xref: /openbmc/linux/kernel/trace/blktrace.c (revision 4800cd83)
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/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 	/*
142 	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
143 	 * message to the trace.
144 	 */
145 	if (!(bt->act_mask & BLK_TC_NOTIFY))
146 		return;
147 
148 	local_irq_save(flags);
149 	buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
150 	va_start(args, fmt);
151 	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
152 	va_end(args);
153 
154 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
155 	local_irq_restore(flags);
156 }
157 EXPORT_SYMBOL_GPL(__trace_note_message);
158 
159 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
160 			 pid_t pid)
161 {
162 	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
163 		return 1;
164 	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
165 		return 1;
166 	if (bt->pid && pid != bt->pid)
167 		return 1;
168 
169 	return 0;
170 }
171 
172 /*
173  * Data direction bit lookup
174  */
175 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
176 				 BLK_TC_ACT(BLK_TC_WRITE) };
177 
178 #define BLK_TC_RAHEAD		BLK_TC_AHEAD
179 
180 /* The ilog2() calls fall out because they're constant */
181 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
182 	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
183 
184 /*
185  * The worker for the various blk_add_trace*() types. Fills out a
186  * blk_io_trace structure and places it in a per-cpu subbuffer.
187  */
188 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
189 		     int rw, u32 what, int error, int pdu_len, void *pdu_data)
190 {
191 	struct task_struct *tsk = current;
192 	struct ring_buffer_event *event = NULL;
193 	struct ring_buffer *buffer = NULL;
194 	struct blk_io_trace *t;
195 	unsigned long flags = 0;
196 	unsigned long *sequence;
197 	pid_t pid;
198 	int cpu, pc = 0;
199 	bool blk_tracer = blk_tracer_enabled;
200 
201 	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
202 		return;
203 
204 	what |= ddir_act[rw & WRITE];
205 	what |= MASK_TC_BIT(rw, SYNC);
206 	what |= MASK_TC_BIT(rw, RAHEAD);
207 	what |= MASK_TC_BIT(rw, META);
208 	what |= MASK_TC_BIT(rw, DISCARD);
209 
210 	pid = tsk->pid;
211 	if (act_log_check(bt, what, sector, pid))
212 		return;
213 	cpu = raw_smp_processor_id();
214 
215 	if (blk_tracer) {
216 		tracing_record_cmdline(current);
217 
218 		buffer = blk_tr->buffer;
219 		pc = preempt_count();
220 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
221 						  sizeof(*t) + pdu_len,
222 						  0, pc);
223 		if (!event)
224 			return;
225 		t = ring_buffer_event_data(event);
226 		goto record_it;
227 	}
228 
229 	/*
230 	 * A word about the locking here - we disable interrupts to reserve
231 	 * some space in the relay per-cpu buffer, to prevent an irq
232 	 * from coming in and stepping on our toes.
233 	 */
234 	local_irq_save(flags);
235 
236 	if (unlikely(tsk->btrace_seq != blktrace_seq))
237 		trace_note_tsk(bt, tsk);
238 
239 	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
240 	if (t) {
241 		sequence = per_cpu_ptr(bt->sequence, cpu);
242 
243 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
244 		t->sequence = ++(*sequence);
245 		t->time = ktime_to_ns(ktime_get());
246 record_it:
247 		/*
248 		 * These two are not needed in ftrace as they are in the
249 		 * generic trace_entry, filled by tracing_generic_entry_update,
250 		 * but for the trace_event->bin() synthesizer benefit we do it
251 		 * here too.
252 		 */
253 		t->cpu = cpu;
254 		t->pid = pid;
255 
256 		t->sector = sector;
257 		t->bytes = bytes;
258 		t->action = what;
259 		t->device = bt->dev;
260 		t->error = error;
261 		t->pdu_len = pdu_len;
262 
263 		if (pdu_len)
264 			memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
265 
266 		if (blk_tracer) {
267 			trace_buffer_unlock_commit(buffer, event, 0, pc);
268 			return;
269 		}
270 	}
271 
272 	local_irq_restore(flags);
273 }
274 
275 static struct dentry *blk_tree_root;
276 static DEFINE_MUTEX(blk_tree_mutex);
277 
278 static void blk_trace_free(struct blk_trace *bt)
279 {
280 	debugfs_remove(bt->msg_file);
281 	debugfs_remove(bt->dropped_file);
282 	relay_close(bt->rchan);
283 	debugfs_remove(bt->dir);
284 	free_percpu(bt->sequence);
285 	free_percpu(bt->msg_data);
286 	kfree(bt);
287 }
288 
289 static void blk_trace_cleanup(struct blk_trace *bt)
290 {
291 	blk_trace_free(bt);
292 	if (atomic_dec_and_test(&blk_probes_ref))
293 		blk_unregister_tracepoints();
294 }
295 
296 int blk_trace_remove(struct request_queue *q)
297 {
298 	struct blk_trace *bt;
299 
300 	bt = xchg(&q->blk_trace, NULL);
301 	if (!bt)
302 		return -EINVAL;
303 
304 	if (bt->trace_state != Blktrace_running)
305 		blk_trace_cleanup(bt);
306 
307 	return 0;
308 }
309 EXPORT_SYMBOL_GPL(blk_trace_remove);
310 
311 static int blk_dropped_open(struct inode *inode, struct file *filp)
312 {
313 	filp->private_data = inode->i_private;
314 
315 	return 0;
316 }
317 
318 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
319 				size_t count, loff_t *ppos)
320 {
321 	struct blk_trace *bt = filp->private_data;
322 	char buf[16];
323 
324 	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
325 
326 	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
327 }
328 
329 static const struct file_operations blk_dropped_fops = {
330 	.owner =	THIS_MODULE,
331 	.open =		blk_dropped_open,
332 	.read =		blk_dropped_read,
333 	.llseek =	default_llseek,
334 };
335 
336 static int blk_msg_open(struct inode *inode, struct file *filp)
337 {
338 	filp->private_data = inode->i_private;
339 
340 	return 0;
341 }
342 
343 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
344 				size_t count, loff_t *ppos)
345 {
346 	char *msg;
347 	struct blk_trace *bt;
348 
349 	if (count >= BLK_TN_MAX_MSG)
350 		return -EINVAL;
351 
352 	msg = kmalloc(count + 1, GFP_KERNEL);
353 	if (msg == NULL)
354 		return -ENOMEM;
355 
356 	if (copy_from_user(msg, buffer, count)) {
357 		kfree(msg);
358 		return -EFAULT;
359 	}
360 
361 	msg[count] = '\0';
362 	bt = filp->private_data;
363 	__trace_note_message(bt, "%s", msg);
364 	kfree(msg);
365 
366 	return count;
367 }
368 
369 static const struct file_operations blk_msg_fops = {
370 	.owner =	THIS_MODULE,
371 	.open =		blk_msg_open,
372 	.write =	blk_msg_write,
373 	.llseek =	noop_llseek,
374 };
375 
376 /*
377  * Keep track of how many times we encountered a full subbuffer, to aid
378  * the user space app in telling how many lost events there were.
