projects/linux·blob v5.7 · refs

1981 lines · 46.7 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2006 Jens Axboe <[email protected]>
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
28static unsigned int blktrace_seq __read_mostly = 1;
29
30static struct trace_array *blk_tr;
31static bool blk_tracer_enabled __read_mostly;
32
33static LIST_HEAD(running_trace_list);
34static __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
41static 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
51static struct tracer_flags blk_tracer_flags = {
52 .val = 0,
53 .opts = blk_tracer_opts,
54};
55
56/* Global reference count of probes */
57static DEFINE_MUTEX(blk_probe_mutex);
58static int blk_probes_ref;
59
60static void blk_register_tracepoints(void);
61static void blk_unregister_tracepoints(void);
62
63/*
64 * Send out a notify message.
65 */
66static void trace_note(struct blk_trace *bt, pid_t pid, int action,
67 const void *data, size_t len, u64 cgid)
68{
69 struct blk_io_trace *t;
70 struct ring_buffer_event *event = NULL;
71 struct trace_buffer *buffer = NULL;
72 int pc = 0;
73 int cpu = smp_processor_id();
74 bool blk_tracer = blk_tracer_enabled;
75 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
76
77 if (blk_tracer) {
78 buffer = blk_tr->array_buffer.buffer;
79 pc = preempt_count();
80 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
81 sizeof(*t) + len + cgid_len,
82 0, pc);
83 if (!event)
84 return;
85 t = ring_buffer_event_data(event);
86 goto record_it;
87 }
88
89 if (!bt->rchan)
90 return;
91
92 t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
93 if (t) {
94 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
95 t->time = ktime_to_ns(ktime_get());
96record_it:
97 t->device = bt->dev;
98 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
99 t->pid = pid;
100 t->cpu = cpu;
101 t->pdu_len = len + cgid_len;
102 if (cgid_len)
103 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
104 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
105
106 if (blk_tracer)
107 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
108 }
109}
110
111/*
112 * Send out a notify for this process, if we haven't done so since a trace
113 * started
114 */
115static void trace_note_tsk(struct task_struct *tsk)
116{
117 unsigned long flags;
118 struct blk_trace *bt;
119
120 tsk->btrace_seq = blktrace_seq;
121 spin_lock_irqsave(&running_trace_lock, flags);
122 list_for_each_entry(bt, &running_trace_list, running_list) {
123 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
124 sizeof(tsk->comm), 0);
125 }
126 spin_unlock_irqrestore(&running_trace_lock, flags);
127}
128
129static void trace_note_time(struct blk_trace *bt)
130{
131 struct timespec64 now;
132 unsigned long flags;
133 u32 words[2];
134
135 /* need to check user space to see if this breaks in y2038 or y2106 */
136 ktime_get_real_ts64(&now);
137 words[0] = (u32)now.tv_sec;
138 words[1] = now.tv_nsec;
139
140 local_irq_save(flags);
141 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
142 local_irq_restore(flags);
143}
144
145void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
146 const char *fmt, ...)
147{
148 int n;
149 va_list args;
150 unsigned long flags;
151 char *buf;
152
153 if (unlikely(bt->trace_state != Blktrace_running &&
154 !blk_tracer_enabled))
155 return;
156
157 /*
158 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
159 * message to the trace.
160 */
161 if (!(bt->act_mask & BLK_TC_NOTIFY))
162 return;
163
164 local_irq_save(flags);
165 buf = this_cpu_ptr(bt->msg_data);
166 va_start(args, fmt);
167 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
168 va_end(args);
169
170 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
171 blkcg = NULL;
172#ifdef CONFIG_BLK_CGROUP
173 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
174 blkcg ? cgroup_id(blkcg->css.cgroup) : 1);
175#else
176 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, 0);
177#endif
178 local_irq_restore(flags);
179}
180EXPORT_SYMBOL_GPL(__trace_note_message);
181
182static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
183 pid_t pid)
184{
185 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
186 return 1;
187 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
188 return 1;
189 if (bt->pid && pid != bt->pid)
190 return 1;
191
192 return 0;
193}
194
195/*
196 * Data direction bit lookup
197 */
198static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
199 BLK_TC_ACT(BLK_TC_WRITE) };
200
201#define BLK_TC_RAHEAD BLK_TC_AHEAD
202#define BLK_TC_PREFLUSH BLK_TC_FLUSH
203
204/* The ilog2() calls fall out because they're constant */
205#define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
206 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
207
208/*
209 * The worker for the various blk_add_trace*() types. Fills out a
210 * blk_io_trace structure and places it in a per-cpu subbuffer.
211 */
212static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
213 int op, int op_flags, u32 what, int error, int pdu_len,
214 void *pdu_data, u64 cgid)
215{
216 struct task_struct *tsk = current;
217 struct ring_buffer_event *event = NULL;
218 struct trace_buffer *buffer = NULL;
219 struct blk_io_trace *t;
220 unsigned long flags = 0;
221 unsigned long *sequence;
222 pid_t pid;
223 int cpu, pc = 0;
224 bool blk_tracer = blk_tracer_enabled;
225 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
226
227 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
228 return;
229
230 what |= ddir_act[op_is_write(op) ? WRITE : READ];
231 what |= MASK_TC_BIT(op_flags, SYNC);
232 what |= MASK_TC_BIT(op_flags, RAHEAD);
233 what |= MASK_TC_BIT(op_flags, META);
234 what |= MASK_TC_BIT(op_flags, PREFLUSH);
235 what |= MASK_TC_BIT(op_flags, FUA);
236 if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
237 what |= BLK_TC_ACT(BLK_TC_DISCARD);
238 if (op == REQ_OP_FLUSH)
239 what |= BLK_TC_ACT(BLK_TC_FLUSH);
240 if (cgid)
241 what |= __BLK_TA_CGROUP;
242
243 pid = tsk->pid;
244 if (act_log_check(bt, what, sector, pid))
245 return;
246 cpu = raw_smp_processor_id();
247
248 if (blk_tracer) {
249 tracing_record_cmdline(current);
250
251 buffer = blk_tr->array_buffer.buffer;
252 pc = preempt_count();
253 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
254 sizeof(*t) + pdu_len + cgid_len,
255 0, pc);
256 if (!event)
257 return;
258 t = ring_buffer_event_data(event);
259 goto record_it;
260 }
261
262 if (unlikely(tsk->btrace_seq != blktrace_seq))
263 trace_note_tsk(tsk);
264
265 /*
266 * A word about the locking here - we disable interrupts to reserve
267 * some space in the relay per-cpu buffer, to prevent an irq
268 * from coming in and stepping on our toes.
