xref: /openbmc/linux/kernel/trace/blktrace.c (revision 6dfcd296)
1 /*
2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
16  *
17  */
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/slab.h>
25 #include <linux/debugfs.h>
26 #include <linux/export.h>
27 #include <linux/time.h>
28 #include <linux/uaccess.h>
29 #include <linux/list.h>
30 
31 #include <trace/events/block.h>
32 
33 #include "trace_output.h"
34 
35 #ifdef CONFIG_BLK_DEV_IO_TRACE
36 
37 static unsigned int blktrace_seq __read_mostly = 1;
38 
39 static struct trace_array *blk_tr;
40 static bool blk_tracer_enabled __read_mostly;
41 
42 static LIST_HEAD(running_trace_list);
43 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
44 
45 /* Select an alternative, minimalistic output than the original one */
46 #define TRACE_BLK_OPT_CLASSIC	0x1
47 
48 static struct tracer_opt blk_tracer_opts[] = {
49 	/* Default disable the minimalistic output */
50 	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
51 	{ }
52 };
53 
54 static struct tracer_flags blk_tracer_flags = {
55 	.val  = 0,
56 	.opts = blk_tracer_opts,
57 };
58 
59 /* Global reference count of probes */
60 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
61 
62 static void blk_register_tracepoints(void);
63 static void blk_unregister_tracepoints(void);
64 
65 /*
66  * Send out a notify message.
67  */
68 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
69 		       const void *data, size_t len)
70 {
71 	struct blk_io_trace *t;
72 	struct ring_buffer_event *event = NULL;
73 	struct ring_buffer *buffer = NULL;
74 	int pc = 0;
75 	int cpu = smp_processor_id();
76 	bool blk_tracer = blk_tracer_enabled;
77 
78 	if (blk_tracer) {
79 		buffer = blk_tr->trace_buffer.buffer;
80 		pc = preempt_count();
81 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
82 						  sizeof(*t) + len,
83 						  0, pc);
84 		if (!event)
85 			return;
86 		t = ring_buffer_event_data(event);
87 		goto record_it;
88 	}
89 
90 	if (!bt->rchan)
91 		return;
92 
93 	t = relay_reserve(bt->rchan, sizeof(*t) + len);
94 	if (t) {
95 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
96 		t->time = ktime_to_ns(ktime_get());
97 record_it:
98 		t->device = bt->dev;
99 		t->action = action;
100 		t->pid = pid;
101 		t->cpu = cpu;
102 		t->pdu_len = len;
103 		memcpy((void *) t + sizeof(*t), data, len);
104 
105 		if (blk_tracer)
106 			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
107 	}
108 }
109 
110 /*
111  * Send out a notify for this process, if we haven't done so since a trace
112  * started
113  */
114 static void trace_note_tsk(struct task_struct *tsk)
115 {
116 	unsigned long flags;
117 	struct blk_trace *bt;
118 
119 	tsk->btrace_seq = blktrace_seq;
120 	spin_lock_irqsave(&running_trace_lock, flags);
121 	list_for_each_entry(bt, &running_trace_list, running_list) {
122 		trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
123 			   sizeof(tsk->comm));
124 	}
125 	spin_unlock_irqrestore(&running_trace_lock, flags);
126 }
127 
128 static void trace_note_time(struct blk_trace *bt)
129 {
130 	struct timespec64 now;
131 	unsigned long flags;
132 	u32 words[2];
133 
134 	/* need to check user space to see if this breaks in y2038 or y2106 */
135 	ktime_get_real_ts64(&now);
136 	words[0] = (u32)now.tv_sec;
137 	words[1] = now.tv_nsec;
138 
139 	local_irq_save(flags);
140 	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
141 	local_irq_restore(flags);
142 }
143 
144 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
145 {
146 	int n;
147 	va_list args;
148 	unsigned long flags;
149 	char *buf;
150 
151 	if (unlikely(bt->trace_state != Blktrace_running &&
152 		     !blk_tracer_enabled))
153 		return;
154 
155 	/*
156 	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
157 	 * message to the trace.
158 	 */
159 	if (!(bt->act_mask & BLK_TC_NOTIFY))
160 		return;
161 
162 	local_irq_save(flags);
163 	buf = this_cpu_ptr(bt->msg_data);
164 	va_start(args, fmt);
165 	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
166 	va_end(args);
167 
168 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
169 	local_irq_restore(flags);
170 }
171 EXPORT_SYMBOL_GPL(__trace_note_message);
172 
173 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
174 			 pid_t pid)
175 {
176 	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
177 		return 1;
178 	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
179 		return 1;
180 	if (bt->pid && pid != bt->pid)
181 		return 1;
182 
183 	return 0;
184 }
185 
186 /*
187  * Data direction bit lookup
188  */
189 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
190 				 BLK_TC_ACT(BLK_TC_WRITE) };
191 
192 #define BLK_TC_RAHEAD		BLK_TC_AHEAD
193 #define BLK_TC_PREFLUSH		BLK_TC_FLUSH
194 
195 /* The ilog2() calls fall out because they're constant */
196 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
197 	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
198 
199 /*
200  * The worker for the various blk_add_trace*() types. Fills out a
201  * blk_io_trace structure and places it in a per-cpu subbuffer.
202  */
203 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
204 		     int op, int op_flags, u32 what, int error, int pdu_len,
205 		     void *pdu_data)
206 {
207 	struct task_struct *tsk = current;
208 	struct ring_buffer_event *event = NULL;
209 	struct ring_buffer *buffer = NULL;
210 	struct blk_io_trace *t;
211 	unsigned long flags = 0;
212 	unsigned long *sequence;
213 	pid_t pid;
214 	int cpu, pc = 0;
215 	bool blk_tracer = blk_tracer_enabled;
216 
217 	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
218 		return;
219 
220 	what |= ddir_act[op_is_write(op) ? WRITE : READ];
221 	what |= MASK_TC_BIT(op_flags, SYNC);
222 	what |= MASK_TC_BIT(op_flags, RAHEAD);
223 	what |= MASK_TC_BIT(op_flags, META);
224 	what |= MASK_TC_BIT(op_flags, PREFLUSH);
225 	what |= MASK_TC_BIT(op_flags, FUA);
226 	if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
227 		what |= BLK_TC_ACT(BLK_TC_DISCARD);
228 	if (op == REQ_OP_FLUSH)
229 		what |= BLK_TC_ACT(BLK_TC_FLUSH);
230 
231 	pid = tsk->pid;
232 	if (act_log_check(bt, what, sector, pid))
233 		return;
234 	cpu = raw_smp_processor_id();
235 
236 	if (blk_tracer) {
237 		tracing_record_cmdline(current);
238 
239 		buffer = blk_tr->trace_buffer.buffer;
240 		pc = preempt_count();
241 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
242 						  sizeof(*t) + pdu_len,
243 						  0, pc);
244 		if (!event)
245 			return;
246 		t = ring_buffer_event_data(event);
247 		goto record_it;
248 	}
249 
250 	if (unlikely(tsk->btrace_seq != blktrace_seq))
251 		trace_note_tsk(tsk);
252 
253 	/*
254 	 * A word about the locking here - we disable interrupts to reserve
255 	 * some space in the relay per-cpu buffer, to prevent an irq
256 	 * from coming in and stepping on our toes.