379  */
380 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
381 				     void *prev_subbuf, size_t prev_padding)
382 {
383 	struct blk_trace *bt;
384 
385 	if (!relay_buf_full(buf))
386 		return 1;
387 
388 	bt = buf->chan->private_data;
389 	atomic_inc(&bt->dropped);
390 	return 0;
391 }
392 
393 static int blk_remove_buf_file_callback(struct dentry *dentry)
394 {
395 	debugfs_remove(dentry);
396 
397 	return 0;
398 }
399 
400 static struct dentry *blk_create_buf_file_callback(const char *filename,
401 						   struct dentry *parent,
402 						   int mode,
403 						   struct rchan_buf *buf,
404 						   int *is_global)
405 {
406 	return debugfs_create_file(filename, mode, parent, buf,
407 					&relay_file_operations);
408 }
409 
410 static struct rchan_callbacks blk_relay_callbacks = {
411 	.subbuf_start		= blk_subbuf_start_callback,
412 	.create_buf_file	= blk_create_buf_file_callback,
413 	.remove_buf_file	= blk_remove_buf_file_callback,
414 };
415 
416 static void blk_trace_setup_lba(struct blk_trace *bt,
417 				struct block_device *bdev)
418 {
419 	struct hd_struct *part = NULL;
420 
421 	if (bdev)
422 		part = bdev->bd_part;
423 
424 	if (part) {
425 		bt->start_lba = part->start_sect;
426 		bt->end_lba = part->start_sect + part->nr_sects;
427 	} else {
428 		bt->start_lba = 0;
429 		bt->end_lba = -1ULL;
430 	}
431 }
432 
433 /*
434  * Setup everything required to start tracing
435  */
436 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
437 		       struct block_device *bdev,
438 		       struct blk_user_trace_setup *buts)
439 {
440 	struct blk_trace *old_bt, *bt = NULL;
441 	struct dentry *dir = NULL;
442 	int ret, i;
443 
444 	if (!buts->buf_size || !buts->buf_nr)
445 		return -EINVAL;
446 
447 	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
448 	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
449 
450 	/*
451 	 * some device names have larger paths - convert the slashes
452 	 * to underscores for this to work as expected
453 	 */
454 	for (i = 0; i < strlen(buts->name); i++)
455 		if (buts->name[i] == '/')
456 			buts->name[i] = '_';
457 
458 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
459 	if (!bt)
460 		return -ENOMEM;
461 
462 	ret = -ENOMEM;
463 	bt->sequence = alloc_percpu(unsigned long);
464 	if (!bt->sequence)
465 		goto err;
466 
467 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
468 	if (!bt->msg_data)
469 		goto err;
470 
471 	ret = -ENOENT;
472 
473 	mutex_lock(&blk_tree_mutex);
474 	if (!blk_tree_root) {
475 		blk_tree_root = debugfs_create_dir("block", NULL);
476 		if (!blk_tree_root) {
477 			mutex_unlock(&blk_tree_mutex);
478 			goto err;
479 		}
480 	}
481 	mutex_unlock(&blk_tree_mutex);
482 
483 	dir = debugfs_create_dir(buts->name, blk_tree_root);
484 
485 	if (!dir)
486 		goto err;
487 
488 	bt->dir = dir;
489 	bt->dev = dev;
490 	atomic_set(&bt->dropped, 0);
491 
492 	ret = -EIO;
493 	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
494 					       &blk_dropped_fops);
495 	if (!bt->dropped_file)
496 		goto err;
497 
498 	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
499 	if (!bt->msg_file)
500 		goto err;
501 
502 	bt->rchan = relay_open("trace", dir, buts->buf_size,
503 				buts->buf_nr, &blk_relay_callbacks, bt);
504 	if (!bt->rchan)
505 		goto err;
506 
507 	bt->act_mask = buts->act_mask;
508 	if (!bt->act_mask)
509 		bt->act_mask = (u16) -1;
510 
511 	blk_trace_setup_lba(bt, bdev);
512 
513 	/* overwrite with user settings */
514 	if (buts->start_lba)
515 		bt->start_lba = buts->start_lba;
516 	if (buts->end_lba)
517 		bt->end_lba = buts->end_lba;
518 
519 	bt->pid = buts->pid;
520 	bt->trace_state = Blktrace_setup;
521 
522 	ret = -EBUSY;
523 	old_bt = xchg(&q->blk_trace, bt);
524 	if (old_bt) {
525 		(void) xchg(&q->blk_trace, old_bt);
526 		goto err;
527 	}
528 
529 	if (atomic_inc_return(&blk_probes_ref) == 1)
530 		blk_register_tracepoints();
531 
532 	return 0;
533 err:
534 	blk_trace_free(bt);
535 	return ret;
536 }
537 
538 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
539 		    struct block_device *bdev,
540 		    char __user *arg)
541 {
542 	struct blk_user_trace_setup buts;
543 	int ret;
544 
545 	ret = copy_from_user(&buts, arg, sizeof(buts));
546 	if (ret)
547 		return -EFAULT;
548 
549 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
550 	if (ret)
551 		return ret;
552 
553 	if (copy_to_user(arg, &buts, sizeof(buts))) {
554 		blk_trace_remove(q);
555 		return -EFAULT;
556 	}
557 	return 0;
558 }
559 EXPORT_SYMBOL_GPL(blk_trace_setup);
560 
561 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
562 static int compat_blk_trace_setup(struct request_queue *q, char *name,
563 				  dev_t dev, struct block_device *bdev,
564 				  char __user *arg)
565 {
566 	struct blk_user_trace_setup buts;
567 	struct compat_blk_user_trace_setup cbuts;
568 	int ret;
569 
570 	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
571 		return -EFAULT;
572 
573 	buts = (struct blk_user_trace_setup) {