269 */
270 local_irq_save(flags);
271 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
272 if (t) {
273 sequence = per_cpu_ptr(bt->sequence, cpu);
274
275 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
276 t->sequence = ++(*sequence);
277 t->time = ktime_to_ns(ktime_get());
278record_it:
279 /*
280 * These two are not needed in ftrace as they are in the
281 * generic trace_entry, filled by tracing_generic_entry_update,
282 * but for the trace_event->bin() synthesizer benefit we do it
283 * here too.
284 */
285 t->cpu = cpu;
286 t->pid = pid;
287
288 t->sector = sector;
289 t->bytes = bytes;
290 t->action = what;
291 t->device = bt->dev;
292 t->error = error;
293 t->pdu_len = pdu_len + cgid_len;
294
295 if (cgid_len)
296 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
297 if (pdu_len)
298 memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
299
300 if (blk_tracer) {
301 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
302 return;
303 }
304 }
305
306 local_irq_restore(flags);
307}
308
309static void blk_trace_free(struct blk_trace *bt)
310{
311 debugfs_remove(bt->msg_file);
312 debugfs_remove(bt->dropped_file);
313 relay_close(bt->rchan);
314 debugfs_remove(bt->dir);
315 free_percpu(bt->sequence);
316 free_percpu(bt->msg_data);
317 kfree(bt);
318}
319
320static void get_probe_ref(void)
321{
322 mutex_lock(&blk_probe_mutex);
323 if (++blk_probes_ref == 1)
324 blk_register_tracepoints();
325 mutex_unlock(&blk_probe_mutex);
326}
327
328static void put_probe_ref(void)
329{
330 mutex_lock(&blk_probe_mutex);
331 if (!--blk_probes_ref)
332 blk_unregister_tracepoints();
333 mutex_unlock(&blk_probe_mutex);
334}
335
336static void blk_trace_cleanup(struct blk_trace *bt)
337{
338 synchronize_rcu();
339 blk_trace_free(bt);
340 put_probe_ref();
341}
342
343static 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
357int 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}
367EXPORT_SYMBOL_GPL(blk_trace_remove);
368
369static 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
380static 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
387static 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
407static 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 */
418static 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
431static int blk_remove_buf_file_callback(struct dentry *dentry)
432{
433 debugfs_remove(dentry);
434
435 return 0;
436}
437
438static 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
448static 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
454static 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 */
474static 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
516 bt->dev = dev;
517 atomic_set(&bt->dropped, 0);
518 INIT_LIST_HEAD(&bt->running_list);
519
520 ret = -EIO;
521 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
522 &blk_dropped_fops);
523
524 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
525
526 bt->rchan = relay_open("trace", dir, buts->buf_size,
527 buts->buf_nr, &blk_relay_callbacks, bt);
528 if (!bt->rchan)
529 goto err;
530
531 bt->act_mask = buts->act_mask;
532 if (!bt->act_mask)
533 bt->act_mask = (u16) -1;
534
535 blk_trace_setup_lba(bt, bdev);
536
537 /* overwrite with user settings */
538 if (buts->start_lba)
539 bt->start_lba = buts->start_lba;
540 if (buts->end_lba)
541 bt->end_lba = buts->end_lba;
542
543 bt->pid = buts->pid;
544 bt->trace_state = Blktrace_setup;
545
546 ret = -EBUSY;
547 if (cmpxchg(&q->blk_trace, NULL, bt))
548 goto err;
549
550 get_probe_ref();
551
552 ret = 0;
553err:
554 if (dir && !bt->dir)
555 dput(dir);
556 if (ret)
557 blk_trace_free(bt);
558 return ret;
559}
560
561static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
562 struct block_device *bdev, char __user *arg)
563{
564 struct blk_user_trace_setup buts;
565 int ret;
566
567 ret = copy_from_user(&buts, arg, sizeof(buts));
568 if (ret)
569 return -EFAULT;
570
571 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
572 if (ret)
573 return ret;
574
575 if (copy_to_user(arg, &buts, sizeof(buts))) {
576 __blk_trace_remove(q);
577 return -EFAULT;
578 }
579 return 0;
580}
581
582int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
583 struct block_device *bdev,
584 char __user *arg)
585{
586 int ret;
587
588 mutex_lock(&q->blk_trace_mutex);
589 ret = __blk_trace_setup(q, name, dev, bdev, arg);
590 mutex_unlock(&q->blk_trace_mutex);
591
592 return ret;
593}
594EXPORT_SYMBOL_GPL(blk_trace_setup);
595
596#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
597static int compat_blk_trace_setup(struct request_queue *q, char *name,
598 dev_t dev, struct block_device *bdev,
599 char __user *arg)
600{
601 struct blk_user_trace_setup buts;
602 struct compat_blk_user_trace_setup cbuts;
603 int ret;
604
605 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
606 return -EFAULT;
607
608 buts = (struct blk_user_trace_setup) {
609 .act_mask = cbuts.act_mask,
610 .buf_size = cbuts.buf_size,
611 .buf_nr = cbuts.buf_nr,
612 .start_lba = cbuts.start_lba,
613 .end_lba = cbuts.end_lba,
614 .pid = cbuts.pid,
615 };
616
617 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
618 if (ret)
619 return ret;
620
621 if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
622 __blk_trace_remove(q);
623 return -EFAULT;
624 }
625
626 return 0;
627}
628#endif
629
630static int __blk_trace_startstop(struct request_queue *q, int start)
631{
632 int ret;
633 struct blk_trace *bt;
634
635 bt = rcu_dereference_protected(q->blk_trace,
636 lockdep_is_held(&q->blk_trace_mutex));
637 if (bt == NULL)
638 return -EINVAL;
639
640 /*
641 * For starting a trace, we can transition from a setup or stopped
642 * trace. For stopping a trace, the state must be running
643 */
644 ret = -EINVAL;
645 if (start) {
646 if (bt->trace_state == Blktrace_setup ||
647 bt->trace_state == Blktrace_stopped) {
648 blktrace_seq++;
649 smp_mb();
650 bt->trace_state = Blktrace_running;
651 spin_lock_irq(&running_trace_lock);
652 list_add(&bt->running_list, &running_trace_list);
653 spin_unlock_irq(&running_trace_lock);
654
655 trace_note_time(bt);
656 ret = 0;
657 }
658 } else {
659 if (bt->trace_state == Blktrace_running) {
660 bt->trace_state = Blktrace_stopped;
661 spin_lock_irq(&running_trace_lock);
662 list_del_init(&bt->running_list);
663 spin_unlock_irq(&running_trace_lock);
664 relay_flush(bt->rchan);
665 ret = 0;
666 }
667 }
668
669 return ret;
670}
671
672int blk_trace_startstop(struct request_queue *q, int start)
673{
674 int ret;
675
676 mutex_lock(&q->blk_trace_mutex);
677 ret = __blk_trace_startstop(q, start);
678 mutex_unlock(&q->blk_trace_mutex);
679
680 return ret;
681}
682EXPORT_SYMBOL_GPL(blk_trace_startstop);
683
684/*
685 * When reading or writing the blktrace sysfs files, the references to the
686 * opened sysfs or device files should prevent the underlying block device
687 * from being removed. So no further delete protection is really needed.