257 	 */
258 	local_irq_save(flags);
259 	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
260 	if (t) {
261 		sequence = per_cpu_ptr(bt->sequence, cpu);
262 
263 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
264 		t->sequence = ++(*sequence);
265 		t->time = ktime_to_ns(ktime_get());
266 record_it:
267 		/*
268 		 * These two are not needed in ftrace as they are in the
269 		 * generic trace_entry, filled by tracing_generic_entry_update,
270 		 * but for the trace_event->bin() synthesizer benefit we do it
271 		 * here too.
272 		 */
273 		t->cpu = cpu;
274 		t->pid = pid;
275 
276 		t->sector = sector;
277 		t->bytes = bytes;
278 		t->action = what;
279 		t->device = bt->dev;
280 		t->error = error;
281 		t->pdu_len = pdu_len;
282 
283 		if (pdu_len)
284 			memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
285 
286 		if (blk_tracer) {
287 			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
288 			return;
289 		}
290 	}
291 
292 	local_irq_restore(flags);
293 }
294 
295 static struct dentry *blk_tree_root;
296 static DEFINE_MUTEX(blk_tree_mutex);
297 
298 static void blk_trace_free(struct blk_trace *bt)
299 {
300 	debugfs_remove(bt->msg_file);
301 	debugfs_remove(bt->dropped_file);
302 	relay_close(bt->rchan);
303 	debugfs_remove(bt->dir);
304 	free_percpu(bt->sequence);
305 	free_percpu(bt->msg_data);
306 	kfree(bt);
307 }
308 
309 static void blk_trace_cleanup(struct blk_trace *bt)
310 {
311 	blk_trace_free(bt);
312 	if (atomic_dec_and_test(&blk_probes_ref))
313 		blk_unregister_tracepoints();
314 }
315 
316 int blk_trace_remove(struct request_queue *q)
317 {
318 	struct blk_trace *bt;
319 
320 	bt = xchg(&q->blk_trace, NULL);
321 	if (!bt)
322 		return -EINVAL;
323 
324 	if (bt->trace_state != Blktrace_running)
325 		blk_trace_cleanup(bt);
326 
327 	return 0;
328 }
329 EXPORT_SYMBOL_GPL(blk_trace_remove);
330 
331 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
332 				size_t count, loff_t *ppos)
333 {
334 	struct blk_trace *bt = filp->private_data;
335 	char buf[16];
336 
337 	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
338 
339 	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
340 }
341 
342 static const struct file_operations blk_dropped_fops = {
343 	.owner =	THIS_MODULE,
344 	.open =		simple_open,
345 	.read =		blk_dropped_read,
346 	.llseek =	default_llseek,
347 };
348 
349 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
350 				size_t count, loff_t *ppos)
351 {
352 	char *msg;
353 	struct blk_trace *bt;
354 
355 	if (count >= BLK_TN_MAX_MSG)
356 		return -EINVAL;
357 
358 	msg = memdup_user_nul(buffer, count);
359 	if (IS_ERR(msg))
360 		return PTR_ERR(msg);
361 
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 =		simple_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 						   umode_t 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 *bt = NULL;
441 	struct dentry *dir = NULL;
442 	int ret;
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 	strreplace(buts->name, '/', '_');
455 
456 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
457 	if (!bt)
458 		return -ENOMEM;
459 
460 	ret = -ENOMEM;
461 	bt->sequence = alloc_percpu(unsigned long);
462 	if (!bt->sequence)
463 		goto err;
464 
465 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
466 	if (!bt->msg_data)
467 		goto err;
468 
469 	ret = -ENOENT;
470 
471 	mutex_lock(&blk_tree_mutex);
472 	if (!blk_tree_root) {
473 		blk_tree_root = debugfs_create_dir("block", NULL);
474 		if (!blk_tree_root) {
475 			mutex_unlock(&blk_tree_mutex);
476 			goto err;
477 		}
478 	}
479 	mutex_unlock(&blk_tree_mutex);
480 
481 	dir = debugfs_create_dir(buts->name, blk_tree_root);
482 
483 	if (!dir)
484 		goto err;
485 
486 	bt->dir = dir;
487 	bt->dev = dev;
488 	atomic_set(&bt->dropped, 0);
489 	INIT_LIST_HEAD(&bt->running_list);
490 
491 	ret = -EIO;
492 	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
493 					       &blk_dropped_fops);
494 	if (!bt->dropped_file)
495 		goto err;
496 
497 	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
498 	if (!bt->msg_file)
499 		goto err;
500 
501 	bt->rchan = relay_open("trace", dir, buts->buf_size,
502 				buts->buf_nr, &blk_relay_callbacks, bt);
503 	if (!bt->rchan)
504 		goto err;
505 
506 	bt->act_mask = buts->act_mask;
507 	if (!bt->act_mask)
508 		bt->act_mask = (u16) -1;
509 
510 	blk_trace_setup_lba(bt, bdev);
511 
512 	/* overwrite with user settings */
513 	if (buts->start_lba)
514 		bt->start_lba = buts->start_lba;
515 	if (buts->end_lba)
516 		bt->end_lba = buts->end_lba;
517 
518 	bt->pid = buts->pid;
519 	bt->trace_state = Blktrace_setup;
520 
521 	ret = -EBUSY;
522 	if (cmpxchg(&q->blk_trace, NULL, bt))
523 		goto err;
524 
525 	if (atomic_inc_return(&blk_probes_ref) == 1)
526 		blk_register_tracepoints();
527 
528 	return 0;
529 err:
530 	blk_trace_free(bt);
531 	return ret;
532 }
533 
534 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
535 		    struct block_device *bdev,
536 		    char __user *arg)
537 {
538 	struct blk_user_trace_setup buts;
539 	int ret;
540 
541 	ret = copy_from_user(&buts, arg, sizeof(buts));
542 	if (ret)
543 		return -EFAULT;
544 
545 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
546 	if (ret)
547 		return ret;
548 
549 	if (copy_to_user(arg, &buts, sizeof(buts))) {
550 		blk_trace_remove(q);
551 		return -EFAULT;
552 	}
553 	return 0;
554 }
555 EXPORT_SYMBOL_GPL(blk_trace_setup);
556 
557 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
558 static int compat_blk_trace_setup(struct request_queue *q, char *name,
