xref: /openbmc/linux/kernel/trace/blktrace.c (revision be709d48)
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 		/* fall through */
727 	case BLKTRACESTOP:
728 		ret = __blk_trace_startstop(q, start);
729 		break;
730 	case BLKTRACETEARDOWN:
731 		ret = __blk_trace_remove(q);
732 		break;
733 	default:
734 		ret = -ENOTTY;
735 		break;
736 	}
737 
738 	mutex_unlock(&q->blk_trace_mutex);
739 	return ret;
740 }
741 
742 /**
743  * blk_trace_shutdown: - stop and cleanup trace structures
744  * @q:    the request queue associated with the device
745  *
746  **/
747 void blk_trace_shutdown(struct request_queue *q)
748 {
749 	mutex_lock(&q->blk_trace_mutex);
750 
751 	if (q->blk_trace) {
752 		__blk_trace_startstop(q, 0);
753 		__blk_trace_remove(q);
754 	}
755 
756 	mutex_unlock(&q->blk_trace_mutex);
757 }
758 
759 #ifdef CONFIG_BLK_CGROUP
760 static union kernfs_node_id *
761 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
762 {
763 	struct blk_trace *bt = q->blk_trace;
764 
765 	if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
766 		return NULL;
767 
768 	if (!bio->bi_blkg)
769 		return NULL;
770 	return cgroup_get_kernfs_id(bio_blkcg(bio)->css.cgroup);
771 }
772 #else
773 static union kernfs_node_id *
774 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
775 {
776 	return NULL;
777 }
778 #endif
779 
780 static union kernfs_node_id *
781 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
782 {
783 	if (!rq->bio)
784 		return NULL;
785 	/* Use the first bio */
786 	return blk_trace_bio_get_cgid(q, rq->bio);
787 }
788 
789 /*
790  * blktrace probes
791  */
792 
793 /**
794  * blk_add_trace_rq - Add a trace for a request oriented action
795  * @rq:		the source request
796  * @error:	return status to log
797  * @nr_bytes:	number of completed bytes
798  * @what:	the action
799  * @cgid:	the cgroup info
800  *
801  * Description:
802  *     Records an action against a request. Will log the bio offset + size.
803  *
804  **/
805 static void blk_add_trace_rq(struct request *rq, int error,
806 			     unsigned int nr_bytes, u32 what,
807 			     union kernfs_node_id *cgid)
808 {
809 	struct blk_trace *bt = rq->q->blk_trace;
810 
811 	if (likely(!bt))
812 		return;
813 
814 	if (blk_rq_is_passthrough(rq))
815 		what |= BLK_TC_ACT(BLK_TC_PC);
816 	else
817 		what |= BLK_TC_ACT(BLK_TC_FS);
818 
819 	__blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
820 			rq->cmd_flags, what, error, 0, NULL, cgid);
821 }
822 
823 static void blk_add_trace_rq_insert(void *ignore,
824 				    struct request_queue *q, struct request *rq)
825 {
826 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
827 			 blk_trace_request_get_cgid(q, rq));
828 }
829 
830 static void blk_add_trace_rq_issue(void *ignore,
831 				   struct request_queue *q, struct request *rq)
832 {
833 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
834 			 blk_trace_request_get_cgid(q, rq));
835 }
836 
837 static void blk_add_trace_rq_requeue(void *ignore,
838 				     struct request_queue *q,
839 				     struct request *rq)
840 {
841 	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
842 			 blk_trace_request_get_cgid(q, rq));
843 }
844 
845 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
846 			int error, unsigned int nr_bytes)
847 {
848 	blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
849 			 blk_trace_request_get_cgid(rq->q, rq));
850 }
851 
852 /**
853  * blk_add_trace_bio - Add a trace for a bio oriented action
854  * @q:		queue the io is for
855  * @bio:	the source bio
856  * @what:	the action
857  * @error:	error, if any
858  *
859  * Description:
860  *     Records an action against a bio. Will log the bio offset + size.
