);
277 	free_percpu(bt->msg_data);
278 	kfree(bt);
279 }
280 
281 static void blk_trace_cleanup(struct blk_trace *bt)
282 {
283 	blk_trace_free(bt);
284 	if (atomic_dec_and_test(&blk_probes_ref))
285 		blk_unregister_tracepoints();
286 }
287 
288 int blk_trace_remove(struct request_queue *q)
289 {
290 	struct blk_trace *bt;
291 
292 	bt = xchg(&q->blk_trace, NULL);
293 	if (!bt)
294 		return -EINVAL;
295 
296 	if (bt->trace_state != Blktrace_running)
297 		blk_trace_cleanup(bt);
298 
299 	return 0;
300 }
301 EXPORT_SYMBOL_GPL(blk_trace_remove);
302 
303 static int blk_dropped_open(struct inode *inode, struct file *filp)
304 {
305 	filp->private_data = inode->i_private;
306 
307 	return 0;
308 }
309 
310 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
311 				size_t count, loff_t *ppos)
312 {
313 	struct blk_trace *bt = filp->private_data;
314 	char buf[16];
315 
316 	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
317 
318 	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
319 }
320 
321 static const struct file_operations blk_dropped_fops = {
322 	.owner =	THIS_MODULE,
323 	.open =		blk_dropped_open,
324 	.read =		blk_dropped_read,
325 };
326 
327 static int blk_msg_open(struct inode *inode, struct file *filp)
328 {
329 	filp->private_data = inode->i_private;
330 
331 	return 0;
332 }
333 
334 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
335 				size_t count, loff_t *ppos)
336 {
337 	char *msg;
338 	struct blk_trace *bt;
339 
340 	if (count >= BLK_TN_MAX_MSG)
341 		return -EINVAL;
342 
343 	msg = kmalloc(count + 1, GFP_KERNEL);
344 	if (msg == NULL)
345 		return -ENOMEM;
346 
347 	if (copy_from_user(msg, buffer, count)) {
348 		kfree(msg);
349 		return -EFAULT;
350 	}
351 
352 	msg[count] = '\0';
353 	bt = filp->private_data;
354 	__trace_note_message(bt, "%s", msg);
355 	kfree(msg);
356 
357 	return count;
358 }
359 
360 static const struct file_operations blk_msg_fops = {
361 	.owner =	THIS_MODULE,
362 	.open =		blk_msg_open,
363 	.write =	blk_msg_write,
364 };
365 
366 /*
367  * Keep track of how many times we encountered a full subbuffer, to aid
368  * the user space app in telling how many lost events there were.
369  */
370 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
371 				     void *prev_subbuf, size_t prev_padding)
372 {
373 	struct blk_trace *bt;
374 
375 	if (!relay_buf_full(buf))
376 		return 1;
377 
378 	bt = buf->chan->private_data;
379 	atomic_inc(&bt->dropped);
380 	return 0;
381 }
382 
383 static int blk_remove_buf_file_callback(struct dentry *dentry)
384 {
385 	debugfs_remove(dentry);
386 
387 	return 0;
388 }
389 
390 static struct dentry *blk_create_buf_file_callback(const char *filename,
391 						   struct dentry *parent,
392 						   int mode,
393 						   struct rchan_buf *buf,
394 						   int *is_global)
395 {
396 	return debugfs_create_file(filename, mode, parent, buf,
397 					&relay_file_operations);
398 }
399 
400 static struct rchan_callbacks blk_relay_callbacks = {
401 	.subbuf_start		= blk_subbuf_start_callback,
402 	.create_buf_file	= blk_create_buf_file_callback,
403 	.remove_buf_file	= blk_remove_buf_file_callback,
404 };
405 
406 static void blk_trace_setup_lba(struct blk_trace *bt,
407 				struct block_device *bdev)
408 {
409 	struct hd_struct *part = NULL;
410 
411 	if (bdev)
412 		part = bdev->bd_part;
413 
414 	if (part) {
415 		bt->start_lba = part->start_sect;
416 		bt->end_lba = part->start_sect + part->nr_sects;
417 	} else {
418 		bt->start_lba = 0;
419 		bt->end_lba = -1ULL;
420 	}
421 }
422 
423 /*
424  * Setup everything required to start tracing
425  */
426 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
427 		       struct block_device *bdev,
428 		       struct blk_user_trace_setup *buts)
429 {
430 	struct blk_trace *old_bt, *bt = NULL;
431 	struct dentry *dir = NULL;
432 	int ret, i;
433 
434 	if (!buts->buf_size || !buts->buf_nr)
435 		return -EINVAL;
436 
437 	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
438 	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
439 
440 	/*
441 	 * some device names have larger paths - convert the slashes
442 	 * to underscores for this to work as expected
443 	 */
444 	for (i = 0; i < strlen(buts->name); i++)
445 		if (buts->name[i] == '/')
446 			buts->name[i] = '_';
447 
448 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
449 	if (!bt)
450 		return -ENOMEM;
451 
452 	ret = -ENOMEM;
453 	bt->sequence = alloc_percpu(unsigned long);
454 	if (!bt->sequence)
455 		goto err;
456 
457 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