559 				  dev_t dev, struct block_device *bdev,
560 				  char __user *arg)
561 {
562 	struct blk_user_trace_setup buts;
563 	struct compat_blk_user_trace_setup cbuts;
564 	int ret;
565 
566 	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
567 		return -EFAULT;
568 
569 	buts = (struct blk_user_trace_setup) {
570 		.act_mask = cbuts.act_mask,
571 		.buf_size = cbuts.buf_size,
572 		.buf_nr = cbuts.buf_nr,
573 		.start_lba = cbuts.start_lba,
574 		.end_lba = cbuts.end_lba,
575 		.pid = cbuts.pid,
576 	};
577 
578 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
579 	if (ret)
580 		return ret;
581 
582 	if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
583 		blk_trace_remove(q);
584 		return -EFAULT;
585 	}
586 
587 	return 0;
588 }
589 #endif
590 
591 int blk_trace_startstop(struct request_queue *q, int start)
592 {
593 	int ret;
594 	struct blk_trace *bt = q->blk_trace;
595 
596 	if (bt == NULL)
597 		return -EINVAL;
598 
599 	/*
600 	 * For starting a trace, we can transition from a setup or stopped
601 	 * trace. For stopping a trace, the state must be running
602 	 */
603 	ret = -EINVAL;
604 	if (start) {
605 		if (bt->trace_state == Blktrace_setup ||
606 		    bt->trace_state == Blktrace_stopped) {
607 			blktrace_seq++;
608 			smp_mb();
609 			bt->trace_state = Blktrace_running;
610 			spin_lock_irq(&running_trace_lock);
611 			list_add(&bt->running_list, &running_trace_list);
612 			spin_unlock_irq(&running_trace_lock);
613 
614 			trace_note_time(bt);
615 			ret = 0;
616 		}
617 	} else {
618 		if (bt->trace_state == Blktrace_running) {
619 			bt->trace_state = Blktrace_stopped;
620 			spin_lock_irq(&running_trace_lock);
621 			list_del_init(&bt->running_list);
622 			spin_unlock_irq(&running_trace_lock);
623 			relay_flush(bt->rchan);
624 			ret = 0;
625 		}
626 	}
627 
628 	return ret;
629 }
630 EXPORT_SYMBOL_GPL(blk_trace_startstop);
631 
632 /**
633  * blk_trace_ioctl: - handle the ioctls associated with tracing
634  * @bdev:	the block device
635  * @cmd:	the ioctl cmd
636  * @arg:	the argument data, if any
637  *
638  **/
639 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
640 {
641 	struct request_queue *q;
642 	int ret, start = 0;
643 	char b[BDEVNAME_SIZE];
644 
645 	q = bdev_get_queue(bdev);
646 	if (!q)
647 		return -ENXIO;
648 
649 	mutex_lock(&bdev->bd_mutex);
650 
651 	switch (cmd) {
652 	case BLKTRACESETUP:
653 		bdevname(bdev, b);
654 		ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
655 		break;
656 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
657 	case BLKTRACESETUP32:
658 		bdevname(bdev, b);
659 		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
660 		break;
661 #endif
662 	case BLKTRACESTART:
663 		start = 1;
664 	case BLKTRACESTOP:
665 		ret = blk_trace_startstop(q, start);
666 		break;
667 	case BLKTRACETEARDOWN:
668 		ret = blk_trace_remove(q);
669 		break;
670 	default:
671 		ret = -ENOTTY;
672 		break;
673 	}
674 
675 	mutex_unlock(&bdev->bd_mutex);
676 	return ret;
677 }
678 
679 /**
680  * blk_trace_shutdown: - stop and cleanup trace structures
681  * @q:    the request queue associated with the device
682  *
683  **/
684 void blk_trace_shutdown(struct request_queue *q)
685 {
686 	if (q->blk_trace) {
687 		blk_trace_startstop(q, 0);
688 		blk_trace_remove(q);
689 	}
690 }
691 
692 /*
693  * blktrace probes
694  */
695 
696 /**
697  * blk_add_trace_rq - Add a trace for a request oriented action
698  * @q:		queue the io is for
699  * @rq:		the source request
700  * @nr_bytes:	number of completed bytes
701  * @what:	the action
702  *
703  * Description:
704  *     Records an action against a request. Will log the bio offset + size.
705  *
706  **/
707 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
708 			     unsigned int nr_bytes, u32 what)
709 {
710 	struct blk_trace *bt = q->blk_trace;
711 
712 	if (likely(!bt))
713 		return;
714 
715 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
716 		what |= BLK_TC_ACT(BLK_TC_PC);
717 		__blk_add_trace(bt, 0, nr_bytes, req_op(rq), rq->cmd_flags,
718 				what, rq->errors, rq->cmd_len, rq->cmd);
719 	} else  {
720 		what |= BLK_TC_ACT(BLK_TC_FS);
721 		__blk_add_trace(bt, blk_rq_pos(rq), nr_bytes, req_op(rq),
722 				rq->cmd_flags, what, rq->errors, 0, NULL);
723 	}
724 }
725 
726 static void blk_add_trace_rq_abort(void *ignore,
727 				   struct request_queue *q, struct request *rq)
728 {
729 	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
730 }
731 
732 static void blk_add_trace_rq_insert(void *ignore,
733 				    struct request_queue *q, struct request *rq)
734 {
735 	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
736 }
737 
738 static void blk_add_trace_rq_issue(void *ignore,
739 				   struct request_queue *q, struct request *rq)
740 {
741 	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
742 }
743 
744 static void blk_add_trace_rq_requeue(void *ignore,
745 				     struct request_queue *q,
746 				     struct request *rq)
747 {
748 	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
749 }
750 
751 static void blk_add_trace_rq_complete(void *ignore,
752 				      struct request_queue *q,
753 				      struct request *rq,
754 				      unsigned int nr_bytes)
755 {
756 	blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
757 }
758 
759 /**
760  * blk_add_trace_bio - Add a trace for a bio oriented action
761  * @q:		queue the io is for
762  * @bio:	the source bio
763  * @what:	the action
764  * @error:	error, if any
765  *
766  * Description:
767  *     Records an action against a bio. Will log the bio offset + size.
