xref: /openbmc/linux/kernel/trace/blktrace.c (revision 5d331b7f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
4  *
5  */
6 #include <linux/kernel.h>
7 #include <linux/blkdev.h>
8 #include <linux/blktrace_api.h>
9 #include <linux/percpu.h>
10 #include <linux/init.h>
11 #include <linux/mutex.h>
12 #include <linux/slab.h>
13 #include <linux/debugfs.h>
14 #include <linux/export.h>
15 #include <linux/time.h>
16 #include <linux/uaccess.h>
17 #include <linux/list.h>
18 #include <linux/blk-cgroup.h>
19 
20 #include "../../block/blk.h"
21 
22 #include <trace/events/block.h>
23 
24 #include "trace_output.h"
25 
26 #ifdef CONFIG_BLK_DEV_IO_TRACE
27 
28 static unsigned int blktrace_seq __read_mostly = 1;
29 
30 static struct trace_array *blk_tr;
31 static bool blk_tracer_enabled __read_mostly;
32 
33 static LIST_HEAD(running_trace_list);
34 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
35 
36 /* Select an alternative, minimalistic output than the original one */
37 #define TRACE_BLK_OPT_CLASSIC	0x1
38 #define TRACE_BLK_OPT_CGROUP	0x2
39 #define TRACE_BLK_OPT_CGNAME	0x4
40 
41 static struct tracer_opt blk_tracer_opts[] = {
42 	/* Default disable the minimalistic output */
43 	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
44 #ifdef CONFIG_BLK_CGROUP
45 	{ TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
46 	{ TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
47 #endif
48 	{ }
49 };
50 
51 static struct tracer_flags blk_tracer_flags = {
52 	.val  = 0,
53 	.opts = blk_tracer_opts,
54 };
55 
56 /* Global reference count of probes */
57 static DEFINE_MUTEX(blk_probe_mutex);
58 static int blk_probes_ref;
59 
60 static void blk_register_tracepoints(void);
61 static void blk_unregister_tracepoints(void);
62 
63 /*
64  * Send out a notify message.
65  */
66 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
67 		       const void *data, size_t len,
68 		       union kernfs_node_id *cgid)
69 {
70 	struct blk_io_trace *t;
71 	struct ring_buffer_event *event = NULL;
72 	struct ring_buffer *buffer = NULL;
73 	int pc = 0;
74 	int cpu = smp_processor_id();
75 	bool blk_tracer = blk_tracer_enabled;
76 	ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
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 + cgid_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 + cgid_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 | (cgid ? __BLK_TN_CGROUP : 0);
100 		t->pid = pid;
101 		t->cpu = cpu;
102 		t->pdu_len = len + cgid_len;
103 		if (cgid)
104 			memcpy((void *)t + sizeof(*t), cgid, cgid_len);
105 		memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
106 
107 		if (blk_tracer)
108 			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
109 	}
110 }
111 
112 /*
113  * Send out a notify for this process, if we haven't done so since a trace
114  * started
115  */
116 static void trace_note_tsk(struct task_struct *tsk)
117 {
118 	unsigned long flags;
119 	struct blk_trace *bt;
120 
121 	tsk->btrace_seq = blktrace_seq;
122 	spin_lock_irqsave(&running_trace_lock, flags);
123 	list_for_each_entry(bt, &running_trace_list, running_list) {
124 		trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
125 			   sizeof(tsk->comm), NULL);
126 	}
127 	spin_unlock_irqrestore(&running_trace_lock, flags);
128 }
129 
130 static void trace_note_time(struct blk_trace *bt)
131 {
132 	struct timespec64 now;
133 	unsigned long flags;
134 	u32 words[2];
135 
136 	/* need to check user space to see if this breaks in y2038 or y2106 */
137 	ktime_get_real_ts64(&now);
138 	words[0] = (u32)now.tv_sec;
139 	words[1] = now.tv_nsec;
140 
141 	local_irq_save(flags);
142 	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
143 	local_irq_restore(flags);
144 }
145 
146 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
147 	const char *fmt, ...)
148 {
149 	int n;
150 	va_list args;
151 	unsigned long flags;
152 	char *buf;
153 
154 	if (unlikely(bt->trace_state != Blktrace_running &&
155 		     !blk_tracer_enabled))
156 		return;
157 
158 	/*
159 	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
160 	 * message to the trace.
161 	 */
162 	if (!(bt->act_mask & BLK_TC_NOTIFY))
163 		return;
164 
165 	local_irq_save(flags);
166 	buf = this_cpu_ptr(bt->msg_data);
167 	va_start(args, fmt);
168 	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
169 	va_end(args);
170 
171 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
172 		blkcg = NULL;
173 #ifdef CONFIG_BLK_CGROUP
174 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
175 		blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
176 #else
177 	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
178 #endif
179 	local_irq_restore(flags);
180 }
181 EXPORT_SYMBOL_GPL(__trace_note_message);
182 
183 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
184 			 pid_t pid)
185 {
186 	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
187 		return 1;
188 	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
189 		return 1;
190 	if (bt->pid && pid != bt->pid)
191 		return 1;
192 
193 	return 0;
194 }
195 
196 /*
197  * Data direction bit lookup
198  */
199 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
200 				 BLK_TC_ACT(BLK_TC_WRITE) };
201 
202 #define BLK_TC_RAHEAD		BLK_TC_AHEAD
203 #define BLK_TC_PREFLUSH		BLK_TC_FLUSH
204 
205 /* The ilog2() calls fall out because they're constant */
206 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
207 	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
208 
209 /*
210  * The worker for the various blk_add_trace*() types. Fills out a
211  * blk_io_trace structure and places it in a per-cpu subbuffer.
212  */
213 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
214 		     int op, int op_flags, u32 what, int error, int pdu_len,
215 		     void *pdu_data, union kernfs_node_id *cgid)
216 {
217 	struct task_struct *tsk = current;
218 	struct ring_buffer_event *event = NULL;
219 	struct ring_buffer *buffer = NULL;
220 	struct blk_io_trace *t;
221 	unsigned long flags = 0;
222 	unsigned long *sequence;
223 	pid_t pid;
224 	int cpu, pc = 0;
225 	bool blk_tracer = blk_tracer_enabled;
226 	ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
227 
228 	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
229 		return;
230 
231 	what |= ddir_act[op_is_write(op) ? WRITE : READ];
232 	what |= MASK_TC_BIT(op_flags, SYNC);
233 	what |= MASK_TC_BIT(op_flags, RAHEAD);
234 	what |= MASK_TC_BIT(op_flags, META);
235 	what |= MASK_TC_BIT(op_flags, PREFLUSH);
236 	what |= MASK_TC_BIT(op_flags, FUA);
237 	if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
238 		what |= BLK_TC_ACT(BLK_TC_DISCARD);
239 	if (op == REQ_OP_FLUSH)
240 		what |= BLK_TC_ACT(BLK_TC_FLUSH);
241 	if (cgid)
242 		what |= __BLK_TA_CGROUP;
243 
244 	pid = tsk->pid;
245 	if (act_log_check(bt, what, sector, pid))
246 		return;
247 	cpu = raw_smp_processor_id();
248 
249 	if (blk_tracer) {
250 		tracing_record_cmdline(current);
251 
252 		buffer = blk_tr->trace_buffer.buffer;
253 		pc = preempt_count();
254 		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
255 						  sizeof(*t) + pdu_len + cgid_len,
256 						  0, pc);
257 		if (!event)
258 			return;
259 		t = ring_buffer_event_data(event);
260 		goto record_it;
261 	}
262 
263 	if (unlikely(tsk->btrace_seq != blktrace_seq))
264 		trace_note_tsk(tsk);
265 
266 	/*
267 	 * A word about the locking here - we disable interrupts to reserve
268 	 * some space in the relay per-cpu buffer, to prevent an irq
269 	 * from coming in and stepping on our toes.