574 		.act_mask = cbuts.act_mask,
575 		.buf_size = cbuts.buf_size,
576 		.buf_nr = cbuts.buf_nr,
577 		.start_lba = cbuts.start_lba,
578 		.end_lba = cbuts.end_lba,
579 		.pid = cbuts.pid,
580 	};
581 	memcpy(&buts.name, &cbuts.name, 32);
582 
583 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
584 	if (ret)
585 		return ret;
586 
587 	if (copy_to_user(arg, &buts.name, 32)) {
588 		blk_trace_remove(q);
589 		return -EFAULT;
590 	}
591 
592 	return 0;
593 }
594 #endif
595 
596 int blk_trace_startstop(struct request_queue *q, int start)
597 {
598 	int ret;
599 	struct blk_trace *bt = q->blk_trace;
600 
601 	if (bt == NULL)
602 		return -EINVAL;
603 
604 	/*
605 	 * For starting a trace, we can transition from a setup or stopped
606 	 * trace. For stopping a trace, the state must be running
607 	 */
608 	ret = -EINVAL;
609 	if (start) {
610 		if (bt->trace_state == Blktrace_setup ||
611 		    bt->trace_state == Blktrace_stopped) {
612 			blktrace_seq++;
613 			smp_mb();
614 			bt->trace_state = Blktrace_running;
615 
616 			trace_note_time(bt);
617 			ret = 0;
618 		}
619 	} else {
620 		if (bt->trace_state == Blktrace_running) {
621 			bt->trace_state = Blktrace_stopped;
622 			relay_flush(bt->rchan);
623 			ret = 0;
624 		}
625 	}
626 
627 	return ret;
628 }
629 EXPORT_SYMBOL_GPL(blk_trace_startstop);
630 
631 /**
632  * blk_trace_ioctl: - handle the ioctls associated with tracing
633  * @bdev:	the block device
634  * @cmd:	the ioctl cmd
635  * @arg:	the argument data, if any
636  *
637  **/
638 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
639 {
640 	struct request_queue *q;
641 	int ret, start = 0;
642 	char b[BDEVNAME_SIZE];
643 
644 	q = bdev_get_queue(bdev);
645 	if (!q)
646 		return -ENXIO;
647 
648 	mutex_lock(&bdev->bd_mutex);
649 
650 	switch (cmd) {
651 	case BLKTRACESETUP:
652 		bdevname(bdev, b);
653 		ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
654 		break;
655 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
656 	case BLKTRACESETUP32:
657 		bdevname(bdev, b);
658 		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
659 		break;
660 #endif
661 	case BLKTRACESTART:
662 		start = 1;
663 	case BLKTRACESTOP:
664 		ret = blk_trace_startstop(q, start);
665 		break;
666 	case BLKTRACETEARDOWN:
667 		ret = blk_trace_remove(q);
668 		break;
669 	default:
670 		ret = -ENOTTY;
671 		break;
672 	}
673 
674 	mutex_unlock(&bdev->bd_mutex);
675 	return ret;
676 }
677 
678 /**
679  * blk_trace_shutdown: - stop and cleanup trace structures
680  * @q:    the request queue associated with the device
681  *
682  **/
683 void blk_trace_shutdown(struct request_queue *q)
684 {
685 	if (q->blk_trace) {
686 		blk_trace_startstop(q, 0);
687 		blk_trace_remove(q);
688 	}
689 }
690 
691 /*
692  * blktrace probes
693  */
694 
695 /**
696  * blk_add_trace_rq - Add a trace for a request oriented action
697  * @q:		queue the io is for
698  * @rq:		the source request
699  * @what:	the action
700  *
701  * Description:
702  *     Records an action against a request. Will log the bio offset + size.
703  *
704  **/
705 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
706 				    u32 what)
707 {
708 	struct blk_trace *bt = q->blk_trace;
709 	int rw = rq->cmd_flags & 0x03;
710 
711 	if (likely(!bt))
712 		return;
713 
714 	if (rq->cmd_flags & REQ_DISCARD)
715 		rw |= REQ_DISCARD;
716 
717 	if (rq->cmd_flags & REQ_SECURE)
718 		rw |= REQ_SECURE;
719 
720 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
721 		what |= BLK_TC_ACT(BLK_TC_PC);
722 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
723 				what, rq->errors, rq->cmd_len, rq->cmd);
724 	} else  {
725 		what |= BLK_TC_ACT(BLK_TC_FS);
726 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
727 				what, rq->errors, 0, NULL);
728 	}
729 }
730 
731 static void blk_add_trace_rq_abort(void *ignore,
732 				   struct request_queue *q, struct request *rq)
733 {
734 	blk_add_trace_rq(q, rq, BLK_TA_ABORT);
735 }
736 
737 static void blk_add_trace_rq_insert(void *ignore,
738 				    struct request_queue *q, struct request *rq)
739 {
740 	blk_add_trace_rq(q, rq, BLK_TA_INSERT);
741 }
742 
743 static void blk_add_trace_rq_issue(void *ignore,
744 				   struct request_queue *q, struct request *rq)
745 {
746 	blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
747 }
748 
749 static void blk_add_trace_rq_requeue(void *ignore,
750 				     struct request_queue *q,
751 				     struct request *rq)
752 {
753 	blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
754 }
755 
756 static void blk_add_trace_rq_complete(void *ignore,
757 				      struct request_queue *q,
758 				      struct request *rq)
759 {
760 	blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
761 }
762 
763 /**
764  * blk_add_trace_bio - Add a trace for a bio oriented action
765  * @q:		queue the io is for
766  * @bio:	the source bio
767  * @what:	the action
768  * @error:	error, if any
769  *
770  * Description:
771  *     Records an action against a bio. Will log the bio offset + size.