861  *
862  **/
863 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
864 			      u32 what, int error)
865 {
866 	struct blk_trace *bt = q->blk_trace;
867 
868 	if (likely(!bt))
869 		return;
870 
871 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
872 			bio_op(bio), bio->bi_opf, what, error, 0, NULL,
873 			blk_trace_bio_get_cgid(q, bio));
874 }
875 
876 static void blk_add_trace_bio_bounce(void *ignore,
877 				     struct request_queue *q, struct bio *bio)
878 {
879 	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
880 }
881 
882 static void blk_add_trace_bio_complete(void *ignore,
883 				       struct request_queue *q, struct bio *bio,
884 				       int error)
885 {
886 	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
887 }
888 
889 static void blk_add_trace_bio_backmerge(void *ignore,
890 					struct request_queue *q,
891 					struct request *rq,
892 					struct bio *bio)
893 {
894 	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
895 }
896 
897 static void blk_add_trace_bio_frontmerge(void *ignore,
898 					 struct request_queue *q,
899 					 struct request *rq,
900 					 struct bio *bio)
901 {
902 	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
903 }
904 
905 static void blk_add_trace_bio_queue(void *ignore,
906 				    struct request_queue *q, struct bio *bio)
907 {
908 	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
909 }
910 
911 static void blk_add_trace_getrq(void *ignore,
912 				struct request_queue *q,
913 				struct bio *bio, int rw)
914 {
915 	if (bio)
916 		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
917 	else {
918 		struct blk_trace *bt = q->blk_trace;
919 
920 		if (bt)
921 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
922 					NULL, NULL);
923 	}
924 }
925 
926 
927 static void blk_add_trace_sleeprq(void *ignore,
928 				  struct request_queue *q,
929 				  struct bio *bio, int rw)
930 {
931 	if (bio)
932 		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
933 	else {
934 		struct blk_trace *bt = q->blk_trace;
935 
936 		if (bt)
937 			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
938 					0, 0, NULL, NULL);
939 	}
940 }
941 
942 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
943 {
944 	struct blk_trace *bt = q->blk_trace;
945 
946 	if (bt)
947 		__blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
948 }
949 
950 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
951 				    unsigned int depth, bool explicit)
952 {
953 	struct blk_trace *bt = q->blk_trace;
954 
955 	if (bt) {
956 		__be64 rpdu = cpu_to_be64(depth);
957 		u32 what;
958 
959 		if (explicit)
960 			what = BLK_TA_UNPLUG_IO;
961 		else
962 			what = BLK_TA_UNPLUG_TIMER;
963 
964 		__blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
965 	}
966 }
967 
968 static void blk_add_trace_split(void *ignore,
969 				struct request_queue *q, struct bio *bio,
970 				unsigned int pdu)
971 {
972 	struct blk_trace *bt = q->blk_trace;
973 
974 	if (bt) {
975 		__be64 rpdu = cpu_to_be64(pdu);
976 
977 		__blk_add_trace(bt, bio->bi_iter.bi_sector,
978 				bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
979 				BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
980 				&rpdu, blk_trace_bio_get_cgid(q, bio));
981 	}
982 }
983 
984 /**
985  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
986  * @ignore:	trace callback data parameter (not used)
987  * @q:		queue the io is for
988  * @bio:	the source bio
989  * @dev:	target device
990  * @from:	source sector
991  *
992  * Description:
993  *     Device mapper or raid target sometimes need to split a bio because
994  *     it spans a stripe (or similar). Add a trace for that action.
995  *
996  **/
997 static void blk_add_trace_bio_remap(void *ignore,
998 				    struct request_queue *q, struct bio *bio,
999 				    dev_t dev, sector_t from)
1000 {
1001 	struct blk_trace *bt = q->blk_trace;
1002 	struct blk_io_trace_remap r;
1003 
1004 	if (likely(!bt))
1005 		return;
1006 
1007 	r.device_from = cpu_to_be32(dev);
1008 	r.device_to   = cpu_to_be32(bio_dev(bio));
1009 	r.sector_from = cpu_to_be64(from);
1010 
1011 	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1012 			bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1013 			sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1014 }
1015 
1016 /**
1017  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1018  * @ignore:	trace callback data parameter (not used)
1019  * @q:		queue the io is for
1020  * @rq:		the source request
1021  * @dev:	target device
1022  * @from:	source sector
1023  *
1024  * Description:
1025  *     Device mapper remaps request to other devices.
1026  *     Add a trace for that action.
1027  *
1028  **/
1029 static void blk_add_trace_rq_remap(void *ignore,
1030 				   struct request_queue *q,
1031 				   struct request *rq, dev_t dev,
1032 				   sector_t from)
1033 {
1034 	struct blk_trace *bt = q->blk_trace;
1035 	struct blk_io_trace_remap r;
1036 
1037 	if (likely(!bt))
1038 		return;
1039 
1040 	r.device_from = cpu_to_be32(dev);
1041 	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1042 	r.sector_from = cpu_to_be64(from);
1043 
1044 	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1045 			rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1046 			sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1047 }
1048 
1049 /**
1050  * blk_add_driver_data - Add binary message with driver-specific data
1051  * @q:		queue the io is for
1052  * @rq:		io request
1053  * @data:	driver-specific data
1054  * @len:	length of driver-specific data
1055  *
1056  * Description:
1057  *     Some drivers might want to write driver-specific data per request.
