xref: /openbmc/linux/kernel/trace/blktrace.c (revision 2359ccdd)
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 #include <linux/blk-cgroup.h>
31 
32 #include "../../block/blk.h"
33 
34 #include <trace/events/block.h>
35 
36 #include "trace_output.h"
37 
38 #ifdef CONFIG_BLK_DEV_IO_TRACE
39 
40 static unsigned int blktrace_seq __read_mostly = 1;
41 
42 static struct trace_array *blk_tr;
43 static bool blk_tracer_enabled __read_mostly;
44 
45 static LIST_HEAD(running_trace_list);
46 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
47 
48 /* Select an alternative, minimalistic output than the original one */
49 #define TRACE_BLK_OPT_CLASSIC	0x1
50 #define TRACE_BLK_OPT_CGROUP	0x2
51 #define TRACE_BLK_OPT_CGNAME	0x4
52 
53 static struct tracer_opt blk_tracer_opts[] = {
54 	/* Default disable the minimalistic output */
55 	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
56 #ifdef CONFIG_BLK_CGROUP
57 	{ TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
58 	{ TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
59 #endif
60 	{ }
61 };
62 
63 static struct tracer_flags blk_tracer_flags = {
64 	.val  = 0,
65 	.opts = blk_tracer_opts,
66 };
67 
68 /* Global reference count of probes */
69 static DEFINE_MUTEX(blk_probe_mutex);
70 static int blk_probes_ref;
71 
72 static void blk_register_tracepoints(void);
73 static void blk_unregister_tracepoints(void);
74 
75 /*
76  * Send out a notify message.
77  */
78 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
79 		       const void *data, size_t len,
80 		       union kernfs_node_id *cgid)
81 {
82 	struct blk_io_trace *t;
83 	struct ring_buffer_event *event = NULL;
84 	struct ring_buffer *buffer = NULL;
85 	int pc = 0;
86 	int cpu = smp_processor_id();
87 	bool blk_tracer = blk_tracer_enabled;
88 	ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
89 
90 	if (blk_tracer) {
91 		buffer = blk_tr->trace_buffer.buffer;
92 		pc = preempt_count();
93 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
94 						  sizeof(*t) + len + cgid_len,
95 						  0, pc);
96 		if (!event)
97 			return;
98 		t = ring_buffer_event_data(event);
99 		goto record_it;
100 	}
101 
102 	if (!bt->rchan)
103 		return;
104 
105 	t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
106 	if (t) {
107 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
108 		t->time = ktime_to_ns(ktime_get());
109 record_it:
110 		t->device = bt->dev;
111 		t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
112 		t->pid = pid;
113 		t->cpu = cpu;
114 		t->pdu_len = len + cgid_len;
115 		if (cgid)
116 			memcpy((void *)t + sizeof(*t), cgid, cgid_len);
117 		memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
118 
119 		if (blk_tracer)
120 			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
121 	}
122 }
123 
124 /*
125  * Send out a notify for this process, if we haven't done so since a trace
126  * started
127  */
128 static void trace_note_tsk(struct task_struct *tsk)
129 {
130 	unsigned long flags;
131 	struct blk_trace *bt;
132 
133 	tsk->btrace_seq = blktrace_seq;
134 	spin_lock_irqsave(&running_trace_lock, flags);
135 	list_for_each_entry(bt, &running_trace_list, running_list) {
136 		trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
137 			   sizeof(tsk->comm), NULL);
138 	}
139 	spin_unlock_irqrestore(&running_trace_lock, flags);
140 }
141 
142 static void trace_note_time(struct blk_trace *bt)
143 {
144 	struct timespec64 now;
145 	unsigned long flags;
146 	u32 words[2];
147 
148 	/* need to check user space to see if this breaks in y2038 or y2106 */
149 	ktime_get_real_ts64(&now);
150 	words[0] = (u32)now.tv_sec;
151 	words[1] = now.tv_nsec;
152 
153 	local_irq_save(flags);
154 	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
155 	local_irq_restore(flags);
156 }
157 
158 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
159 	const char *fmt, ...)
160 {
161 	int n;
162 	va_list args;
163 	unsigned long flags;
164 	char *buf;
165 
166 	if (unlikely(bt->trace_state != Blktrace_running &&
167 		     !blk_tracer_enabled))
168 		return;
169 
170 	/*
171 	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
172 	 * message to the trace.
173 	 */
174 	if (!(bt->act_mask & BLK_TC_NOTIFY))
175 		return;
176 
177 	local_irq_save(flags);
178 	buf = this_cpu_ptr(bt->msg_data);
179 	va_start(args, fmt);
180 	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
181 	va_end(args);
182 
183 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
184 		blkcg = NULL;
185 #ifdef CONFIG_BLK_CGROUP
186 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
187 		blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
188 #else
189 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
190 #endif
191 	local_irq_restore(flags);
192 }
193 EXPORT_SYMBOL_GPL(__trace_note_message);
194 
195 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
196 			 pid_t pid)
197 {
198 	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
199 		return 1;
200 	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
201 		return 1;
202 	if (bt->pid && pid != bt->pid)
203 		return 1;
204 
205 	return 0;
206 }
207 
208 /*
209  * Data direction bit lookup
210  */
211 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
212 				 BLK_TC_ACT(BLK_TC_WRITE) };
213 
214 #define BLK_TC_RAHEAD		BLK_TC_AHEAD
215 #define BLK_TC_PREFLUSH		BLK_TC_FLUSH
216 
217 /* The ilog2() calls fall out because they're constant */
218 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
219 	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
220 
221 /*
222  * The worker for the various blk_add_trace*() types. Fills out a
223  * blk_io_trace structure and places it in a per-cpu subbuffer.
224  */
225 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
226 		     int op, int op_flags, u32 what, int error, int pdu_len,
227 		     void *pdu_data, union kernfs_node_id *cgid)
228 {
229 	struct task_struct *tsk = current;
230 	struct ring_buffer_event *event = NULL;
231 	struct ring_buffer *buffer = NULL;
232 	struct blk_io_trace *t;
233 	unsigned long flags = 0;
234 	unsigned long *sequence;
235 	pid_t pid;
236 	int cpu, pc = 0;
237 	bool blk_tracer = blk_tracer_enabled;
238 	ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
239 
240 	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
241 		return;
242 
243 	what |= ddir_act[op_is_write(op) ? WRITE : READ];
244 	what |= MASK_TC_BIT(op_flags, SYNC);
245 	what |= MASK_TC_BIT(op_flags, RAHEAD);
246 	what |= MASK_TC_BIT(op_flags, META);
247 	what |= MASK_TC_BIT(op_flags, PREFLUSH);
248 	what |= MASK_TC_BIT(op_flags, FUA);
249 	if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
250 		what |= BLK_TC_ACT(BLK_TC_DISCARD);
251 	if (op == REQ_OP_FLUSH)
252 		what |= BLK_TC_ACT(BLK_TC_FLUSH);
253 	if (cgid)
254 		what |= __BLK_TA_CGROUP;
255 
256 	pid = tsk->pid;
257 	if (act_log_check(bt, what, sector, pid))
258 		return;
259 	cpu = raw_smp_processor_id();
260 
261 	if (blk_tracer) {
262 		tracing_record_cmdline(current);
263 
264 		buffer = blk_tr->trace_buffer.buffer;
265 		pc = preempt_count();
266 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
267 						  sizeof(*t) + pdu_len + cgid_len,
268 						  0, pc);
269 		if (!event)
270 			return;
271 		t = ring_buffer_event_data(event);
272 		goto record_it;
273 	}
274 
275 	if (unlikely(tsk->btrace_seq != blktrace_seq))
276 		trace_note_tsk(tsk);
277 
278 	/*
279 	 * A word about the locking here - we disable interrupts to reserve
280 	 * some space in the relay per-cpu buffer, to prevent an irq
281 	 * from coming in and stepping on our toes.