458 	if (!bt->msg_data)
459 		goto err;
460 
461 	ret = -ENOENT;
462 
463 	mutex_lock(&blk_tree_mutex);
464 	if (!blk_tree_root) {
465 		blk_tree_root = debugfs_create_dir("block", NULL);
466 		if (!blk_tree_root) {
467 			mutex_unlock(&blk_tree_mutex);
468 			goto err;
469 		}
470 	}
471 	mutex_unlock(&blk_tree_mutex);
472 
473 	dir = debugfs_create_dir(buts->name, blk_tree_root);
474 
475 	if (!dir)
476 		goto err;
477 
478 	bt->dir = dir;
479 	bt->dev = dev;
480 	atomic_set(&bt->dropped, 0);
481 
482 	ret = -EIO;
483 	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
484 					       &blk_dropped_fops);
485 	if (!bt->dropped_file)
486 		goto err;
487 
488 	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
489 	if (!bt->msg_file)
490 		goto err;
491 
492 	bt->rchan = relay_open("trace", dir, buts->buf_size,
493 				buts->buf_nr, &blk_relay_callbacks, bt);
494 	if (!bt->rchan)
495 		goto err;
496 
497 	bt->act_mask = buts->act_mask;
498 	if (!bt->act_mask)
499 		bt->act_mask = (u16) -1;
500 
501 	blk_trace_setup_lba(bt, bdev);
502 
503 	/* overwrite with user settings */
504 	if (buts->start_lba)
505 		bt->start_lba = buts->start_lba;
506 	if (buts->end_lba)
507 		bt->end_lba = buts->end_lba;
508 
509 	bt->pid = buts->pid;
510 	bt->trace_state = Blktrace_setup;
511 
512 	ret = -EBUSY;
513 	old_bt = xchg(&q->blk_trace, bt);
514 	if (old_bt) {
515 		(void) xchg(&q->blk_trace, old_bt);
516 		goto err;
517 	}
518 
519 	if (atomic_inc_return(&blk_probes_ref) == 1)
520 		blk_register_tracepoints();
521 
522 	return 0;
523 err:
524 	blk_trace_free(bt);
525 	return ret;
526 }
527 
528 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
529 		    struct block_device *bdev,
530 		    char __user *arg)
531 {
532 	struct blk_user_trace_setup buts;
533 	int ret;
534 
535 	ret = copy_from_user(&buts, arg, sizeof(buts));
536 	if (ret)
537 		return -EFAULT;
538 
539 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
540 	if (ret)
541 		return ret;
542 
543 	if (copy_to_user(arg, &buts, sizeof(buts)))
544 		return -EFAULT;
545 
546 	return 0;
547 }
548 EXPORT_SYMBOL_GPL(blk_trace_setup);
549 
550 int blk_trace_startstop(struct request_queue *q, int start)
551 {
552 	int ret;
553 	struct blk_trace *bt = q->blk_trace;
554 
555 	if (bt == NULL)
556 		return -EINVAL;
557 
558 	/*
559 	 * For starting a trace, we can transition from a setup or stopped
560 	 * trace. For stopping a trace, the state must be running
561 	 */
562 	ret = -EINVAL;
563 	if (start) {
564 		if (bt->trace_state == Blktrace_setup ||
565 		    bt->trace_state == Blktrace_stopped) {
566 			blktrace_seq++;
567 			smp_mb();
568 			bt->trace_state = Blktrace_running;
569 
570 			trace_note_time(bt);
571 			ret = 0;
572 		}
573 	} else {
574 		if (bt->trace_state == Blktrace_running) {
575 			bt->trace_state = Blktrace_stopped;
576 			relay_flush(bt->rchan);
577 			ret = 0;
578 		}
579 	}
580 
581 	return ret;
582 }
583 EXPORT_SYMBOL_GPL(blk_trace_startstop);
584 
585 /**
586  * blk_trace_ioctl: - handle the ioctls associated with tracing
587  * @bdev:	the block device
588  * @cmd:	the ioctl cmd
589  * @arg:	the argument data, if any
590  *
591  **/
592 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
593 {
594 	struct request_queue *q;
595 	int ret, start = 0;
596 	char b[BDEVNAME_SIZE];
597 
598 	q = bdev_get_queue(bdev);
599 	if (!q)
600 		return -ENXIO;
601 
602 	mutex_lock(&bdev->bd_mutex);
603 
604 	switch (cmd) {
605 	case BLKTRACESETUP:
606 		bdevname(bdev, b);
607 		ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
608 		break;
609 	case BLKTRACESTART:
610 		start = 1;
611 	case BLKTRACESTOP:
612 		ret = blk_trace_startstop(q, start);
613 		break;
614 	case BLKTRACETEARDOWN:
615 		ret = blk_trace_remove(q);
616 		break;
617 	default:
618 		ret = -ENOTTY;
619 		break;
620 	}
621 
622 	mutex_unlock(&bdev->bd_mutex);
623 	return ret;
624 }
625 
626 /**
627  * blk_trace_shutdown: - stop and cleanup trace structures
628  * @q:    the request queue associated with the device
629  *
630  **/
631 void blk_trace_shutdown(struct request_queue *q)
632 {
633 	if (q->blk_trace) {
634 		blk_trace_startstop(q, 0);
635 		blk_trace_remove(q);
636 	}
637 }
638 
639 /*
640  * blktrace probes
641  */
642 
643 /**
644  * blk_add_trace_rq - Add a trace for a request oriented action
645  * @q:		queue the io is for
646  * @rq:		the source request
647  * @what:	the action
648  *
649  * Description:
650  *     Records an action against a request. Will log the bio offset + size.