1058  *
1059  **/
1060 void blk_add_driver_data(struct request_queue *q,
1061 			 struct request *rq,
1062 			 void *data, size_t len)
1063 {
1064 	struct blk_trace *bt = q->blk_trace;
1065 
1066 	if (likely(!bt))
1067 		return;
1068 
1069 	__blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1070 				BLK_TA_DRV_DATA, 0, len, data,
1071 				blk_trace_request_get_cgid(q, rq));
1072 }
1073 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1074 
1075 static void blk_register_tracepoints(void)
1076 {
1077 	int ret;
1078 
1079 	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1080 	WARN_ON(ret);
1081 	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1082 	WARN_ON(ret);
1083 	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1084 	WARN_ON(ret);
1085 	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1086 	WARN_ON(ret);
1087 	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1088 	WARN_ON(ret);
1089 	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1090 	WARN_ON(ret);
1091 	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1092 	WARN_ON(ret);
1093 	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1094 	WARN_ON(ret);
1095 	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1096 	WARN_ON(ret);
1097 	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1098 	WARN_ON(ret);
1099 	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1100 	WARN_ON(ret);
1101 	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1102 	WARN_ON(ret);
1103 	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1104 	WARN_ON(ret);
1105 	ret = register_trace_block_split(blk_add_trace_split, NULL);
1106 	WARN_ON(ret);
1107 	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1108 	WARN_ON(ret);
1109 	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1110 	WARN_ON(ret);
1111 }
1112 
1113 static void blk_unregister_tracepoints(void)
1114 {
1115 	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1116 	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1117 	unregister_trace_block_split(blk_add_trace_split, NULL);
1118 	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1119 	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1120 	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1121 	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1122 	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1123 	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1124 	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1125 	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1126 	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1127 	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1128 	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1129 	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1130 	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1131 
1132 	tracepoint_synchronize_unregister();
1133 }
1134 
1135 /*
1136  * struct blk_io_tracer formatting routines
1137  */
1138 
1139 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1140 {
1141 	int i = 0;
1142 	int tc = t->action >> BLK_TC_SHIFT;
1143 
1144 	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1145 		rwbs[i++] = 'N';
1146 		goto out;
1147 	}
1148 
1149 	if (tc & BLK_TC_FLUSH)
1150 		rwbs[i++] = 'F';
1151 
1152 	if (tc & BLK_TC_DISCARD)
1153 		rwbs[i++] = 'D';
1154 	else if (tc & BLK_TC_WRITE)
1155 		rwbs[i++] = 'W';
1156 	else if (t->bytes)
1157 		rwbs[i++] = 'R';
1158 	else
1159 		rwbs[i++] = 'N';
1160 
1161 	if (tc & BLK_TC_FUA)
1162 		rwbs[i++] = 'F';
1163 	if (tc & BLK_TC_AHEAD)
1164 		rwbs[i++] = 'A';
1165 	if (tc & BLK_TC_SYNC)
1166 		rwbs[i++] = 'S';
1167 	if (tc & BLK_TC_META)
1168 		rwbs[i++] = 'M';
1169 out:
1170 	rwbs[i] = '\0';
1171 }
1172 
1173 static inline
1174 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1175 {
1176 	return (const struct blk_io_trace *)ent;
1177 }
1178 
1179 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1180 {
1181 	return (void *)(te_blk_io_trace(ent) + 1) +
1182 		(has_cg ? sizeof(union kernfs_node_id) : 0);
1183 }
1184 
1185 static inline const void *cgid_start(const struct trace_entry *ent)