282 	 */
283 	local_irq_save(flags);
284 	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
285 	if (t) {
286 		sequence = per_cpu_ptr(bt->sequence, cpu);
287 
288 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
289 		t->sequence = ++(*sequence);
290 		t->time = ktime_to_ns(ktime_get());
291 record_it:
292 		/*
293 		 * These two are not needed in ftrace as they are in the
294 		 * generic trace_entry, filled by tracing_generic_entry_update,
295 		 * but for the trace_event->bin() synthesizer benefit we do it
296 		 * here too.
297 		 */
298 		t->cpu = cpu;
299 		t->pid = pid;
300 
301 		t->sector = sector;
302 		t->bytes = bytes;
303 		t->action = what;
304 		t->device = bt->dev;
305 		t->error = error;
306 		t->pdu_len = pdu_len + cgid_len;
307 
308 		if (cgid_len)
309 			memcpy((void *)t + sizeof(*t), cgid, cgid_len);
310 		if (pdu_len)
311 			memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
312 
313 		if (blk_tracer) {
314 			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
315 			return;
316 		}
317 	}
318 
319 	local_irq_restore(flags);
320 }
321 
322 static void blk_trace_free(struct blk_trace *bt)
323 {
324 	debugfs_remove(bt->msg_file);
325 	debugfs_remove(bt->dropped_file);
326 	relay_close(bt->rchan);
327 	debugfs_remove(bt->dir);
328 	free_percpu(bt->sequence);
329 	free_percpu(bt->msg_data);
330 	kfree(bt);
331 }
332 
333 static void get_probe_ref(void)
334 {
335 	mutex_lock(&blk_probe_mutex);
336 	if (++blk_probes_ref == 1)
337 		blk_register_tracepoints();
338 	mutex_unlock(&blk_probe_mutex);
339 }
340 
341 static void put_probe_ref(void)
342 {
343 	mutex_lock(&blk_probe_mutex);
344 	if (!--blk_probes_ref)
345 		blk_unregister_tracepoints();
346 	mutex_unlock(&blk_probe_mutex);
347 }
348 
349 static void blk_trace_cleanup(struct blk_trace *bt)
350 {
351 	blk_trace_free(bt);
352 	put_probe_ref();
353 }
354 
355 static int __blk_trace_remove(struct request_queue *q)
356 {
357 	struct blk_trace *bt;
358 
359 	bt = xchg(&q->blk_trace, NULL);
360 	if (!bt)
361 		return -EINVAL;
362 
363 	if (bt->trace_state != Blktrace_running)
364 		blk_trace_cleanup(bt);
365 
366 	return 0;
367 }
368 
369 int blk_trace_remove(struct request_queue *q)
370 {
371 	int ret;
372 
373 	mutex_lock(&q->blk_trace_mutex);
374 	ret = __blk_trace_remove(q);
375 	mutex_unlock(&q->blk_trace_mutex);
376 
377 	return ret;
378 }
379 EXPORT_SYMBOL_GPL(blk_trace_remove);
380 
381 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
382 				size_t count, loff_t *ppos)
383 {
384 	struct blk_trace *bt = filp->private_data;
385 	char buf[16];
386 
387 	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
388 
389 	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
390 }
391 
392 static const struct file_operations blk_dropped_fops = {
393 	.owner =	THIS_MODULE,
394 	.open =		simple_open,
395 	.read =		blk_dropped_read,
396 	.llseek =	default_llseek,
397 };
398 
399 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
400 				size_t count, loff_t *ppos)
401 {
402 	char *msg;
403 	struct blk_trace *bt;
404 
405 	if (count >= BLK_TN_MAX_MSG)
406 		return -EINVAL;
407 
408 	msg = memdup_user_nul(buffer, count);
409 	if (IS_ERR(msg))
410 		return PTR_ERR(msg);
411 
412 	bt = filp->private_data;
413 	__trace_note_message(bt, NULL, "%s", msg);
414 	kfree(msg);
415 
416 	return count;
417 }
418 
419 static const struct file_operations blk_msg_fops = {
420 	.owner =	THIS_MODULE,
421 	.open =		simple_open,
422 	.write =	blk_msg_write,
423 	.llseek =	noop_llseek,
424 };
425 
426 /*
427  * Keep track of how many times we encountered a full subbuffer, to aid
428  * the user space app in telling how many lost events there were.