270 	 */
271 	local_irq_save(flags);
272 	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
273 	if (t) {
274 		sequence = per_cpu_ptr(bt->sequence, cpu);
275 
276 		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
277 		t->sequence = ++(*sequence);
278 		t->time = ktime_to_ns(ktime_get());
279 record_it:
280 		/*
281 		 * These two are not needed in ftrace as they are in the
282 		 * generic trace_entry, filled by tracing_generic_entry_update,
283 		 * but for the trace_event->bin() synthesizer benefit we do it
284 		 * here too.
285 		 */
286 		t->cpu = cpu;
287 		t->pid = pid;
288 
289 		t->sector = sector;
290 		t->bytes = bytes;
291 		t->action = what;
292 		t->device = bt->dev;
293 		t->error = error;
294 		t->pdu_len = pdu_len + cgid_len;
295 
296 		if (cgid_len)
297 			memcpy((void *)t + sizeof(*t), cgid, cgid_len);
298 		if (pdu_len)
299 			memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
300 
301 		if (blk_tracer) {
302 			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
303 			return;
304 		}
305 	}
306 
307 	local_irq_restore(flags);
308 }
309 
310 static void blk_trace_free(struct blk_trace *bt)
311 {
312 	debugfs_remove(bt->msg_file);
313 	debugfs_remove(bt->dropped_file);
314 	relay_close(bt->rchan);
315 	debugfs_remove(bt->dir);
316 	free_percpu(bt->sequence);
317 	free_percpu(bt->msg_data);
318 	kfree(bt);
319 }
320 
321 static void get_probe_ref(void)
322 {
323 	mutex_lock(&blk_probe_mutex);
324 	if (++blk_probes_ref == 1)
325 		blk_register_tracepoints();
326 	mutex_unlock(&blk_probe_mutex);
327 }
328 
329 static void put_probe_ref(void)
330 {
331 	mutex_lock(&blk_probe_mutex);
332 	if (!--blk_probes_ref)
333 		blk_unregister_tracepoints();
334 	mutex_unlock(&blk_probe_mutex);
335 }
336 
337 static void blk_trace_cleanup(struct blk_trace *bt)
338 {
339 	blk_trace_free(bt);
340 	put_probe_ref();
341 }
342 
343 static int __blk_trace_remove(struct request_queue *q)
344 {
345 	struct blk_trace *bt;
346 
347 	bt = xchg(&q->blk_trace, NULL);
348 	if (!bt)
349 		return -EINVAL;
350 
351 	if (bt->trace_state != Blktrace_running)
352 		blk_trace_cleanup(bt);
353 
354 	return 0;
355 }
356 
357 int blk_trace_remove(struct request_queue *q)
358 {
359 	int ret;
360 
361 	mutex_lock(&q->blk_trace_mutex);
362 	ret = __blk_trace_remove(q);
363 	mutex_unlock(&q->blk_trace_mutex);
364 
365 	return ret;
366 }
367 EXPORT_SYMBOL_GPL(blk_trace_remove);
368 
369 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
370 				size_t count, loff_t *ppos)
371 {
372 	struct blk_trace *bt = filp->private_data;
373 	char buf[16];
374 
375 	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
376 
377 	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
378 }
379 
380 static const struct file_operations blk_dropped_fops = {
381 	.owner =	THIS_MODULE,
382 	.open =		simple_open,
383 	.read =		blk_dropped_read,
384 	.llseek =	default_llseek,
385 };
386 
387 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
388 				size_t count, loff_t *ppos)
389 {
390 	char *msg;
391 	struct blk_trace *bt;
392 
393 	if (count >= BLK_TN_MAX_MSG)
394 		return -EINVAL;
395 
396 	msg = memdup_user_nul(buffer, count);
397 	if (IS_ERR(msg))
398 		return PTR_ERR(msg);
399 
400 	bt = filp->private_data;
401 	__trace_note_message(bt, NULL, "%s", msg);
402 	kfree(msg);
403 
404 	return count;
405 }
406 
407 static const struct file_operations blk_msg_fops = {
408 	.owner =	THIS_MODULE,
409 	.open =		simple_open,
410 	.write =	blk_msg_write,
411 	.llseek =	noop_llseek,
412 };
413 
414 /*
415  * Keep track of how many times we encountered a full subbuffer, to aid
416  * the user space app in telling how many lost events there were.