1186 {
1187 	return (void *)(te_blk_io_trace(ent) + 1);
1188 }
1189 
1190 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1191 {
1192 	return te_blk_io_trace(ent)->pdu_len -
1193 			(has_cg ? sizeof(union kernfs_node_id) : 0);
1194 }
1195 
1196 static inline u32 t_action(const struct trace_entry *ent)
1197 {
1198 	return te_blk_io_trace(ent)->action;
1199 }
1200 
1201 static inline u32 t_bytes(const struct trace_entry *ent)
1202 {
1203 	return te_blk_io_trace(ent)->bytes;
1204 }
1205 
1206 static inline u32 t_sec(const struct trace_entry *ent)
1207 {
1208 	return te_blk_io_trace(ent)->bytes >> 9;
1209 }
1210 
1211 static inline unsigned long long t_sector(const struct trace_entry *ent)
1212 {
1213 	return te_blk_io_trace(ent)->sector;
1214 }
1215 
1216 static inline __u16 t_error(const struct trace_entry *ent)
1217 {
1218 	return te_blk_io_trace(ent)->error;
1219 }
1220 
1221 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1222 {
1223 	const __u64 *val = pdu_start(ent, has_cg);
1224 	return be64_to_cpu(*val);
1225 }
1226 
1227 static void get_pdu_remap(const struct trace_entry *ent,
1228 			  struct blk_io_trace_remap *r, bool has_cg)
1229 {
1230 	const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1231 	__u64 sector_from = __r->sector_from;
1232 
1233 	r->device_from = be32_to_cpu(__r->device_from);
1234 	r->device_to   = be32_to_cpu(__r->device_to);
1235 	r->sector_from = be64_to_cpu(sector_from);
1236 }
1237 
1238 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1239 	bool has_cg);
1240 
1241 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1242 	bool has_cg)
1243 {
1244 	char rwbs[RWBS_LEN];
1245 	unsigned long long ts  = iter->ts;
1246 	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1247 	unsigned secs	       = (unsigned long)ts;
1248 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1249 
1250 	fill_rwbs(rwbs, t);
1251 
1252 	trace_seq_printf(&iter->seq,
1253 			 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1254 			 MAJOR(t->device), MINOR(t->device), iter->cpu,
1255 			 secs, nsec_rem, iter->ent->pid, act, rwbs);
1256 }
1257 
1258 static void blk_log_action(struct trace_iterator *iter, const char *act,
1259 	bool has_cg)
1260 {
1261 	char rwbs[RWBS_LEN];
1262 	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1263 
1264 	fill_rwbs(rwbs, t);
1265 	if (has_cg) {
1266 		const union kernfs_node_id *id = cgid_start(iter->ent);
1267 
1268 		if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1269 			char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1270 
1271 			cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1272 				sizeof(blkcg_name_buf));
1273 			trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1274 				 MAJOR(t->device), MINOR(t->device),
1275 				 blkcg_name_buf, act, rwbs);
1276 		} else
1277 			trace_seq_printf(&iter->seq,
1278 				 "%3d,%-3d %x,%-x %2s %3s ",
1279 				 MAJOR(t->device), MINOR(t->device),
1280 				 id->ino, id->generation, act, rwbs);
1281 	} else
1282 		trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1283 				 MAJOR(t->device), MINOR(t->device), act, rwbs);
1284 }
1285 
1286 static void blk_log_dump_pdu(struct trace_seq *s,
1287 	const struct trace_entry *ent, bool has_cg)
1288 {
1289 	const unsigned char *pdu_buf;
1290 	int pdu_len;
1291 	int i, end;
1292 
1293 	pdu_buf = pdu_start(ent, has_cg);
1294 	pdu_len = pdu_real_len(ent, has_cg);
1295 
1296 	if (!pdu_len)
1297 		return;
1298 
1299 	/* find the last zero that needs to be printed */
1300 	for (end = pdu_len - 1; end >= 0; end--)
1301 		if (pdu_buf[end])
1302 			break;
1303 	end++;
1304 
1305 	trace_seq_putc(s, '(');
1306 
1307 	for (i = 0; i < pdu_len; i++) {
1308 
1309 		trace_seq_printf(s, "%s%02x",
1310 				 i == 0 ? "" : " ", pdu_buf[i]);
1311 
1312 		/*
1313 		 * stop when the rest is just zeroes and indicate so
1314 		 * with a ".." appended
1315 		 */
1316 		if (i == end && end != pdu_len - 1) {
1317 			trace_seq_puts(s, " ..) ");
1318 			return;
1319 		}
1320 	}
1321 
1322 	trace_seq_puts(s, ") ");
1323 }
1324 
1325 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1326 {