429  */
430 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
431 				     void *prev_subbuf, size_t prev_padding)
432 {
433 	struct blk_trace *bt;
434 
435 	if (!relay_buf_full(buf))
436 		return 1;
437 
438 	bt = buf->chan->private_data;
439 	atomic_inc(&bt->dropped);
440 	return 0;
441 }
442 
443 static int blk_remove_buf_file_callback(struct dentry *dentry)
444 {
445 	debugfs_remove(dentry);
446 
447 	return 0;
448 }
449 
450 static struct dentry *blk_create_buf_file_callback(const char *filename,
451 						   struct dentry *parent,
452 						   umode_t mode,
453 						   struct rchan_buf *buf,
454 						   int *is_global)
455 {
456 	return debugfs_create_file(filename, mode, parent, buf,
457 					&relay_file_operations);
458 }
459 
460 static struct rchan_callbacks blk_relay_callbacks = {
461 	.subbuf_start		= blk_subbuf_start_callback,
462 	.create_buf_file	= blk_create_buf_file_callback,
463 	.remove_buf_file	= blk_remove_buf_file_callback,
464 };
465 
466 static void blk_trace_setup_lba(struct blk_trace *bt,
467 				struct block_device *bdev)
468 {
469 	struct hd_struct *part = NULL;
470 
471 	if (bdev)
472 		part = bdev->bd_part;
473 
474 	if (part) {
475 		bt->start_lba = part->start_sect;
476 		bt->end_lba = part->start_sect + part->nr_sects;
477 	} else {
478 		bt->start_lba = 0;
479 		bt->end_lba = -1ULL;
480 	}
481 }
482 
483 /*
484  * Setup everything required to start tracing
485  */
486 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
487 			      struct block_device *bdev,
488 			      struct blk_user_trace_setup *buts)
489 {
490 	struct blk_trace *bt = NULL;
491 	struct dentry *dir = NULL;
492 	int ret;
493 
494 	if (!buts->buf_size || !buts->buf_nr)
495 		return -EINVAL;
496 
497 	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
498 	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
499 
500 	/*
501 	 * some device names have larger paths - convert the slashes
502 	 * to underscores for this to work as expected
503 	 */
504 	strreplace(buts->name, '/', '_');
505 
506 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
507 	if (!bt)
508 		return -ENOMEM;
509 
510 	ret = -ENOMEM;
511 	bt->sequence = alloc_percpu(unsigned long);
512 	if (!bt->sequence)
513 		goto err;
514 
515 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
516 	if (!bt->msg_data)
517 		goto err;
518 
519 	ret = -ENOENT;
520 
521 	if (!blk_debugfs_root)
522 		goto err;
523 
524 	dir = debugfs_lookup(buts->name, blk_debugfs_root);
525 	if (!dir)
526 		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
527 	if (!dir)
528 		goto err;
529 
530 	bt->dev = dev;
531 	atomic_set(&bt->dropped, 0);
532 	INIT_LIST_HEAD(&bt->running_list);
533 
534 	ret = -EIO;
535 	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
536 					       &blk_dropped_fops);
537 	if (!bt->dropped_file)
538 		goto err;
539 
540 	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
541 	if (!bt->msg_file)
542 		goto err;
543 
544 	bt->rchan = relay_open("trace", dir, buts->buf_size,
545 				buts->buf_nr, &blk_relay_callbacks, bt);
546 	if (!bt->rchan)
547 		goto err;
548 
549 	bt->act_mask = buts->act_mask;
550 	if (!bt->act_mask)
551 		bt->act_mask = (u16) -1;
552 
553 	blk_trace_setup_lba(bt, bdev);
554 
555 	/* overwrite with user settings */
556 	if (buts->start_lba)
557 		bt->start_lba = buts->start_lba;
558 	if (buts->end_lba)
559 		bt->end_lba = buts->end_lba;
560 
561 	bt->pid = buts->pid;
562 	bt->trace_state = Blktrace_setup;
563 
564 	ret = -EBUSY;
565 	if (cmpxchg(&q->blk_trace, NULL, bt))
566 		goto err;
567 
568 	get_probe_ref();
569 
570 	ret = 0;
571 err:
572 	if (dir && !bt->dir)
573 		dput(dir);
574 	if (ret)
575 		blk_trace_free(bt);
576 	return ret;
577 }
578 
579 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
580 			     struct block_device *bdev, char __user *arg)
581 {
582 	struct blk_user_trace_setup buts;
583 	int ret;
584 
585 	ret = copy_from_user(&buts, arg, sizeof(buts));
586 	if (ret)
587 		return -EFAULT;
588 
589 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
590 	if (ret)
591 		return ret;
592 
593 	if (copy_to_user(arg, &buts, sizeof(buts))) {
594 		__blk_trace_remove(q);
595 		return -EFAULT;
596 	}
597 	return 0;
598 }
599 
600 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
601 		    struct block_device *bdev,
602 		    char __user *arg)
603 {
604 	int ret;
605 
606 	mutex_lock(&q->blk_trace_mutex);
607 	ret = __blk_trace_setup(q, name, dev, bdev, arg);
608 	mutex_unlock(&q->blk_trace_mutex);
609 
610 	return ret;
611 }
612 EXPORT_SYMBOL_GPL(blk_trace_setup);
613 
614 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
615 static int compat_blk_trace_setup(struct request_queue *q, char *name,
616 				  dev_t dev, struct block_device *bdev,
617 				  char __user *arg)
618 {
619 	struct blk_user_trace_setup buts;
620 	struct compat_blk_user_trace_setup cbuts;
621 	int ret;
622 
623 	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
624 		return -EFAULT;
625 
626 	buts = (struct blk_user_trace_setup) {
627 		.act_mask = cbuts.act_mask,
628 		.buf_size = cbuts.buf_size,
629 		.buf_nr = cbuts.buf_nr,
630 		.start_lba = cbuts.start_lba,
631 		.end_lba = cbuts.end_lba,
632 		.pid = cbuts.pid,
633 	};
634 
635 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
636 	if (ret)
637 		return ret;
638 
639 	if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
640 		__blk_trace_remove(q);
641 		return -EFAULT;
642 	}
643 
644 	return 0;
645 }
646 #endif
647 
648 static int __blk_trace_startstop(struct request_queue *q, int start)
649 {
650 	int ret;
651 	struct blk_trace *bt = q->blk_trace;
652 
653 	if (bt == NULL)
654 		return -EINVAL;
655 
656 	/*
657 	 * For starting a trace, we can transition from a setup or stopped
658 	 * trace. For stopping a trace, the state must be running
659 	 */
660 	ret = -EINVAL;
661 	if (start) {
662 		if (bt->trace_state == Blktrace_setup ||
663 		    bt->trace_state == Blktrace_stopped) {
664 			blktrace_seq++;
665 			smp_mb();
666 			bt->trace_state = Blktrace_running;
667 			spin_lock_irq(&running_trace_lock);
668 			list_add(&bt->running_list, &running_trace_list);
669 			spin_unlock_irq(&running_trace_lock);
670 
671 			trace_note_time(bt);
672 			ret = 0;
673 		}
674 	} else {
675 		if (bt->trace_state == Blktrace_running) {
676 			bt->trace_state = Blktrace_stopped;
677 			spin_lock_irq(&running_trace_lock);
678 			list_del_init(&bt->running_list);
679 			spin_unlock_irq(&running_trace_lock);
680 			relay_flush(bt->rchan);
681 			ret = 0;
682 		}
683 	}
684 
685 	return ret;
686 }
687 
688 int blk_trace_startstop(struct request_queue *q, int start)
689 {
690 	int ret;
691 
692 	mutex_lock(&q->blk_trace_mutex);
693 	ret = __blk_trace_startstop(q, start);
694 	mutex_unlock(&q->blk_trace_mutex);
695 
696 	return ret;
697 }
698 EXPORT_SYMBOL_GPL(blk_trace_startstop);
699 
700 /*
701  * When reading or writing the blktrace sysfs files, the references to the
702  * opened sysfs or device files should prevent the underlying block device
703  * from being removed. So no further delete protection is really needed.