688 */
689
690/**
691 * blk_trace_ioctl: - handle the ioctls associated with tracing
692 * @bdev: the block device
693 * @cmd: the ioctl cmd
694 * @arg: the argument data, if any
695 *
696 **/
697int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
698{
699 struct request_queue *q;
700 int ret, start = 0;
701 char b[BDEVNAME_SIZE];
702
703 q = bdev_get_queue(bdev);
704 if (!q)
705 return -ENXIO;
706
707 mutex_lock(&q->blk_trace_mutex);
708
709 switch (cmd) {
710 case BLKTRACESETUP:
711 bdevname(bdev, b);
712 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
713 break;
714#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
715 case BLKTRACESETUP32:
716 bdevname(bdev, b);
717 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
718 break;
719#endif
720 case BLKTRACESTART:
721 start = 1;
722 /* fall through */
723 case BLKTRACESTOP:
724 ret = __blk_trace_startstop(q, start);
725 break;
726 case BLKTRACETEARDOWN:
727 ret = __blk_trace_remove(q);
728 break;
729 default:
730 ret = -ENOTTY;
731 break;
732 }
733
734 mutex_unlock(&q->blk_trace_mutex);
735 return ret;
736}
737
738/**
739 * blk_trace_shutdown: - stop and cleanup trace structures
740 * @q: the request queue associated with the device
741 *
742 **/
743void blk_trace_shutdown(struct request_queue *q)
744{
745 mutex_lock(&q->blk_trace_mutex);
746 if (rcu_dereference_protected(q->blk_trace,
747 lockdep_is_held(&q->blk_trace_mutex))) {
748 __blk_trace_startstop(q, 0);
749 __blk_trace_remove(q);
750 }
751
752 mutex_unlock(&q->blk_trace_mutex);
753}
754
755#ifdef CONFIG_BLK_CGROUP
756static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
757{
758 struct blk_trace *bt;
759
760 /* We don't use the 'bt' value here except as an optimization... */
761 bt = rcu_dereference_protected(q->blk_trace, 1);
762 if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
763 return 0;
764
765 if (!bio->bi_blkg)
766 return 0;
767 return cgroup_id(bio_blkcg(bio)->css.cgroup);
768}
769#else
770u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
771{
772 return 0;
773}
774#endif
775
776static u64
777blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
778{
779 if (!rq->bio)
780 return 0;
781 /* Use the first bio */
782 return blk_trace_bio_get_cgid(q, rq->bio);
783}
784
785/*
786 * blktrace probes
787 */
788
789/**
790 * blk_add_trace_rq - Add a trace for a request oriented action
791 * @rq: the source request
792 * @error: return status to log
793 * @nr_bytes: number of completed bytes
794 * @what: the action
795 * @cgid: the cgroup info
796 *
797 * Description:
798 * Records an action against a request. Will log the bio offset + size.