651  *
652  **/
653 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
654 				    u32 what)
655 {
656 	struct blk_trace *bt = q->blk_trace;
657 	int rw = rq->cmd_flags & 0x03;
658 
659 	if (likely(!bt))
660 		return;
661 
662 	if (blk_discard_rq(rq))
663 		rw |= (1 << BIO_RW_DISCARD);
664 
665 	if (blk_pc_request(rq)) {
666 		what |= BLK_TC_ACT(BLK_TC_PC);
667 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
668 				what, rq->errors, rq->cmd_len, rq->cmd);
669 	} else  {
670 		what |= BLK_TC_ACT(BLK_TC_FS);
671 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
672 				what, rq->errors, 0, NULL);
673 	}
674 }
675 
676 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
677 {
678 	blk_add_trace_rq(q, rq, BLK_TA_ABORT);
679 }
680 
681 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
682 {
683 	blk_add_trace_rq(q, rq, BLK_TA_INSERT);
684 }
685 
686 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
687 {
688 	blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
689 }
690 
691 static void blk_add_trace_rq_requeue(struct request_queue *q,
692 				     struct request *rq)
693 {
694 	blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
695 }
696 
697 static void blk_add_trace_rq_complete(struct request_queue *q,
698 				      struct request *rq)
699 {
700 	blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
701 }
702 
703 /**
704  * blk_add_trace_bio - Add a trace for a bio oriented action
705  * @q:		queue the io is for
706  * @bio:	the source bio
707  * @what:	the action
708  *
709  * Description:
710  *     Records an action against a bio. Will log the bio offset + size.
711  *
712  **/
713 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
714 				     u32 what)
715 {
716 	struct blk_trace *bt = q->blk_trace;
717 
718 	if (likely(!bt))
719 		return;
720 
721 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
722 			!bio_flagged(bio, BIO_UPTODATE), 0, NULL);
723 }
724 
725 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
726 {
727 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
728 }
729 
730 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
731 {
732 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
733 }
734 
735 static void blk_add_trace_bio_backmerge(struct request_queue *q,
736 					struct bio *bio)
737 {
738 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
739 }
740 
741 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
742 					 struct bio *bio)
743 {
744 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
745 }
746 
747 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
748 {
749 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
750 }
751 
752 static void blk_add_trace_getrq(struct request_queue *q,
753 				struct bio *bio, int rw)
754 {
755 	if (bio)
756 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
757 	else {
758 		struct blk_trace *bt = q->blk_trace;
759 
760 		if (bt)
761 			__blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
762 	}
763 }
764 
765 
766 static void blk_add_trace_sleeprq(struct request_queue *q,
767 				  struct bio *bio, int rw)
768 {
769 	if (bio)
770 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
771 	else {
772 		struct blk_trace *bt = q->blk_trace;
773 
774 		if (bt)
775 			__blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
776 					0, 0, NULL);
777 	}
778 }
779 
780 static void blk_add_trace_plug(struct request_queue *q)
781 {
782 	struct blk_trace *bt = q->blk_trace;
783 
784 	if (bt)
785 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
786 }
787 
788 static void blk_add_trace_unplug_io(struct request_queue *q)
789 {
790 	struct blk_trace *bt = q->blk_trace;
791 
792 	if (bt) {
793 		unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
794 		__be64 rpdu = cpu_to_be64(pdu);
795 
796 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
797 				sizeof(rpdu), &rpdu);
798 	}
799 }
800 
801 static void blk_add_trace_unplug_timer(struct request_queue *q)
802 {
803 	struct blk_trace *bt = q->blk_trace;
804 
805 	if (bt) {
806 		unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
807 		__be64 rpdu = cpu_to_be64(pdu);
808 
809 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
810 				sizeof(rpdu), &rpdu);
811 	}
812 }
813 
814 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
815 				unsigned int pdu)
816 {
817 	struct blk_trace *bt = q->blk_trace;
818 
819 	if (bt) {
820 		__be64 rpdu = cpu_to_be64(pdu);
821 
822 		__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
823 				BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
824 				sizeof(rpdu), &rpdu);
825 	}
826 }
827 
828 /**
829  * blk_add_trace_remap - Add a trace for a remap operation
830  * @q:		queue the io is for
831  * @bio:	the source bio
832  * @dev:	target device
833  * @from:	source sector
834  *
835  * Description:
836  *     Device mapper or raid target sometimes need to split a bio because
837  *     it spans a stripe (or similar). Add a trace for that action.
838  *
839  **/
840 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
841 				       dev_t dev, sector_t from)
842 {
843 	struct blk_trace *bt = q->blk_trace;
844 	struct blk_io_trace_remap r;
845 
846 	if (likely(!bt))
847 		return;
848 
849 	r.device_from = cpu_to_be32(dev);
850 	r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
851 	r.sector_from = cpu_to_be64(from);
852 
853 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
854 			BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
855 			sizeof(r), &r);
856 }
857 
858 /**
859  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
860  * @q:		queue the io is for
861  * @rq:		the source request
862  * @dev:	target device
863  * @from:	source sector
864  *
865  * Description:
866  *     Device mapper remaps request to other devices.
867  *     Add a trace for that action.
868  *
869  **/
870 static void blk_add_trace_rq_remap(struct request_queue *q,
871 				   struct request *rq, dev_t dev,
872 				   sector_t from)
873 {
874 	struct blk_trace *bt = q->blk_trace;
875 	struct blk_io_trace_remap r;
876 
877 	if (likely(!bt))
878 		return;
879 
880 	r.device_from = cpu_to_be32(dev);
881 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
882 	r.sector_from = cpu_to_be64(from);
883 
884 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
885 			rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
886 			sizeof(r), &r);
887 }
888 
889 /**
890  * blk_add_driver_data - Add binary message with driver-specific data
891  * @q:		queue the io is for
892  * @rq:		io request
893  * @data:	driver-specific data
894  * @len:	length of driver-specific data
895  *
896  * Description:
897  *     Some drivers might want to write driver-specific data per request.