704  */
705 
706 /**
707  * blk_trace_ioctl: - handle the ioctls associated with tracing
708  * @bdev:	the block device
709  * @cmd:	the ioctl cmd
710  * @arg:	the argument data, if any
711  *
712  **/
713 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
714 {
715 	struct request_queue *q;
716 	int ret, start = 0;
717 	char b[BDEVNAME_SIZE];
718 
719 	q = bdev_get_queue(bdev);
720 	if (!q)
721 		return -ENXIO;
722 
723 	mutex_lock(&q->blk_trace_mutex);
724 
725 	switch (cmd) {
726 	case BLKTRACESETUP:
727 		bdevname(bdev, b);
728 		ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
729 		break;
730 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
731 	case BLKTRACESETUP32:
732 		bdevname(bdev, b);
733 		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
734 		break;
735 #endif
736 	case BLKTRACESTART:
737 		start = 1;
738 	case BLKTRACESTOP:
739 		ret = __blk_trace_startstop(q, start);
740 		break;
741 	case BLKTRACETEARDOWN:
742 		ret = __blk_trace_remove(q);
743 		break;
744 	default:
745 		ret = -ENOTTY;
746 		break;
747 	}
748 
749 	mutex_unlock(&q->blk_trace_mutex);
750 	return ret;
751 }
752 
753 /**
754  * blk_trace_shutdown: - stop and cleanup trace structures
755  * @q:    the request queue associated with the device
756  *
757  **/
758 void blk_trace_shutdown(struct request_queue *q)
759 {
760 	mutex_lock(&q->blk_trace_mutex);
761 
762 	if (q->blk_trace) {
763 		__blk_trace_startstop(q, 0);
764 		__blk_trace_remove(q);
765 	}
766 
767 	mutex_unlock(&q->blk_trace_mutex);
768 }
769 
770 #ifdef CONFIG_BLK_CGROUP
771 static union kernfs_node_id *
772 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
773 {
774 	struct blk_trace *bt = q->blk_trace;
775 
776 	if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
777 		return NULL;
778 
779 	if (!bio->bi_css)
780 		return NULL;
781 	return cgroup_get_kernfs_id(bio->bi_css->cgroup);
782 }
783 #else
784 static union kernfs_node_id *
785 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
786 {
787 	return NULL;
788 }
789 #endif
790 
791 static union kernfs_node_id *
792 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
793 {
794 	if (!rq->bio)
795 		return NULL;
796 	/* Use the first bio */
797 	return blk_trace_bio_get_cgid(q, rq->bio);
798 }
799 
800 /*
801  * blktrace probes
802  */
803 
804 /**
805  * blk_add_trace_rq - Add a trace for a request oriented action
806  * @rq:		the source request
807  * @error:	return status to log
808  * @nr_bytes:	number of completed bytes
809  * @what:	the action
810  * @cgid:	the cgroup info
811  *
812  * Description:
813  *     Records an action against a request. Will log the bio offset + size.
814  *
815  **/
816 static void blk_add_trace_rq(struct request *rq, int error,
817 			     unsigned int nr_bytes, u32 what,
818 			     union kernfs_node_id *cgid)
819 {
820 	struct blk_trace *bt = rq->q->blk_trace;
821 
822 	if (likely(!bt))
823 		return;
824 
825 	if (blk_rq_is_passthrough(rq))
826 		what |= BLK_TC_ACT(BLK_TC_PC);
827 	else
828 		what |= BLK_TC_ACT(BLK_TC_FS);
829 
830 	__blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
831 			rq->cmd_flags, what, error, 0, NULL, cgid);
832 }
833 
834 static void blk_add_trace_rq_insert(void *ignore,
835 				    struct request_queue *q, struct request *rq)
836 {
837 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
838 			 blk_trace_request_get_cgid(q, rq));
839 }
840 
841 static void blk_add_trace_rq_issue(void *ignore,
842 				   struct request_queue *q, struct request *rq)
843 {
844 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
845 			 blk_trace_request_get_cgid(q, rq));
846 }
847 
848 static void blk_add_trace_rq_requeue(void *ignore,
849 				     struct request_queue *q,
850 				     struct request *rq)
851 {
852 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
853 			 blk_trace_request_get_cgid(q, rq));
854 }
855 
856 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
857 			int error, unsigned int nr_bytes)
858 {
859 	blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
860 			 blk_trace_request_get_cgid(rq->q, rq));
861 }
862 
863 /**
864  * blk_add_trace_bio - Add a trace for a bio oriented action
865  * @q:		queue the io is for
866  * @bio:	the source bio
867  * @what:	the action
868  * @error:	error, if any
869  *
870  * Description:
871  *     Records an action against a bio. Will log the bio offset + size.
872  *
873  **/
874 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
875 			      u32 what, int error)
876 {
877 	struct blk_trace *bt = q->blk_trace;
878 
879 	if (likely(!bt))
880 		return;
881 
882 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
883 			bio_op(bio), bio->bi_opf, what, error, 0, NULL,
884 			blk_trace_bio_get_cgid(q, bio));
885 }
886 
887 static void blk_add_trace_bio_bounce(void *ignore,
888 				     struct request_queue *q, struct bio *bio)
889 {
890 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
891 }
892 
893 static void blk_add_trace_bio_complete(void *ignore,
894 				       struct request_queue *q, struct bio *bio,
895 				       int error)
896 {
897 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
898 }
899 
900 static void blk_add_trace_bio_backmerge(void *ignore,
901 					struct request_queue *q,
902 					struct request *rq,
903 					struct bio *bio)
904 {
905 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
906 }
907 
908 static void blk_add_trace_bio_frontmerge(void *ignore,
909 					 struct request_queue *q,
910 					 struct request *rq,
911 					 struct bio *bio)
912 {
913 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
914 }
915 
916 static void blk_add_trace_bio_queue(void *ignore,
917 				    struct request_queue *q, struct bio *bio)
918 {
919 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
920 }
921 
922 static void blk_add_trace_getrq(void *ignore,
923 				struct request_queue *q,
924 				struct bio *bio, int rw)
925 {
926 	if (bio)
927 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
928 	else {
929 		struct blk_trace *bt = q->blk_trace;
930 
931 		if (bt)
932 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
933 					NULL, NULL);
934 	}
935 }
936 
937 
938 static void blk_add_trace_sleeprq(void *ignore,
939 				  struct request_queue *q,
940 				  struct bio *bio, int rw)
941 {
942 	if (bio)
943 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
944 	else {
945 		struct blk_trace *bt = q->blk_trace;
946 
947 		if (bt)
948 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
949 					0, 0, NULL, NULL);
950 	}
951 }
952 
953 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
954 {
955 	struct blk_trace *bt = q->blk_trace;
956 
957 	if (bt)
958 		__blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
959 }
960 
961 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
962 				    unsigned int depth, bool explicit)
963 {
964 	struct blk_trace *bt = q->blk_trace;
965 
966 	if (bt) {
967 		__be64 rpdu = cpu_to_be64(depth);
968 		u32 what;
969 
970 		if (explicit)
971 			what = BLK_TA_UNPLUG_IO;
972 		else
973 			what = BLK_TA_UNPLUG_TIMER;
974 
975 		__blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
976 	}
977 }
978 
979 static void blk_add_trace_split(void *ignore,
980 				struct request_queue *q, struct bio *bio,
981 				unsigned int pdu)
982 {
983 	struct blk_trace *bt = q->blk_trace;
984 
985 	if (bt) {
986 		__be64 rpdu = cpu_to_be64(pdu);
987 
988 		__blk_add_trace(bt, bio->bi_iter.bi_sector,
989 				bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
990 				BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
991 				&rpdu, blk_trace_bio_get_cgid(q, bio));
992 	}
993 }
994 
995 /**
996  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
997  * @ignore:	trace callback data parameter (not used)
998  * @q:		queue the io is for
999  * @bio:	the source bio
1000  * @dev:	target device
1001  * @from:	source sector
1002  *
1003  * Description:
1004  *     Device mapper or raid target sometimes need to split a bio because
1005  *     it spans a stripe (or similar). Add a trace for that action.