772  *
773  **/
774 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
775 			      u32 what, int error)
776 {
777 	struct blk_trace *bt = q->blk_trace;
778 
779 	if (likely(!bt))
780 		return;
781 
782 	if (!error && !bio_flagged(bio, BIO_UPTODATE))
783 		error = EIO;
784 
785 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
786 			error, 0, NULL);
787 }
788 
789 static void blk_add_trace_bio_bounce(void *ignore,
790 				     struct request_queue *q, struct bio *bio)
791 {
792 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
793 }
794 
795 static void blk_add_trace_bio_complete(void *ignore,
796 				       struct request_queue *q, struct bio *bio,
797 				       int error)
798 {
799 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
800 }
801 
802 static void blk_add_trace_bio_backmerge(void *ignore,
803 					struct request_queue *q,
804 					struct bio *bio)
805 {
806 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
807 }
808 
809 static void blk_add_trace_bio_frontmerge(void *ignore,
810 					 struct request_queue *q,
811 					 struct bio *bio)
812 {
813 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
814 }
815 
816 static void blk_add_trace_bio_queue(void *ignore,
817 				    struct request_queue *q, struct bio *bio)
818 {
819 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
820 }
821 
822 static void blk_add_trace_getrq(void *ignore,
823 				struct request_queue *q,
824 				struct bio *bio, int rw)
825 {
826 	if (bio)
827 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
828 	else {
829 		struct blk_trace *bt = q->blk_trace;
830 
831 		if (bt)
832 			__blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
833 	}
834 }
835 
836 
837 static void blk_add_trace_sleeprq(void *ignore,
838 				  struct request_queue *q,
839 				  struct bio *bio, int rw)
840 {
841 	if (bio)
842 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
843 	else {
844 		struct blk_trace *bt = q->blk_trace;
845 
846 		if (bt)
847 			__blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
848 					0, 0, NULL);
849 	}
850 }
851 
852 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
853 {
854 	struct blk_trace *bt = q->blk_trace;
855 
856 	if (bt)
857 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
858 }
859 
860 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
861 {
862 	struct blk_trace *bt = q->blk_trace;
863 
864 	if (bt) {
865 		unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
866 		__be64 rpdu = cpu_to_be64(pdu);
867 
868 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
869 				sizeof(rpdu), &rpdu);
870 	}
871 }
872 
873 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q)
874 {
875 	struct blk_trace *bt = q->blk_trace;
876 
877 	if (bt) {
878 		unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
879 		__be64 rpdu = cpu_to_be64(pdu);
880 
881 		__blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
882 				sizeof(rpdu), &rpdu);
883 	}
884 }
885 
886 static void blk_add_trace_split(void *ignore,
887 				struct request_queue *q, struct bio *bio,
888 				unsigned int pdu)
889 {
890 	struct blk_trace *bt = q->blk_trace;
891 
892 	if (bt) {
893 		__be64 rpdu = cpu_to_be64(pdu);
894 
895 		__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
896 				BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
897 				sizeof(rpdu), &rpdu);
898 	}
899 }
900 
901 /**
902  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
903  * @ignore:	trace callback data parameter (not used)
904  * @q:		queue the io is for
905  * @bio:	the source bio
906  * @dev:	target device
907  * @from:	source sector
908  *
909  * Description:
910  *     Device mapper or raid target sometimes need to split a bio because
911  *     it spans a stripe (or similar). Add a trace for that action.
912  *
913  **/
914 static void blk_add_trace_bio_remap(void *ignore,
915 				    struct request_queue *q, struct bio *bio,
916 				    dev_t dev, sector_t from)
917 {
918 	struct blk_trace *bt = q->blk_trace;
919 	struct blk_io_trace_remap r;
920 
921 	if (likely(!bt))
922 		return;
923 
924 	r.device_from = cpu_to_be32(dev);
925 	r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
926 	r.sector_from = cpu_to_be64(from);
927 
928 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
929 			BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
930 			sizeof(r), &r);
931 }
932 
933 /**
934  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
935  * @ignore:	trace callback data parameter (not used)
936  * @q:		queue the io is for
937  * @rq:		the source request
938  * @dev:	target device
939  * @from:	source sector
940  *
941  * Description:
942  *     Device mapper remaps request to other devices.
943  *     Add a trace for that action.
944  *
945  **/
946 static void blk_add_trace_rq_remap(void *ignore,
947 				   struct request_queue *q,
948 				   struct request *rq, dev_t dev,
949 				   sector_t from)
950 {
951 	struct blk_trace *bt = q->blk_trace;
952 	struct blk_io_trace_remap r;
953 
954 	if (likely(!bt))
955 		return;
956 
957 	r.device_from = cpu_to_be32(dev);
958 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
959 	r.sector_from = cpu_to_be64(from);
960 
961 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
962 			rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
963 			sizeof(r), &r);
964 }
965 
966 /**
967  * blk_add_driver_data - Add binary message with driver-specific data
968  * @q:		queue the io is for
969  * @rq:		io request
970  * @data:	driver-specific data
971  * @len:	length of driver-specific data
972  *
973  * Description:
974  *     Some drivers might want to write driver-specific data per request.