799 *
800 **/
801static void blk_add_trace_rq(struct request *rq, int error,
802 unsigned int nr_bytes, u32 what, u64 cgid)
803{
804 struct blk_trace *bt;
805
806 rcu_read_lock();
807 bt = rcu_dereference(rq->q->blk_trace);
808 if (likely(!bt)) {
809 rcu_read_unlock();
810 return;
811 }
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 rcu_read_unlock();
821}
822
823static 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
830static 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
837static 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
845static 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 **/
863static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
864 u32 what, int error)
865{
866 struct blk_trace *bt;
867
868 rcu_read_lock();
869 bt = rcu_dereference(q->blk_trace);
870 if (likely(!bt)) {
871 rcu_read_unlock();
872 return;
873 }
874
875 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
876 bio_op(bio), bio->bi_opf, what, error, 0, NULL,
877 blk_trace_bio_get_cgid(q, bio));
878 rcu_read_unlock();
879}
880
881static void blk_add_trace_bio_bounce(void *ignore,
882 struct request_queue *q, struct bio *bio)
883{
884 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
885}
886
887static void blk_add_trace_bio_complete(void *ignore,
888 struct request_queue *q, struct bio *bio,
889 int error)
890{
891 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
892}
893
894static void blk_add_trace_bio_backmerge(void *ignore,
895 struct request_queue *q,
896 struct request *rq,
897 struct bio *bio)
898{
899 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
900}
901
902static void blk_add_trace_bio_frontmerge(void *ignore,
903 struct request_queue *q,
904 struct request *rq,
905 struct bio *bio)
906{
907 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
908}
909
910static void blk_add_trace_bio_queue(void *ignore,
911 struct request_queue *q, struct bio *bio)
912{
913 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
914}
915
916static void blk_add_trace_getrq(void *ignore,
917 struct request_queue *q,
918 struct bio *bio, int rw)
919{
920 if (bio)
921 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
922 else {
923 struct blk_trace *bt;
924
925 rcu_read_lock();
926 bt = rcu_dereference(q->blk_trace);
927 if (bt)
928 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
929 NULL, 0);
930 rcu_read_unlock();
931 }
932}
933
934
935static void blk_add_trace_sleeprq(void *ignore,
936 struct request_queue *q,
937 struct bio *bio, int rw)
938{
939 if (bio)
940 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
941 else {
942 struct blk_trace *bt;
943
944 rcu_read_lock();
945 bt = rcu_dereference(q->blk_trace);
946 if (bt)
947 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
948 0, 0, NULL, 0);
949 rcu_read_unlock();
950 }
951}
952
953static void blk_add_trace_plug(void *ignore, struct request_queue *q)
954{
955 struct blk_trace *bt;
956
957 rcu_read_lock();
958 bt = rcu_dereference(q->blk_trace);
959 if (bt)
960 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
961 rcu_read_unlock();
962}
963
964static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
965 unsigned int depth, bool explicit)
966{
967 struct blk_trace *bt;
968
969 rcu_read_lock();
970 bt = rcu_dereference(q->blk_trace);
971 if (bt) {
972 __be64 rpdu = cpu_to_be64(depth);
973 u32 what;
974
975 if (explicit)
976 what = BLK_TA_UNPLUG_IO;
977 else
978 what = BLK_TA_UNPLUG_TIMER;
979
980 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
981 }
982 rcu_read_unlock();
983}
984
985static void blk_add_trace_split(void *ignore,
986 struct request_queue *q, struct bio *bio,
987 unsigned int pdu)
988{
989 struct blk_trace *bt;
990
991 rcu_read_lock();
992 bt = rcu_dereference(q->blk_trace);
993 if (bt) {
994 __be64 rpdu = cpu_to_be64(pdu);
995
996 __blk_add_trace(bt, bio->bi_iter.bi_sector,
997 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
998 BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
999 &rpdu, blk_trace_bio_get_cgid(q, bio));
1000 }
1001 rcu_read_unlock();
1002}
1003
1004/**
1005 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1006 * @ignore: trace callback data parameter (not used)
1007 * @q: queue the io is for
1008 * @bio: the source bio
1009 * @dev: target device
1010 * @from: source sector
1011 *
1012 * Description:
1013 * Device mapper or raid target sometimes need to split a bio because
1014 * it spans a stripe (or similar). Add a trace for that action.
1015 *
1016 **/
1017static void blk_add_trace_bio_remap(void *ignore,
1018 struct request_queue *q, struct bio *bio,
1019 dev_t dev, sector_t from)
1020{
1021 struct blk_trace *bt;
1022 struct blk_io_trace_remap r;
1023
1024 rcu_read_lock();
1025 bt = rcu_dereference(q->blk_trace);
1026 if (likely(!bt)) {
1027 rcu_read_unlock();
1028 return;
1029 }
1030
1031 r.device_from = cpu_to_be32(dev);
1032 r.device_to = cpu_to_be32(bio_dev(bio));
1033 r.sector_from = cpu_to_be64(from);
1034
1035 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1036 bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1037 sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1038 rcu_read_unlock();
1039}
1040
1041/**
1042 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1043 * @ignore: trace callback data parameter (not used)
1044 * @q: queue the io is for
1045 * @rq: the source request
1046 * @dev: target device
1047 * @from: source sector
1048 *
1049 * Description:
1050 * Device mapper remaps request to other devices.
1051 * Add a trace for that action.
1052 *
1053 **/
1054static void blk_add_trace_rq_remap(void *ignore,
1055 struct request_queue *q,
1056 struct request *rq, dev_t dev,
1057 sector_t from)
1058{
1059 struct blk_trace *bt;
1060 struct blk_io_trace_remap r;
1061
1062 rcu_read_lock();
1063 bt = rcu_dereference(q->blk_trace);
1064 if (likely(!bt)) {
1065 rcu_read_unlock();
1066 return;
1067 }
1068
1069 r.device_from = cpu_to_be32(dev);
1070 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
1071 r.sector_from = cpu_to_be64(from);
1072
1073 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1074 rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1075 sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1076 rcu_read_unlock();
1077}
1078
1079/**
1080 * blk_add_driver_data - Add binary message with driver-specific data
1081 * @q: queue the io is for
1082 * @rq: io request
1083 * @data: driver-specific data
1084 * @len: length of driver-specific data
1085 *
1086 * Description:
1087 * Some drivers might want to write driver-specific data per request.