1006  *
1007  **/
1008 static void blk_add_trace_bio_remap(void *ignore,
1009 				    struct request_queue *q, struct bio *bio,
1010 				    dev_t dev, sector_t from)
1011 {
1012 	struct blk_trace *bt = q->blk_trace;
1013 	struct blk_io_trace_remap r;
1014 
1015 	if (likely(!bt))
1016 		return;
1017 
1018 	r.device_from = cpu_to_be32(dev);
1019 	r.device_to   = cpu_to_be32(bio_dev(bio));
1020 	r.sector_from = cpu_to_be64(from);
1021 
1022 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1023 			bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1024 			sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1025 }
1026 
1027 /**
1028  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1029  * @ignore:	trace callback data parameter (not used)
1030  * @q:		queue the io is for
1031  * @rq:		the source request
1032  * @dev:	target device
1033  * @from:	source sector
1034  *
1035  * Description:
1036  *     Device mapper remaps request to other devices.
1037  *     Add a trace for that action.
1038  *
1039  **/
1040 static void blk_add_trace_rq_remap(void *ignore,
1041 				   struct request_queue *q,
1042 				   struct request *rq, dev_t dev,
1043 				   sector_t from)
1044 {
1045 	struct blk_trace *bt = q->blk_trace;
1046 	struct blk_io_trace_remap r;
1047 
1048 	if (likely(!bt))
1049 		return;
1050 
1051 	r.device_from = cpu_to_be32(dev);
1052 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1053 	r.sector_from = cpu_to_be64(from);
1054 
1055 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1056 			rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1057 			sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1058 }
1059 
1060 /**
1061  * blk_add_driver_data - Add binary message with driver-specific data
1062  * @q:		queue the io is for
1063  * @rq:		io request
1064  * @data:	driver-specific data
1065  * @len:	length of driver-specific data
1066  *
1067  * Description:
1068  *     Some drivers might want to write driver-specific data per request.
1069  *
1070  **/
1071 void blk_add_driver_data(struct request_queue *q,
1072 			 struct request *rq,
1073 			 void *data, size_t len)
1074 {
1075 	struct blk_trace *bt = q->blk_trace;
1076 
1077 	if (likely(!bt))
1078 		return;
1079 
1080 	__blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1081 				BLK_TA_DRV_DATA, 0, len, data,
1082 				blk_trace_request_get_cgid(q, rq));
1083 }
1084 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1085 
1086 static void blk_register_tracepoints(void)
1087 {
1088 	int ret;
1089 
1090 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1091 	WARN_ON(ret);
1092 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1093 	WARN_ON(ret);
1094 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1095 	WARN_ON(ret);
1096 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1097 	WARN_ON(ret);
1098 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1099 	WARN_ON(ret);
1100 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1101 	WARN_ON(ret);
1102 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1103 	WARN_ON(ret);
1104 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1105 	WARN_ON(ret);
1106 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1107 	WARN_ON(ret);
1108 	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1109 	WARN_ON(ret);
1110 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1111 	WARN_ON(ret);
1112 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1113 	WARN_ON(ret);
1114 	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1115 	WARN_ON(ret);
1116 	ret = register_trace_block_split(blk_add_trace_split, NULL);
1117 	WARN_ON(ret);
1118 	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1119 	WARN_ON(ret);
1120 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1121 	WARN_ON(ret);
1122 }
1123 
1124 static void blk_unregister_tracepoints(void)
1125 {
1126 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1127 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1128 	unregister_trace_block_split(blk_add_trace_split, NULL);
1129 	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1130 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1131 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1132 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1133 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1134 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1135 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1136 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1137 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1138 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1139 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1140 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1141 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1142 
1143 	tracepoint_synchronize_unregister();
1144 }
1145 
1146 /*
1147  * struct blk_io_tracer formatting routines
1148  */
1149 
1150 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1151 {
1152 	int i = 0;
1153 	int tc = t->action >> BLK_TC_SHIFT;
1154 
1155 	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1156 		rwbs[i++] = 'N';
1157 		goto out;
1158 	}
1159 
1160 	if (tc & BLK_TC_FLUSH)
1161 		rwbs[i++] = 'F';
1162 
1163 	if (tc & BLK_TC_DISCARD)
1164 		rwbs[i++] = 'D';
1165 	else if (tc & BLK_TC_WRITE)
1166 		rwbs[i++] = 'W';
1167 	else if (t->bytes)
1168 		rwbs[i++] = 'R';
1169 	else
1170 		rwbs[i++] = 'N';
1171 
1172 	if (tc & BLK_TC_FUA)
1173 		rwbs[i++] = 'F';
1174 	if (tc & BLK_TC_AHEAD)
1175 		rwbs[i++] = 'A';
1176 	if (tc & BLK_TC_SYNC)
1177 		rwbs[i++] = 'S';
1178 	if (tc & BLK_TC_META)
1179 		rwbs[i++] = 'M';
1180 out:
1181 	rwbs[i] = '\0';
1182 }
1183 
1184 static inline
1185 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1186 {
1187 	return (const struct blk_io_trace *)ent;
1188 }
1189 
1190 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1191 {
1192 	return (void *)(te_blk_io_trace(ent) + 1) +
1193 		(has_cg ? sizeof(union kernfs_node_id) : 0);
1194 }
1195 
1196 static inline const void *cgid_start(const struct trace_entry *ent)
1197 {
1198 	return (void *)(te_blk_io_trace(ent) + 1);
1199 }
1200 
1201 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1202 {
1203 	return te_blk_io_trace(ent)->pdu_len -
1204 			(has_cg ? sizeof(union kernfs_node_id) : 0);
1205 }
1206 
1207 static inline u32 t_action(const struct trace_entry *ent)
1208 {
1209 	return te_blk_io_trace(ent)->action;
1210 }
1211 
1212 static inline u32 t_bytes(const struct trace_entry *ent)
1213 {
1214 	return te_blk_io_trace(ent)->bytes;
1215 }
1216 
1217 static inline u32 t_sec(const struct trace_entry *ent)