975  *
976  **/
977 void blk_add_driver_data(struct request_queue *q,
978 			 struct request *rq,
979 			 void *data, size_t len)
980 {
981 	struct blk_trace *bt = q->blk_trace;
982 
983 	if (likely(!bt))
984 		return;
985 
986 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
987 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
988 				BLK_TA_DRV_DATA, rq->errors, len, data);
989 	else
990 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
991 				BLK_TA_DRV_DATA, rq->errors, len, data);
992 }
993 EXPORT_SYMBOL_GPL(blk_add_driver_data);
994 
995 static void blk_register_tracepoints(void)
996 {
997 	int ret;
998 
999 	ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1000 	WARN_ON(ret);
1001 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1002 	WARN_ON(ret);
1003 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1004 	WARN_ON(ret);
1005 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1006 	WARN_ON(ret);
1007 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1008 	WARN_ON(ret);
1009 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1010 	WARN_ON(ret);
1011 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1012 	WARN_ON(ret);
1013 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1014 	WARN_ON(ret);
1015 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1016 	WARN_ON(ret);
1017 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1018 	WARN_ON(ret);
1019 	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1020 	WARN_ON(ret);
1021 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1022 	WARN_ON(ret);
1023 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1024 	WARN_ON(ret);
1025 	ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1026 	WARN_ON(ret);
1027 	ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1028 	WARN_ON(ret);
1029 	ret = register_trace_block_split(blk_add_trace_split, NULL);
1030 	WARN_ON(ret);
1031 	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1032 	WARN_ON(ret);
1033 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1034 	WARN_ON(ret);
1035 }
1036 
1037 static void blk_unregister_tracepoints(void)
1038 {
1039 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1040 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1041 	unregister_trace_block_split(blk_add_trace_split, NULL);
1042 	unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1043 	unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1044 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1045 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1046 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1047 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1048 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1049 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1050 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1051 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1052 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1053 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1054 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1055 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1056 	unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1057 
1058 	tracepoint_synchronize_unregister();
1059 }
1060 
1061 /*
1062  * struct blk_io_tracer formatting routines
1063  */
1064 
1065 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1066 {
1067 	int i = 0;
1068 	int tc = t->action >> BLK_TC_SHIFT;
1069 
1070 	if (t->action == BLK_TN_MESSAGE) {
1071 		rwbs[i++] = 'N';
1072 		goto out;
1073 	}
1074 
1075 	if (tc & BLK_TC_DISCARD)
1076 		rwbs[i++] = 'D';
1077 	else if (tc & BLK_TC_WRITE)
1078 		rwbs[i++] = 'W';
1079 	else if (t->bytes)
1080 		rwbs[i++] = 'R';
1081 	else
1082 		rwbs[i++] = 'N';
1083 
1084 	if (tc & BLK_TC_AHEAD)
1085 		rwbs[i++] = 'A';
1086 	if (tc & BLK_TC_BARRIER)
1087 		rwbs[i++] = 'B';
1088 	if (tc & BLK_TC_SYNC)
1089 		rwbs[i++] = 'S';
1090 	if (tc & BLK_TC_META)
1091 		rwbs[i++] = 'M';
1092 out:
1093 	rwbs[i] = '\0';
1094 }
1095 
1096 static inline
1097 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1098 {
1099 	return (const struct blk_io_trace *)ent;
1100 }
1101 
1102 static inline const void *pdu_start(const struct trace_entry *ent)
1103 {
1104 	return te_blk_io_trace(ent) + 1;
1105 }
1106 
1107 static inline u32 t_action(const struct trace_entry *ent)
1108 {
1109 	return te_blk_io_trace(ent)->action;
1110 }
1111 
1112 static inline u32 t_bytes(const struct trace_entry *ent)
1113 {
1114 	return te_blk_io_trace(ent)->bytes;
1115 }
1116 
1117 static inline u32 t_sec(const struct trace_entry *ent)
1118 {
1119 	return te_blk_io_trace(ent)->bytes >> 9;
1120 }
1121 
1122 static inline unsigned long long t_sector(const struct trace_entry *ent)
1123 {
1124 	return te_blk_io_trace(ent)->sector;
1125 }
1126 
1127 static inline __u16 t_error(const struct trace_entry *ent)
1128 {
1129 	return te_blk_io_trace(ent)->error;
1130 }
1131 
1132 static __u64 get_pdu_int(const struct trace_entry *ent)
1133 {
1134 	const __u64 *val = pdu_start(ent);
1135 	return be64_to_cpu(*val);
1136 }
1137 
1138 static void get_pdu_remap(const struct trace_entry *ent,
1139 			  struct blk_io_trace_remap *r)
1140 {
1141 	const struct blk_io_trace_remap *__r = pdu_start(ent);
1142 	__u64 sector_from = __r->sector_from;
1143 
1144 	r->device_from = be32_to_cpu(__r->device_from);
1145 	r->device_to   = be32_to_cpu(__r->device_to);
1146 	r->sector_from = be64_to_cpu(sector_from);
1147 }
1148 
1149 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1150 
1151 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1152 {
1153 	char rwbs[6];
1154 	unsigned long long ts  = iter->ts;
1155 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1156 	unsigned secs	       = (unsigned long)ts;
1157 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1158 
1159 	fill_rwbs(rwbs, t);
1160 
1161 	return trace_seq_printf(&iter->seq,
1162 				"%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1163 				MAJOR(t->device), MINOR(t->device), iter->cpu,
1164 				secs, nsec_rem, iter->ent->pid, act, rwbs);
1165 }
1166 
1167 static int blk_log_action(struct trace_iterator *iter, const char *act)
1168 {
1169 	char rwbs[6];
1170 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1171 
1172 	fill_rwbs(rwbs, t);
1173 	return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1174 				MAJOR(t->device), MINOR(t->device), act, rwbs);
1175 }
1176 
1177 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1178 {
1179 	const unsigned char *pdu_buf;
1180 	int pdu_len;
1181 	int i, end, ret;
1182 
1183 	pdu_buf = pdu_start(ent);
1184 	pdu_len = te_blk_io_trace(ent)->pdu_len;
1185 
1186 	if (!pdu_len)
1187 		return 1;
1188 
1189 	/* find the last zero that needs to be printed */
1190 	for (end = pdu_len - 1; end >= 0; end--)
1191 		if (pdu_buf[end])
1192 			break;
1193 	end++;
1194 