1088 *
1089 **/
1090void blk_add_driver_data(struct request_queue *q,
1091 struct request *rq,
1092 void *data, size_t len)
1093{
1094 struct blk_trace *bt;
1095
1096 rcu_read_lock();
1097 bt = rcu_dereference(q->blk_trace);
1098 if (likely(!bt)) {
1099 rcu_read_unlock();
1100 return;
1101 }
1102
1103 __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1104 BLK_TA_DRV_DATA, 0, len, data,
1105 blk_trace_request_get_cgid(q, rq));
1106 rcu_read_unlock();
1107}
1108EXPORT_SYMBOL_GPL(blk_add_driver_data);
1109
1110static void blk_register_tracepoints(void)
1111{
1112 int ret;
1113
1114 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1115 WARN_ON(ret);
1116 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1117 WARN_ON(ret);
1118 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1119 WARN_ON(ret);
1120 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1121 WARN_ON(ret);
1122 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1123 WARN_ON(ret);
1124 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1125 WARN_ON(ret);
1126 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1127 WARN_ON(ret);
1128 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1129 WARN_ON(ret);
1130 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1131 WARN_ON(ret);
1132 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1133 WARN_ON(ret);
1134 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1135 WARN_ON(ret);
1136 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1137 WARN_ON(ret);
1138 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1139 WARN_ON(ret);
1140 ret = register_trace_block_split(blk_add_trace_split, NULL);
1141 WARN_ON(ret);
1142 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1143 WARN_ON(ret);
1144 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1145 WARN_ON(ret);
1146}
1147
1148static void blk_unregister_tracepoints(void)
1149{
1150 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1151 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1152 unregister_trace_block_split(blk_add_trace_split, NULL);
1153 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1154 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1155 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1156 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1157 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1158 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1159 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1160 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1161 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1162 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1163 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1164 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1165 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1166
1167 tracepoint_synchronize_unregister();
1168}
1169
1170/*
1171 * struct blk_io_tracer formatting routines
1172 */
1173
1174static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1175{
1176 int i = 0;
1177 int tc = t->action >> BLK_TC_SHIFT;
1178
1179 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1180 rwbs[i++] = 'N';
1181 goto out;
1182 }
1183
1184 if (tc & BLK_TC_FLUSH)
1185 rwbs[i++] = 'F';
1186
1187 if (tc & BLK_TC_DISCARD)
1188 rwbs[i++] = 'D';
1189 else if (tc & BLK_TC_WRITE)
1190 rwbs[i++] = 'W';
1191 else if (t->bytes)
1192 rwbs[i++] = 'R';
1193 else
1194 rwbs[i++] = 'N';
1195
1196 if (tc & BLK_TC_FUA)
1197 rwbs[i++] = 'F';
1198 if (tc & BLK_TC_AHEAD)
1199 rwbs[i++] = 'A';
1200 if (tc & BLK_TC_SYNC)
1201 rwbs[i++] = 'S';
1202 if (tc & BLK_TC_META)
1203 rwbs[i++] = 'M';
1204out:
1205 rwbs[i] = '\0';
1206}
1207
1208static inline
1209const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1210{
1211 return (const struct blk_io_trace *)ent;
1212}
1213
1214static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1215{
1216 return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1217}
1218
1219static inline u64 t_cgid(const struct trace_entry *ent)
1220{
1221 return *(u64 *)(te_blk_io_trace(ent) + 1);
1222}
1223
1224static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1225{
1226 return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1227}
1228
1229static inline u32 t_action(const struct trace_entry *ent)
1230{
1231 return te_blk_io_trace(ent)->action;
1232}
1233
1234static inline u32 t_bytes(const struct trace_entry *ent)
1235{
1236 return te_blk_io_trace(ent)->bytes;
1237}
1238
1239static inline u32 t_sec(const struct trace_entry *ent)
1240{
1241 return te_blk_io_trace(ent)->bytes >> 9;
1242}
1243
1244static inline unsigned long long t_sector(const struct trace_entry *ent)
1245{
1246 return te_blk_io_trace(ent)->sector;
1247}
1248
1249static inline __u16 t_error(const struct trace_entry *ent)
1250{
1251 return te_blk_io_trace(ent)->error;
1252}
1253
1254static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1255{
1256 const __u64 *val = pdu_start(ent, has_cg);
1257 return be64_to_cpu(*val);
1258}
1259
1260static void get_pdu_remap(const struct trace_entry *ent,
1261 struct blk_io_trace_remap *r, bool has_cg)
1262{
1263 const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1264 __u64 sector_from = __r->sector_from;
1265
1266 r->device_from = be32_to_cpu(__r->device_from);
1267 r->device_to = be32_to_cpu(__r->device_to);
1268 r->sector_from = be64_to_cpu(sector_from);
1269}
1270
1271typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1272 bool has_cg);
1273
1274static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1275 bool has_cg)
1276{
1277 char rwbs[RWBS_LEN];
1278 unsigned long long ts = iter->ts;
1279 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1280 unsigned secs = (unsigned long)ts;
1281 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1282
1283 fill_rwbs(rwbs, t);
1284
1285 trace_seq_printf(&iter->seq,
1286 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1287 MAJOR(t->device), MINOR(t->device), iter->cpu,
1288 secs, nsec_rem, iter->ent->pid, act, rwbs);
1289}
1290
1291static void blk_log_action(struct trace_iterator *iter, const char *act,
1292 bool has_cg)
1293{
1294 char rwbs[RWBS_LEN];
1295 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1296
1297 fill_rwbs(rwbs, t);
1298 if (has_cg) {
1299 u64 id = t_cgid(iter->ent);
1300
1301 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1302 char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1303
1304 cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1305 sizeof(blkcg_name_buf));
1306 trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1307 MAJOR(t->device), MINOR(t->device),
1308 blkcg_name_buf, act, rwbs);
1309 } else {
1310 /*
1311 * The cgid portion used to be "INO,GEN". Userland
1312 * builds a FILEID_INO32_GEN fid out of them and
1313 * opens the cgroup using open_by_handle_at(2).
1314 * While 32bit ino setups are still the same, 64bit
1315 * ones now use the 64bit ino as the whole ID and
1316 * no longer use generation.
1317 *
1318 * Regarldess of the content, always output
1319 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1320 * be mapped back to @id on both 64 and 32bit ino
1321 * setups. See __kernfs_fh_to_dentry().