898  *
899  **/
900 void blk_add_driver_data(struct request_queue *q,
901 			 struct request *rq,
902 			 void *data, size_t len)
903 {
904 	struct blk_trace *bt = q->blk_trace;
905 
906 	if (likely(!bt))
907 		return;
908 
909 	if (blk_pc_request(rq))
910 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
911 				BLK_TA_DRV_DATA, rq->errors, len, data);
912 	else
913 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
914 				BLK_TA_DRV_DATA, rq->errors, len, data);
915 }
916 EXPORT_SYMBOL_GPL(blk_add_driver_data);
917 
918 static void blk_register_tracepoints(void)
919 {
920 	int ret;
921 
922 	ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
923 	WARN_ON(ret);
924 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
925 	WARN_ON(ret);
926 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
927 	WARN_ON(ret);
928 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
929 	WARN_ON(ret);
930 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
931 	WARN_ON(ret);
932 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
933 	WARN_ON(ret);
934 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
935 	WARN_ON(ret);
936 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
937 	WARN_ON(ret);
938 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
939 	WARN_ON(ret);
940 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
941 	WARN_ON(ret);
942 	ret = register_trace_block_getrq(blk_add_trace_getrq);
943 	WARN_ON(ret);
944 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
945 	WARN_ON(ret);
946 	ret = register_trace_block_plug(blk_add_trace_plug);
947 	WARN_ON(ret);
948 	ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
949 	WARN_ON(ret);
950 	ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
951 	WARN_ON(ret);
952 	ret = register_trace_block_split(blk_add_trace_split);
953 	WARN_ON(ret);
954 	ret = register_trace_block_remap(blk_add_trace_remap);
955 	WARN_ON(ret);
956 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap);
957 	WARN_ON(ret);
958 }
959 
960 static void blk_unregister_tracepoints(void)
961 {
962 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap);
963 	unregister_trace_block_remap(blk_add_trace_remap);
964 	unregister_trace_block_split(blk_add_trace_split);
965 	unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
966 	unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
967 	unregister_trace_block_plug(blk_add_trace_plug);
968 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
969 	unregister_trace_block_getrq(blk_add_trace_getrq);
970 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
971 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
972 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
973 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
974 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
975 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
976 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
977 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
978 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
979 	unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
980 
981 	tracepoint_synchronize_unregister();
982 }
983 
984 /*
985  * struct blk_io_tracer formatting routines
986  */
987 
988 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
989 {
990 	int i = 0;
991 	int tc = t->action >> BLK_TC_SHIFT;
992 
993 	if (t->action == BLK_TN_MESSAGE) {
994 		rwbs[i++] = 'N';
995 		goto out;
996 	}
997 
998 	if (tc & BLK_TC_DISCARD)
999 		rwbs[i++] = 'D';
1000 	else if (tc & BLK_TC_WRITE)
1001 		rwbs[i++] = 'W';
1002 	else if (t->bytes)
1003 		rwbs[i++] = 'R';
1004 	else
1005 		rwbs[i++] = 'N';
1006 
1007 	if (tc & BLK_TC_AHEAD)
1008 		rwbs[i++] = 'A';
1009 	if (tc & BLK_TC_BARRIER)
1010 		rwbs[i++] = 'B';
1011 	if (tc & BLK_TC_SYNC)
1012 		rwbs[i++] = 'S';
1013 	if (tc & BLK_TC_META)
1014 		rwbs[i++] = 'M';
1015 out:
1016 	rwbs[i] = '\0';
1017 }
1018 
1019 static inline
1020 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1021 {
1022 	return (const struct blk_io_trace *)ent;
1023 }
1024 
1025 static inline const void *pdu_start(const struct trace_entry *ent)
1026 {
1027 	return te_blk_io_trace(ent) + 1;
1028 }
1029 
1030 static inline u32 t_action(const struct trace_entry *ent)
1031 {
1032 	return te_blk_io_trace(ent)->action;
1033 }
1034 
1035 static inline u32 t_bytes(const struct trace_entry *ent)
1036 {
1037 	return te_blk_io_trace(ent)->bytes;
1038 }
1039 
1040 static inline u32 t_sec(const struct trace_entry *ent)
1041 {
1042 	return te_blk_io_trace(ent)->bytes >> 9;
1043 }
1044 
1045 static inline unsigned long long t_sector(const struct trace_entry *ent)
1046 {
1047 	return te_blk_io_trace(ent)->sector;
1048 }
1049 
1050 static inline __u16 t_error(const struct trace_entry *ent)
1051 {
1052 	return te_blk_io_trace(ent)->error;
1053 }
1054 
1055 static __u64 get_pdu_int(const struct trace_entry *ent)