417  */
418 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
419 				     void *prev_subbuf, size_t prev_padding)
420 {
421 	struct blk_trace *bt;
422 
423 	if (!relay_buf_full(buf))
424 		return 1;
425 
426 	bt = buf->chan->private_data;
427 	atomic_inc(&bt->dropped);
428 	return 0;
429 }
430 
431 static int blk_remove_buf_file_callback(struct dentry *dentry)
432 {
433 	debugfs_remove(dentry);
434 
435 	return 0;
436 }
437 
438 static struct dentry *blk_create_buf_file_callback(const char *filename,
439 						   struct dentry *parent,
440 						   umode_t mode,
441 						   struct rchan_buf *buf,
442 						   int *is_global)
443 {
444 	return debugfs_create_file(filename, mode, parent, buf,
445 					&relay_file_operations);
446 }
447 
448 static struct rchan_callbacks blk_relay_callbacks = {
449 	.subbuf_start		= blk_subbuf_start_callback,
450 	.create_buf_file	= blk_create_buf_file_callback,
451 	.remove_buf_file	= blk_remove_buf_file_callback,
452 };
453 
454 static void blk_trace_setup_lba(struct blk_trace *bt,
455 				struct block_device *bdev)
456 {
457 	struct hd_struct *part = NULL;
458 
459 	if (bdev)
460 		part = bdev->bd_part;
461 
462 	if (part) {
463 		bt->start_lba = part->start_sect;
464 		bt->end_lba = part->start_sect + part->nr_sects;
465 	} else {
466 		bt->start_lba = 0;
467 		bt->end_lba = -1ULL;
468 	}
469 }
470 
471 /*
472  * Setup everything required to start tracing
473  */
474 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
475 			      struct block_device *bdev,
476 			      struct blk_user_trace_setup *buts)
477 {
478 	struct blk_trace *bt = NULL;
479 	struct dentry *dir = NULL;
480 	int ret;
481 
482 	if (!buts->buf_size || !buts->buf_nr)
483 		return -EINVAL;
484 
485 	if (!blk_debugfs_root)
486 		return -ENOENT;
487 
488 	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
489 	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
490 
491 	/*
492 	 * some device names have larger paths - convert the slashes
493 	 * to underscores for this to work as expected
494 	 */
495 	strreplace(buts->name, '/', '_');
496 
497 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
498 	if (!bt)
499 		return -ENOMEM;
500 
501 	ret = -ENOMEM;
502 	bt->sequence = alloc_percpu(unsigned long);
503 	if (!bt->sequence)
504 		goto err;
505 
506 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
507 	if (!bt->msg_data)
508 		goto err;
509 
510 	ret = -ENOENT;
511 
512 	dir = debugfs_lookup(buts->name, blk_debugfs_root);
513 	if (!dir)
514 		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
515 	if (!dir)
516 		goto err;
517 
518 	bt->dev = dev;
519 	atomic_set(&bt->dropped, 0);
520 	INIT_LIST_HEAD(&bt->running_list);
521 
522 	ret = -EIO;
523 	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
524 					       &blk_dropped_fops);
525 	if (!bt->dropped_file)
526 		goto err;
527 
528 	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
529 	if (!bt->msg_file)
530 		goto err;
531 
532 	bt->rchan = relay_open("trace", dir, buts->buf_size,
533 				buts->buf_nr, &blk_relay_callbacks, bt);
534 	if (!bt->rchan)
535 		goto err;
536 
537 	bt->act_mask = buts->act_mask;
538 	if (!bt->act_mask)
539 		bt->act_mask = (u16) -1;
540 
541 	blk_trace_setup_lba(bt, bdev);
542 
543 	/* overwrite with user settings */
544 	if (buts->start_lba)
545 		bt->start_lba = buts->start_lba;
546 	if (buts->end_lba)
547 		bt->end_lba = buts->end_lba;
548 
549 	bt->pid = buts->pid;
550 	bt->trace_state = Blktrace_setup;
551 
552 	ret = -EBUSY;
553 	if (cmpxchg(&q->blk_trace, NULL, bt))
554 		goto err;
555 
556 	get_probe_ref();
557 
558 	ret = 0;
559 err:
560 	if (dir && !bt->dir)
561 		dput(dir);
562 	if (ret)
563 		blk_trace_free(bt);
564 	return ret;
565 }
566 
567 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
568 			     struct block_device *bdev, char __user *arg)
569 {
570 	struct blk_user_trace_setup buts;
571 	int ret;
572 
573 	ret = copy_from_user(&buts, arg, sizeof(buts));
574 	if (ret)
575 		return -EFAULT;
576 
577 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
578 	if (ret)
579 		return ret;
580 
581 	if (copy_to_user(arg, &buts, sizeof(buts))) {
582 		__blk_trace_remove(q);
583 		return -EFAULT;
584 	}
585 	return 0;
586 }
587 
588 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
589 		    struct block_device *bdev,
590 		    char __user *arg)
591 {
592 	int ret;
593 
594 	mutex_lock(&q->blk_trace_mutex);
595 	ret = __blk_trace_setup(q, name, dev, bdev, arg);
596 	mutex_unlock(&q->blk_trace_mutex);
597 
598 	return ret;
599 }
600 EXPORT_SYMBOL_GPL(blk_trace_setup);
601 
602 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
603 static int compat_blk_trace_setup(struct request_queue *q, char *name,
604 				  dev_t dev, struct block_device *bdev,
605 				  char __user *arg)
606 {
607 	struct blk_user_trace_setup buts;
608 	struct compat_blk_user_trace_setup cbuts;
609 	int ret;
610 
611 	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
612 		return -EFAULT;
613 
614 	buts = (struct blk_user_trace_setup) {
615 		.act_mask = cbuts.act_mask,
616 		.buf_size = cbuts.buf_size,
617 		.buf_nr = cbuts.buf_nr,
618 		.start_lba = cbuts.start_lba,
619 		.end_lba = cbuts.end_lba,
620 		.pid = cbuts.pid,
621 	};
622 
623 	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
624 	if (ret)
625 		return ret;
626 
627 	if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
628 		__blk_trace_remove(q);
629 		return -EFAULT;
630 	}
631 
632 	return 0;
633 }
634 #endif
635 
636 static int __blk_trace_startstop(struct request_queue *q, int start)
637 {
638 	int ret;
639 	struct blk_trace *bt = q->blk_trace;
640 
641 	if (bt == NULL)
642 		return -EINVAL;
643 
644 	/*
645 	 * For starting a trace, we can transition from a setup or stopped
646 	 * trace. For stopping a trace, the state must be running
647 	 */
648 	ret = -EINVAL;
649 	if (start) {
650 		if (bt->trace_state == Blktrace_setup ||
651 		    bt->trace_state == Blktrace_stopped) {
652 			blktrace_seq++;
653 			smp_mb();
654 			bt->trace_state = Blktrace_running;
655 			spin_lock_irq(&running_trace_lock);
656 			list_add(&bt->running_list, &running_trace_list);
657 			spin_unlock_irq(&running_trace_lock);
658 
659 			trace_note_time(bt);
660 			ret = 0;
661 		}
662 	} else {
663 		if (bt->trace_state == Blktrace_running) {
664 			bt->trace_state = Blktrace_stopped;
665 			spin_lock_irq(&running_trace_lock);
666 			list_del_init(&bt->running_list);
667 			spin_unlock_irq(&running_trace_lock);
668 			relay_flush(bt->rchan);
669 			ret = 0;
670 		}
671 	}
672 
673 	return ret;
674 }
675 
676 int blk_trace_startstop(struct request_queue *q, int start)
677 {
678 	int ret;
679 
680 	mutex_lock(&q->blk_trace_mutex);
681 	ret = __blk_trace_startstop(q, start);
682 	mutex_unlock(&q->blk_trace_mutex);
683 
684 	return ret;
685 }
686 EXPORT_SYMBOL_GPL(blk_trace_startstop);
687 
688 /*
689  * When reading or writing the blktrace sysfs files, the references to the
690  * opened sysfs or device files should prevent the underlying block device
691  * from being removed. So no further delete protection is really needed.