1322 */
1323 trace_seq_printf(&iter->seq,
1324 "%3d,%-3d %llx,%-llx %2s %3s ",
1325 MAJOR(t->device), MINOR(t->device),
1326 id & U32_MAX, id >> 32, act, rwbs);
1327 }
1328 } else
1329 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1330 MAJOR(t->device), MINOR(t->device), act, rwbs);
1331}
1332
1333static void blk_log_dump_pdu(struct trace_seq *s,
1334 const struct trace_entry *ent, bool has_cg)
1335{
1336 const unsigned char *pdu_buf;
1337 int pdu_len;
1338 int i, end;
1339
1340 pdu_buf = pdu_start(ent, has_cg);
1341 pdu_len = pdu_real_len(ent, has_cg);
1342
1343 if (!pdu_len)
1344 return;
1345
1346 /* find the last zero that needs to be printed */
1347 for (end = pdu_len - 1; end >= 0; end--)
1348 if (pdu_buf[end])
1349 break;
1350 end++;
1351
1352 trace_seq_putc(s, '(');
1353
1354 for (i = 0; i < pdu_len; i++) {
1355
1356 trace_seq_printf(s, "%s%02x",
1357 i == 0 ? "" : " ", pdu_buf[i]);
1358
1359 /*
1360 * stop when the rest is just zeroes and indicate so
1361 * with a ".." appended
1362 */
1363 if (i == end && end != pdu_len - 1) {
1364 trace_seq_puts(s, " ..) ");
1365 return;
1366 }
1367 }
1368
1369 trace_seq_puts(s, ") ");
1370}
1371
1372static void blk_log_generic(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 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1379 trace_seq_printf(s, "%u ", t_bytes(ent));
1380 blk_log_dump_pdu(s, ent, has_cg);
1381 trace_seq_printf(s, "[%s]\n", cmd);
1382 } else {
1383 if (t_sec(ent))
1384 trace_seq_printf(s, "%llu + %u [%s]\n",
1385 t_sector(ent), t_sec(ent), cmd);
1386 else
1387 trace_seq_printf(s, "[%s]\n", cmd);
1388 }
1389}
1390
1391static void blk_log_with_error(struct trace_seq *s,
1392 const struct trace_entry *ent, bool has_cg)
1393{
1394 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1395 blk_log_dump_pdu(s, ent, has_cg);
1396 trace_seq_printf(s, "[%d]\n", t_error(ent));
1397 } else {
1398 if (t_sec(ent))
1399 trace_seq_printf(s, "%llu + %u [%d]\n",
1400 t_sector(ent),
1401 t_sec(ent), t_error(ent));
1402 else
1403 trace_seq_printf(s, "%llu [%d]\n",
1404 t_sector(ent), t_error(ent));
1405 }
1406}
1407
1408static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1409{
1410 struct blk_io_trace_remap r = { .device_from = 0, };
1411
1412 get_pdu_remap(ent, &r, has_cg);
1413 trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1414 t_sector(ent), t_sec(ent),
1415 MAJOR(r.device_from), MINOR(r.device_from),
1416 (unsigned long long)r.sector_from);
1417}
1418
1419static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1420{
1421 char cmd[TASK_COMM_LEN];
1422
1423 trace_find_cmdline(ent->pid, cmd);
1424
1425 trace_seq_printf(s, "[%s]\n", cmd);
1426}
1427
1428static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1429{
1430 char cmd[TASK_COMM_LEN];
1431
1432 trace_find_cmdline(ent->pid, cmd);
1433
1434 trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1435}
1436
1437static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1438{
1439 char cmd[TASK_COMM_LEN];
1440
1441 trace_find_cmdline(ent->pid, cmd);
1442
1443 trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1444 get_pdu_int(ent, has_cg), cmd);
1445}
1446
1447static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1448 bool has_cg)
1449{
1450
1451 trace_seq_putmem(s, pdu_start(ent, has_cg),
1452 pdu_real_len(ent, has_cg));
1453 trace_seq_putc(s, '\n');
1454}
1455
1456/*
1457 * struct tracer operations
1458 */
1459
1460static void blk_tracer_print_header(struct seq_file *m)
1461{
1462 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1463 return;
1464 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1465 "# | | | | | |\n");
1466}
1467
1468static void blk_tracer_start(struct trace_array *tr)
1469{
1470 blk_tracer_enabled = true;
1471}
1472
1473static int blk_tracer_init(struct trace_array *tr)
1474{
1475 blk_tr = tr;
1476 blk_tracer_start(tr);
1477 return 0;
1478}
1479
1480static void blk_tracer_stop(struct trace_array *tr)
1481{
1482 blk_tracer_enabled = false;
1483}
1484
1485static void blk_tracer_reset(struct trace_array *tr)
1486{
1487 blk_tracer_stop(tr);
1488}
1489
1490static const struct {
1491 const char *act[2];
1492 void (*print)(struct trace_seq *s, const struct trace_entry *ent,
1493 bool has_cg);
1494} what2act[] = {
1495 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1496 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1497 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1498 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1499 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1500 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1501 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1502 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1503 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1504 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1505 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1506 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1507 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1508 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1509 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1510};
1511
1512static enum print_line_t print_one_line(struct trace_iterator *iter,
1513 bool classic)
1514{
1515 struct trace_array *tr = iter->tr;
1516 struct trace_seq *s = &iter->seq;
1517 const struct blk_io_trace *t;
1518 u16 what;
1519 bool long_act;
1520 blk_log_action_t *log_action;
1521 bool has_cg;
1522
1523 t = te_blk_io_trace(iter->ent);
1524 what = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1525 long_act = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1526 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1527 has_cg = t->action & __BLK_TA_CGROUP;