1056 {
1057 	const __u64 *val = pdu_start(ent);
1058 	return be64_to_cpu(*val);
1059 }
1060 
1061 static void get_pdu_remap(const struct trace_entry *ent,
1062 			  struct blk_io_trace_remap *r)
1063 {
1064 	const struct blk_io_trace_remap *__r = pdu_start(ent);
1065 	__u64 sector_from = __r->sector_from;
1066 
1067 	r->device_from = be32_to_cpu(__r->device_from);
1068 	r->device_to   = be32_to_cpu(__r->device_to);
1069 	r->sector_from = be64_to_cpu(sector_from);
1070 }
1071 
1072 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1073 
1074 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1075 {
1076 	char rwbs[6];
1077 	unsigned long long ts  = iter->ts;
1078 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1079 	unsigned secs	       = (unsigned long)ts;
1080 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1081 
1082 	fill_rwbs(rwbs, t);
1083 
1084 	return trace_seq_printf(&iter->seq,
1085 				"%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1086 				MAJOR(t->device), MINOR(t->device), iter->cpu,
1087 				secs, nsec_rem, iter->ent->pid, act, rwbs);
1088 }
1089 
1090 static int blk_log_action(struct trace_iterator *iter, const char *act)
1091 {
1092 	char rwbs[6];
1093 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1094 
1095 	fill_rwbs(rwbs, t);
1096 	return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1097 				MAJOR(t->device), MINOR(t->device), act, rwbs);
1098 }
1099 
1100 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1101 {
1102 	const unsigned char *pdu_buf;
1103 	int pdu_len;
1104 	int i, end, ret;
1105 
1106 	pdu_buf = pdu_start(ent);
1107 	pdu_len = te_blk_io_trace(ent)->pdu_len;
1108 
1109 	if (!pdu_len)
1110 		return 1;
1111 
1112 	/* find the last zero that needs to be printed */
1113 	for (end = pdu_len - 1; end >= 0; end--)
1114 		if (pdu_buf[end])
1115 			break;
1116 	end++;
1117 
1118 	if (!trace_seq_putc(s, '('))
1119 		return 0;
1120 
1121 	for (i = 0; i < pdu_len; i++) {
1122 
1123 		ret = trace_seq_printf(s, "%s%02x",
1124 				       i == 0 ? "" : " ", pdu_buf[i]);
1125 		if (!ret)
1126 			return ret;
1127 
1128 		/*
1129 		 * stop when the rest is just zeroes and indicate so
1130 		 * with a ".." appended
1131 		 */
1132 		if (i == end && end != pdu_len - 1)
1133 			return trace_seq_puts(s, " ..) ");
1134 	}
1135 
1136 	return trace_seq_puts(s, ") ");
1137 }
1138 
1139 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1140 {
1141 	char cmd[TASK_COMM_LEN];
1142 
1143 	trace_find_cmdline(ent->pid, cmd);
1144 
1145 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1146 		int ret;
1147 
1148 		ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1149 		if (!ret)
1150 			return 0;
1151 		ret = blk_log_dump_pdu(s, ent);
1152 		if (!ret)
1153 			return 0;
1154 		return trace_seq_printf(s, "[%s]\n", cmd);
1155 	} else {
1156 		if (t_sec(ent))
1157 			return trace_seq_printf(s, "%llu + %u [%s]\n",
1158 						t_sector(ent), t_sec(ent), cmd);
1159 		return trace_seq_printf(s, "[%s]\n", cmd);
1160 	}
1161 }
1162 
1163 static int blk_log_with_error(struct trace_seq *s,
1164 			      const struct trace_entry *ent)
1165 {
1166 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1167 		int ret;
1168 
1169 		ret = blk_log_dump_pdu(s, ent);
1170 		if (ret)
1171 			return trace_seq_printf(s, "[%d]\n", t_error(ent));
1172 		return 0;
1173 	} else {
1174 		if (t_sec(ent))
1175 			return trace_seq_printf(s, "%llu + %u [%d]\n",
1176 						t_sector(ent),
1177 						t_sec(ent), t_error(ent));
1178 		return trace_seq_printf(s, "%llu [%d]\n",
1179 					t_sector(ent), t_error(ent));
1180 	}
1181 }
1182 
1183 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1184 {
1185 	struct blk_io_trace_remap r = { .device_from = 0, };
1186 
1187 	get_pdu_remap(ent, &r);
1188 	return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1189 				t_sector(ent), t_sec(ent),
1190 				MAJOR(r.device_from), MINOR(r.device_from),
1191 				(unsigned long long)r.sector_from);
1192 }
1193 
1194 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1195 {
1196 	char cmd[TASK_COMM_LEN];
1197 
1198 	trace_find_cmdline(ent->pid, cmd);
1199 
1200 	return trace_seq_printf(s, "[%s]\n", cmd);
1201 }
1202 
1203 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1204 {
1205 	char cmd[TASK_COMM_LEN];
1206 
1207 	trace_find_cmdline(ent->pid, cmd);
1208 
1209 	return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1210 }
1211 
1212 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1213 {
1214 	char cmd[TASK_COMM_LEN];
1215 
1216 	trace_find_cmdline(ent->pid, cmd);
1217 
1218 	return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1219 				get_pdu_int(ent), cmd);
1220 }
1221 
1222 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1223 {
1224 	int ret;
1225 	const struct blk_io_trace *t = te_blk_io_trace(ent);
1226 
1227 	ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1228 	if (ret)