1218 {
1219 	return te_blk_io_trace(ent)->bytes >> 9;
1220 }
1221 
1222 static inline unsigned long long t_sector(const struct trace_entry *ent)
1223 {
1224 	return te_blk_io_trace(ent)->sector;
1225 }
1226 
1227 static inline __u16 t_error(const struct trace_entry *ent)
1228 {
1229 	return te_blk_io_trace(ent)->error;
1230 }
1231 
1232 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1233 {
1234 	const __u64 *val = pdu_start(ent, has_cg);
1235 	return be64_to_cpu(*val);
1236 }
1237 
1238 static void get_pdu_remap(const struct trace_entry *ent,
1239 			  struct blk_io_trace_remap *r, bool has_cg)
1240 {
1241 	const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1242 	__u64 sector_from = __r->sector_from;
1243 
1244 	r->device_from = be32_to_cpu(__r->device_from);
1245 	r->device_to   = be32_to_cpu(__r->device_to);
1246 	r->sector_from = be64_to_cpu(sector_from);
1247 }
1248 
1249 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1250 	bool has_cg);
1251 
1252 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1253 	bool has_cg)
1254 {
1255 	char rwbs[RWBS_LEN];
1256 	unsigned long long ts  = iter->ts;
1257 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1258 	unsigned secs	       = (unsigned long)ts;
1259 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1260 
1261 	fill_rwbs(rwbs, t);
1262 
1263 	trace_seq_printf(&iter->seq,
1264 			 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1265 			 MAJOR(t->device), MINOR(t->device), iter->cpu,
1266 			 secs, nsec_rem, iter->ent->pid, act, rwbs);
1267 }
1268 
1269 static void blk_log_action(struct trace_iterator *iter, const char *act,
1270 	bool has_cg)
1271 {
1272 	char rwbs[RWBS_LEN];
1273 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1274 
1275 	fill_rwbs(rwbs, t);
1276 	if (has_cg) {
1277 		const union kernfs_node_id *id = cgid_start(iter->ent);
1278 
1279 		if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1280 			char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1281 
1282 			cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1283 				sizeof(blkcg_name_buf));
1284 			trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1285 				 MAJOR(t->device), MINOR(t->device),
1286 				 blkcg_name_buf, act, rwbs);
1287 		} else
1288 			trace_seq_printf(&iter->seq,
1289 				 "%3d,%-3d %x,%-x %2s %3s ",
1290 				 MAJOR(t->device), MINOR(t->device),
1291 				 id->ino, id->generation, act, rwbs);
1292 	} else
1293 		trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1294 				 MAJOR(t->device), MINOR(t->device), act, rwbs);
1295 }
1296 
1297 static void blk_log_dump_pdu(struct trace_seq *s,
1298 	const struct trace_entry *ent, bool has_cg)
1299 {
1300 	const unsigned char *pdu_buf;
1301 	int pdu_len;
1302 	int i, end;
1303 
1304 	pdu_buf = pdu_start(ent, has_cg);
1305 	pdu_len = pdu_real_len(ent, has_cg);
1306 
1307 	if (!pdu_len)
1308 		return;
1309 
1310 	/* find the last zero that needs to be printed */
1311 	for (end = pdu_len - 1; end >= 0; end--)
1312 		if (pdu_buf[end])
1313 			break;
1314 	end++;
1315 
1316 	trace_seq_putc(s, '(');
1317 
1318 	for (i = 0; i < pdu_len; i++) {
1319 
1320 		trace_seq_printf(s, "%s%02x",
1321 				 i == 0 ? "" : " ", pdu_buf[i]);
1322 
1323 		/*
1324 		 * stop when the rest is just zeroes and indicate so
1325 		 * with a ".." appended
1326 		 */
1327 		if (i == end && end != pdu_len - 1) {
1328 			trace_seq_puts(s, " ..) ");
1329 			return;
1330 		}
1331 	}
1332 
1333 	trace_seq_puts(s, ") ");
1334 }
1335 
1336 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1337 {
1338 	char cmd[TASK_COMM_LEN];
1339 
1340 	trace_find_cmdline(ent->pid, cmd);
1341 
1342 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1343 		trace_seq_printf(s, "%u ", t_bytes(ent));
1344 		blk_log_dump_pdu(s, ent, has_cg);
1345 		trace_seq_printf(s, "[%s]\n", cmd);
1346 	} else {
1347 		if (t_sec(ent))
1348 			trace_seq_printf(s, "%llu + %u [%s]\n",
1349 						t_sector(ent), t_sec(ent), cmd);
1350 		else
1351 			trace_seq_printf(s, "[%s]\n", cmd);
1352 	}
1353 }
1354 
1355 static void blk_log_with_error(struct trace_seq *s,
1356 			      const struct trace_entry *ent, bool has_cg)
1357 {
1358 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1359 		blk_log_dump_pdu(s, ent, has_cg);
1360 		trace_seq_printf(s, "[%d]\n", t_error(ent));
1361 	} else {
1362 		if (t_sec(ent))
1363 			trace_seq_printf(s, "%llu + %u [%d]\n",
1364 					 t_sector(ent),
1365 					 t_sec(ent), t_error(ent));
1366 		else
1367 			trace_seq_printf(s, "%llu [%d]\n",
1368 					 t_sector(ent), t_error(ent));
1369 	}
1370 }
1371 
1372 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1373 {
1374 	struct blk_io_trace_remap r = { .device_from = 0, };
1375 
1376 	get_pdu_remap(ent, &r, has_cg);
1377 	trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1378 			 t_sector(ent), t_sec(ent),
1379 			 MAJOR(r.device_from), MINOR(r.device_from),
1380 			 (unsigned long long)r.sector_from);
1381 }
1382 
1383 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1384 {
1385 	char cmd[TASK_COMM_LEN];
1386 
1387 	trace_find_cmdline(ent->pid, cmd);
1388 
1389 	trace_seq_printf(s, "[%s]\n", cmd);
1390 }
1391 
1392 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1393 {
1394 	char cmd[TASK_COMM_LEN];
1395 
1396 	trace_find_cmdline(ent->pid, cmd);
1397 
1398 	trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1399 }
1400 
1401 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1402 {
1403 	char cmd[TASK_COMM_LEN];
1404 
1405 	trace_find_cmdline(ent->pid, cmd);
1406 
1407 	trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1408 			 get_pdu_int(ent, has_cg), cmd);
1409 }
1410 
1411 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1412 			bool has_cg)
1413 {
1414 
1415 	trace_seq_putmem(s, pdu_start(ent, has_cg),
1416 		pdu_real_len(ent, has_cg));
1417 	trace_seq_putc(s, '\n');
1418 }
1419 
1420 /*
1421  * struct tracer operations
1422  */
1423 
1424 static void blk_tracer_print_header(struct seq_file *m)
1425 {
1426 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1427 		return;
1428 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1429 		    "#  |     |     |           |   |   |\n");
1430 }
1431 
1432 static void blk_tracer_start(struct trace_array *tr)
1433 {
1434 	blk_tracer_enabled = true;
1435 }
1436 
1437 static int blk_tracer_init(struct trace_array *tr)
1438 {
1439 	blk_tr = tr;
1440 	blk_tracer_start(tr);
1441 	return 0;
1442 }
1443 
1444 static void blk_tracer_stop(struct trace_array *tr)
1445 {
1446 	blk_tracer_enabled = false;
1447 }
1448 
1449 static void blk_tracer_reset(struct trace_array *tr)
1450 {