1195 	if (!trace_seq_putc(s, '('))
1196 		return 0;
1197 
1198 	for (i = 0; i < pdu_len; i++) {
1199 
1200 		ret = trace_seq_printf(s, "%s%02x",
1201 				       i == 0 ? "" : " ", pdu_buf[i]);
1202 		if (!ret)
1203 			return ret;
1204 
1205 		/*
1206 		 * stop when the rest is just zeroes and indicate so
1207 		 * with a ".." appended
1208 		 */
1209 		if (i == end && end != pdu_len - 1)
1210 			return trace_seq_puts(s, " ..) ");
1211 	}
1212 
1213 	return trace_seq_puts(s, ") ");
1214 }
1215 
1216 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1217 {
1218 	char cmd[TASK_COMM_LEN];
1219 
1220 	trace_find_cmdline(ent->pid, cmd);
1221 
1222 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1223 		int ret;
1224 
1225 		ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1226 		if (!ret)
1227 			return 0;
1228 		ret = blk_log_dump_pdu(s, ent);
1229 		if (!ret)
1230 			return 0;
1231 		return trace_seq_printf(s, "[%s]\n", cmd);
1232 	} else {
1233 		if (t_sec(ent))
1234 			return trace_seq_printf(s, "%llu + %u [%s]\n",
1235 						t_sector(ent), t_sec(ent), cmd);
1236 		return trace_seq_printf(s, "[%s]\n", cmd);
1237 	}
1238 }
1239 
1240 static int blk_log_with_error(struct trace_seq *s,
1241 			      const struct trace_entry *ent)
1242 {
1243 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1244 		int ret;
1245 
1246 		ret = blk_log_dump_pdu(s, ent);
1247 		if (ret)
1248 			return trace_seq_printf(s, "[%d]\n", t_error(ent));
1249 		return 0;
1250 	} else {
1251 		if (t_sec(ent))
1252 			return trace_seq_printf(s, "%llu + %u [%d]\n",
1253 						t_sector(ent),
1254 						t_sec(ent), t_error(ent));
1255 		return trace_seq_printf(s, "%llu [%d]\n",
1256 					t_sector(ent), t_error(ent));
1257 	}
1258 }
1259 
1260 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1261 {
1262 	struct blk_io_trace_remap r = { .device_from = 0, };
1263 
1264 	get_pdu_remap(ent, &r);
1265 	return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1266 				t_sector(ent), t_sec(ent),
1267 				MAJOR(r.device_from), MINOR(r.device_from),
1268 				(unsigned long long)r.sector_from);
1269 }
1270 
1271 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1272 {
1273 	char cmd[TASK_COMM_LEN];
1274 
1275 	trace_find_cmdline(ent->pid, cmd);
1276 
1277 	return trace_seq_printf(s, "[%s]\n", cmd);
1278 }
1279 
1280 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1281 {
1282 	char cmd[TASK_COMM_LEN];
1283 
1284 	trace_find_cmdline(ent->pid, cmd);
1285 
1286 	return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1287 }
1288 
1289 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1290 {
1291 	char cmd[TASK_COMM_LEN];
1292 
1293 	trace_find_cmdline(ent->pid, cmd);
1294 
1295 	return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1296 				get_pdu_int(ent), cmd);
1297 }
1298 
1299 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1300 {
1301 	int ret;
1302 	const struct blk_io_trace *t = te_blk_io_trace(ent);
1303 
1304 	ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1305 	if (ret)
1306 		return trace_seq_putc(s, '\n');
1307 	return ret;
1308 }
1309 
1310 /*
1311  * struct tracer operations
1312  */
1313 
1314 static void blk_tracer_print_header(struct seq_file *m)
1315 {
1316 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1317 		return;
1318 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1319 		    "#  |     |     |           |   |   |\n");
1320 }
1321 
1322 static void blk_tracer_start(struct trace_array *tr)
1323 {
1324 	blk_tracer_enabled = true;
1325 }
1326 
1327 static int blk_tracer_init(struct trace_array *tr)
1328 {
1329 	blk_tr = tr;
1330 	blk_tracer_start(tr);
1331 	return 0;
1332 }
1333 
1334 static void blk_tracer_stop(struct trace_array *tr)
1335 {
1336 	blk_tracer_enabled = false;
1337 }
1338 
1339 static void blk_tracer_reset(struct trace_array *tr)
1340 {
1341 	blk_tracer_stop(tr);
1342 }
1343 
1344 static const struct {
1345 	const char *act[2];
1346 	int	   (*print)(struct trace_seq *s, const struct trace_entry *ent);
1347 } what2act[] = {
1348 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1349 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1350 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1351 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1352 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1353 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1354 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1355 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1356 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1357 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1358 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1359 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1360 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1361 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1362 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1363 };
1364 
1365 static enum print_line_t print_one_line(struct trace_iterator *iter,
1366 					bool classic)
1367 {
1368 	struct trace_seq *s = &iter->seq;
1369 	const struct blk_io_trace *t;
1370 	u16 what;
1371 	int ret;
1372 	bool long_act;
1373 	blk_log_action_t *log_action;
1374 
1375 	t	   = te_blk_io_trace(iter->ent);
1376 	what	   = t->action & ((1 << BLK_TC_SHIFT) - 1);
1377 	long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1378 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1379 
1380 	if (t->action == BLK_TN_MESSAGE) {
1381 		ret = log_action(iter, long_act ? "message" : "m");
1382 		if (ret)
1383 			ret = blk_log_msg(s, iter->ent);
1384 		goto out;
1385 	}
1386 
1387 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1388 		ret = trace_seq_printf(s, "Unknown action %x\n", what);
1389 	else {
1390 		ret = log_action(iter, what2act[what].act[long_act]);
1391 		if (ret)
1392 			ret = what2act[what].print(s, iter->ent);
1393 	}
1394 out:
1395 	return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1396 }
1397 
1398 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1399 					       int flags, struct trace_event *event)
1400 {
1401 	return print_one_line(iter, false);
1402 }
1403 
1404 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1405 {
1406 	struct trace_seq *s = &iter->seq;
1407 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1408 	const int offset = offsetof(struct blk_io_trace, sector);
1409 	struct blk_io_trace old = {
1410 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1411 		.time     = iter->ts,
1412 	};
1413 
1414 	if (!trace_seq_putmem(s, &old, offset))
1415 		return 0;
1416 	return trace_seq_putmem(s, &t->sector,
1417 				sizeof(old) - offset + t->pdu_len);
1418 }
1419 
1420 static enum print_line_t
1421 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1422 			     struct trace_event *event)
1423 {
1424 	return blk_trace_synthesize_old_trace(iter) ?
