projects/linux·blob v5.8 · refs

1983 lines · 46.8 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2006 Jens Axboe <[email protected]>
4 *
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/kernel.h>
10#include <linux/blkdev.h>
11#include <linux/blktrace_api.h>
12#include <linux/percpu.h>
13#include <linux/init.h>
14#include <linux/mutex.h>
15#include <linux/slab.h>
16#include <linux/debugfs.h>
17#include <linux/export.h>
18#include <linux/time.h>
19#include <linux/uaccess.h>
20#include <linux/list.h>
21#include <linux/blk-cgroup.h>
22
23#include "../../block/blk.h"
24
25#include <trace/events/block.h>
26
27#include "trace_output.h"
28
29#ifdef CONFIG_BLK_DEV_IO_TRACE
30
31static unsigned int blktrace_seq __read_mostly = 1;
32
33static struct trace_array *blk_tr;
34static bool blk_tracer_enabled __read_mostly;
35
36static LIST_HEAD(running_trace_list);
37static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
38
39/* Select an alternative, minimalistic output than the original one */
40#define TRACE_BLK_OPT_CLASSIC 0x1
41#define TRACE_BLK_OPT_CGROUP 0x2
42#define TRACE_BLK_OPT_CGNAME 0x4
43
44static struct tracer_opt blk_tracer_opts[] = {
45 /* Default disable the minimalistic output */
46 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
47#ifdef CONFIG_BLK_CGROUP
48 { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
49 { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
50#endif
51 { }
52};
53
54static struct tracer_flags blk_tracer_flags = {
55 .val = 0,
56 .opts = blk_tracer_opts,
57};
58
59/* Global reference count of probes */
60static DEFINE_MUTEX(blk_probe_mutex);
61static int blk_probes_ref;
62
63static void blk_register_tracepoints(void);
64static void blk_unregister_tracepoints(void);
65
66/*
67 * Send out a notify message.
68 */
69static void trace_note(struct blk_trace *bt, pid_t pid, int action,
70 const void *data, size_t len, u64 cgid)
71{
72 struct blk_io_trace *t;
73 struct ring_buffer_event *event = NULL;
74 struct trace_buffer *buffer = NULL;
75 int pc = 0;
76 int cpu = smp_processor_id();
77 bool blk_tracer = blk_tracer_enabled;
78 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
79
80 if (blk_tracer) {
81 buffer = blk_tr->array_buffer.buffer;
82 pc = preempt_count();
83 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
84 sizeof(*t) + len + cgid_len,
85 0, pc);
86 if (!event)
87 return;
88 t = ring_buffer_event_data(event);
89 goto record_it;
90 }
91
92 if (!bt->rchan)
93 return;
94
95 t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
96 if (t) {
97 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
98 t->time = ktime_to_ns(ktime_get());
99record_it:
100 t->device = bt->dev;
101 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
102 t->pid = pid;
103 t->cpu = cpu;
104 t->pdu_len = len + cgid_len;
105 if (cgid_len)
106 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
107 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
108
109 if (blk_tracer)
110 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
111 }
112}
113
114/*
115 * Send out a notify for this process, if we haven't done so since a trace
116 * started
117 */
118static void trace_note_tsk(struct task_struct *tsk)
119{
120 unsigned long flags;
121 struct blk_trace *bt;
122
123 tsk->btrace_seq = blktrace_seq;
124 spin_lock_irqsave(&running_trace_lock, flags);
125 list_for_each_entry(bt, &running_trace_list, running_list) {
126 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
127 sizeof(tsk->comm), 0);
128 }
129 spin_unlock_irqrestore(&running_trace_lock, flags);
130}
131
132static void trace_note_time(struct blk_trace *bt)
133{
134 struct timespec64 now;
135 unsigned long flags;
136 u32 words[2];
137
138 /* need to check user space to see if this breaks in y2038 or y2106 */
139 ktime_get_real_ts64(&now);
140 words[0] = (u32)now.tv_sec;
141 words[1] = now.tv_nsec;
142
143 local_irq_save(flags);
144 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
145 local_irq_restore(flags);
146}
147
148void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
149 const char *fmt, ...)
150{
151 int n;
152 va_list args;
153 unsigned long flags;
154 char *buf;
155
156 if (unlikely(bt->trace_state != Blktrace_running &&
157 !blk_tracer_enabled))
158 return;
159
160 /*
161 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
162 * message to the trace.