1451 	blk_tracer_stop(tr);
1452 }
1453 
1454 static const struct {
1455 	const char *act[2];
1456 	void	   (*print)(struct trace_seq *s, const struct trace_entry *ent,
1457 			    bool has_cg);
1458 } what2act[] = {
1459 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1460 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1461 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1462 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1463 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1464 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1465 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1466 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1467 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1468 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1469 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1470 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1471 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1472 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1473 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1474 };
1475 
1476 static enum print_line_t print_one_line(struct trace_iterator *iter,
1477 					bool classic)
1478 {
1479 	struct trace_array *tr = iter->tr;
1480 	struct trace_seq *s = &iter->seq;
1481 	const struct blk_io_trace *t;
1482 	u16 what;
1483 	bool long_act;
1484 	blk_log_action_t *log_action;
1485 	bool has_cg;
1486 
1487 	t	   = te_blk_io_trace(iter->ent);
1488 	what	   = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1489 	long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1490 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1491 	has_cg	   = t->action & __BLK_TA_CGROUP;
1492 
1493 	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1494 		log_action(iter, long_act ? "message" : "m", has_cg);
1495 		blk_log_msg(s, iter->ent, has_cg);
1496 		return trace_handle_return(s);
1497 	}
1498 
1499 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1500 		trace_seq_printf(s, "Unknown action %x\n", what);
1501 	else {
1502 		log_action(iter, what2act[what].act[long_act], has_cg);
1503 		what2act[what].print(s, iter->ent, has_cg);
1504 	}
1505 
1506 	return trace_handle_return(s);
1507 }
1508 
1509 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1510 					       int flags, struct trace_event *event)
1511 {
1512 	return print_one_line(iter, false);
1513 }
1514 
1515 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1516 {
1517 	struct trace_seq *s = &iter->seq;
1518 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1519 	const int offset = offsetof(struct blk_io_trace, sector);
1520 	struct blk_io_trace old = {
1521 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1522 		.time     = iter->ts,
1523 	};
1524 
1525 	trace_seq_putmem(s, &old, offset);
1526 	trace_seq_putmem(s, &t->sector,
1527 			 sizeof(old) - offset + t->pdu_len);
1528 }
1529 
1530 static enum print_line_t
1531 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1532 			     struct trace_event *event)
1533 {
1534 	blk_trace_synthesize_old_trace(iter);
1535 
1536 	return trace_handle_return(&iter->seq);
1537 }
1538 
1539 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1540 {
1541 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1542 		return TRACE_TYPE_UNHANDLED;
1543 
1544 	return print_one_line(iter, true);
1545 }
1546 
1547 static int
1548 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1549 {
1550 	/* don't output context-info for blk_classic output */
1551 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1552 		if (set)
1553 			tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1554 		else
1555 			tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1556 	}
1557 	return 0;
1558 }
1559 
1560 static struct tracer blk_tracer __read_mostly = {
1561 	.name		= "blk",
1562 	.init		= blk_tracer_init,
1563 	.reset		= blk_tracer_reset,
1564 	.start		= blk_tracer_start,
1565 	.stop		= blk_tracer_stop,
1566 	.print_header	= blk_tracer_print_header,
1567 	.print_line	= blk_tracer_print_line,
1568 	.flags		= &blk_tracer_flags,
1569 	.set_flag	= blk_tracer_set_flag,
1570 };
1571 
1572 static struct trace_event_functions trace_blk_event_funcs = {
1573 	.trace		= blk_trace_event_print,
1574 	.binary		= blk_trace_event_print_binary,
1575 };
1576 
1577 static struct trace_event trace_blk_event = {
1578 	.type		= TRACE_BLK,
1579 	.funcs		= &trace_blk_event_funcs,
1580 };
1581 
1582 static int __init init_blk_tracer(void)
1583 {
1584 	if (!register_trace_event(&trace_blk_event)) {
1585 		pr_warn("Warning: could not register block events\n");
1586 		return 1;
1587 	}
1588 
1589 	if (register_tracer(&blk_tracer) != 0) {
1590 		pr_warn("Warning: could not register the block tracer\n");
1591 		unregister_trace_event(&trace_blk_event);
1592 		return 1;
1593 	}
1594 
1595 	return 0;
1596 }
1597 
1598 device_initcall(init_blk_tracer);
1599 
1600 static int blk_trace_remove_queue(struct request_queue *q)
1601 {
1602 	struct blk_trace *bt;
1603 
1604 	bt = xchg(&q->blk_trace, NULL);
1605 	if (bt == NULL)
1606 		return -EINVAL;
1607 
1608 	put_probe_ref();
1609 	blk_trace_free(bt);
1610 	return 0;
1611 }
1612 
1613 /*
1614  * Setup everything required to start tracing
1615  */
1616 static int blk_trace_setup_queue(struct request_queue *q,
1617 				 struct block_device *bdev)
1618 {
1619 	struct blk_trace *bt = NULL;
1620 	int ret = -ENOMEM;
1621 
1622 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1623 	if (!bt)
1624 		return -ENOMEM;
1625 
1626 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1627 	if (!bt->msg_data)
1628 		goto free_bt;
1629 
1630 	bt->dev = bdev->bd_dev;
1631 	bt->act_mask = (u16)-1;
1632 
1633 	blk_trace_setup_lba(bt, bdev);
1634 
1635 	ret = -EBUSY;
1636 	if (cmpxchg(&q->blk_trace, NULL, bt))
1637 		goto free_bt;
1638 
1639 	get_probe_ref();
1640 	return 0;
1641 
1642 free_bt:
1643 	blk_trace_free(bt);
1644 	return ret;
1645 }
1646 
1647 /*
1648  * sysfs interface to enable and configure tracing
1649  */
1650 
1651 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1652 					 struct device_attribute *attr,
1653 					 char *buf);
1654 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1655 					  struct device_attribute *attr,
1656 					  const char *buf, size_t count);
1657 #define BLK_TRACE_DEVICE_ATTR(_name) \
1658 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1659 		    sysfs_blk_trace_attr_show, \
1660 		    sysfs_blk_trace_attr_store)
1661 
1662 static BLK_TRACE_DEVICE_ATTR(enable);
1663 static BLK_TRACE_DEVICE_ATTR(act_mask);
1664 static BLK_TRACE_DEVICE_ATTR(pid);
1665 static BLK_TRACE_DEVICE_ATTR(start_lba);