768  *
769  **/
770 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
771 			      u32 what, int error)
772 {
773 	struct blk_trace *bt = q->blk_trace;
774 
775 	if (likely(!bt))
776 		return;
777 
778 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
779 			bio_op(bio), bio->bi_opf, what, error, 0, NULL);
780 }
781 
782 static void blk_add_trace_bio_bounce(void *ignore,
783 				     struct request_queue *q, struct bio *bio)
784 {
785 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
786 }
787 
788 static void blk_add_trace_bio_complete(void *ignore,
789 				       struct request_queue *q, struct bio *bio,
790 				       int error)
791 {
792 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
793 }
794 
795 static void blk_add_trace_bio_backmerge(void *ignore,
796 					struct request_queue *q,
797 					struct request *rq,
798 					struct bio *bio)
799 {
800 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
801 }
802 
803 static void blk_add_trace_bio_frontmerge(void *ignore,
804 					 struct request_queue *q,
805 					 struct request *rq,
806 					 struct bio *bio)
807 {
808 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
809 }
810 
811 static void blk_add_trace_bio_queue(void *ignore,
812 				    struct request_queue *q, struct bio *bio)
813 {
814 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
815 }
816 
817 static void blk_add_trace_getrq(void *ignore,
818 				struct request_queue *q,
819 				struct bio *bio, int rw)
820 {
821 	if (bio)
822 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
823 	else {
824 		struct blk_trace *bt = q->blk_trace;
825 
826 		if (bt)
827 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
828 					NULL);
829 	}
830 }
831 
832 
833 static void blk_add_trace_sleeprq(void *ignore,
834 				  struct request_queue *q,
835 				  struct bio *bio, int rw)
836 {
837 	if (bio)
838 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
839 	else {
840 		struct blk_trace *bt = q->blk_trace;
841 
842 		if (bt)
843 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
844 					0, 0, NULL);
845 	}
846 }
847 
848 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
849 {
850 	struct blk_trace *bt = q->blk_trace;
851 
852 	if (bt)
853 		__blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
854 }
855 
856 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
857 				    unsigned int depth, bool explicit)
858 {
859 	struct blk_trace *bt = q->blk_trace;
860 
861 	if (bt) {
862 		__be64 rpdu = cpu_to_be64(depth);
863 		u32 what;
864 
865 		if (explicit)
866 			what = BLK_TA_UNPLUG_IO;
867 		else
868 			what = BLK_TA_UNPLUG_TIMER;
869 
870 		__blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
871 	}
872 }
873 
874 static void blk_add_trace_split(void *ignore,
875 				struct request_queue *q, struct bio *bio,
876 				unsigned int pdu)
877 {
878 	struct blk_trace *bt = q->blk_trace;
879 
880 	if (bt) {
881 		__be64 rpdu = cpu_to_be64(pdu);
882 
883 		__blk_add_trace(bt, bio->bi_iter.bi_sector,
884 				bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
885 				BLK_TA_SPLIT, bio->bi_error, sizeof(rpdu),
886 				&rpdu);
887 	}
888 }
889 
890 /**
891  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
892  * @ignore:	trace callback data parameter (not used)
893  * @q:		queue the io is for
894  * @bio:	the source bio
895  * @dev:	target device
896  * @from:	source sector
897  *
898  * Description:
899  *     Device mapper or raid target sometimes need to split a bio because
900  *     it spans a stripe (or similar). Add a trace for that action.
901  *
902  **/
903 static void blk_add_trace_bio_remap(void *ignore,
904 				    struct request_queue *q, struct bio *bio,
905 				    dev_t dev, sector_t from)
906 {
907 	struct blk_trace *bt = q->blk_trace;
908 	struct blk_io_trace_remap r;
909 
910 	if (likely(!bt))
911 		return;
912 
913 	r.device_from = cpu_to_be32(dev);
914 	r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
915 	r.sector_from = cpu_to_be64(from);
916 
917 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
918 			bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_error,
919 			sizeof(r), &r);
920 }
921 
922 /**
923  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
924  * @ignore:	trace callback data parameter (not used)
925  * @q:		queue the io is for
926  * @rq:		the source request
927  * @dev:	target device
928  * @from:	source sector
929  *
930  * Description:
931  *     Device mapper remaps request to other devices.
932  *     Add a trace for that action.
933  *
934  **/
935 static void blk_add_trace_rq_remap(void *ignore,
936 				   struct request_queue *q,
937 				   struct request *rq, dev_t dev,
938 				   sector_t from)
939 {
940 	struct blk_trace *bt = q->blk_trace;
941 	struct blk_io_trace_remap r;
942 
943 	if (likely(!bt))
944 		return;
945 
946 	r.device_from = cpu_to_be32(dev);
947 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
948 	r.sector_from = cpu_to_be64(from);
949 
950 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
951 			rq_data_dir(rq), 0, BLK_TA_REMAP, !!rq->errors,
952 			sizeof(r), &r);
953 }
954 
955 /**
956  * blk_add_driver_data - Add binary message with driver-specific data
957  * @q:		queue the io is for
958  * @rq:		io request
959  * @data:	driver-specific data
960  * @len:	length of driver-specific data
961  *
962  * Description:
963  *     Some drivers might want to write driver-specific data per request.