1327 	char cmd[TASK_COMM_LEN];
1328 
1329 	trace_find_cmdline(ent->pid, cmd);
1330 
1331 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1332 		trace_seq_printf(s, "%u ", t_bytes(ent));
1333 		blk_log_dump_pdu(s, ent, has_cg);
1334 		trace_seq_printf(s, "[%s]\n", cmd);
1335 	} else {
1336 		if (t_sec(ent))
1337 			trace_seq_printf(s, "%llu + %u [%s]\n",
1338 						t_sector(ent), t_sec(ent), cmd);
1339 		else
1340 			trace_seq_printf(s, "[%s]\n", cmd);
1341 	}
1342 }
1343 
1344 static void blk_log_with_error(struct trace_seq *s,
1345 			      const struct trace_entry *ent, bool has_cg)
1346 {
1347 	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1348 		blk_log_dump_pdu(s, ent, has_cg);
1349 		trace_seq_printf(s, "[%d]\n", t_error(ent));
1350 	} else {
1351 		if (t_sec(ent))
1352 			trace_seq_printf(s, "%llu + %u [%d]\n",
1353 					 t_sector(ent),
1354 					 t_sec(ent), t_error(ent));
1355 		else
1356 			trace_seq_printf(s, "%llu [%d]\n",
1357 					 t_sector(ent), t_error(ent));
1358 	}
1359 }
1360 
1361 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1362 {
1363 	struct blk_io_trace_remap r = { .device_from = 0, };
1364 
1365 	get_pdu_remap(ent, &r, has_cg);
1366 	trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1367 			 t_sector(ent), t_sec(ent),
1368 			 MAJOR(r.device_from), MINOR(r.device_from),
1369 			 (unsigned long long)r.sector_from);
1370 }
1371 
1372 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1373 {
1374 	char cmd[TASK_COMM_LEN];
1375 
1376 	trace_find_cmdline(ent->pid, cmd);
1377 
1378 	trace_seq_printf(s, "[%s]\n", cmd);
1379 }
1380 
1381 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1382 {
1383 	char cmd[TASK_COMM_LEN];
1384 
1385 	trace_find_cmdline(ent->pid, cmd);
1386 
1387 	trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1388 }
1389 
1390 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1391 {
1392 	char cmd[TASK_COMM_LEN];
1393 
1394 	trace_find_cmdline(ent->pid, cmd);
1395 
1396 	trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1397 			 get_pdu_int(ent, has_cg), cmd);
1398 }
1399 
1400 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1401 			bool has_cg)
1402 {
1403 
1404 	trace_seq_putmem(s, pdu_start(ent, has_cg),
1405 		pdu_real_len(ent, has_cg));
1406 	trace_seq_putc(s, '\n');
1407 }
1408 
1409 /*
1410  * struct tracer operations
1411  */
1412 
1413 static void blk_tracer_print_header(struct seq_file *m)
1414 {
1415 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1416 		return;
1417 	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1418 		    "#  |     |     |           |   |   |\n");
1419 }
1420 
1421 static void blk_tracer_start(struct trace_array *tr)
1422 {
1423 	blk_tracer_enabled = true;
1424 }
1425 
1426 static int blk_tracer_init(struct trace_array *tr)
1427 {
1428 	blk_tr = tr;
1429 	blk_tracer_start(tr);
1430 	return 0;
1431 }
1432 
1433 static void blk_tracer_stop(struct trace_array *tr)
1434 {
1435 	blk_tracer_enabled = false;
1436 }
1437 
1438 static void blk_tracer_reset(struct trace_array *tr)
1439 {
1440 	blk_tracer_stop(tr);
1441 }
1442 
1443 static const struct {
1444 	const char *act[2];
1445 	void	   (*print)(struct trace_seq *s, const struct trace_entry *ent,
1446 			    bool has_cg);
1447 } what2act[] = {
1448 	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1449 	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1450 	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1451 	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1452 	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1453 	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1454 	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1455 	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1456 	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1457 	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1458 	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1459 	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1460 	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1461 	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1462 	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1463 };