692  */
693 
694 /**
695  * blk_trace_ioctl: - handle the ioctls associated with tracing
696  * @bdev:	the block device
697  * @cmd:	the ioctl cmd
698  * @arg:	the argument data, if any
699  *
700  **/
701 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
702 {
703 	struct request_queue *q;
704 	int ret, start = 0;
705 	char b[BDEVNAME_SIZE];
706 
707 	q = bdev_get_queue(bdev);
708 	if (!q)
709 		return -ENXIO;
710 
711 	mutex_lock(&q->blk_trace_mutex);
712 
713 	switch (cmd) {
714 	case BLKTRACESETUP:
715 		bdevname(bdev, b);
716 		ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
717 		break;
718 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
719 	case BLKTRACESETUP32:
720 		bdevname(bdev, b);
721 		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
722 		break;
723 #endif
724 	case BLKTRACESTART:
725 		start = 1;
726 	case BLKTRACESTOP:
727 		ret = __blk_trace_startstop(q, start);
728 		break;
729 	case BLKTRACETEARDOWN:
730 		ret = __blk_trace_remove(q);
731 		break;
732 	default:
733 		ret = -ENOTTY;
734 		break;
735 	}
736 
737 	mutex_unlock(&q->blk_trace_mutex);
738 	return ret;
739 }
740 
741 /**
742  * blk_trace_shutdown: - stop and cleanup trace structures
743  * @q:    the request queue associated with the device
744  *
745  **/
746 void blk_trace_shutdown(struct request_queue *q)
747 {
748 	mutex_lock(&q->blk_trace_mutex);
749 
750 	if (q->blk_trace) {
751 		__blk_trace_startstop(q, 0);
752 		__blk_trace_remove(q);
753 	}
754 
755 	mutex_unlock(&q->blk_trace_mutex);
756 }
757 
758 #ifdef CONFIG_BLK_CGROUP
759 static union kernfs_node_id *
760 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
761 {
762 	struct blk_trace *bt = q->blk_trace;
763 
764 	if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
765 		return NULL;
766 
767 	if (!bio->bi_css)
768 		return NULL;
769 	return cgroup_get_kernfs_id(bio->bi_css->cgroup);
770 }
771 #else
772 static union kernfs_node_id *
773 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
774 {
775 	return NULL;
776 }
777 #endif
778 
779 static union kernfs_node_id *
780 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
781 {
782 	if (!rq->bio)
783 		return NULL;
784 	/* Use the first bio */
785 	return blk_trace_bio_get_cgid(q, rq->bio);
786 }
787 
788 /*
789  * blktrace probes
790  */
791 
792 /**
793  * blk_add_trace_rq - Add a trace for a request oriented action
794  * @rq:		the source request
795  * @error:	return status to log
796  * @nr_bytes:	number of completed bytes
797  * @what:	the action
798  * @cgid:	the cgroup info
799  *
800  * Description:
801  *     Records an action against a request. Will log the bio offset + size.
802  *
803  **/
804 static void blk_add_trace_rq(struct request *rq, int error,
805 			     unsigned int nr_bytes, u32 what,
806 			     union kernfs_node_id *cgid)
807 {
808 	struct blk_trace *bt = rq->q->blk_trace;
809 
810 	if (likely(!bt))
811 		return;
812 
813 	if (blk_rq_is_passthrough(rq))
814 		what |= BLK_TC_ACT(BLK_TC_PC);
815 	else
816 		what |= BLK_TC_ACT(BLK_TC_FS);
817 
818 	__blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
819 			rq->cmd_flags, what, error, 0, NULL, cgid);
820 }
821 
822 static void blk_add_trace_rq_insert(void *ignore,
823 				    struct request_queue *q, struct request *rq)
824 {
825 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
826 			 blk_trace_request_get_cgid(q, rq));
827 }
828 
829 static void blk_add_trace_rq_issue(void *ignore,
830 				   struct request_queue *q, struct request *rq)
831 {
832 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
833 			 blk_trace_request_get_cgid(q, rq));
834 }
835 
836 static void blk_add_trace_rq_requeue(void *ignore,
837 				     struct request_queue *q,
838 				     struct request *rq)
839 {
840 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
841 			 blk_trace_request_get_cgid(q, rq));
842 }
843 
844 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
845 			int error, unsigned int nr_bytes)
846 {
847 	blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
848 			 blk_trace_request_get_cgid(rq->q, rq));
849 }
850 
851 /**
852  * blk_add_trace_bio - Add a trace for a bio oriented action
853  * @q:		queue the io is for
854  * @bio:	the source bio
855  * @what:	the action
856  * @error:	error, if any
857  *
858  * Description:
859  *     Records an action against a bio. Will log the bio offset + size.
860  *
861  **/
862 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
863 			      u32 what, int error)
864 {
865 	struct blk_trace *bt = q->blk_trace;
866 
867 	if (likely(!bt))
868 		return;
869 
870 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
871 			bio_op(bio), bio->bi_opf, what, error, 0, NULL,
872 			blk_trace_bio_get_cgid(q, bio));
873 }
874 
875 static void blk_add_trace_bio_bounce(void *ignore,
876 				     struct request_queue *q, struct bio *bio)
877 {
878 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
879 }
880 
881 static void blk_add_trace_bio_complete(void *ignore,
882 				       struct request_queue *q, struct bio *bio,
883 				       int error)
884 {
885 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
886 }
887 
888 static void blk_add_trace_bio_backmerge(void *ignore,
889 					struct request_queue *q,
890 					struct request *rq,
891 					struct bio *bio)
892 {
893 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
894 }
895 
896 static void blk_add_trace_bio_frontmerge(void *ignore,
897 					 struct request_queue *q,
898 					 struct request *rq,
899 					 struct bio *bio)
900 {
901 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
902 }
903 
904 static void blk_add_trace_bio_queue(void *ignore,
905 				    struct request_queue *q, struct bio *bio)
906 {
907 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
908 }
909 
910 static void blk_add_trace_getrq(void *ignore,
911 				struct request_queue *q,
912 				struct bio *bio, int rw)
913 {
914 	if (bio)
915 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
916 	else {
917 		struct blk_trace *bt = q->blk_trace;
918 
919 		if (bt)
920 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
921 					NULL, NULL);
922 	}
923 }
924 
925 
926 static void blk_add_trace_sleeprq(void *ignore,
927 				  struct request_queue *q,
928 				  struct bio *bio, int rw)
929 {
930 	if (bio)
931 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
932 	else {
933 		struct blk_trace *bt = q->blk_trace;
934 
935 		if (bt)
936 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
937 					0, 0, NULL, NULL);
938 	}
939 }
940 
941 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
942 {
943 	struct blk_trace *bt = q->blk_trace;
944 
945 	if (bt)
946 		__blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
947 }
948 
949 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
950 				    unsigned int depth, bool explicit)
951 {
952 	struct blk_trace *bt = q->blk_trace;
953 
954 	if (bt) {
955 		__be64 rpdu = cpu_to_be64(depth);
956 		u32 what;
957 
958 		if (explicit)
959 			what = BLK_TA_UNPLUG_IO;
960 		else
961 			what = BLK_TA_UNPLUG_TIMER;
962 
963 		__blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
964 	}
965 }
966 
967 static void blk_add_trace_split(void *ignore,
968 				struct request_queue *q, struct bio *bio,
969 				unsigned int pdu)
970 {
971 	struct blk_trace *bt = q->blk_trace;
972 
973 	if (bt) {
974 		__be64 rpdu = cpu_to_be64(pdu);
975 
976 		__blk_add_trace(bt, bio->bi_iter.bi_sector,
977 				bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
978 				BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
979 				&rpdu, blk_trace_bio_get_cgid(q, bio));
980 	}
981 }
982 
983 /**
984  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
985  * @ignore:	trace callback data parameter (not used)
986  * @q:		queue the io is for
987  * @bio:	the source bio
988  * @dev:	target device
989  * @from:	source sector
990  *
991  * Description:
992  *     Device mapper or raid target sometimes need to split a bio because
993  *     it spans a stripe (or similar). Add a trace for that action.