964  *
965  **/
966 void blk_add_driver_data(struct request_queue *q,
967 			 struct request *rq,
968 			 void *data, size_t len)
969 {
970 	struct blk_trace *bt = q->blk_trace;
971 
972 	if (likely(!bt))
973 		return;
974 
975 	if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
976 		__blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, 0,
977 				BLK_TA_DRV_DATA, rq->errors, len, data);
978 	else
979 		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, 0,
980 				BLK_TA_DRV_DATA, rq->errors, len, data);
981 }
982 EXPORT_SYMBOL_GPL(blk_add_driver_data);
983 
984 static void blk_register_tracepoints(void)
985 {
986 	int ret;
987 
988 	ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
989 	WARN_ON(ret);
990 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
991 	WARN_ON(ret);
992 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
993 	WARN_ON(ret);
994 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
995 	WARN_ON(ret);
996 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
997 	WARN_ON(ret);
998 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
999 	WARN_ON(ret);
1000 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1001 	WARN_ON(ret);
1002 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1003 	WARN_ON(ret);
1004 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1005 	WARN_ON(ret);
1006 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1007 	WARN_ON(ret);
1008 	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1009 	WARN_ON(ret);
1010 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1011 	WARN_ON(ret);
1012 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1013 	WARN_ON(ret);
1014 	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1015 	WARN_ON(ret);
1016 	ret = register_trace_block_split(blk_add_trace_split, NULL);
1017 	WARN_ON(ret);
1018 	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1019 	WARN_ON(ret);
1020 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1021 	WARN_ON(ret);
1022 }
1023 
1024 static void blk_unregister_tracepoints(void)
1025 {
1026 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1027 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1028 	unregister_trace_block_split(blk_add_trace_split, NULL);
1029 	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1030 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1031 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1032 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1033 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1034 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1035 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1036 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1037 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1038 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1039 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1040 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1041 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1042 	unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1043 
1044 	tracepoint_synchronize_unregister();
1045 }
1046 
1047 /*
1048  * struct blk_io_tracer formatting routines
1049  */
1050 
1051 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1052 {
1053 	int i = 0;
1054 	int tc = t->action >> BLK_TC_SHIFT;
1055 
1056 	if (t->action == BLK_TN_MESSAGE) {
1057 		rwbs[i++] = 'N';
1058 		goto out;
1059 	}
1060 
1061 	if (tc & BLK_TC_FLUSH)
1062 		rwbs[i++] = 'F';
1063 
1064 	if (tc & BLK_TC_DISCARD)
1065 		rwbs[i++] = 'D';
1066 	else if (tc & BLK_TC_WRITE)
1067 		rwbs[i++] = 'W';
1068 	else if (t->bytes)
1069 		rwbs[i++] = 'R';
1070 	else
1071 		rwbs[i++] = 'N';
1072 
1073 	if (tc & BLK_TC_FUA)
1074 		rwbs[i++] = 'F';
1075 	if (tc & BLK_TC_AHEAD)
1076 		rwbs[i++] = 'A';
1077 	if (tc & BLK_TC_SYNC)
1078 		rwbs[i++] = 'S';
1079 	if (tc & BLK_TC_META)
1080 		rwbs[i++] = 'M';
1081 out:
1082 	rwbs[i] = '\0';
1083 }
1084 
1085 static inline
1086 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1087 {
1088 	return (const struct blk_io_trace *)ent;
1089 }
1090 
1091 static inline const void *pdu_start(const struct trace_entry *ent)
1092 {
1093 	return te_blk_io_trace(ent) + 1;
1094 }
1095 
1096 static inline u32 t_action(const struct trace_entry *ent)
1097 {
1098 	return te_blk_io_trace(ent)->action;
1099 }
1100 
1101 static inline u32 t_bytes(const struct trace_entry *ent)
1102 {
1103 	return te_blk_io_trace(ent)->bytes;
1104 }
1105 
1106 static inline u32 t_sec(const struct trace_entry *ent)
1107 {
1108 	return te_blk_io_trace(ent)->bytes >> 9;
1109 }
1110 
1111 static inline unsigned long long t_sector(const struct trace_entry *ent)
1112 {
1113 	return te_blk_io_trace(ent)->sector;
1114 }
1115 
1116 static inline __u16 t_error(const struct trace_entry *ent)
1117 {
1118 	return te_blk_io_trace(ent)->error;
1119 }
1120 
1121 static __u64 get_pdu_int(const struct trace_entry *ent)
1122 {
1123 	const __u64 *val = pdu_start(ent);
1124 	return be64_to_cpu(*val);
1125 }
1126 
1127 static void get_pdu_remap(const struct trace_entry *ent,
1128 			  struct blk_io_trace_remap *r)
1129 {
1130 	const struct blk_io_trace_remap *__r = pdu_start(ent);
1131 	__u64 sector_from = __r->sector_from;
1132 
1133 	r->device_from = be32_to_cpu(__r->device_from);
1134 	r->device_to   = be32_to_cpu(__r->device_to);
1135 	r->sector_from = be64_to_cpu(sector_from);
1136 }
1137 
1138 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1139 
1140 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1141 {
1142 	char rwbs[RWBS_LEN];
1143 	unsigned long long ts  = iter->ts;
1144 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1145 	unsigned secs	       = (unsigned long)ts;
1146 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1147 
1148 	fill_rwbs(rwbs, t);
1149 
1150 	trace_seq_printf(&iter->seq,
1151 			 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1152 			 MAJOR(t->device), MINOR(t->device), iter->cpu,
1153 			 secs, nsec_rem, iter->ent->pid, act, rwbs);
1154 }
1155 