1464 
1465 static enum print_line_t print_one_line(struct trace_iterator *iter,
1466 					bool classic)
1467 {
1468 	struct trace_array *tr = iter->tr;
1469 	struct trace_seq *s = &iter->seq;
1470 	const struct blk_io_trace *t;
1471 	u16 what;
1472 	bool long_act;
1473 	blk_log_action_t *log_action;
1474 	bool has_cg;
1475 
1476 	t	   = te_blk_io_trace(iter->ent);
1477 	what	   = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1478 	long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1479 	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1480 	has_cg	   = t->action & __BLK_TA_CGROUP;
1481 
1482 	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1483 		log_action(iter, long_act ? "message" : "m", has_cg);
1484 		blk_log_msg(s, iter->ent, has_cg);
1485 		return trace_handle_return(s);
1486 	}
1487 
1488 	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1489 		trace_seq_printf(s, "Unknown action %x\n", what);
1490 	else {
1491 		log_action(iter, what2act[what].act[long_act], has_cg);
1492 		what2act[what].print(s, iter->ent, has_cg);
1493 	}
1494 
1495 	return trace_handle_return(s);
1496 }
1497 
1498 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1499 					       int flags, struct trace_event *event)
1500 {
1501 	return print_one_line(iter, false);
1502 }
1503 
1504 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1505 {
1506 	struct trace_seq *s = &iter->seq;
1507 	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1508 	const int offset = offsetof(struct blk_io_trace, sector);
1509 	struct blk_io_trace old = {
1510 		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1511 		.time     = iter->ts,
1512 	};
1513 
1514 	trace_seq_putmem(s, &old, offset);
1515 	trace_seq_putmem(s, &t->sector,
1516 			 sizeof(old) - offset + t->pdu_len);
1517 }
1518 
1519 static enum print_line_t
1520 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1521 			     struct trace_event *event)
1522 {
1523 	blk_trace_synthesize_old_trace(iter);
1524 
1525 	return trace_handle_return(&iter->seq);
1526 }
1527 
1528 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1529 {
1530 	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1531 		return TRACE_TYPE_UNHANDLED;
1532 
1533 	return print_one_line(iter, true);
1534 }
1535 
1536 static int
1537 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1538 {
1539 	/* don't output context-info for blk_classic output */
1540 	if (bit == TRACE_BLK_OPT_CLASSIC) {
1541 		if (set)
1542 			tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1543 		else
1544 			tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1545 	}
1546 	return 0;
1547 }
1548 
1549 static struct tracer blk_tracer __read_mostly = {
1550 	.name		= "blk",
1551 	.init		= blk_tracer_init,
1552 	.reset		= blk_tracer_reset,
1553 	.start		= blk_tracer_start,
1554 	.stop		= blk_tracer_stop,
1555 	.print_header	= blk_tracer_print_header,
1556 	.print_line	= blk_tracer_print_line,
1557 	.flags		= &blk_tracer_flags,
1558 	.set_flag	= blk_tracer_set_flag,
1559 };
1560 
1561 static struct trace_event_functions trace_blk_event_funcs = {
1562 	.trace		= blk_trace_event_print,
1563 	.binary		= blk_trace_event_print_binary,
1564 };
1565 
1566 static struct trace_event trace_blk_event = {
1567 	.type		= TRACE_BLK,
1568 	.funcs		= &trace_blk_event_funcs,
1569 };
1570 
1571 static int __init init_blk_tracer(void)
1572 {
1573 	if (!register_trace_event(&trace_blk_event)) {
1574 		pr_warn("Warning: could not register block events\n");
1575 		return 1;
1576 	}
1577 
1578 	if (register_tracer(&blk_tracer) != 0) {
1579 		pr_warn("Warning: could not register the block tracer\n");
1580 		unregister_trace_event(&trace_blk_event);
1581 		return 1;
1582 	}
1583 
1584 	return 0;
1585 }
1586 
1587 device_initcall(init_blk_tracer);
1588 
1589 static int blk_trace_remove_queue(struct request_queue *q)
1590 {
1591 	struct blk_trace *bt;
1592 
1593 	bt = xchg(&q->blk_trace, NULL);
1594 	if (bt == NULL)
1595 		return -EINVAL;
1596 
1597 	put_probe_ref();
1598 	blk_trace_free(bt);