1229 		return trace_seq_putc(s, '\n');
1230 	return ret;
1231 }
1232 
1233 /*
1234  * struct tracer operations
1235  */
1236 
1237 static void blk_tracer_print_header(struct seq_file *m)
1238 {
1239 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1240 		return;
1241 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1242 		    "#  |     |     |           |   |   |\n");
1243 }
1244 
1245 static void blk_tracer_start(struct trace_array *tr)
1246 {
1247 	blk_tracer_enabled = true;
1248 }
1249 
1250 static int blk_tracer_init(struct trace_array *tr)
1251 {
1252 	blk_tr = tr;
1253 	blk_tracer_start(tr);
1254 	return 0;
1255 }
1256 
1257 static void blk_tracer_stop(struct trace_array *tr)
1258 {
1259 	blk_tracer_enabled = false;
1260 }
1261 
1262 static void blk_tracer_reset(struct trace_array *tr)
1263 {
1264 	blk_tracer_stop(tr);
1265 }
1266 
1267 static const struct {
1268 	const char *act[2];
1269 	int	   (*print)(struct trace_seq *s, const struct trace_entry *ent);
1270 } what2act[] = {
1271 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1272 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1273 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1274 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1275 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1276 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1277 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1278 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1279 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1280 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1281 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1282 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1283 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1284 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1285 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1286 };
1287 
1288 static enum print_line_t print_one_line(struct trace_iterator *iter,
1289 					bool classic)
1290 {
1291 	struct trace_seq *s = &iter->seq;
1292 	const struct blk_io_trace *t;
1293 	u16 what;
1294 	int ret;
1295 	bool long_act;
1296 	blk_log_action_t *log_action;
1297 
1298 	t	   = te_blk_io_trace(iter->ent);
1299 	what	   = t->action & ((1 << BLK_TC_SHIFT) - 1);
1300 	long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1301 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1302 
1303 	if (t->action == BLK_TN_MESSAGE) {
1304 		ret = log_action(iter, long_act ? "message" : "m");
1305 		if (ret)
1306 			ret = blk_log_msg(s, iter->ent);
1307 		goto out;
1308 	}
1309 
1310 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1311 		ret = trace_seq_printf(s, "Unknown action %x\n", what);
1312 	else {
1313 		ret = log_action(iter, what2act[what].act[long_act]);
1314 		if (ret)
1315 			ret = what2act[what].print(s, iter->ent);
1316 	}
1317 out:
1318 	return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1319 }
1320 
1321 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1322 					       int flags)
1323 {
1324 	return print_one_line(iter, false);
1325 }
1326 
1327 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1328 {
1329 	struct trace_seq *s = &iter->seq;
1330 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1331 	const int offset = offsetof(struct blk_io_trace, sector);
1332 	struct blk_io_trace old = {
1333 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1334 		.time     = iter->ts,
1335 	};
1336 
1337 	if (!trace_seq_putmem(s, &old, offset))
1338 		return 0;
1339 	return trace_seq_putmem(s, &t->sector,
1340 				sizeof(old) - offset + t->pdu_len);
1341 }
1342 
1343 static enum print_line_t
1344 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1345 {
1346 	return blk_trace_synthesize_old_trace(iter) ?
1347 			TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1348 }
1349 
1350 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1351 {
1352 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1353 		return TRACE_TYPE_UNHANDLED;
1354 
1355 	return print_one_line(iter, true);
1356 }
1357 
1358 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1359 {
1360 	/* don't output context-info for blk_classic output */
1361 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1362 		if (set)
1363 			trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1364 		else
1365 			trace_flags |= TRACE_ITER_CONTEXT_INFO;
1366 	}
1367 	return 0;
1368 }
1369 
1370 static struct tracer blk_tracer __read_mostly = {
1371 	.name		= "blk",
1372 	.init		= blk_tracer_init,
1373 	.reset		= blk_tracer_reset,
1374 	.start		= blk_tracer_start,
1375 	.stop		= blk_tracer_stop,
1376 	.print_header	= blk_tracer_print_header,
1377 	.print_line	= blk_tracer_print_line,
1378 	.flags		= &blk_tracer_flags,
1379 	.set_flag	= blk_tracer_set_flag,
1380 };
1381 
1382 static struct trace_event trace_blk_event = {