994  *
995  **/
996 static void blk_add_trace_bio_remap(void *ignore,
997 				    struct request_queue *q, struct bio *bio,
998 				    dev_t dev, sector_t from)
999 {
1000 	struct blk_trace *bt = q->blk_trace;
1001 	struct blk_io_trace_remap r;
1002 
1003 	if (likely(!bt))
1004 		return;
1005 
1006 	r.device_from = cpu_to_be32(dev);
1007 	r.device_to   = cpu_to_be32(bio_dev(bio));
1008 	r.sector_from = cpu_to_be64(from);
1009 
1010 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1011 			bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1012 			sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1013 }
1014 
1015 /**
1016  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1017  * @ignore:	trace callback data parameter (not used)
1018  * @q:		queue the io is for
1019  * @rq:		the source request
1020  * @dev:	target device
1021  * @from:	source sector
1022  *
1023  * Description:
1024  *     Device mapper remaps request to other devices.
1025  *     Add a trace for that action.
1026  *
1027  **/
1028 static void blk_add_trace_rq_remap(void *ignore,
1029 				   struct request_queue *q,
1030 				   struct request *rq, dev_t dev,
1031 				   sector_t from)
1032 {
1033 	struct blk_trace *bt = q->blk_trace;
1034 	struct blk_io_trace_remap r;
1035 
1036 	if (likely(!bt))
1037 		return;
1038 
1039 	r.device_from = cpu_to_be32(dev);
1040 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1041 	r.sector_from = cpu_to_be64(from);
1042 
1043 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1044 			rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1045 			sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1046 }
1047 
1048 /**
1049  * blk_add_driver_data - Add binary message with driver-specific data
1050  * @q:		queue the io is for
1051  * @rq:		io request
1052  * @data:	driver-specific data
1053  * @len:	length of driver-specific data
1054  *
1055  * Description:
1056  *     Some drivers might want to write driver-specific data per request.
1057  *
1058  **/
1059 void blk_add_driver_data(struct request_queue *q,
1060 			 struct request *rq,
1061 			 void *data, size_t len)
1062 {
1063 	struct blk_trace *bt = q->blk_trace;
1064 
1065 	if (likely(!bt))
1066 		return;
1067 
1068 	__blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1069 				BLK_TA_DRV_DATA, 0, len, data,
1070 				blk_trace_request_get_cgid(q, rq));
1071 }
1072 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1073 
1074 static void blk_register_tracepoints(void)
1075 {
1076 	int ret;
1077 
1078 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1079 	WARN_ON(ret);
1080 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1081 	WARN_ON(ret);
1082 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1083 	WARN_ON(ret);
1084 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1085 	WARN_ON(ret);
1086 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1087 	WARN_ON(ret);
1088 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1089 	WARN_ON(ret);
1090 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1091 	WARN_ON(ret);
1092 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1093 	WARN_ON(ret);
1094 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1095 	WARN_ON(ret);
1096 	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1097 	WARN_ON(ret);
1098 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1099 	WARN_ON(ret);
1100 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1101 	WARN_ON(ret);
1102 	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1103 	WARN_ON(ret);
1104 	ret = register_trace_block_split(blk_add_trace_split, NULL);
1105 	WARN_ON(ret);
1106 	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1107 	WARN_ON(ret);
1108 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1109 	WARN_ON(ret);
1110 }
1111 
1112 static void blk_unregister_tracepoints(void)
1113 {
1114 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1115 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1116 	unregister_trace_block_split(blk_add_trace_split, NULL);
1117 	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1118 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1119 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1120 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1121 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1122 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1123 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1124 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1125 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1126 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1127 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1128 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1129 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1130 
1131 	tracepoint_synchronize_unregister();
1132 }
1133 
1134 /*
1135  * struct blk_io_tracer formatting routines
1136  */
1137 
1138 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1139 {
1140 	int i = 0;
1141 	int tc = t->action >> BLK_TC_SHIFT;
1142 
1143 	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1144 		rwbs[i++] = 'N';
1145 		goto out;
1146 	}
1147 
1148 	if (tc & BLK_TC_FLUSH)
1149 		rwbs[i++] = 'F';
1150 
1151 	if (tc & BLK_TC_DISCARD)
1152 		rwbs[i++] = 'D';
1153 	else if (tc & BLK_TC_WRITE)
1154 		rwbs[i++] = 'W';
1155 	else if (t->bytes)
1156 		rwbs[i++] = 'R';
1157 	else
1158 		rwbs[i++] = 'N';
1159 
1160 	if (tc & BLK_TC_FUA)
1161 		rwbs[i++] = 'F';
1162 	if (tc & BLK_TC_AHEAD)
1163 		rwbs[i++] = 'A';
1164 	if (tc & BLK_TC_SYNC)
1165 		rwbs[i++] = 'S';
1166 	if (tc & BLK_TC_META)
1167 		rwbs[i++] = 'M';
1168 out:
1169 	rwbs[i] = '\0';
1170 }
1171 
1172 static inline
1173 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1174 {
1175 	return (const struct blk_io_trace *)ent;
1176 }
1177 
1178 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1179 {
1180 	return (void *)(te_blk_io_trace(ent) + 1) +
1181 		(has_cg ? sizeof(union kernfs_node_id) : 0);
1182 }
1183 
1184 static inline const void *cgid_start(const struct trace_entry *ent)
1185 {
1186 	return (void *)(te_blk_io_trace(ent) + 1);
1187 }
1188 
1189 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1190 {
1191 	return te_blk_io_trace(ent)->pdu_len -
1192 			(has_cg ? sizeof(union kernfs_node_id) : 0);
1193 }
1194 
1195 static inline u32 t_action(const struct trace_entry *ent)
1196 {
1197 	return te_blk_io_trace(ent)->action;
1198 }
1199 
1200 static inline u32 t_bytes(const struct trace_entry *ent)
1201 {
1202 	return te_blk_io_trace(ent)->bytes;
1203 }
1204 
1205 static inline u32 t_sec(const struct trace_entry *ent)
1206 {