1599 	return 0;
1600 }
1601 
1602 /*
1603  * Setup everything required to start tracing
1604  */
1605 static int blk_trace_setup_queue(struct request_queue *q,
1606 				 struct block_device *bdev)
1607 {
1608 	struct blk_trace *bt = NULL;
1609 	int ret = -ENOMEM;
1610 
1611 	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1612 	if (!bt)
1613 		return -ENOMEM;
1614 
1615 	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1616 	if (!bt->msg_data)
1617 		goto free_bt;
1618 
1619 	bt->dev = bdev->bd_dev;
1620 	bt->act_mask = (u16)-1;
1621 
1622 	blk_trace_setup_lba(bt, bdev);
1623 
1624 	ret = -EBUSY;
1625 	if (cmpxchg(&q->blk_trace, NULL, bt))
1626 		goto free_bt;
1627 
1628 	get_probe_ref();
1629 	return 0;
1630 
1631 free_bt:
1632 	blk_trace_free(bt);
1633 	return ret;
1634 }
1635 
1636 /*
1637  * sysfs interface to enable and configure tracing
1638  */
1639 
1640 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1641 					 struct device_attribute *attr,
1642 					 char *buf);
1643 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1644 					  struct device_attribute *attr,
1645 					  const char *buf, size_t count);
1646 #define BLK_TRACE_DEVICE_ATTR(_name) \
1647 	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1648 		    sysfs_blk_trace_attr_show, \
1649 		    sysfs_blk_trace_attr_store)
1650 
1651 static BLK_TRACE_DEVICE_ATTR(enable);
1652 static BLK_TRACE_DEVICE_ATTR(act_mask);
1653 static BLK_TRACE_DEVICE_ATTR(pid);
1654 static BLK_TRACE_DEVICE_ATTR(start_lba);
1655 static BLK_TRACE_DEVICE_ATTR(end_lba);
1656 
1657 static struct attribute *blk_trace_attrs[] = {
1658 	&dev_attr_enable.attr,
1659 	&dev_attr_act_mask.attr,
1660 	&dev_attr_pid.attr,
1661 	&dev_attr_start_lba.attr,
1662 	&dev_attr_end_lba.attr,
1663 	NULL
1664 };
1665 
1666 struct attribute_group blk_trace_attr_group = {
1667 	.name  = "trace",
1668 	.attrs = blk_trace_attrs,
1669 };
1670 
1671 static const struct {
1672 	int mask;
1673 	const char *str;
1674 } mask_maps[] = {
1675 	{ BLK_TC_READ,		"read"		},
1676 	{ BLK_TC_WRITE,		"write"		},
1677 	{ BLK_TC_FLUSH,		"flush"		},
1678 	{ BLK_TC_SYNC,		"sync"		},
1679 	{ BLK_TC_QUEUE,		"queue"		},
1680 	{ BLK_TC_REQUEUE,	"requeue"	},
1681 	{ BLK_TC_ISSUE,		"issue"		},
1682 	{ BLK_TC_COMPLETE,	"complete"	},
1683 	{ BLK_TC_FS,		"fs"		},
1684 	{ BLK_TC_PC,		"pc"		},
1685 	{ BLK_TC_NOTIFY,	"notify"	},
1686 	{ BLK_TC_AHEAD,		"ahead"		},
1687 	{ BLK_TC_META,		"meta"		},
1688 	{ BLK_TC_DISCARD,	"discard"	},
1689 	{ BLK_TC_DRV_DATA,	"drv_data"	},
1690 	{ BLK_TC_FUA,		"fua"		},
1691 };
1692 
1693 static int blk_trace_str2mask(const char *str)
1694 {
1695 	int i;
1696 	int mask = 0;
1697 	char *buf, *s, *token;
1698 
1699 	buf = kstrdup(str, GFP_KERNEL);
1700 	if (buf == NULL)
1701 		return -ENOMEM;
1702 	s = strstrip(buf);
1703 
1704 	while (1) {
1705 		token = strsep(&s, ",");
1706 		if (token == NULL)
1707 			break;
1708 
1709 		if (*token == '\0')
1710 			continue;
1711 
1712 		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1713 			if (strcasecmp(token, mask_maps[i].str) == 0) {
1714 				mask |= mask_maps[i].mask;
1715 				break;
1716 			}
1717 		}
1718 		if (i == ARRAY_SIZE(mask_maps)) {
1719 			mask = -EINVAL;
1720 			break;
1721 		}
1722 	}
1723 	kfree(buf);
1724 
1725 	return mask;
1726 }
1727 
1728 static ssize_t blk_trace_mask2str(char *buf, int mask)
1729 {
1730 	int i;
1731 	char *p = buf;
1732 
1733 	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1734 		if (mask & mask_maps[i].mask) {
1735 			p += sprintf(p, "%s%s",
1736 				    (p == buf) ? "" : ",", mask_maps[i].str);
1737 		}
1738 	}
1739 	*p++ = '\n';
1740 
1741 	return p - buf;
1742 }
1743 
1744 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1745 {
1746 	if (bdev->bd_disk == NULL)
1747 		return NULL;
1748 
1749 	return bdev_get_queue(bdev);
1750 }
1751 
1752 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1753 					 struct device_attribute *attr,
1754 					 char *buf)
1755 {
1756 	struct hd_struct *p = dev_to_part(dev);
1757 	struct request_queue *q;
1758 	struct block_device *bdev;
1759 	ssize_t ret = -ENXIO;
1760 