163 */
164 if (!(bt->act_mask & BLK_TC_NOTIFY))
165 return;
166
167 local_irq_save(flags);
168 buf = this_cpu_ptr(bt->msg_data);
169 va_start(args, fmt);
170 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
171 va_end(args);
172
173 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
174 blkcg = NULL;
175#ifdef CONFIG_BLK_CGROUP
176 trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n,
177 blkcg ? cgroup_id(blkcg->css.cgroup) : 1);
178#else
179 trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, 0);
180#endif
181 local_irq_restore(flags);
182}
183EXPORT_SYMBOL_GPL(__trace_note_message);
184
185static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
186 pid_t pid)
187{
188 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
189 return 1;
190 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
191 return 1;
192 if (bt->pid && pid != bt->pid)
193 return 1;
194
195 return 0;
196}
197
198/*
199 * Data direction bit lookup
200 */
201static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
202 BLK_TC_ACT(BLK_TC_WRITE) };
203
204#define BLK_TC_RAHEAD BLK_TC_AHEAD
205#define BLK_TC_PREFLUSH BLK_TC_FLUSH
206
207/* The ilog2() calls fall out because they're constant */
208#define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
209 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
210
211/*
212 * The worker for the various blk_add_trace*() types. Fills out a
213 * blk_io_trace structure and places it in a per-cpu subbuffer.
214 */
215static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
216 int op, int op_flags, u32 what, int error, int pdu_len,
217 void *pdu_data, u64 cgid)
218{
219 struct task_struct *tsk = current;
220 struct ring_buffer_event *event = NULL;
221 struct trace_buffer *buffer = NULL;
222 struct blk_io_trace *t;
223 unsigned long flags = 0;
224 unsigned long *sequence;
225 pid_t pid;
226 int cpu, pc = 0;
227 bool blk_tracer = blk_tracer_enabled;
228 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
229
230 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
231 return;
232
233 what |= ddir_act[op_is_write(op) ? WRITE : READ];
234 what |= MASK_TC_BIT(op_flags, SYNC);
235 what |= MASK_TC_BIT(op_flags, RAHEAD);
236 what |= MASK_TC_BIT(op_flags, META);
237 what |= MASK_TC_BIT(op_flags, PREFLUSH);
238 what |= MASK_TC_BIT(op_flags, FUA);
239 if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
240 what |= BLK_TC_ACT(BLK_TC_DISCARD);
241 if (op == REQ_OP_FLUSH)
242 what |= BLK_TC_ACT(BLK_TC_FLUSH);
243 if (cgid)
244 what |= __BLK_TA_CGROUP;
245
246 pid = tsk->pid;
247 if (act_log_check(bt, what, sector, pid))
248 return;
249 cpu = raw_smp_processor_id();
250
251 if (blk_tracer) {
252 tracing_record_cmdline(current);
253
254 buffer = blk_tr->array_buffer.buffer;
255 pc = preempt_count();
256 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
257 sizeof(*t) + pdu_len + cgid_len,
258 0, pc);
259 if (!event)
260 return;
261 t = ring_buffer_event_data(event);
262 goto record_it;
263 }
264
265 if (unlikely(tsk->btrace_seq != blktrace_seq))
266 trace_note_tsk(tsk);
267
268 /*
269 * A word about the locking here - we disable interrupts to reserve
270 * some space in the relay per-cpu buffer, to prevent an irq
271 * from coming in and stepping on our toes.
272 */
273 local_irq_save(flags);
274 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
275 if (t) {
276 sequence = per_cpu_ptr(bt->sequence, cpu);
277
278 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
279 t->sequence = ++(*sequence);
280 t->time = ktime_to_ns(ktime_get());
281record_it:
282 /*
283 * These two are not needed in ftrace as they are in the
284 * generic trace_entry, filled by tracing_generic_entry_update,
285 * but for the trace_event->bin() synthesizer benefit we do it
286 * here too.