1383 	.type		= TRACE_BLK,
1384 	.trace		= blk_trace_event_print,
1385 	.binary		= blk_trace_event_print_binary,
1386 };
1387 
1388 static int __init init_blk_tracer(void)
1389 {
1390 	if (!register_ftrace_event(&trace_blk_event)) {
1391 		pr_warning("Warning: could not register block events\n");
1392 		return 1;
1393 	}
1394 
1395 	if (register_tracer(&blk_tracer) != 0) {
1396 		pr_warning("Warning: could not register the block tracer\n");
1397 		unregister_ftrace_event(&trace_blk_event);
1398 		return 1;
1399 	}
1400 
1401 	return 0;
1402 }
1403 
1404 device_initcall(init_blk_tracer);
1405 
1406 static int blk_trace_remove_queue(struct request_queue *q)
1407 {
1408 	struct blk_trace *bt;
1409 
1410 	bt = xchg(&q->blk_trace, NULL);
1411 	if (bt == NULL)
1412 		return -EINVAL;
1413 
1414 	if (atomic_dec_and_test(&blk_probes_ref))
1415 		blk_unregister_tracepoints();
1416 
1417 	blk_trace_free(bt);
1418 	return 0;
1419 }
1420 
1421 /*
1422  * Setup everything required to start tracing
1423  */
1424 static int blk_trace_setup_queue(struct request_queue *q,
1425 				 struct block_device *bdev)
1426 {
1427 	struct blk_trace *old_bt, *bt = NULL;
1428 	int ret = -ENOMEM;
1429 
1430 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1431 	if (!bt)
1432 		return -ENOMEM;
1433 
1434 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1435 	if (!bt->msg_data)
1436 		goto free_bt;
1437 
1438 	bt->dev = bdev->bd_dev;
1439 	bt->act_mask = (u16)-1;
1440 
1441 	blk_trace_setup_lba(bt, bdev);
1442 
1443 	old_bt = xchg(&q->blk_trace, bt);
1444 	if (old_bt != NULL) {
1445 		(void)xchg(&q->blk_trace, old_bt);
1446 		ret = -EBUSY;
1447 		goto free_bt;
1448 	}
1449 
1450 	if (atomic_inc_return(&blk_probes_ref) == 1)
1451 		blk_register_tracepoints();
1452 	return 0;
1453 
1454 free_bt:
1455 	blk_trace_free(bt);
1456 	return ret;
1457 }
1458 
1459 /*
1460  * sysfs interface to enable and configure tracing
1461  */
1462 
1463 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1464 					 struct device_attribute *attr,
1465 					 char *buf);
1466 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1467 					  struct device_attribute *attr,
1468 					  const char *buf, size_t count);
1469 #define BLK_TRACE_DEVICE_ATTR(_name) \
1470 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1471 		    sysfs_blk_trace_attr_show, \
1472 		    sysfs_blk_trace_attr_store)
1473 
1474 static BLK_TRACE_DEVICE_ATTR(enable);
1475 static BLK_TRACE_DEVICE_ATTR(act_mask);
1476 static BLK_TRACE_DEVICE_ATTR(pid);
1477 static BLK_TRACE_DEVICE_ATTR(start_lba);
1478 static BLK_TRACE_DEVICE_ATTR(end_lba);
1479 
1480 static struct attribute *blk_trace_attrs[] = {
1481 	&dev_attr_enable.attr,
1482 	&dev_attr_act_mask.attr,
1483 	&dev_attr_pid.attr,
1484 	&dev_attr_start_lba.attr,
1485 	&dev_attr_end_lba.attr,
1486 	NULL
1487 };
1488 
1489 struct attribute_group blk_trace_attr_group = {
1490 	.name  = "trace",
1491 	.attrs = blk_trace_attrs,
1492 };
1493 
1494 static const struct {
1495 	int mask;
1496 	const char *str;
1497 } mask_maps[] = {
1498 	{ BLK_TC_READ,		"read"		},
1499 	{ BLK_TC_WRITE,		"write"		},
1500 	{ BLK_TC_BARRIER,	"barrier"	},
1501 	{ BLK_TC_SYNC,		"sync"		},
1502 	{ BLK_TC_QUEUE,		"queue"		},
1503 	{ BLK_TC_REQUEUE,	"requeue"	},
1504 	{ BLK_TC_ISSUE,		"issue"		},
1505 	{ BLK_TC_COMPLETE,	"complete"	},
1506 	{ BLK_TC_FS,		"fs"		},
1507 	{ BLK_TC_PC,		"pc"		},
1508 	{ BLK_TC_AHEAD,		"ahead"		},
1509 	{ BLK_TC_META,		"meta"		},
1510 	{ BLK_TC_DISCARD,	"discard"	},
1511 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1512 };
1513 
1514 static int blk_trace_str2mask(const char *str)
1515 {
1516 	int i;
1517 	int mask = 0;
1518 	char *buf, *s, *token;
1519 
1520 	buf = kstrdup(str, GFP_KERNEL);
1521 	if (buf == NULL)
1522 		return -ENOMEM;
1523 	s = strstrip(buf);
1524 
1525 	while (1) {
1526 		token = strsep(&s, ",");
1527 		if (token == NULL)
1528 			break;
1529 
1530 		if (*token == '\0')
1531 			continue;
1532 
1533 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1534 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1535 				mask |= mask_maps[i].mask;
1536 				break;
1537 			}
1538 		}
1539 		if (i == ARRAY_SIZE(mask_maps)) {
1540 			mask = -EINVAL;
1541 			break;
1542 		}
1543 	}
1544 	kfree(buf);
1545 
1546 	return mask;
1547 }
1548 
1549 static ssize_t blk_trace_mask2str(char *buf, int mask)
1550 {
1551 	int i;
1552 	char *p = buf;
1553 
1554 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1555 		if (mask & mask_maps[i].mask) {
1556 			p += sprintf(p, "%s%s",
1557 				    (p == buf) ? "" : ",", mask_maps[i].str);
1558 		}
1559 	}
1560 	*p++ = '\n';
1561 
1562 	return p - buf;
1563 }
1564 
1565 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1566 {
1567 	if (bdev->bd_disk == NULL)
1568 		return NULL;
1569 
1570 	return bdev_get_queue(bdev);
1571 }
1572 