1425 			TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1426 }
1427 
1428 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1429 {
1430 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1431 		return TRACE_TYPE_UNHANDLED;
1432 
1433 	return print_one_line(iter, true);
1434 }
1435 
1436 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1437 {
1438 	/* don't output context-info for blk_classic output */
1439 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1440 		if (set)
1441 			trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1442 		else
1443 			trace_flags |= TRACE_ITER_CONTEXT_INFO;
1444 	}
1445 	return 0;
1446 }
1447 
1448 static struct tracer blk_tracer __read_mostly = {
1449 	.name		= "blk",
1450 	.init		= blk_tracer_init,
1451 	.reset		= blk_tracer_reset,
1452 	.start		= blk_tracer_start,
1453 	.stop		= blk_tracer_stop,
1454 	.print_header	= blk_tracer_print_header,
1455 	.print_line	= blk_tracer_print_line,
1456 	.flags		= &blk_tracer_flags,
1457 	.set_flag	= blk_tracer_set_flag,
1458 };
1459 
1460 static struct trace_event_functions trace_blk_event_funcs = {
1461 	.trace		= blk_trace_event_print,
1462 	.binary		= blk_trace_event_print_binary,
1463 };
1464 
1465 static struct trace_event trace_blk_event = {
1466 	.type		= TRACE_BLK,
1467 	.funcs		= &trace_blk_event_funcs,
1468 };
1469 
1470 static int __init init_blk_tracer(void)
1471 {
1472 	if (!register_ftrace_event(&trace_blk_event)) {
1473 		pr_warning("Warning: could not register block events\n");
1474 		return 1;
1475 	}
1476 
1477 	if (register_tracer(&blk_tracer) != 0) {
1478 		pr_warning("Warning: could not register the block tracer\n");
1479 		unregister_ftrace_event(&trace_blk_event);
1480 		return 1;
1481 	}
1482 
1483 	return 0;
1484 }
1485 
1486 device_initcall(init_blk_tracer);
1487 
1488 static int blk_trace_remove_queue(struct request_queue *q)
1489 {
1490 	struct blk_trace *bt;
1491 
1492 	bt = xchg(&q->blk_trace, NULL);
1493 	if (bt == NULL)
1494 		return -EINVAL;
1495 
1496 	if (atomic_dec_and_test(&blk_probes_ref))
1497 		blk_unregister_tracepoints();
1498 
1499 	blk_trace_free(bt);
1500 	return 0;
1501 }
1502 
1503 /*
1504  * Setup everything required to start tracing
1505  */
1506 static int blk_trace_setup_queue(struct request_queue *q,
1507 				 struct block_device *bdev)
1508 {
1509 	struct blk_trace *old_bt, *bt = NULL;
1510 	int ret = -ENOMEM;
1511 
1512 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1513 	if (!bt)
1514 		return -ENOMEM;
1515 
1516 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1517 	if (!bt->msg_data)
1518 		goto free_bt;
1519 
1520 	bt->dev = bdev->bd_dev;
1521 	bt->act_mask = (u16)-1;
1522 
1523 	blk_trace_setup_lba(bt, bdev);
1524 
1525 	old_bt = xchg(&q->blk_trace, bt);
1526 	if (old_bt != NULL) {
1527 		(void)xchg(&q->blk_trace, old_bt);
1528 		ret = -EBUSY;
1529 		goto free_bt;
1530 	}
1531 
1532 	if (atomic_inc_return(&blk_probes_ref) == 1)
1533 		blk_register_tracepoints();
1534 	return 0;
1535 
1536 free_bt:
1537 	blk_trace_free(bt);
1538 	return ret;
1539 }
1540 
1541 /*
1542  * sysfs interface to enable and configure tracing
1543  */
1544 
1545 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1546 					 struct device_attribute *attr,
1547 					 char *buf);
1548 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1549 					  struct device_attribute *attr,
1550 					  const char *buf, size_t count);
1551 #define BLK_TRACE_DEVICE_ATTR(_name) \
1552 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1553 		    sysfs_blk_trace_attr_show, \
1554 		    sysfs_blk_trace_attr_store)
1555 
1556 static BLK_TRACE_DEVICE_ATTR(enable);
1557 static BLK_TRACE_DEVICE_ATTR(act_mask);
1558 static BLK_TRACE_DEVICE_ATTR(pid);
1559 static BLK_TRACE_DEVICE_ATTR(start_lba);
1560 static BLK_TRACE_DEVICE_ATTR(end_lba);
1561 
1562 static struct attribute *blk_trace_attrs[] = {
1563 	&dev_attr_enable.attr,
1564 	&dev_attr_act_mask.attr,
1565 	&dev_attr_pid.attr,
1566 	&dev_attr_start_lba.attr,
1567 	&dev_attr_end_lba.attr,
1568 	NULL
1569 };
1570 
1571 struct attribute_group blk_trace_attr_group = {
1572 	.name  = "trace",
1573 	.attrs = blk_trace_attrs,
1574 };
1575 
1576 static const struct {
1577 	int mask;
1578 	const char *str;
1579 } mask_maps[] = {
1580 	{ BLK_TC_READ,		"read"		},
1581 	{ BLK_TC_WRITE,		"write"		},
1582 	{ BLK_TC_BARRIER,	"barrier"	},
1583 	{ BLK_TC_SYNC,		"sync"		},
1584 	{ BLK_TC_QUEUE,		"queue"		},
1585 	{ BLK_TC_REQUEUE,	"requeue"	},
1586 	{ BLK_TC_ISSUE,		"issue"		},
1587 	{ BLK_TC_COMPLETE,	"complete"	},
1588 	{ BLK_TC_FS,		"fs"		},
1589 	{ BLK_TC_PC,		"pc"		},
1590 	{ BLK_TC_AHEAD,		"ahead"		},
1591 	{ BLK_TC_META,		"meta"		},
1592 	{ BLK_TC_DISCARD,	"discard"	},
1593 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1594 };
1595 
1596 static int blk_trace_str2mask(const char *str)
1597 {
1598 	int i;
1599 	int mask = 0;
1600 	char *buf, *s, *token;
1601 
1602 	buf = kstrdup(str, GFP_KERNEL);
1603 	if (buf == NULL)
1604 		return -ENOMEM;
1605 	s = strstrip(buf);
1606 
1607 	while (1) {
1608 		token = strsep(&s, ",");
1609 		if (token == NULL)
1610 			break;
1611 
1612 		if (*token == '\0')
1613 			continue;
1614 
1615 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1616 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1617 				mask |= mask_maps[i].mask;