287 */
288 t->cpu = cpu;
289 t->pid = pid;
290
291 t->sector = sector;
292 t->bytes = bytes;
293 t->action = what;
294 t->device = bt->dev;
295 t->error = error;
296 t->pdu_len = pdu_len + cgid_len;
297
298 if (cgid_len)
299 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
300 if (pdu_len)
301 memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
302
303 if (blk_tracer) {
304 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
305 return;
306 }
307 }
308
309 local_irq_restore(flags);
310}
311
312static void blk_trace_free(struct blk_trace *bt)
313{
314 debugfs_remove(bt->msg_file);
315 debugfs_remove(bt->dropped_file);
316 relay_close(bt->rchan);
317 debugfs_remove(bt->dir);
318 free_percpu(bt->sequence);
319 free_percpu(bt->msg_data);
320 kfree(bt);
321}
322
323static void get_probe_ref(void)
324{
325 mutex_lock(&blk_probe_mutex);
326 if (++blk_probes_ref == 1)
327 blk_register_tracepoints();
328 mutex_unlock(&blk_probe_mutex);
329}
330
331static void put_probe_ref(void)
332{
333 mutex_lock(&blk_probe_mutex);
334 if (!--blk_probes_ref)
335 blk_unregister_tracepoints();
336 mutex_unlock(&blk_probe_mutex);
337}
338
339static void blk_trace_cleanup(struct blk_trace *bt)
340{
341 synchronize_rcu();
342 blk_trace_free(bt);
343 put_probe_ref();
344}
345
346static int __blk_trace_remove(struct request_queue *q)
347{
348 struct blk_trace *bt;
349
350 bt = rcu_replace_pointer(q->blk_trace, NULL,
351 lockdep_is_held(&q->blk_trace_mutex));
352 if (!bt)
353 return -EINVAL;
354
355 if (bt->trace_state != Blktrace_running)
356 blk_trace_cleanup(bt);
357
358 return 0;
359}
360
361int blk_trace_remove(struct request_queue *q)
362{
363 int ret;
364
365 mutex_lock(&q->blk_trace_mutex);
366 ret = __blk_trace_remove(q);
367 mutex_unlock(&q->blk_trace_mutex);
368
369 return ret;
370}
371EXPORT_SYMBOL_GPL(blk_trace_remove);
372
373static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
374 size_t count, loff_t *ppos)
375{
376 struct blk_trace *bt = filp->private_data;
377 char buf[16];
378
379 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
380
381 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
382}
383
384static const struct file_operations blk_dropped_fops = {
385 .owner = THIS_MODULE,
386 .open = simple_open,
387 .read = blk_dropped_read,
388 .llseek = default_llseek,
389};
390
391static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
392 size_t count, loff_t *ppos)
393{
394 char *msg;
395 struct blk_trace *bt;
396
397 if (count >= BLK_TN_MAX_MSG)
398 return -EINVAL;
399
400 msg = memdup_user_nul(buffer, count);
401 if (IS_ERR(msg))
402 return PTR_ERR(msg);
403
404 bt = filp->private_data;
405 __trace_note_message(bt, NULL, "%s", msg);
406 kfree(msg);
407
408 return count;
409}
410
411static const struct file_operations blk_msg_fops = {
412 .owner = THIS_MODULE,
413 .open = simple_open,
414 .write = blk_msg_write,
415 .llseek = noop_llseek,
416};
417
418/*
419 * Keep track of how many times we encountered a full subbuffer, to aid
420 * the user space app in telling how many lost events there were.
421 */
422static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
423 void *prev_subbuf, size_t prev_padding)
424{
425 struct blk_trace *bt;
426
427 if (!relay_buf_full(buf))
428 return 1;
429
430 bt = buf->chan->private_data;
431 atomic_inc(&bt->dropped);
432 return 0;
433}
434
435static int blk_remove_buf_file_callback(struct dentry *dentry)
436{
437 debugfs_remove(dentry);
438
439 return 0;
440}
441
442static struct dentry *blk_create_buf_file_callback(const char *filename,
443 struct dentry *parent,
444 umode_t mode,
445 struct rchan_buf *buf,
446 int *is_global)
447{
448 return debugfs_create_file(filename, mode, parent, buf,
449 &relay_file_operations);
450}
451
452static struct rchan_callbacks blk_relay_callbacks = {
453 .subbuf_start = blk_subbuf_start_callback,
454 .create_buf_file = blk_create_buf_file_callback,
455 .remove_buf_file = blk_remove_buf_file_callback,
456};
457
458static void blk_trace_setup_lba(struct blk_trace *bt,
459 struct block_device *bdev)
460{
461 struct hd_struct *part = NULL;
462
463 if (bdev)
464 part = bdev->bd_part;
465
466 if (part) {
467 bt->start_lba = part->start_sect;
468 bt->end_lba = part->start_sect + part->nr_sects;
469 } else {
470 bt->start_lba = 0;
471 bt->end_lba = -1ULL;
472 }
473}
474
475/*
476 * Setup everything required to start tracing
477 */
478static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
479 struct block_device *bdev,
480 struct blk_user_trace_setup *buts)
481{
482 struct blk_trace *bt = NULL;
483 struct dentry *dir = NULL;
484 int ret;
485
486 if (!buts->buf_size || !buts->buf_nr)
487 return -EINVAL;
488
489 if (!blk_debugfs_root)
490 return -ENOENT;
491
492 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
493 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
494
495 /*
496 * some device names have larger paths - convert the slashes
497 * to underscores for this to work as expected
498 */
499 strreplace(buts->name, '/', '_');
500
501 /*
502 * bdev can be NULL, as with scsi-generic, this is a helpful as
503 * we can be.