1666 static BLK_TRACE_DEVICE_ATTR(end_lba);
1667 
1668 static struct attribute *blk_trace_attrs[] = {
1669 	&dev_attr_enable.attr,
1670 	&dev_attr_act_mask.attr,
1671 	&dev_attr_pid.attr,
1672 	&dev_attr_start_lba.attr,
1673 	&dev_attr_end_lba.attr,
1674 	NULL
1675 };
1676 
1677 struct attribute_group blk_trace_attr_group = {
1678 	.name  = "trace",
1679 	.attrs = blk_trace_attrs,
1680 };
1681 
1682 static const struct {
1683 	int mask;
1684 	const char *str;
1685 } mask_maps[] = {
1686 	{ BLK_TC_READ,		"read"		},
1687 	{ BLK_TC_WRITE,		"write"		},
1688 	{ BLK_TC_FLUSH,		"flush"		},
1689 	{ BLK_TC_SYNC,		"sync"		},
1690 	{ BLK_TC_QUEUE,		"queue"		},
1691 	{ BLK_TC_REQUEUE,	"requeue"	},
1692 	{ BLK_TC_ISSUE,		"issue"		},
1693 	{ BLK_TC_COMPLETE,	"complete"	},
1694 	{ BLK_TC_FS,		"fs"		},
1695 	{ BLK_TC_PC,		"pc"		},
1696 	{ BLK_TC_NOTIFY,	"notify"	},
1697 	{ BLK_TC_AHEAD,		"ahead"		},
1698 	{ BLK_TC_META,		"meta"		},
1699 	{ BLK_TC_DISCARD,	"discard"	},
1700 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1701 	{ BLK_TC_FUA,		"fua"		},
1702 };
1703 
1704 static int blk_trace_str2mask(const char *str)
1705 {
1706 	int i;
1707 	int mask = 0;
1708 	char *buf, *s, *token;
1709 
1710 	buf = kstrdup(str, GFP_KERNEL);
1711 	if (buf == NULL)
1712 		return -ENOMEM;
1713 	s = strstrip(buf);
1714 
1715 	while (1) {
1716 		token = strsep(&s, ",");
1717 		if (token == NULL)
1718 			break;
1719 
1720 		if (*token == '\0')
1721 			continue;
1722 
1723 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1724 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1725 				mask |= mask_maps[i].mask;
1726 				break;
1727 			}
1728 		}
1729 		if (i == ARRAY_SIZE(mask_maps)) {
1730 			mask = -EINVAL;
1731 			break;
1732 		}
1733 	}
1734 	kfree(buf);
1735 
1736 	return mask;
1737 }
1738 
1739 static ssize_t blk_trace_mask2str(char *buf, int mask)
1740 {
1741 	int i;
1742 	char *p = buf;
1743 
1744 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1745 		if (mask & mask_maps[i].mask) {
1746 			p += sprintf(p, "%s%s",
1747 				    (p == buf) ? "" : ",", mask_maps[i].str);
1748 		}
1749 	}
1750 	*p++ = '\n';
1751 
1752 	return p - buf;
1753 }
1754 
1755 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1756 {
1757 	if (bdev->bd_disk == NULL)
1758 		return NULL;
1759 
1760 	return bdev_get_queue(bdev);
1761 }
1762 
1763 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1764 					 struct device_attribute *attr,
1765 					 char *buf)
1766 {
1767 	struct hd_struct *p = dev_to_part(dev);
1768 	struct request_queue *q;
1769 	struct block_device *bdev;
1770 	ssize_t ret = -ENXIO;
1771 
1772 	bdev = bdget(part_devt(p));
1773 	if (bdev == NULL)
1774 		goto out;
1775 
1776 	q = blk_trace_get_queue(bdev);
1777 	if (q == NULL)
1778 		goto out_bdput;
1779 
1780 	mutex_lock(&q->blk_trace_mutex);
1781 
1782 	if (attr == &dev_attr_enable) {
1783 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1784 		goto out_unlock_bdev;
1785 	}
1786 
1787 	if (q->blk_trace == NULL)
1788 		ret = sprintf(buf, "disabled\n");
1789 	else if (attr == &dev_attr_act_mask)
1790 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1791 	else if (attr == &dev_attr_pid)
1792 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1793 	else if (attr == &dev_attr_start_lba)
1794 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1795 	else if (attr == &dev_attr_end_lba)
1796 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1797 
1798 out_unlock_bdev:
1799 	mutex_unlock(&q->blk_trace_mutex);
1800 out_bdput:
1801 	bdput(bdev);
1802 out:
1803 	return ret;
1804 }
1805 
1806 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1807 					  struct device_attribute *attr,
1808 					  const char *buf, size_t count)
1809 {
1810 	struct block_device *bdev;
1811 	struct request_queue *q;
1812 	struct hd_struct *p;
1813 	u64 value;
1814 	ssize_t ret = -EINVAL;
1815 
1816 	if (count == 0)
1817 		goto out;
1818 
1819 	if (attr == &dev_attr_act_mask) {
1820 		if (kstrtoull(buf, 0, &value)) {
1821 			/* Assume it is a list of trace category names */
1822 			ret = blk_trace_str2mask(buf);
1823 			if (ret < 0)
1824 				goto out;
1825 			value = ret;
1826 		}
1827 	} else if (kstrtoull(buf, 0, &value))
1828 		goto out;
1829 
1830 	ret = -ENXIO;
1831 
1832 	p = dev_to_part(dev);
1833 	bdev = bdget(part_devt(p));
1834 	if (bdev == NULL)
1835 		goto out;
1836 
1837 	q = blk_trace_get_queue(bdev);
1838 	if (q == NULL)
1839 		goto out_bdput;
1840 
1841 	mutex_lock(&q->blk_trace_mutex);
1842 
1843 	if (attr == &dev_attr_enable) {
1844 		if (value)
1845 			ret = blk_trace_setup_queue(q, bdev);
1846 		else
1847 			ret = blk_trace_remove_queue(q);
1848 		goto out_unlock_bdev;
1849 	}
1850 
1851 	ret = 0;
1852 	if (q->blk_trace == NULL)
1853 		ret = blk_trace_setup_queue(q, bdev);
1854 
1855 	if (ret == 0) {
1856 		if (attr == &dev_attr_act_mask)
1857 			q->blk_trace->act_mask = value;
1858 		else if (attr == &dev_attr_pid)
1859 			q->blk_trace->pid = value;
1860 		else if (attr == &dev_attr_start_lba)
1861 			q->blk_trace->start_lba = value;
1862 		else if (attr == &dev_attr_end_lba)
1863 			q->blk_trace->end_lba = value;
1864 	}
1865 
1866 out_unlock_bdev:
1867 	mutex_unlock(&q->blk_trace_mutex);
1868 out_bdput:
1869 	bdput(bdev);
1870 out:
1871 	return ret ? ret : count;
1872 }
1873 
1874 int blk_trace_init_sysfs(struct device *dev)
1875 {
1876 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1877 }
1878 
1879 void blk_trace_remove_sysfs(struct device *dev)
1880 {
1881 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1882 }
1883 
1884 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1885 
1886 #ifdef CONFIG_EVENT_TRACING
1887 
1888 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1889 {
1890 	int i = 0;
1891 
1892 	if (op & REQ_PREFLUSH)
1893 		rwbs[i++] = 'F';
1894 
1895 	switch (op & REQ_OP_MASK) {
1896 	case REQ_OP_WRITE:
1897 	case REQ_OP_WRITE_SAME:
1898 		rwbs[i++] = 'W';
1899 		break;
1900 	case REQ_OP_DISCARD:
1901 		rwbs[i++] = 'D';
1902 		break;
1903 	case REQ_OP_SECURE_ERASE:
1904 		rwbs[i++] = 'D';
1905 		rwbs[i++] = 'E';
1906 		break;
1907 	case REQ_OP_FLUSH:
1908 		rwbs[i++] = 'F';
1909 		break;
1910 	case REQ_OP_READ:
1911 		rwbs[i++] = 'R';
1912 		break;
1913 	default:
1914 		rwbs[i++] = 'N';
1915 	}
1916 
1917 	if (op & REQ_FUA)
1918 		rwbs[i++] = 'F';
1919 	if (op & REQ_RAHEAD)
1920 		rwbs[i++] = 'A';
1921 	if (op & REQ_SYNC)
1922 		rwbs[i++] = 'S';
1923 	if (op & REQ_META)
1924 		rwbs[i++] = 'M';
1925 
1926 	rwbs[i] = '\0';
1927 }
1928 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1929 
1930 #endif /* CONFIG_EVENT_TRACING */
1931 
1932