1207 	return te_blk_io_trace(ent)->bytes >> 9;
1208 }
1209 
1210 static inline unsigned long long t_sector(const struct trace_entry *ent)
1211 {
1212 	return te_blk_io_trace(ent)->sector;
1213 }
1214 
1215 static inline __u16 t_error(const struct trace_entry *ent)
1216 {
1217 	return te_blk_io_trace(ent)->error;
1218 }
1219 
1220 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1221 {
1222 	const __u64 *val = pdu_start(ent, has_cg);
1223 	return be64_to_cpu(*val);
1224 }
1225 
1226 static void get_pdu_remap(const struct trace_entry *ent,
1227 			  struct blk_io_trace_remap *r, bool has_cg)
1228 {
1229 	const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1230 	__u64 sector_from = __r->sector_from;
1231 
1232 	r->device_from = be32_to_cpu(__r->device_from);
1233 	r->device_to   = be32_to_cpu(__r->device_to);
1234 	r->sector_from = be64_to_cpu(sector_from);
1235 }
1236 
1237 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1238 	bool has_cg);
1239 
1240 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1241 	bool has_cg)
1242 {
1243 	char rwbs[RWBS_LEN];
1244 	unsigned long long ts  = iter->ts;
1245 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1246 	unsigned secs	       = (unsigned long)ts;
1247 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1248 
1249 	fill_rwbs(rwbs, t);
1250 
1251 	trace_seq_printf(&iter->seq,
1252 			 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1253 			 MAJOR(t->device), MINOR(t->device), iter->cpu,
1254 			 secs, nsec_rem, iter->ent->pid, act, rwbs);
1255 }
1256 
1257 static void blk_log_action(struct trace_iterator *iter, const char *act,
1258 	bool has_cg)
1259 {
1260 	char rwbs[RWBS_LEN];
1261 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1262 
1263 	fill_rwbs(rwbs, t);
1264 	if (has_cg) {
1265 		const union kernfs_node_id *id = cgid_start(iter->ent);
1266 
1267 		if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1268 			char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1269 
1270 			cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1271 				sizeof(blkcg_name_buf));
1272 			trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1273 				 MAJOR(t->device), MINOR(t->device),
1274 				 blkcg_name_buf, act, rwbs);
1275 		} else
1276 			trace_seq_printf(&iter->seq,
1277 				 "%3d,%-3d %x,%-x %2s %3s ",
1278 				 MAJOR(t->device), MINOR(t->device),
1279 				 id->ino, id->generation, act, rwbs);
1280 	} else
1281 		trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1282 				 MAJOR(t->device), MINOR(t->device), act, rwbs);
1283 }
1284 
1285 static void blk_log_dump_pdu(struct trace_seq *s,
1286 	const struct trace_entry *ent, bool has_cg)
1287 {
1288 	const unsigned char *pdu_buf;
1289 	int pdu_len;
1290 	int i, end;
1291 
1292 	pdu_buf = pdu_start(ent, has_cg);
1293 	pdu_len = pdu_real_len(ent, has_cg);
1294 
1295 	if (!pdu_len)
1296 		return;
1297 
1298 	/* find the last zero that needs to be printed */
1299 	for (end = pdu_len - 1; end >= 0; end--)
1300 		if (pdu_buf[end])
1301 			break;
1302 	end++;
1303 
1304 	trace_seq_putc(s, '(');
1305 
1306 	for (i = 0; i < pdu_len; i++) {
1307 
1308 		trace_seq_printf(s, "%s%02x",
1309 				 i == 0 ? "" : " ", pdu_buf[i]);
1310 
1311 		/*
1312 		 * stop when the rest is just zeroes and indicate so
1313 		 * with a ".." appended
1314 		 */
1315 		if (i == end && end != pdu_len - 1) {
1316 			trace_seq_puts(s, " ..) ");
1317 			return;
1318 		}
1319 	}
1320 
1321 	trace_seq_puts(s, ") ");
1322 }
1323 
1324 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1325 {
1326 	char cmd[TASK_COMM_LEN];
1327 
1328 	trace_find_cmdline(ent->pid, cmd);
1329 
1330 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1331 		trace_seq_printf(s, "%u ", t_bytes(ent));
1332 		blk_log_dump_pdu(s, ent, has_cg);
1333 		trace_seq_printf(s, "[%s]\n", cmd);
1334 	} else {
1335 		if (t_sec(ent))
1336 			trace_seq_printf(s, "%llu + %u [%s]\n",
1337 						t_sector(ent), t_sec(ent), cmd);
1338 		else
1339 			trace_seq_printf(s, "[%s]\n", cmd);
1340 	}
1341 }
1342 
1343 static void blk_log_with_error(struct trace_seq *s,
1344 			      const struct trace_entry *ent, bool has_cg)
1345 {
1346 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1347 		blk_log_dump_pdu(s, ent, has_cg);
1348 		trace_seq_printf(s, "[%d]\n", t_error(ent));
1349 	} else {
1350 		if (t_sec(ent))
1351 			trace_seq_printf(s, "%llu + %u [%d]\n",
1352 					 t_sector(ent),
1353 					 t_sec(ent), t_error(ent));
1354 		else
1355 			trace_seq_printf(s, "%llu [%d]\n",
1356 					 t_sector(ent), t_error(ent));
1357 	}
1358 }
1359 
1360 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1361 {
1362 	struct blk_io_trace_remap r = { .device_from = 0, };
1363 
1364 	get_pdu_remap(ent, &r, has_cg);
1365 	trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1366 			 t_sector(ent), t_sec(ent),
1367 			 MAJOR(r.device_from), MINOR(r.device_from),
1368 			 (unsigned long long)r.sector_from);
1369 }
1370 
1371 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1372 {
1373 	char cmd[TASK_COMM_LEN];
1374 
1375 	trace_find_cmdline(ent->pid, cmd);
1376 
1377 	trace_seq_printf(s, "[%s]\n", cmd);
1378 }
1379 
1380 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1381 {
1382 	char cmd[TASK_COMM_LEN];
1383 
1384 	trace_find_cmdline(ent->pid, cmd);
1385 
1386 	trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1387 }
1388 
1389 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1390 {
1391 	char cmd[TASK_COMM_LEN];
1392 
1393 	trace_find_cmdline(ent->pid, cmd);
1394 
1395 	trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1396 			 get_pdu_int(ent, has_cg), cmd);
1397 }
1398 
1399 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1400 			bool has_cg)
1401 {
1402 
1403 	trace_seq_putmem(s, pdu_start(ent, has_cg),
1404 		pdu_real_len(ent, has_cg));
1405 	trace_seq_putc(s, '\n');
1406 }
1407 
1408 /*
1409  * struct tracer operations
1410  */
1411 
1412 static void blk_tracer_print_header(struct seq_file *m)
1413 {
1414 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1415 		return;
1416 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1417 		    "#  |     |     |           |   |   |\n");
1418 }
1419 
1420 static void blk_tracer_start(struct trace_array *tr)
1421 {
1422 	blk_tracer_enabled = true;
1423 }
1424 
1425 static int blk_tracer_init(struct trace_array *tr)
1426 {
1427 	blk_tr = tr;
1428 	blk_tracer_start(tr);
1429 	return 0;
1430 }
1431 
1432 static void blk_tracer_stop(struct trace_array *tr)
1433 {
1434 	blk_tracer_enabled = false;
1435 }
1436 
1437 static void blk_tracer_reset(struct trace_array *tr)
1438 {
1439 	blk_tracer_stop(tr);
1440 }
1441 
1442 static const struct {