504 */
505 if (rcu_dereference_protected(q->blk_trace,
506 lockdep_is_held(&q->blk_trace_mutex))) {
507 pr_warn("Concurrent blktraces are not allowed on %s\n",
508 buts->name);
509 return -EBUSY;
510 }
511
512 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
513 if (!bt)
514 return -ENOMEM;
515
516 ret = -ENOMEM;
517 bt->sequence = alloc_percpu(unsigned long);
518 if (!bt->sequence)
519 goto err;
520
521 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
522 if (!bt->msg_data)
523 goto err;
524
525 ret = -ENOENT;
526
527 dir = debugfs_lookup(buts->name, blk_debugfs_root);
528 if (!dir)
529 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
530
531 bt->dev = dev;
532 atomic_set(&bt->dropped, 0);
533 INIT_LIST_HEAD(&bt->running_list);
534
535 ret = -EIO;
536 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
537 &blk_dropped_fops);
538
539 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
540
541 bt->rchan = relay_open("trace", dir, buts->buf_size,
542 buts->buf_nr, &blk_relay_callbacks, bt);
543 if (!bt->rchan)
544 goto err;
545
546 bt->act_mask = buts->act_mask;
547 if (!bt->act_mask)
548 bt->act_mask = (u16) -1;
549
550 blk_trace_setup_lba(bt, bdev);
551
552 /* overwrite with user settings */
553 if (buts->start_lba)
554 bt->start_lba = buts->start_lba;
555 if (buts->end_lba)
556 bt->end_lba = buts->end_lba;
557
558 bt->pid = buts->pid;
559 bt->trace_state = Blktrace_setup;
560
561 rcu_assign_pointer(q->blk_trace, bt);
562 get_probe_ref();
563
564 ret = 0;
565err:
566 if (dir && !bt->dir)
567 dput(dir);
568 if (ret)
569 blk_trace_free(bt);
570 return ret;
571}
572
573static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
574 struct block_device *bdev, char __user *arg)
575{
576 struct blk_user_trace_setup buts;
577 int ret;
578
579 ret = copy_from_user(&buts, arg, sizeof(buts));
580 if (ret)
581 return -EFAULT;
582
583 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
584 if (ret)
585 return ret;
586
587 if (copy_to_user(arg, &buts, sizeof(buts))) {
588 __blk_trace_remove(q);
589 return -EFAULT;
590 }
591 return 0;
592}
593
594int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
595 struct block_device *bdev,
596 char __user *arg)
597{
598 int ret;
599
600 mutex_lock(&q->blk_trace_mutex);
601 ret = __blk_trace_setup(q, name, dev, bdev, arg);
602 mutex_unlock(&q->blk_trace_mutex);
603
604 return ret;
605}
606EXPORT_SYMBOL_GPL(blk_trace_setup);
607
608#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
609static int compat_blk_trace_setup(struct request_queue *q, char *name,
610 dev_t dev, struct block_device *bdev,
611 char __user *arg)
612{
613 struct blk_user_trace_setup buts;
614 struct compat_blk_user_trace_setup cbuts;
615 int ret;
616
617 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
618 return -EFAULT;
619
620 buts = (struct blk_user_trace_setup) {
621 .act_mask = cbuts.act_mask,
622 .buf_size = cbuts.buf_size,
623 .buf_nr = cbuts.buf_nr,
624 .start_lba = cbuts.start_lba,
625 .end_lba = cbuts.end_lba,
626 .pid = cbuts.pid,
627 };
628
629 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
630 if (ret)
631 return ret;
632
633 if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
634 __blk_trace_remove(q);
635 return -EFAULT;
636 }
637
638 return 0;
639}
640#endif
641
642static int __blk_trace_startstop(struct request_queue *q, int start)
643{
644 int ret;
645 struct blk_trace *bt;
646
647 bt = rcu_dereference_protected(q->blk_trace,
648 lockdep_is_held(&q->blk_trace_mutex));
649 if (bt == NULL)
650 return -EINVAL;
651
652 /*
653 * For starting a trace, we can transition from a setup or stopped
654 * trace. For stopping a trace, the state must be running
655 */
656 ret = -EINVAL;
657 if (start) {
658 if (bt->trace_state == Blktrace_setup ||