1573 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1574 					 struct device_attribute *attr,
1575 					 char *buf)
1576 {
1577 	struct hd_struct *p = dev_to_part(dev);
1578 	struct request_queue *q;
1579 	struct block_device *bdev;
1580 	ssize_t ret = -ENXIO;
1581 
1582 	lock_kernel();
1583 	bdev = bdget(part_devt(p));
1584 	if (bdev == NULL)
1585 		goto out_unlock_kernel;
1586 
1587 	q = blk_trace_get_queue(bdev);
1588 	if (q == NULL)
1589 		goto out_bdput;
1590 
1591 	mutex_lock(&bdev->bd_mutex);
1592 
1593 	if (attr == &dev_attr_enable) {
1594 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1595 		goto out_unlock_bdev;
1596 	}
1597 
1598 	if (q->blk_trace == NULL)
1599 		ret = sprintf(buf, "disabled\n");
1600 	else if (attr == &dev_attr_act_mask)
1601 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1602 	else if (attr == &dev_attr_pid)
1603 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1604 	else if (attr == &dev_attr_start_lba)
1605 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1606 	else if (attr == &dev_attr_end_lba)
1607 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1608 
1609 out_unlock_bdev:
1610 	mutex_unlock(&bdev->bd_mutex);
1611 out_bdput:
1612 	bdput(bdev);
1613 out_unlock_kernel:
1614 	unlock_kernel();
1615 	return ret;
1616 }
1617 
1618 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1619 					  struct device_attribute *attr,
1620 					  const char *buf, size_t count)
1621 {
1622 	struct block_device *bdev;
1623 	struct request_queue *q;
1624 	struct hd_struct *p;
1625 	u64 value;
1626 	ssize_t ret = -EINVAL;
1627 
1628 	if (count == 0)
1629 		goto out;
1630 
1631 	if (attr == &dev_attr_act_mask) {
1632 		if (sscanf(buf, "%llx", &value) != 1) {
1633 			/* Assume it is a list of trace category names */
1634 			ret = blk_trace_str2mask(buf);
1635 			if (ret < 0)
1636 				goto out;
1637 			value = ret;
1638 		}
1639 	} else if (sscanf(buf, "%llu", &value) != 1)
1640 		goto out;
1641 
1642 	ret = -ENXIO;
1643 
1644 	lock_kernel();
1645 	p = dev_to_part(dev);
1646 	bdev = bdget(part_devt(p));
1647 	if (bdev == NULL)
1648 		goto out_unlock_kernel;
1649 
1650 	q = blk_trace_get_queue(bdev);
1651 	if (q == NULL)
1652 		goto out_bdput;
1653 
1654 	mutex_lock(&bdev->bd_mutex);
1655 
1656 	if (attr == &dev_attr_enable) {
1657 		if (value)
1658 			ret = blk_trace_setup_queue(q, bdev);
1659 		else
1660 			ret = blk_trace_remove_queue(q);
1661 		goto out_unlock_bdev;
1662 	}
1663 
1664 	ret = 0;
1665 	if (q->blk_trace == NULL)
1666 		ret = blk_trace_setup_queue(q, bdev);
1667 
1668 	if (ret == 0) {
1669 		if (attr == &dev_attr_act_mask)
1670 			q->blk_trace->act_mask = value;
1671 		else if (attr == &dev_attr_pid)
1672 			q->blk_trace->pid = value;
1673 		else if (attr == &dev_attr_start_lba)
1674 			q->blk_trace->start_lba = value;
1675 		else if (attr == &dev_attr_end_lba)
1676 			q->blk_trace->end_lba = value;
1677 	}
1678 
1679 out_unlock_bdev:
1680 	mutex_unlock(&bdev->bd_mutex);
1681 out_bdput:
1682 	bdput(bdev);
1683 out_unlock_kernel:
1684 	unlock_kernel();
1685 out:
1686 	return ret ? ret : count;
1687 }
1688 
1689 int blk_trace_init_sysfs(struct device *dev)
1690 {
1691 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1692 }
1693 
1694 void blk_trace_remove_sysfs(struct device *dev)
1695 {
1696 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1697 }
1698 
1699 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1700 
1701 #ifdef CONFIG_EVENT_TRACING
1702 
1703 void blk_dump_cmd(char *buf, struct request *rq)
1704 {
1705 	int i, end;
1706 	int len = rq->cmd_len;
1707 	unsigned char *cmd = rq->cmd;
1708 
1709 	if (!blk_pc_request(rq)) {
1710 		buf[0] = '\0';
1711 		return;
1712 	}
1713 
1714 	for (end = len - 1; end >= 0; end--)
1715 		if (cmd[end])
1716 			break;
1717 	end++;
1718 
1719 	for (i = 0; i < len; i++) {
1720 		buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1721 		if (i == end && end != len - 1) {
1722 			sprintf(buf, " ..");
1723 			break;
1724 		}
1725 	}
1726 }
1727 
1728 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1729 {
1730 	int i = 0;
1731 
1732 	if (rw & WRITE)
1733 		rwbs[i++] = 'W';
1734 	else if (rw & 1 << BIO_RW_DISCARD)
1735 		rwbs[i++] = 'D';
1736 	else if (bytes)
1737 		rwbs[i++] = 'R';
1738 	else
1739 		rwbs[i++] = 'N';
1740 
1741 	if (rw & 1 << BIO_RW_AHEAD)
1742 		rwbs[i++] = 'A';
1743 	if (rw & 1 << BIO_RW_BARRIER)
1744 		rwbs[i++] = 'B';
1745 	if (rw & 1 << BIO_RW_SYNCIO)
1746 		rwbs[i++] = 'S';
1747 	if (rw & 1 << BIO_RW_META)
1748 		rwbs[i++] = 'M';
1749 
1750 	rwbs[i] = '\0';
1751 }
1752 
1753 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1754 {
1755 	int rw = rq->cmd_flags & 0x03;
1756 	int bytes;
1757 
1758 	if (blk_discard_rq(rq))
1759 		rw |= (1 << BIO_RW_DISCARD);
1760 
1761 	bytes = blk_rq_bytes(rq);
1762 
1763 	blk_fill_rwbs(rwbs, rw, bytes);
1764 }
1765 
1766 #endif /* CONFIG_EVENT_TRACING */
1767 
1768