1528
1529 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1530 log_action(iter, long_act ? "message" : "m", has_cg);
1531 blk_log_msg(s, iter->ent, has_cg);
1532 return trace_handle_return(s);
1533 }
1534
1535 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1536 trace_seq_printf(s, "Unknown action %x\n", what);
1537 else {
1538 log_action(iter, what2act[what].act[long_act], has_cg);
1539 what2act[what].print(s, iter->ent, has_cg);
1540 }
1541
1542 return trace_handle_return(s);
1543}
1544
1545static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1546 int flags, struct trace_event *event)
1547{
1548 return print_one_line(iter, false);
1549}
1550
1551static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1552{
1553 struct trace_seq *s = &iter->seq;
1554 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1555 const int offset = offsetof(struct blk_io_trace, sector);
1556 struct blk_io_trace old = {
1557 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1558 .time = iter->ts,
1559 };
1560
1561 trace_seq_putmem(s, &old, offset);
1562 trace_seq_putmem(s, &t->sector,
1563 sizeof(old) - offset + t->pdu_len);
1564}
1565
1566static enum print_line_t
1567blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1568 struct trace_event *event)
1569{
1570 blk_trace_synthesize_old_trace(iter);
1571
1572 return trace_handle_return(&iter->seq);
1573}
1574
1575static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1576{
1577 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1578 return TRACE_TYPE_UNHANDLED;
1579
1580 return print_one_line(iter, true);
1581}
1582
1583static int
1584blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1585{
1586 /* don't output context-info for blk_classic output */
1587 if (bit == TRACE_BLK_OPT_CLASSIC) {
1588 if (set)
1589 tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1590 else
1591 tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1592 }
1593 return 0;
1594}
1595
1596static struct tracer blk_tracer __read_mostly = {
1597 .name = "blk",
1598 .init = blk_tracer_init,
1599 .reset = blk_tracer_reset,
1600 .start = blk_tracer_start,
1601 .stop = blk_tracer_stop,
1602 .print_header = blk_tracer_print_header,
1603 .print_line = blk_tracer_print_line,
1604 .flags = &blk_tracer_flags,
1605 .set_flag = blk_tracer_set_flag,
1606};
1607
1608static struct trace_event_functions trace_blk_event_funcs = {
1609 .trace = blk_trace_event_print,
1610 .binary = blk_trace_event_print_binary,
1611};
1612
1613static struct trace_event trace_blk_event = {
1614 .type = TRACE_BLK,
1615 .funcs = &trace_blk_event_funcs,
1616};
1617
1618static int __init init_blk_tracer(void)
1619{
1620 if (!register_trace_event(&trace_blk_event)) {
1621 pr_warn("Warning: could not register block events\n");
1622 return 1;
1623 }
1624
1625 if (register_tracer(&blk_tracer) != 0) {
1626 pr_warn("Warning: could not register the block tracer\n");
1627 unregister_trace_event(&trace_blk_event);
1628 return 1;
1629 }
1630
1631 return 0;
1632}
1633
1634device_initcall(init_blk_tracer);
1635
1636static int blk_trace_remove_queue(struct request_queue *q)
1637{
1638 struct blk_trace *bt;
1639
1640 bt = xchg(&q->blk_trace, NULL);
1641 if (bt == NULL)
1642 return -EINVAL;
1643
1644 put_probe_ref();
1645 synchronize_rcu();
1646 blk_trace_free(bt);
1647 return 0;
1648}
1649
1650/*
1651 * Setup everything required to start tracing
1652 */
1653static int blk_trace_setup_queue(struct request_queue *q,
1654 struct block_device *bdev)
1655{
1656 struct blk_trace *bt = NULL;
1657 int ret = -ENOMEM;
1658
1659 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1660 if (!bt)
1661 return -ENOMEM;
1662
1663 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1664 if (!bt->msg_data)
1665 goto free_bt;
1666
1667 bt->dev = bdev->bd_dev;
1668 bt->act_mask = (u16)-1;
1669
1670 blk_trace_setup_lba(bt, bdev);
1671
1672 ret = -EBUSY;
1673 if (cmpxchg(&q->blk_trace, NULL, bt))
1674 goto free_bt;
1675
1676 get_probe_ref();
1677 return 0;
1678
1679free_bt:
1680 blk_trace_free(bt);
1681 return ret;
1682}
1683
1684/*
1685 * sysfs interface to enable and configure tracing
1686 */
1687
1688static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1689 struct device_attribute *attr,
1690 char *buf);
1691static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1692 struct device_attribute *attr,
1693 const char *buf, size_t count);
1694#define BLK_TRACE_DEVICE_ATTR(_name) \
1695 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1696 sysfs_blk_trace_attr_show, \
1697 sysfs_blk_trace_attr_store)
1698
1699static BLK_TRACE_DEVICE_ATTR(enable);
1700static BLK_TRACE_DEVICE_ATTR(act_mask);
1701static BLK_TRACE_DEVICE_ATTR(pid);
1702static BLK_TRACE_DEVICE_ATTR(start_lba);
1703static BLK_TRACE_DEVICE_ATTR(end_lba);
1704
1705static struct attribute *blk_trace_attrs[] = {
1706 &dev_attr_enable.attr,
1707 &dev_attr_act_mask.attr,
1708 &dev_attr_pid.attr,
1709 &dev_attr_start_lba.attr,
1710 &dev_attr_end_lba.attr,
1711 NULL
1712};
1713
1714struct attribute_group blk_trace_attr_group = {
1715 .name = "trace",
1716 .attrs = blk_trace_attrs,
1717};
1718
1719static const struct {
1720 int mask;
1721 const char *str;
1722} mask_maps[] = {
1723 { BLK_TC_READ, "read" },
1724 { BLK_TC_WRITE, "write" },
1725 { BLK_TC_FLUSH, "flush" },
1726 { BLK_TC_SYNC, "sync" },
1727 { BLK_TC_QUEUE, "queue" },
1728 { BLK_TC_REQUEUE, "requeue" },
1729 { BLK_TC_ISSUE, "issue" },
1730 { BLK_TC_COMPLETE, "complete" },
1731 { BLK_TC_FS, "fs" },
1732 { BLK_TC_PC, "pc" },
1733 { BLK_TC_NOTIFY, "notify" },
1734 { BLK_TC_AHEAD, "ahead" },