659 bt->trace_state == Blktrace_stopped) {
660 blktrace_seq++;
661 smp_mb();
662 bt->trace_state = Blktrace_running;
663 spin_lock_irq(&running_trace_lock);
664 list_add(&bt->running_list, &running_trace_list);
665 spin_unlock_irq(&running_trace_lock);
666
667 trace_note_time(bt);
668 ret = 0;
669 }
670 } else {
671 if (bt->trace_state == Blktrace_running) {
672 bt->trace_state = Blktrace_stopped;
673 spin_lock_irq(&running_trace_lock);
674 list_del_init(&bt->running_list);
675 spin_unlock_irq(&running_trace_lock);
676 relay_flush(bt->rchan);
677 ret = 0;
678 }
679 }
680
681 return ret;
682}
683
684int blk_trace_startstop(struct request_queue *q, int start)
685{
686 int ret;
687
688 mutex_lock(&q->blk_trace_mutex);
689 ret = __blk_trace_startstop(q, start);
690 mutex_unlock(&q->blk_trace_mutex);
691
692 return ret;
693}
694EXPORT_SYMBOL_GPL(blk_trace_startstop);
695
696/*
697 * When reading or writing the blktrace sysfs files, the references to the
698 * opened sysfs or device files should prevent the underlying block device
699 * from being removed. So no further delete protection is really needed.
700 */
701
702/**
703 * blk_trace_ioctl: - handle the ioctls associated with tracing
704 * @bdev: the block device
705 * @cmd: the ioctl cmd
706 * @arg: the argument data, if any
707 *
708 **/
709int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
710{
711 struct request_queue *q;
712 int ret, start = 0;
713 char b[BDEVNAME_SIZE];
714
715 q = bdev_get_queue(bdev);
716 if (!q)
717 return -ENXIO;
718
719 mutex_lock(&q->blk_trace_mutex);
720
721 switch (cmd) {
722 case BLKTRACESETUP:
723 bdevname(bdev, b);
724 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
725 break;
726#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
727 case BLKTRACESETUP32:
728 bdevname(bdev, b);
729 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
730 break;
731#endif
732 case BLKTRACESTART:
733 start = 1;
734 /* fall through */
735 case BLKTRACESTOP:
736 ret = __blk_trace_startstop(q, start);
737 break;
738 case BLKTRACETEARDOWN:
739 ret = __blk_trace_remove(q);
740 break;
741 default:
742 ret = -ENOTTY;
743 break;
744 }
745
746 mutex_unlock(&q->blk_trace_mutex);
747 return ret;
748}
749
750/**
751 * blk_trace_shutdown: - stop and cleanup trace structures
752 * @q: the request queue associated with the device
753 *
754 **/
755void blk_trace_shutdown(struct request_queue *q)
756{
757 mutex_lock(&q->blk_trace_mutex);
758 if (rcu_dereference_protected(q->blk_trace,
759 lockdep_is_held(&q->blk_trace_mutex))) {
760 __blk_trace_startstop(q, 0);
761 __blk_trace_remove(q);
762 }
763
764 mutex_unlock(&q->blk_trace_mutex);
765}
766
767#ifdef CONFIG_BLK_CGROUP
768static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
769{
770 struct blk_trace *bt;
771
772 /* We don't use the 'bt' value here except as an optimization... */
773 bt = rcu_dereference_protected(q->blk_trace, 1);
774 if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
775 return 0;
776
777 if (!bio->bi_blkg)
778 return 0;
779 return cgroup_id(bio_blkcg(bio)->css.cgroup);
780}
781#else
782u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
783{
784 return 0;
785}
786#endif
787
788static u64
789blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
790{
791 if (!rq->bio)
792 return 0;
793 /* Use the first bio */
794 return blk_trace_bio_get_cgid(q, rq->bio);
795}
796
797/*
798 * blktrace probes
799 */
800
801/**
802 * blk_add_trace_rq - Add a trace for a request oriented action
803 * @rq: the source request
804 * @error: return status to log
805 * @nr_bytes: number of completed bytes
806 * @what: the action
807 * @cgid: the cgroup info
808 *
809 * Description:
810 * Records an action against a request. Will log the bio offset + size.