1156 static void blk_log_action(struct trace_iterator *iter, const char *act)
1157 {
1158 	char rwbs[RWBS_LEN];
1159 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1160 
1161 	fill_rwbs(rwbs, t);
1162 	trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1163 			 MAJOR(t->device), MINOR(t->device), act, rwbs);
1164 }
1165 
1166 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1167 {
1168 	const unsigned char *pdu_buf;
1169 	int pdu_len;
1170 	int i, end;
1171 
1172 	pdu_buf = pdu_start(ent);
1173 	pdu_len = te_blk_io_trace(ent)->pdu_len;
1174 
1175 	if (!pdu_len)
1176 		return;
1177 
1178 	/* find the last zero that needs to be printed */
1179 	for (end = pdu_len - 1; end >= 0; end--)
1180 		if (pdu_buf[end])
1181 			break;
1182 	end++;
1183 
1184 	trace_seq_putc(s, '(');
1185 
1186 	for (i = 0; i < pdu_len; i++) {
1187 
1188 		trace_seq_printf(s, "%s%02x",
1189 				 i == 0 ? "" : " ", pdu_buf[i]);
1190 
1191 		/*
1192 		 * stop when the rest is just zeroes and indicate so
1193 		 * with a ".." appended
1194 		 */
1195 		if (i == end && end != pdu_len - 1) {
1196 			trace_seq_puts(s, " ..) ");
1197 			return;
1198 		}
1199 	}
1200 
1201 	trace_seq_puts(s, ") ");
1202 }
1203 
1204 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1205 {
1206 	char cmd[TASK_COMM_LEN];
1207 
1208 	trace_find_cmdline(ent->pid, cmd);
1209 
1210 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1211 		trace_seq_printf(s, "%u ", t_bytes(ent));
1212 		blk_log_dump_pdu(s, ent);
1213 		trace_seq_printf(s, "[%s]\n", cmd);
1214 	} else {
1215 		if (t_sec(ent))
1216 			trace_seq_printf(s, "%llu + %u [%s]\n",
1217 						t_sector(ent), t_sec(ent), cmd);
1218 		else
1219 			trace_seq_printf(s, "[%s]\n", cmd);
1220 	}
1221 }
1222 
1223 static void blk_log_with_error(struct trace_seq *s,
1224 			      const struct trace_entry *ent)
1225 {
1226 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1227 		blk_log_dump_pdu(s, ent);
1228 		trace_seq_printf(s, "[%d]\n", t_error(ent));
1229 	} else {
1230 		if (t_sec(ent))
1231 			trace_seq_printf(s, "%llu + %u [%d]\n",
1232 					 t_sector(ent),
1233 					 t_sec(ent), t_error(ent));
1234 		else
1235 			trace_seq_printf(s, "%llu [%d]\n",
1236 					 t_sector(ent), t_error(ent));
1237 	}
1238 }
1239 
1240 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1241 {
1242 	struct blk_io_trace_remap r = { .device_from = 0, };
1243 
1244 	get_pdu_remap(ent, &r);
1245 	trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1246 			 t_sector(ent), t_sec(ent),
1247 			 MAJOR(r.device_from), MINOR(r.device_from),
1248 			 (unsigned long long)r.sector_from);
1249 }
1250 
1251 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1252 {
1253 	char cmd[TASK_COMM_LEN];
1254 
1255 	trace_find_cmdline(ent->pid, cmd);
1256 
1257 	trace_seq_printf(s, "[%s]\n", cmd);
1258 }
1259 
1260 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1261 {
1262 	char cmd[TASK_COMM_LEN];
1263 
1264 	trace_find_cmdline(ent->pid, cmd);
1265 
1266 	trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1267 }
1268 
1269 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1270 {
1271 	char cmd[TASK_COMM_LEN];
1272 
1273 	trace_find_cmdline(ent->pid, cmd);
1274 
1275 	trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1276 			 get_pdu_int(ent), cmd);
1277 }
1278 
1279 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1280 {
1281 	const struct blk_io_trace *t = te_blk_io_trace(ent);
1282 
1283 	trace_seq_putmem(s, t + 1, t->pdu_len);
1284 	trace_seq_putc(s, '\n');
1285 }
1286 
1287 /*
1288  * struct tracer operations
1289  */
1290 
1291 static void blk_tracer_print_header(struct seq_file *m)
1292 {
1293 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1294 		return;
1295 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1296 		    "#  |     |     |           |   |   |\n");
1297 }
1298 
1299 static void blk_tracer_start(struct trace_array *tr)
1300 {
1301 	blk_tracer_enabled = true;
1302 }
1303 
1304 static int blk_tracer_init(struct trace_array *tr)
1305 {
1306 	blk_tr = tr;
1307 	blk_tracer_start(tr);
1308 	return 0;
1309 }
1310 
1311 static void blk_tracer_stop(struct trace_array *tr)
1312 {
1313 	blk_tracer_enabled = false;
1314 }
1315 
1316 static void blk_tracer_reset(struct trace_array *tr)
1317 {
1318 	blk_tracer_stop(tr);
1319 }
1320 
1321 static const struct {
1322 	const char *act[2];
1323 	void	   (*print)(struct trace_seq *s, const struct trace_entry *ent);
1324 } what2act[] = {
1325 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1326 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1327 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1328 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1329 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1330 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1331 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1332 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1333 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1334 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1335 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1336 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1337 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1338 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1339 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1340 };
1341 
1342 static enum print_line_t print_one_line(struct trace_iterator *iter,
1343 					bool classic)
1344 {
1345 	struct trace_array *tr = iter->tr;
1346 	struct trace_seq *s = &iter->seq;
1347 	const struct blk_io_trace *t;
1348 	u16 what;
1349 	bool long_act;
1350 	blk_log_action_t *log_action;
1351 
1352 	t	   = te_blk_io_trace(iter->ent);
1353 	what	   = t->action & ((1 << BLK_TC_SHIFT) - 1);
1354 	long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1355 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1356 
1357 	if (t->action == BLK_TN_MESSAGE) {
1358 		log_action(iter, long_act ? "message" : "m");
1359 		blk_log_msg(s, iter->ent);
1360 		return trace_handle_return(s);