1735 { BLK_TC_META, "meta" },
1736 { BLK_TC_DISCARD, "discard" },
1737 { BLK_TC_DRV_DATA, "drv_data" },
1738 { BLK_TC_FUA, "fua" },
1739};
1740
1741static int blk_trace_str2mask(const char *str)
1742{
1743 int i;
1744 int mask = 0;
1745 char *buf, *s, *token;
1746
1747 buf = kstrdup(str, GFP_KERNEL);
1748 if (buf == NULL)
1749 return -ENOMEM;
1750 s = strstrip(buf);
1751
1752 while (1) {
1753 token = strsep(&s, ",");
1754 if (token == NULL)
1755 break;
1756
1757 if (*token == '\0')
1758 continue;
1759
1760 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1761 if (strcasecmp(token, mask_maps[i].str) == 0) {
1762 mask |= mask_maps[i].mask;
1763 break;
1764 }
1765 }
1766 if (i == ARRAY_SIZE(mask_maps)) {
1767 mask = -EINVAL;
1768 break;
1769 }
1770 }
1771 kfree(buf);
1772
1773 return mask;
1774}
1775
1776static ssize_t blk_trace_mask2str(char *buf, int mask)
1777{
1778 int i;
1779 char *p = buf;
1780
1781 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1782 if (mask & mask_maps[i].mask) {
1783 p += sprintf(p, "%s%s",
1784 (p == buf) ? "" : ",", mask_maps[i].str);
1785 }
1786 }
1787 *p++ = '\n';
1788
1789 return p - buf;
1790}
1791
1792static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1793{
1794 if (bdev->bd_disk == NULL)
1795 return NULL;
1796
1797 return bdev_get_queue(bdev);
1798}
1799
1800static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1801 struct device_attribute *attr,
1802 char *buf)
1803{
1804 struct hd_struct *p = dev_to_part(dev);
1805 struct request_queue *q;
1806 struct block_device *bdev;
1807 struct blk_trace *bt;
1808 ssize_t ret = -ENXIO;
1809
1810 bdev = bdget(part_devt(p));
1811 if (bdev == NULL)
1812 goto out;
1813
1814 q = blk_trace_get_queue(bdev);
1815 if (q == NULL)
1816 goto out_bdput;
1817
1818 mutex_lock(&q->blk_trace_mutex);
1819
1820 bt = rcu_dereference_protected(q->blk_trace,
1821 lockdep_is_held(&q->blk_trace_mutex));
1822 if (attr == &dev_attr_enable) {
1823 ret = sprintf(buf, "%u\n", !!bt);
1824 goto out_unlock_bdev;
1825 }
1826
1827 if (bt == NULL)
1828 ret = sprintf(buf, "disabled\n");
1829 else if (attr == &dev_attr_act_mask)
1830 ret = blk_trace_mask2str(buf, bt->act_mask);
1831 else if (attr == &dev_attr_pid)
1832 ret = sprintf(buf, "%u\n", bt->pid);
1833 else if (attr == &dev_attr_start_lba)
1834 ret = sprintf(buf, "%llu\n", bt->start_lba);
1835 else if (attr == &dev_attr_end_lba)
1836 ret = sprintf(buf, "%llu\n", bt->end_lba);
1837
1838out_unlock_bdev:
1839 mutex_unlock(&q->blk_trace_mutex);
1840out_bdput:
1841 bdput(bdev);
1842out:
1843 return ret;
1844}
1845
1846static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1847 struct device_attribute *attr,
1848 const char *buf, size_t count)
1849{
1850 struct block_device *bdev;
1851 struct request_queue *q;
1852 struct hd_struct *p;
1853 struct blk_trace *bt;
1854 u64 value;
1855 ssize_t ret = -EINVAL;
1856
1857 if (count == 0)
1858 goto out;
1859
1860 if (attr == &dev_attr_act_mask) {
1861 if (kstrtoull(buf, 0, &value)) {
1862 /* Assume it is a list of trace category names */
1863 ret = blk_trace_str2mask(buf);
1864 if (ret < 0)
1865 goto out;
1866 value = ret;
1867 }
1868 } else if (kstrtoull(buf, 0, &value))
1869 goto out;
1870
1871 ret = -ENXIO;
1872
1873 p = dev_to_part(dev);
1874 bdev = bdget(part_devt(p));
1875 if (bdev == NULL)
1876 goto out;
1877
1878 q = blk_trace_get_queue(bdev);
1879 if (q == NULL)
1880 goto out_bdput;
1881
1882 mutex_lock(&q->blk_trace_mutex);
1883
1884 bt = rcu_dereference_protected(q->blk_trace,
1885 lockdep_is_held(&q->blk_trace_mutex));
1886 if (attr == &dev_attr_enable) {
1887 if (!!value == !!bt) {
1888 ret = 0;
1889 goto out_unlock_bdev;
1890 }
1891 if (value)
1892 ret = blk_trace_setup_queue(q, bdev);
1893 else
1894 ret = blk_trace_remove_queue(q);
1895 goto out_unlock_bdev;
1896 }
1897
1898 ret = 0;
1899 if (bt == NULL) {
1900 ret = blk_trace_setup_queue(q, bdev);
1901 bt = rcu_dereference_protected(q->blk_trace,
1902 lockdep_is_held(&q->blk_trace_mutex));
1903 }
1904
1905 if (ret == 0) {
1906 if (attr == &dev_attr_act_mask)
1907 bt->act_mask = value;
1908 else if (attr == &dev_attr_pid)
1909 bt->pid = value;
1910 else if (attr == &dev_attr_start_lba)
1911 bt->start_lba = value;
1912 else if (attr == &dev_attr_end_lba)
1913 bt->end_lba = value;
1914 }
1915
1916out_unlock_bdev:
1917 mutex_unlock(&q->blk_trace_mutex);
1918out_bdput:
1919 bdput(bdev);
1920out:
1921 return ret ? ret : count;
1922}
1923
1924int blk_trace_init_sysfs(struct device *dev)
1925{
1926 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1927}
1928
1929void blk_trace_remove_sysfs(struct device *dev)
1930{
1931 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1932}
1933
1934#endif /* CONFIG_BLK_DEV_IO_TRACE */
1935
1936#ifdef CONFIG_EVENT_TRACING
1937
1938void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1939{
1940 int i = 0;
1941
1942 if (op & REQ_PREFLUSH)
1943 rwbs[i++] = 'F';
1944
1945 switch (op & REQ_OP_MASK) {
1946 case REQ_OP_WRITE:
1947 case REQ_OP_WRITE_SAME:
1948 rwbs[i++] = 'W';
1949 break;
1950 case REQ_OP_DISCARD:
1951 rwbs[i++] = 'D';
1952 break;
1953 case REQ_OP_SECURE_ERASE:
1954 rwbs[i++] = 'D';
1955 rwbs[i++] = 'E';
1956 break;
1957 case REQ_OP_FLUSH:
1958 rwbs[i++] = 'F';
1959 break;
1960 case REQ_OP_READ:
1961 rwbs[i++] = 'R';
1962 break;
1963 default:
1964 rwbs[i++] = 'N';
1965 }
1966
1967 if (op & REQ_FUA)
1968 rwbs[i++] = 'F';
1969 if (op & REQ_RAHEAD)
1970 rwbs[i++] = 'A';
1971 if (op & REQ_SYNC)
1972 rwbs[i++] = 'S';
1973 if (op & REQ_META)
1974 rwbs[i++] = 'M';
1975
1976 rwbs[i] = '\0';
1977}
1978EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1979
1980#endif /* CONFIG_EVENT_TRACING */
1981