811 *
812 **/
813static void blk_add_trace_rq(struct request *rq, int error,
814 unsigned int nr_bytes, u32 what, u64 cgid)
815{
816 struct blk_trace *bt;
817
818 rcu_read_lock();
819 bt = rcu_dereference(rq->q->blk_trace);
820 if (likely(!bt)) {
821 rcu_read_unlock();
822 return;
823 }
824
825 if (blk_rq_is_passthrough(rq))
826 what |= BLK_TC_ACT(BLK_TC_PC);
827 else
828 what |= BLK_TC_ACT(BLK_TC_FS);
829
830 __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
831 rq->cmd_flags, what, error, 0, NULL, cgid);
832 rcu_read_unlock();
833}
834
835static void blk_add_trace_rq_insert(void *ignore,
836 struct request_queue *q, struct request *rq)
837{
838 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
839 blk_trace_request_get_cgid(q, rq));
840}
841
842static void blk_add_trace_rq_issue(void *ignore,
843 struct request_queue *q, struct request *rq)
844{
845 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
846 blk_trace_request_get_cgid(q, rq));
847}
848
849static void blk_add_trace_rq_requeue(void *ignore,
850 struct request_queue *q,
851 struct request *rq)
852{
853 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
854 blk_trace_request_get_cgid(q, rq));
855}
856
857static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
858 int error, unsigned int nr_bytes)
859{
860 blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
861 blk_trace_request_get_cgid(rq->q, rq));
862}
863
864/**
865 * blk_add_trace_bio - Add a trace for a bio oriented action
866 * @q: queue the io is for
867 * @bio: the source bio
868 * @what: the action
869 * @error: error, if any
870 *
871 * Description:
872 * Records an action against a bio. Will log the bio offset + size.
873 *
874 **/
875static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
876 u32 what, int error)
877{
878 struct blk_trace *bt;
879
880 rcu_read_lock();
881 bt = rcu_dereference(q->blk_trace);
882 if (likely(!bt)) {
883 rcu_read_unlock();
884 return;
885 }
886
887 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
888 bio_op(bio), bio->bi_opf, what, error, 0, NULL,
889 blk_trace_bio_get_cgid(q, bio));
890 rcu_read_unlock();
891}
892
893static void blk_add_trace_bio_bounce(void *ignore,
894 struct request_queue *q, struct bio *bio)
895{
896 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
897}
898
899static void blk_add_trace_bio_complete(void *ignore,
900 struct request_queue *q, struct bio *bio)
901{
902 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
903 blk_status_to_errno(bio->bi_status));
904}
905
906static void blk_add_trace_bio_backmerge(void *ignore,
907 struct request_queue *q,
908 struct request *rq,
909 struct bio *bio)
910{
911 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
912}
913
914static void blk_add_trace_bio_frontmerge(void *ignore,
915 struct request_queue *q,
916 struct request *rq,
917 struct bio *bio)
918{
919 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
920}
921
922static void blk_add_trace_bio_queue(void *ignore,
923 struct request_queue *q, struct bio *bio)
924{
925 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
926}
927
928static void blk_add_trace_getrq(void *ignore,
929 struct request_queue *q,
930 struct bio *bio, int rw)
931{
932 if (bio)
933 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
934 else {
935 struct blk_trace *bt;
936
937 rcu_read_lock();
938 bt = rcu_dereference(q->blk_trace);
939 if (bt)
940 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
941 NULL, 0);
942 rcu_read_unlock();
943 }
944}
945
946
947static void blk_add_trace_sleeprq(void *ignore,
948 struct request_queue *q,
949 struct bio *bio, int rw)
950{
951 if (bio)
952 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
953 else {
954 struct blk_trace *bt;
955
956 rcu_read_lock();
957 bt = rcu_dereference(q->blk_trace);
958 if (bt)
959 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