1443 	const char *act[2];
1444 	void	   (*print)(struct trace_seq *s, const struct trace_entry *ent,
1445 			    bool has_cg);
1446 } what2act[] = {
1447 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1448 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1449 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1450 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1451 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1452 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1453 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1454 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1455 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1456 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1457 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1458 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1459 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1460 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1461 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1462 };
1463 
1464 static enum print_line_t print_one_line(struct trace_iterator *iter,
1465 					bool classic)
1466 {
1467 	struct trace_array *tr = iter->tr;
1468 	struct trace_seq *s = &iter->seq;
1469 	const struct blk_io_trace *t;
1470 	u16 what;
1471 	bool long_act;
1472 	blk_log_action_t *log_action;
1473 	bool has_cg;
1474 
1475 	t	   = te_blk_io_trace(iter->ent);
1476 	what	   = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1477 	long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1478 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1479 	has_cg	   = t->action & __BLK_TA_CGROUP;
1480 
1481 	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1482 		log_action(iter, long_act ? "message" : "m", has_cg);
1483 		blk_log_msg(s, iter->ent, has_cg);
1484 		return trace_handle_return(s);
1485 	}
1486 
1487 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1488 		trace_seq_printf(s, "Unknown action %x\n", what);
1489 	else {
1490 		log_action(iter, what2act[what].act[long_act], has_cg);
1491 		what2act[what].print(s, iter->ent, has_cg);
1492 	}
1493 
1494 	return trace_handle_return(s);
1495 }
1496 
1497 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1498 					       int flags, struct trace_event *event)
1499 {
1500 	return print_one_line(iter, false);
1501 }
1502 
1503 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1504 {
1505 	struct trace_seq *s = &iter->seq;
1506 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1507 	const int offset = offsetof(struct blk_io_trace, sector);
1508 	struct blk_io_trace old = {
1509 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1510 		.time     = iter->ts,
1511 	};
1512 
1513 	trace_seq_putmem(s, &old, offset);
1514 	trace_seq_putmem(s, &t->sector,
1515 			 sizeof(old) - offset + t->pdu_len);
1516 }
1517 
1518 static enum print_line_t
1519 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1520 			     struct trace_event *event)
1521 {
1522 	blk_trace_synthesize_old_trace(iter);
1523 
1524 	return trace_handle_return(&iter->seq);
1525 }
1526 
1527 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1528 {
1529 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1530 		return TRACE_TYPE_UNHANDLED;
1531 
1532 	return print_one_line(iter, true);
1533 }
1534 
1535 static int
1536 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1537 {
1538 	/* don't output context-info for blk_classic output */
1539 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1540 		if (set)
1541 			tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1542 		else
1543 			tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1544 	}
1545 	return 0;
1546 }
1547 
1548 static struct tracer blk_tracer __read_mostly = {
1549 	.name		= "blk",
1550 	.init		= blk_tracer_init,
1551 	.reset		= blk_tracer_reset,
1552 	.start		= blk_tracer_start,
1553 	.stop		= blk_tracer_stop,
1554 	.print_header	= blk_tracer_print_header,
1555 	.print_line	= blk_tracer_print_line,
1556 	.flags		= &blk_tracer_flags,
1557 	.set_flag	= blk_tracer_set_flag,
1558 };
1559 
1560 static struct trace_event_functions trace_blk_event_funcs = {
1561 	.trace		= blk_trace_event_print,
1562 	.binary		= blk_trace_event_print_binary,
1563 };
1564 
1565 static struct trace_event trace_blk_event = {
1566 	.type		= TRACE_BLK,
1567 	.funcs		= &trace_blk_event_funcs,
1568 };
1569 
1570 static int __init init_blk_tracer(void)
1571 {
1572 	if (!register_trace_event(&trace_blk_event)) {
1573 		pr_warn("Warning: could not register block events\n");
1574 		return 1;
1575 	}
1576 
1577 	if (register_tracer(&blk_tracer) != 0) {
1578 		pr_warn("Warning: could not register the block tracer\n");
1579 		unregister_trace_event(&trace_blk_event);
1580 		return 1;
1581 	}
1582 
1583 	return 0;
1584 }
1585 
1586 device_initcall(init_blk_tracer);
1587 
1588 static int blk_trace_remove_queue(struct request_queue *q)
1589 {
1590 	struct blk_trace *bt;
1591 
1592 	bt = xchg(&q->blk_trace, NULL);
1593 	if (bt == NULL)
1594 		return -EINVAL;
1595 
1596 	put_probe_ref();
1597 	blk_trace_free(bt);
1598 	return 0;
1599 }
1600 
1601 /*
1602  * Setup everything required to start tracing
1603  */
1604 static int blk_trace_setup_queue(struct request_queue *q,
1605 				 struct block_device *bdev)
1606 {
1607 	struct blk_trace *bt = NULL;
1608 	int ret = -ENOMEM;
1609 
1610 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1611 	if (!bt)
1612 		return -ENOMEM;
1613 
1614 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1615 	if (!bt->msg_data)
1616 		goto free_bt;
1617 
1618 	bt->dev = bdev->bd_dev;
1619 	bt->act_mask = (u16)-1;
1620 
1621 	blk_trace_setup_lba(bt, bdev);
1622 
1623 	ret = -EBUSY;
1624 	if (cmpxchg(&q->blk_trace, NULL, bt))
1625 		goto free_bt;
1626 
1627 	get_probe_ref();
1628 	return 0;
1629 
1630 free_bt:
1631 	blk_trace_free(bt);
1632 	return ret;
1633 }
1634 
1635 /*
1636  * sysfs interface to enable and configure tracing
1637  */
1638 
1639 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1640 					 struct device_attribute *attr,
1641 					 char *buf);
1642 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1643 					  struct device_attribute *attr,
1644 					  const char *buf, size_t count);
1645 #define BLK_TRACE_DEVICE_ATTR(_name) \
1646 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1647 		    sysfs_blk_trace_attr_show, \
1648 		    sysfs_blk_trace_attr_store)
1649 
1650 static BLK_TRACE_DEVICE_ATTR(enable);
1651 static BLK_TRACE_DEVICE_ATTR(act_mask);
1652 static BLK_TRACE_DEVICE_ATTR(pid);
1653 static BLK_TRACE_DEVICE_ATTR(start_lba);
1654 static BLK_TRACE_DEVICE_ATTR(end_lba);
1655 