1361 	}
1362 
1363 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1364 		trace_seq_printf(s, "Unknown action %x\n", what);
1365 	else {
1366 		log_action(iter, what2act[what].act[long_act]);
1367 		what2act[what].print(s, iter->ent);
1368 	}
1369 
1370 	return trace_handle_return(s);
1371 }
1372 
1373 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1374 					       int flags, struct trace_event *event)
1375 {
1376 	return print_one_line(iter, false);
1377 }
1378 
1379 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1380 {
1381 	struct trace_seq *s = &iter->seq;
1382 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1383 	const int offset = offsetof(struct blk_io_trace, sector);
1384 	struct blk_io_trace old = {
1385 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1386 		.time     = iter->ts,
1387 	};
1388 
1389 	trace_seq_putmem(s, &old, offset);
1390 	trace_seq_putmem(s, &t->sector,
1391 			 sizeof(old) - offset + t->pdu_len);
1392 }
1393 
1394 static enum print_line_t
1395 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1396 			     struct trace_event *event)
1397 {
1398 	blk_trace_synthesize_old_trace(iter);
1399 
1400 	return trace_handle_return(&iter->seq);
1401 }
1402 
1403 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1404 {
1405 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1406 		return TRACE_TYPE_UNHANDLED;
1407 
1408 	return print_one_line(iter, true);
1409 }
1410 
1411 static int
1412 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1413 {
1414 	/* don't output context-info for blk_classic output */
1415 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1416 		if (set)
1417 			tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1418 		else
1419 			tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1420 	}
1421 	return 0;
1422 }
1423 
1424 static struct tracer blk_tracer __read_mostly = {
1425 	.name		= "blk",
1426 	.init		= blk_tracer_init,
1427 	.reset		= blk_tracer_reset,
1428 	.start		= blk_tracer_start,
1429 	.stop		= blk_tracer_stop,
1430 	.print_header	= blk_tracer_print_header,
1431 	.print_line	= blk_tracer_print_line,
1432 	.flags		= &blk_tracer_flags,
1433 	.set_flag	= blk_tracer_set_flag,
1434 };
1435 
1436 static struct trace_event_functions trace_blk_event_funcs = {
1437 	.trace		= blk_trace_event_print,
1438 	.binary		= blk_trace_event_print_binary,
1439 };
1440 
1441 static struct trace_event trace_blk_event = {
1442 	.type		= TRACE_BLK,
1443 	.funcs		= &trace_blk_event_funcs,
1444 };
1445 
1446 static int __init init_blk_tracer(void)
1447 {
1448 	if (!register_trace_event(&trace_blk_event)) {
1449 		pr_warn("Warning: could not register block events\n");
1450 		return 1;
1451 	}
1452 
1453 	if (register_tracer(&blk_tracer) != 0) {
1454 		pr_warn("Warning: could not register the block tracer\n");
1455 		unregister_trace_event(&trace_blk_event);
1456 		return 1;
1457 	}
1458 
1459 	return 0;
1460 }
1461 
1462 device_initcall(init_blk_tracer);
1463 
1464 static int blk_trace_remove_queue(struct request_queue *q)
1465 {
1466 	struct blk_trace *bt;
1467 
1468 	bt = xchg(&q->blk_trace, NULL);
1469 	if (bt == NULL)
1470 		return -EINVAL;
1471 
1472 	if (atomic_dec_and_test(&blk_probes_ref))
1473 		blk_unregister_tracepoints();
1474 
1475 	blk_trace_free(bt);
1476 	return 0;
1477 }
1478 
1479 /*
1480  * Setup everything required to start tracing
1481  */
1482 static int blk_trace_setup_queue(struct request_queue *q,
1483 				 struct block_device *bdev)
1484 {
1485 	struct blk_trace *bt = NULL;
1486 	int ret = -ENOMEM;
1487 
1488 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1489 	if (!bt)
1490 		return -ENOMEM;
1491 
1492 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1493 	if (!bt->msg_data)
1494 		goto free_bt;
1495 
1496 	bt->dev = bdev->bd_dev;
1497 	bt->act_mask = (u16)-1;
1498 
1499 	blk_trace_setup_lba(bt, bdev);
1500 
1501 	ret = -EBUSY;
1502 	if (cmpxchg(&q->blk_trace, NULL, bt))
1503 		goto free_bt;
1504 
1505 	if (atomic_inc_return(&blk_probes_ref) == 1)
1506 		blk_register_tracepoints();
1507 	return 0;
1508 
1509 free_bt:
1510 	blk_trace_free(bt);
1511 	return ret;
1512 }
1513 
1514 /*
1515  * sysfs interface to enable and configure tracing
1516  */
1517 
1518 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1519 					 struct device_attribute *attr,
1520 					 char *buf);
1521 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1522 					  struct device_attribute *attr,
1523 					  const char *buf, size_t count);
1524 #define BLK_TRACE_DEVICE_ATTR(_name) \
1525 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1526 		    sysfs_blk_trace_attr_show, \
1527 		    sysfs_blk_trace_attr_store)
1528 
1529 static BLK_TRACE_DEVICE_ATTR(enable);
1530 static BLK_TRACE_DEVICE_ATTR(act_mask);
1531 static BLK_TRACE_DEVICE_ATTR(pid);
1532 static BLK_TRACE_DEVICE_ATTR(start_lba);
1533 static BLK_TRACE_DEVICE_ATTR(end_lba);
1534 
1535 static struct attribute *blk_trace_attrs[] = {
1536 	&dev_attr_enable.attr,
1537 	&dev_attr_act_mask.attr,
1538 	&dev_attr_pid.attr,
1539 	&dev_attr_start_lba.attr,
1540 	&dev_attr_end_lba.attr,
1541 	NULL
1542 };
1543 
1544 struct attribute_group blk_trace_attr_group = {
1545 	.name  = "trace",
1546 	.attrs = blk_trace_attrs,
1547 };
1548 
1549 static const struct {
1550 	int mask;
1551 	const char *str;
1552 } mask_maps[] = {
1553 	{ BLK_TC_READ,		"read"		},
1554 	{ BLK_TC_WRITE,		"write"		},
1555 	{ BLK_TC_FLUSH,		"flush"		},
1556 	{ BLK_TC_SYNC,		"sync"		},
1557 	{ BLK_TC_QUEUE,		"queue"		},
1558 	{ BLK_TC_REQUEUE,	"requeue"	},
1559 	{ BLK_TC_ISSUE,		"issue"		},
1560 	{ BLK_TC_COMPLETE,	"complete"	},
1561 	{ BLK_TC_FS,		"fs"		},
1562 	{ BLK_TC_PC,		"pc"		},
1563 	{ BLK_TC_NOTIFY,	"notify"	},
1564 	{ BLK_TC_AHEAD,		"ahead"		},
1565 	{ BLK_TC_META,		"meta"		},
1566 	{ BLK_TC_DISCARD,	"discard"	},
1567 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1568 	{ BLK_TC_FUA,		"fua"		},
1569 };
1570 
1571 static int blk_trace_str2mask(const char *str)
1572 {
1573 	int i;
1574 	int mask = 0;