960 0, 0, NULL, 0);
961 rcu_read_unlock();
962 }
963}
964
965static void blk_add_trace_plug(void *ignore, struct request_queue *q)
966{
967 struct blk_trace *bt;
968
969 rcu_read_lock();
970 bt = rcu_dereference(q->blk_trace);
971 if (bt)
972 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
973 rcu_read_unlock();
974}
975
976static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
977 unsigned int depth, bool explicit)
978{
979 struct blk_trace *bt;
980
981 rcu_read_lock();
982 bt = rcu_dereference(q->blk_trace);
983 if (bt) {
984 __be64 rpdu = cpu_to_be64(depth);
985 u32 what;
986
987 if (explicit)
988 what = BLK_TA_UNPLUG_IO;
989 else
990 what = BLK_TA_UNPLUG_TIMER;
991
992 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
993 }
994 rcu_read_unlock();
995}
996
997static void blk_add_trace_split(void *ignore,
998 struct request_queue *q, struct bio *bio,
999 unsigned int pdu)
1000{
1001 struct blk_trace *bt;
1002
1003 rcu_read_lock();
1004 bt = rcu_dereference(q->blk_trace);
1005 if (bt) {
1006 __be64 rpdu = cpu_to_be64(pdu);
1007
1008 __blk_add_trace(bt, bio->bi_iter.bi_sector,
1009 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
1010 BLK_TA_SPLIT,
1011 blk_status_to_errno(bio->bi_status),
1012 sizeof(rpdu), &rpdu,
1013 blk_trace_bio_get_cgid(q, bio));
1014 }
1015 rcu_read_unlock();
1016}
1017
1018/**
1019 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1020 * @ignore: trace callback data parameter (not used)
1021 * @q: queue the io is for
1022 * @bio: the source bio
1023 * @dev: target device
1024 * @from: source sector
1025 *
1026 * Description:
1027 * Device mapper or raid target sometimes need to split a bio because
1028 * it spans a stripe (or similar). Add a trace for that action.
1029 *
1030 **/
1031static void blk_add_trace_bio_remap(void *ignore,
1032 struct request_queue *q, struct bio *bio,
1033 dev_t dev, sector_t from)
1034{
1035 struct blk_trace *bt;
1036 struct blk_io_trace_remap r;
1037
1038 rcu_read_lock();
1039 bt = rcu_dereference(q->blk_trace);
1040 if (likely(!bt)) {
1041 rcu_read_unlock();
1042 return;
1043 }
1044
1045 r.device_from = cpu_to_be32(dev);
1046 r.device_to = cpu_to_be32(bio_dev(bio));
1047 r.sector_from = cpu_to_be64(from);
1048
1049 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1050 bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1051 blk_status_to_errno(bio->bi_status),
1052 sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1053 rcu_read_unlock();
1054}
1055
1056/**
1057 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1058 * @ignore: trace callback data parameter (not used)
1059 * @q: queue the io is for
1060 * @rq: the source request
1061 * @dev: target device
1062 * @from: source sector
1063 *
1064 * Description:
1065 * Device mapper remaps request to other devices.
1066 * Add a trace for that action.
1067 *
1068 **/
1069static void blk_add_trace_rq_remap(void *ignore,
1070 struct request_queue *q,
1071 struct request *rq, dev_t dev,
1072 sector_t from)
1073{
1074 struct blk_trace *bt;
1075 struct blk_io_trace_remap r;
1076
1077 rcu_read_lock();
1078 bt = rcu_dereference(q->blk_trace);
1079 if (likely(!bt)) {
1080 rcu_read_unlock();
1081 return;
1082 }
1083
1084 r.device_from = cpu_to_be32(dev);
1085 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
1086 r.sector_from = cpu_to_be64(from);
1087
1088 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1089 rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1090 sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1091 rcu_read_unlock();
1092}
1093
1094/**
1095 * blk_add_driver_data - Add binary message with driver-specific data
1096 * @q: queue the io is for
1097 * @rq: io request
1098 * @data: driver-specific data
1099 * @len: length of driver-specific data
1100 *
1101 * Description:
1102 * Some drivers might want to write driver-specific data per request.