1656 static struct attribute *blk_trace_attrs[] = {
1657 	&dev_attr_enable.attr,
1658 	&dev_attr_act_mask.attr,
1659 	&dev_attr_pid.attr,
1660 	&dev_attr_start_lba.attr,
1661 	&dev_attr_end_lba.attr,
1662 	NULL
1663 };
1664 
1665 struct attribute_group blk_trace_attr_group = {
1666 	.name  = "trace",
1667 	.attrs = blk_trace_attrs,
1668 };
1669 
1670 static const struct {
1671 	int mask;
1672 	const char *str;
1673 } mask_maps[] = {
1674 	{ BLK_TC_READ,		"read"		},
1675 	{ BLK_TC_WRITE,		"write"		},
1676 	{ BLK_TC_FLUSH,		"flush"		},
1677 	{ BLK_TC_SYNC,		"sync"		},
1678 	{ BLK_TC_QUEUE,		"queue"		},
1679 	{ BLK_TC_REQUEUE,	"requeue"	},
1680 	{ BLK_TC_ISSUE,		"issue"		},
1681 	{ BLK_TC_COMPLETE,	"complete"	},
1682 	{ BLK_TC_FS,		"fs"		},
1683 	{ BLK_TC_PC,		"pc"		},
1684 	{ BLK_TC_NOTIFY,	"notify"	},
1685 	{ BLK_TC_AHEAD,		"ahead"		},
1686 	{ BLK_TC_META,		"meta"		},
1687 	{ BLK_TC_DISCARD,	"discard"	},
1688 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1689 	{ BLK_TC_FUA,		"fua"		},
1690 };
1691 
1692 static int blk_trace_str2mask(const char *str)
1693 {
1694 	int i;
1695 	int mask = 0;
1696 	char *buf, *s, *token;
1697 
1698 	buf = kstrdup(str, GFP_KERNEL);
1699 	if (buf == NULL)
1700 		return -ENOMEM;
1701 	s = strstrip(buf);
1702 
1703 	while (1) {
1704 		token = strsep(&s, ",");
1705 		if (token == NULL)
1706 			break;
1707 
1708 		if (*token == '\0')
1709 			continue;
1710 
1711 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1712 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1713 				mask |= mask_maps[i].mask;
1714 				break;
1715 			}
1716 		}
1717 		if (i == ARRAY_SIZE(mask_maps)) {
1718 			mask = -EINVAL;
1719 			break;
1720 		}
1721 	}
1722 	kfree(buf);
1723 
1724 	return mask;
1725 }
1726 
1727 static ssize_t blk_trace_mask2str(char *buf, int mask)
1728 {
1729 	int i;
1730 	char *p = buf;
1731 
1732 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1733 		if (mask & mask_maps[i].mask) {
1734 			p += sprintf(p, "%s%s",
1735 				    (p == buf) ? "" : ",", mask_maps[i].str);
1736 		}
1737 	}
1738 	*p++ = '\n';
1739 
1740 	return p - buf;
1741 }
1742 
1743 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1744 {
1745 	if (bdev->bd_disk == NULL)
1746 		return NULL;
1747 
1748 	return bdev_get_queue(bdev);
1749 }
1750 
1751 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1752 					 struct device_attribute *attr,
1753 					 char *buf)
1754 {
1755 	struct hd_struct *p = dev_to_part(dev);
1756 	struct request_queue *q;
1757 	struct block_device *bdev;
1758 	ssize_t ret = -ENXIO;
1759 
1760 	bdev = bdget(part_devt(p));
1761 	if (bdev == NULL)
1762 		goto out;
1763 
1764 	q = blk_trace_get_queue(bdev);
1765 	if (q == NULL)
1766 		goto out_bdput;
1767 
1768 	mutex_lock(&q->blk_trace_mutex);
1769 
1770 	if (attr == &dev_attr_enable) {
1771 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1772 		goto out_unlock_bdev;
1773 	}
1774 
1775 	if (q->blk_trace == NULL)
1776 		ret = sprintf(buf, "disabled\n");
1777 	else if (attr == &dev_attr_act_mask)
1778 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1779 	else if (attr == &dev_attr_pid)
1780 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1781 	else if (attr == &dev_attr_start_lba)
1782 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1783 	else if (attr == &dev_attr_end_lba)
1784 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1785 
1786 out_unlock_bdev:
1787 	mutex_unlock(&q->blk_trace_mutex);
1788 out_bdput:
1789 	bdput(bdev);
1790 out:
1791 	return ret;
1792 }
1793 
1794 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1795 					  struct device_attribute *attr,
1796 					  const char *buf, size_t count)
1797 {
1798 	struct block_device *bdev;
1799 	struct request_queue *q;
1800 	struct hd_struct *p;
1801 	u64 value;
1802 	ssize_t ret = -EINVAL;
1803 
1804 	if (count == 0)
1805 		goto out;
1806 
1807 	if (attr == &dev_attr_act_mask) {
1808 		if (kstrtoull(buf, 0, &value)) {
1809 			/* Assume it is a list of trace category names */
1810 			ret = blk_trace_str2mask(buf);
1811 			if (ret < 0)
1812 				goto out;
1813 			value = ret;
1814 		}
1815 	} else if (kstrtoull(buf, 0, &value))
1816 		goto out;
1817 
1818 	ret = -ENXIO;
1819 
1820 	p = dev_to_part(dev);
1821 	bdev = bdget(part_devt(p));
1822 	if (bdev == NULL)
1823 		goto out;
1824 
1825 	q = blk_trace_get_queue(bdev);
1826 	if (q == NULL)
1827 		goto out_bdput;
1828 
1829 	mutex_lock(&q->blk_trace_mutex);
1830 
1831 	if (attr == &dev_attr_enable) {
1832 		if (!!value == !!q->blk_trace) {
1833 			ret = 0;
1834 			goto out_unlock_bdev;
1835 		}
1836 		if (value)
1837 			ret = blk_trace_setup_queue(q, bdev);
1838 		else
1839 			ret = blk_trace_remove_queue(q);
1840 		goto out_unlock_bdev;
1841 	}
1842 
1843 	ret = 0;
1844 	if (q->blk_trace == NULL)
1845 		ret = blk_trace_setup_queue(q, bdev);
1846 
1847 	if (ret == 0) {
1848 		if (attr == &dev_attr_act_mask)
1849 			q->blk_trace->act_mask = value;
1850 		else if (attr == &dev_attr_pid)
1851 			q->blk_trace->pid = value;
1852 		else if (attr == &dev_attr_start_lba)
1853 			q->blk_trace->start_lba = value;
1854 		else if (attr == &dev_attr_end_lba)
1855 			q->blk_trace->end_lba = value;
1856 	}
1857 
1858 out_unlock_bdev:
1859 	mutex_unlock(&q->blk_trace_mutex);
1860 out_bdput:
1861 	bdput(bdev);
1862 out:
1863 	return ret ? ret : count;
1864 }
1865 
1866 int blk_trace_init_sysfs(struct device *dev)
1867 {
1868 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1869 }
1870 
1871 void blk_trace_remove_sysfs(struct device *dev)
1872 {
1873 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1874 }
1875 
1876 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1877 
1878 #ifdef CONFIG_EVENT_TRACING
1879 
1880 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1881 {
1882 	int i = 0;
1883 
1884 	if (op & REQ_PREFLUSH)
1885 		rwbs[i++] = 'F';
1886 
1887 	switch (op & REQ_OP_MASK) {
1888 	case REQ_OP_WRITE:
1889 	case REQ_OP_WRITE_SAME:
1890 		rwbs[i++] = 'W';
1891 		break;
1892 	case REQ_OP_DISCARD:
1893 		rwbs[i++] = 'D';
1894 		break;
1895 	case REQ_OP_SECURE_ERASE:
1896 		rwbs[i++] = 'D';
1897 		rwbs[i++] = 'E';
1898 		break;
1899 	case REQ_OP_FLUSH:
1900 		rwbs[i++] = 'F';
1901 		break;
1902 	case REQ_OP_READ:
1903 		rwbs[i++] = 'R';
1904 		break;
1905 	default:
1906 		rwbs[i++] = 'N';
1907 	}
1908 
1909 	if (op & REQ_FUA)
1910 		rwbs[i++] = 'F';
1911 	if (op & REQ_RAHEAD)
1912 		rwbs[i++] = 'A';
1913 	if (op & REQ_SYNC)
1914 		rwbs[i++] = 'S';
1915 	if (op & REQ_META)
1916 		rwbs[i++] = 'M';
1917 
1918 	rwbs[i] = '\0';
1919 }
1920 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1921 
1922 #endif /* CONFIG_EVENT_TRACING */
1923 
1924