1575 	char *buf, *s, *token;
1576 
1577 	buf = kstrdup(str, GFP_KERNEL);
1578 	if (buf == NULL)
1579 		return -ENOMEM;
1580 	s = strstrip(buf);
1581 
1582 	while (1) {
1583 		token = strsep(&s, ",");
1584 		if (token == NULL)
1585 			break;
1586 
1587 		if (*token == '\0')
1588 			continue;
1589 
1590 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1591 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1592 				mask |= mask_maps[i].mask;
1593 				break;
1594 			}
1595 		}
1596 		if (i == ARRAY_SIZE(mask_maps)) {
1597 			mask = -EINVAL;
1598 			break;
1599 		}
1600 	}
1601 	kfree(buf);
1602 
1603 	return mask;
1604 }
1605 
1606 static ssize_t blk_trace_mask2str(char *buf, int mask)
1607 {
1608 	int i;
1609 	char *p = buf;
1610 
1611 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1612 		if (mask & mask_maps[i].mask) {
1613 			p += sprintf(p, "%s%s",
1614 				    (p == buf) ? "" : ",", mask_maps[i].str);
1615 		}
1616 	}
1617 	*p++ = '\n';
1618 
1619 	return p - buf;
1620 }
1621 
1622 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1623 {
1624 	if (bdev->bd_disk == NULL)
1625 		return NULL;
1626 
1627 	return bdev_get_queue(bdev);
1628 }
1629 
1630 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1631 					 struct device_attribute *attr,
1632 					 char *buf)
1633 {
1634 	struct hd_struct *p = dev_to_part(dev);
1635 	struct request_queue *q;
1636 	struct block_device *bdev;
1637 	ssize_t ret = -ENXIO;
1638 
1639 	bdev = bdget(part_devt(p));
1640 	if (bdev == NULL)
1641 		goto out;
1642 
1643 	q = blk_trace_get_queue(bdev);
1644 	if (q == NULL)
1645 		goto out_bdput;
1646 
1647 	mutex_lock(&bdev->bd_mutex);
1648 
1649 	if (attr == &dev_attr_enable) {
1650 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1651 		goto out_unlock_bdev;
1652 	}
1653 
1654 	if (q->blk_trace == NULL)
1655 		ret = sprintf(buf, "disabled\n");
1656 	else if (attr == &dev_attr_act_mask)
1657 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1658 	else if (attr == &dev_attr_pid)
1659 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1660 	else if (attr == &dev_attr_start_lba)
1661 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1662 	else if (attr == &dev_attr_end_lba)
1663 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1664 
1665 out_unlock_bdev:
1666 	mutex_unlock(&bdev->bd_mutex);
1667 out_bdput:
1668 	bdput(bdev);
1669 out:
1670 	return ret;
1671 }
1672 
1673 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1674 					  struct device_attribute *attr,
1675 					  const char *buf, size_t count)
1676 {
1677 	struct block_device *bdev;
1678 	struct request_queue *q;
1679 	struct hd_struct *p;
1680 	u64 value;
1681 	ssize_t ret = -EINVAL;
1682 
1683 	if (count == 0)
1684 		goto out;
1685 
1686 	if (attr == &dev_attr_act_mask) {
1687 		if (sscanf(buf, "%llx", &value) != 1) {
1688 			/* Assume it is a list of trace category names */
1689 			ret = blk_trace_str2mask(buf);
1690 			if (ret < 0)
1691 				goto out;
1692 			value = ret;
1693 		}
1694 	} else if (sscanf(buf, "%llu", &value) != 1)
1695 		goto out;
1696 
1697 	ret = -ENXIO;
1698 
1699 	p = dev_to_part(dev);
1700 	bdev = bdget(part_devt(p));
1701 	if (bdev == NULL)
1702 		goto out;
1703 
1704 	q = blk_trace_get_queue(bdev);
1705 	if (q == NULL)
1706 		goto out_bdput;
1707 
1708 	mutex_lock(&bdev->bd_mutex);
1709 
1710 	if (attr == &dev_attr_enable) {
1711 		if (value)
1712 			ret = blk_trace_setup_queue(q, bdev);
1713 		else
1714 			ret = blk_trace_remove_queue(q);
1715 		goto out_unlock_bdev;
1716 	}
1717 
1718 	ret = 0;
1719 	if (q->blk_trace == NULL)
1720 		ret = blk_trace_setup_queue(q, bdev);
1721 
1722 	if (ret == 0) {
1723 		if (attr == &dev_attr_act_mask)
1724 			q->blk_trace->act_mask = value;
1725 		else if (attr == &dev_attr_pid)
1726 			q->blk_trace->pid = value;
1727 		else if (attr == &dev_attr_start_lba)
1728 			q->blk_trace->start_lba = value;
1729 		else if (attr == &dev_attr_end_lba)
1730 			q->blk_trace->end_lba = value;
1731 	}
1732 
1733 out_unlock_bdev:
1734 	mutex_unlock(&bdev->bd_mutex);
1735 out_bdput:
1736 	bdput(bdev);
1737 out:
1738 	return ret ? ret : count;
1739 }
1740 
1741 int blk_trace_init_sysfs(struct device *dev)
1742 {
1743 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1744 }
1745 
1746 void blk_trace_remove_sysfs(struct device *dev)
1747 {
1748 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1749 }
1750 
1751 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1752 
1753 #ifdef CONFIG_EVENT_TRACING
1754 
1755 void blk_dump_cmd(char *buf, struct request *rq)
1756 {
1757 	int i, end;
1758 	int len = rq->cmd_len;
1759 	unsigned char *cmd = rq->cmd;
1760 
1761 	if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1762 		buf[0] = '\0';
1763 		return;
1764 	}
1765 
1766 	for (end = len - 1; end >= 0; end--)
1767 		if (cmd[end])
1768 			break;
1769 	end++;
1770 
1771 	for (i = 0; i < len; i++) {
1772 		buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1773 		if (i == end && end != len - 1) {
1774 			sprintf(buf, " ..");
1775 			break;
1776 		}
1777 	}
1778 }
1779 
1780 void blk_fill_rwbs(char *rwbs, int op, u32 rw, int bytes)
1781 {
1782 	int i = 0;
1783 
1784 	if (rw & REQ_PREFLUSH)
1785 		rwbs[i++] = 'F';
1786 
1787 	switch (op) {
1788 	case REQ_OP_WRITE:
1789 	case REQ_OP_WRITE_SAME:
1790 		rwbs[i++] = 'W';
1791 		break;
1792 	case REQ_OP_DISCARD:
1793 		rwbs[i++] = 'D';
1794 		break;
1795 	case REQ_OP_SECURE_ERASE:
1796 		rwbs[i++] = 'D';
1797 		rwbs[i++] = 'E';
1798 		break;
1799 	case REQ_OP_FLUSH:
1800 		rwbs[i++] = 'F';
1801 		break;
1802 	case REQ_OP_READ:
1803 		rwbs[i++] = 'R';
1804 		break;
1805 	default:
1806 		rwbs[i++] = 'N';
1807 	}
1808 
1809 	if (rw & REQ_FUA)
1810 		rwbs[i++] = 'F';
1811 	if (rw & REQ_RAHEAD)
1812 		rwbs[i++] = 'A';
1813 	if (rw & REQ_SYNC)
1814 		rwbs[i++] = 'S';
1815 	if (rw & REQ_META)
1816 		rwbs[i++] = 'M';
1817 
1818 	rwbs[i] = '\0';
1819 }
1820 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1821 
1822 #endif /* CONFIG_EVENT_TRACING */
1823 
1824