1761 	bdev = bdget(part_devt(p));
1762 	if (bdev == NULL)
1763 		goto out;
1764 
1765 	q = blk_trace_get_queue(bdev);
1766 	if (q == NULL)
1767 		goto out_bdput;
1768 
1769 	mutex_lock(&q->blk_trace_mutex);
1770 
1771 	if (attr == &dev_attr_enable) {
1772 		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1773 		goto out_unlock_bdev;
1774 	}
1775 
1776 	if (q->blk_trace == NULL)
1777 		ret = sprintf(buf, "disabled\n");
1778 	else if (attr == &dev_attr_act_mask)
1779 		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1780 	else if (attr == &dev_attr_pid)
1781 		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1782 	else if (attr == &dev_attr_start_lba)
1783 		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1784 	else if (attr == &dev_attr_end_lba)
1785 		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1786 
1787 out_unlock_bdev:
1788 	mutex_unlock(&q->blk_trace_mutex);
1789 out_bdput:
1790 	bdput(bdev);
1791 out:
1792 	return ret;
1793 }
1794 
1795 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1796 					  struct device_attribute *attr,
1797 					  const char *buf, size_t count)
1798 {
1799 	struct block_device *bdev;
1800 	struct request_queue *q;
1801 	struct hd_struct *p;
1802 	u64 value;
1803 	ssize_t ret = -EINVAL;
1804 
1805 	if (count == 0)
1806 		goto out;
1807 
1808 	if (attr == &dev_attr_act_mask) {
1809 		if (kstrtoull(buf, 0, &value)) {
1810 			/* Assume it is a list of trace category names */
1811 			ret = blk_trace_str2mask(buf);
1812 			if (ret < 0)
1813 				goto out;
1814 			value = ret;
1815 		}
1816 	} else if (kstrtoull(buf, 0, &value))
1817 		goto out;
1818 
1819 	ret = -ENXIO;
1820 
1821 	p = dev_to_part(dev);
1822 	bdev = bdget(part_devt(p));
1823 	if (bdev == NULL)
1824 		goto out;
1825 
1826 	q = blk_trace_get_queue(bdev);
1827 	if (q == NULL)
1828 		goto out_bdput;
1829 
1830 	mutex_lock(&q->blk_trace_mutex);
1831 
1832 	if (attr == &dev_attr_enable) {
1833 		if (!!value == !!q->blk_trace) {
1834 			ret = 0;
1835 			goto out_unlock_bdev;
1836 		}
1837 		if (value)
1838 			ret = blk_trace_setup_queue(q, bdev);
1839 		else
1840 			ret = blk_trace_remove_queue(q);
1841 		goto out_unlock_bdev;
1842 	}
1843 
1844 	ret = 0;
1845 	if (q->blk_trace == NULL)
1846 		ret = blk_trace_setup_queue(q, bdev);
1847 
1848 	if (ret == 0) {
1849 		if (attr == &dev_attr_act_mask)
1850 			q->blk_trace->act_mask = value;
1851 		else if (attr == &dev_attr_pid)
1852 			q->blk_trace->pid = value;
1853 		else if (attr == &dev_attr_start_lba)
1854 			q->blk_trace->start_lba = value;
1855 		else if (attr == &dev_attr_end_lba)
1856 			q->blk_trace->end_lba = value;
1857 	}
1858 
1859 out_unlock_bdev:
1860 	mutex_unlock(&q->blk_trace_mutex);
1861 out_bdput:
1862 	bdput(bdev);
1863 out:
1864 	return ret ? ret : count;
1865 }
1866 
1867 int blk_trace_init_sysfs(struct device *dev)
1868 {
1869 	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1870 }
1871 
1872 void blk_trace_remove_sysfs(struct device *dev)
1873 {
1874 	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1875 }
1876 
1877 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1878 
1879 #ifdef CONFIG_EVENT_TRACING
1880 
1881 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1882 {
1883 	int i = 0;
1884 
1885 	if (op & REQ_PREFLUSH)
1886 		rwbs[i++] = 'F';
1887 
1888 	switch (op & REQ_OP_MASK) {
1889 	case REQ_OP_WRITE:
1890 	case REQ_OP_WRITE_SAME:
1891 		rwbs[i++] = 'W';
1892 		break;
1893 	case REQ_OP_DISCARD:
1894 		rwbs[i++] = 'D';
1895 		break;
1896 	case REQ_OP_SECURE_ERASE:
1897 		rwbs[i++] = 'D';
1898 		rwbs[i++] = 'E';
1899 		break;
1900 	case REQ_OP_FLUSH:
1901 		rwbs[i++] = 'F';
1902 		break;
1903 	case REQ_OP_READ:
1904 		rwbs[i++] = 'R';
1905 		break;
1906 	default:
1907 		rwbs[i++] = 'N';
1908 	}
1909 
1910 	if (op & REQ_FUA)
1911 		rwbs[i++] = 'F';
1912 	if (op & REQ_RAHEAD)
1913 		rwbs[i++] = 'A';
1914 	if (op & REQ_SYNC)
1915 		rwbs[i++] = 'S';
1916 	if (op & REQ_META)
1917 		rwbs[i++] = 'M';
1918 
1919 	rwbs[i] = '\0';
1920 }
1921 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1922 
1923 #endif /* CONFIG_EVENT_TRACING */
1924 
1925