1618 				break;
1619 			}
1620 		}
1621 		if (i == ARRAY_SIZE(mask_maps)) {
1622 			mask = -EINVAL;
1623 			break;
1624 		}
1625 	}
1626 	kfree(buf);
1627 
1628 	return mask;
1629 }
1630 
1631 static ssize_t blk_trace_mask2str(char *buf, int mask)
1632 {
1633 	int i;
1634 	char *p = buf;
1635 
1636 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1637 		if (mask & mask_maps[i].mask) {
1638 			p += sprintf(p, "%s%s",
1639 				    (p == buf) ? "" : ",", mask_maps[i].str);
1640 		}
1641 	}
1642 	*p++ = '\n';
1643 
1644 	return p - buf;
1645 }
1646 
1647 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1648 {
1649 	if (bdev->bd_disk == NULL)
1650 		return NULL;
1651 
1652 	return bdev_get_queue(bdev);
1653 }
1654 
1655 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1656 					 struct device_attribute *attr,
1657 					 char *buf)
1658 {
1659 	struct hd_struct *p = dev_to_part(dev);
1660 	struct request_queue *q;
1661 	struct block_device *bdev;
1662 	ssize_t ret = -ENXIO;
1663 
1664 	bdev = bdget(part_devt(p));
1665 	if (bdev == NULL)
1666 		goto out;
1667 
1668 	q = blk_trace_get_queue(bdev);
1669 	if (q == NULL)
1670 		goto out_bdput;
1671 
1672 	mutex_lock(&bdev->bd_mutex);
1673 
1674 	if (attr == &dev_attr_enable) {
1675 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1676 		goto out_unlock_bdev;
1677 	}
1678 
1679 	if (q->blk_trace == NULL)
1680 		ret = sprintf(buf, "disabled\n");
1681 	else if (attr == &dev_attr_act_mask)
1682 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1683 	else if (attr == &dev_attr_pid)
1684 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1685 	else if (attr == &dev_attr_start_lba)
1686 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1687 	else if (attr == &dev_attr_end_lba)
1688 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1689 
1690 out_unlock_bdev:
1691 	mutex_unlock(&bdev->bd_mutex);
1692 out_bdput:
1693 	bdput(bdev);
1694 out:
1695 	return ret;
1696 }
1697 
1698 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1699 					  struct device_attribute *attr,
1700 					  const char *buf, size_t count)
1701 {
1702 	struct block_device *bdev;
1703 	struct request_queue *q;
1704 	struct hd_struct *p;
1705 	u64 value;
1706 	ssize_t ret = -EINVAL;
1707 
1708 	if (count == 0)
1709 		goto out;
1710 
1711 	if (attr == &dev_attr_act_mask) {
1712 		if (sscanf(buf, "%llx", &value) != 1) {
1713 			/* Assume it is a list of trace category names */
1714 			ret = blk_trace_str2mask(buf);
1715 			if (ret < 0)
1716 				goto out;
1717 			value = ret;
1718 		}
1719 	} else if (sscanf(buf, "%llu", &value) != 1)
1720 		goto out;
1721 
1722 	ret = -ENXIO;
1723 
1724 	p = dev_to_part(dev);
1725 	bdev = bdget(part_devt(p));
1726 	if (bdev == NULL)
1727 		goto out;
1728 
1729 	q = blk_trace_get_queue(bdev);
1730 	if (q == NULL)
1731 		goto out_bdput;
1732 
1733 	mutex_lock(&bdev->bd_mutex);
1734 
1735 	if (attr == &dev_attr_enable) {
1736 		if (value)
1737 			ret = blk_trace_setup_queue(q, bdev);
1738 		else
1739 			ret = blk_trace_remove_queue(q);
1740 		goto out_unlock_bdev;
1741 	}
1742 
1743 	ret = 0;
1744 	if (q->blk_trace == NULL)
1745 		ret = blk_trace_setup_queue(q, bdev);
1746 
1747 	if (ret == 0) {
1748 		if (attr == &dev_attr_act_mask)
1749 			q->blk_trace->act_mask = value;
1750 		else if (attr == &dev_attr_pid)
1751 			q->blk_trace->pid = value;
1752 		else if (attr == &dev_attr_start_lba)
1753 			q->blk_trace->start_lba = value;
1754 		else if (attr == &dev_attr_end_lba)
1755 			q->blk_trace->end_lba = value;
1756 	}
1757 
1758 out_unlock_bdev:
1759 	mutex_unlock(&bdev->bd_mutex);
1760 out_bdput:
1761 	bdput(bdev);
1762 out:
1763 	return ret ? ret : count;
1764 }
1765 
1766 int blk_trace_init_sysfs(struct device *dev)
1767 {
1768 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1769 }
1770 
1771 void blk_trace_remove_sysfs(struct device *dev)
1772 {
1773 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1774 }
1775 
1776 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1777 
1778 #ifdef CONFIG_EVENT_TRACING
1779 
1780 void blk_dump_cmd(char *buf, struct request *rq)
1781 {
1782 	int i, end;
1783 	int len = rq->cmd_len;
1784 	unsigned char *cmd = rq->cmd;
1785 
1786 	if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1787 		buf[0] = '\0';
1788 		return;
1789 	}
1790 
1791 	for (end = len - 1; end >= 0; end--)
1792 		if (cmd[end])
1793 			break;
1794 	end++;
1795 
1796 	for (i = 0; i < len; i++) {
1797 		buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1798 		if (i == end && end != len - 1) {
1799 			sprintf(buf, " ..");
1800 			break;
1801 		}
1802 	}
1803 }
1804 
1805 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1806 {
1807 	int i = 0;
1808 
1809 	if (rw & WRITE)
1810 		rwbs[i++] = 'W';
1811 	else if (rw & REQ_DISCARD)
1812 		rwbs[i++] = 'D';
1813 	else if (bytes)
1814 		rwbs[i++] = 'R';
1815 	else
1816 		rwbs[i++] = 'N';
1817 
1818 	if (rw & REQ_RAHEAD)
1819 		rwbs[i++] = 'A';
1820 	if (rw & REQ_SYNC)
1821 		rwbs[i++] = 'S';
1822 	if (rw & REQ_META)
1823 		rwbs[i++] = 'M';
1824 	if (rw & REQ_SECURE)
1825 		rwbs[i++] = 'E';
1826 
1827 	rwbs[i] = '\0';
1828 }
1829 
1830 #endif /* CONFIG_EVENT_TRACING */
1831 
1832