1103 *
1104 **/
1105void blk_add_driver_data(struct request_queue *q,
1106 struct request *rq,
1107 void *data, size_t len)
1108{
1109 struct blk_trace *bt;
1110
1111 rcu_read_lock();
1112 bt = rcu_dereference(q->blk_trace);
1113 if (likely(!bt)) {
1114 rcu_read_unlock();
1115 return;
1116 }
1117
1118 __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1119 BLK_TA_DRV_DATA, 0, len, data,
1120 blk_trace_request_get_cgid(q, rq));
1121 rcu_read_unlock();
1122}
1123EXPORT_SYMBOL_GPL(blk_add_driver_data);
1124
1125static void blk_register_tracepoints(void)
1126{
1127 int ret;
1128
1129 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1130 WARN_ON(ret);
1131 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1132 WARN_ON(ret);
1133 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1134 WARN_ON(ret);
1135 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1136 WARN_ON(ret);
1137 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1138 WARN_ON(ret);
1139 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1140 WARN_ON(ret);
1141 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1142 WARN_ON(ret);
1143 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1144 WARN_ON(ret);
1145 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1146 WARN_ON(ret);
1147 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1148 WARN_ON(ret);
1149 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1150 WARN_ON(ret);
1151 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1152 WARN_ON(ret);
1153 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1154 WARN_ON(ret);
1155 ret = register_trace_block_split(blk_add_trace_split, NULL);
1156 WARN_ON(ret);
1157 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1158 WARN_ON(ret);
1159 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1160 WARN_ON(ret);
1161}
1162
1163static void blk_unregister_tracepoints(void)
1164{
1165 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1166 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1167 unregister_trace_block_split(blk_add_trace_split, NULL);
1168 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1169 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1170 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1171 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1172 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1173 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1174 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1175 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1176 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1177 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1178 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1179 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1180 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1181
1182 tracepoint_synchronize_unregister();
1183}
1184
1185/*
1186 * struct blk_io_tracer formatting routines
1187 */
1188
1189static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1190{
1191 int i = 0;
1192 int tc = t->action >> BLK_TC_SHIFT;
1193
1194 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1195 rwbs[i++] = 'N';
1196 goto out;
1197 }
1198
1199 if (tc & BLK_TC_FLUSH)
1200 rwbs[i++] = 'F';
1201
1202 if (tc & BLK_TC_DISCARD)
1203 rwbs[i++] = 'D';
1204 else if (tc & BLK_TC_WRITE)
1205 rwbs[i++] = 'W';
1206 else if (t->bytes)
1207 rwbs[i++] = 'R';
1208 else
1209 rwbs[i++] = 'N';
1210
1211 if (tc & BLK_TC_FUA)
1212 rwbs[i++] = 'F';
1213 if (tc & BLK_TC_AHEAD)
1214 rwbs[i++] = 'A';
1215 if (tc & BLK_TC_SYNC)
1216 rwbs[i++] = 'S';
1217 if (tc & BLK_TC_META)
1218 rwbs[i++] = 'M';
1219out:
1220 rwbs[i] = '\0';
1221}
1222
1223static inline
1224const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1225{
1226 return (const struct blk_io_trace *)ent;
1227}
1228
1229static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1230{
1231 return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1232}
1233
1234static inline u64 t_cgid(const struct trace_entry *ent)
1235{
1236 return *(u64 *)(te_blk_io_trace(ent) + 1);
1237}
1238
1239static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1240{
1241 return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1242}
1243
1244static inline u32 t_action(const struct trace_entry *ent)
1245{
1246 return te_blk_io_trace(ent)->action;
1247}
1248
1249static inline u32 t_bytes(const struct trace_entry *ent)
1250{
1251 return te_blk_io_trace(ent)->bytes;
1252}
1253
1254static inline u32 t_sec(const struct trace_entry *ent)
1255{
1256 return te_blk_io_trace(ent)->bytes >> 9;
1257}
1258
1259static inline unsigned long long t_sector(const struct trace_entry *ent)
1260{
1261 return te_blk_io_trace(ent)->sector;
1262}
1263
1264static inline __u16 t_error(const struct trace_entry *ent)
1265{
1266 return te_blk_io_trace(ent)->error;
1267}
1268
1269static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1270{
1271 const __be64 *val = pdu_start(ent, has_cg);
1272 return be64_to_cpu(*val);
1273}
1274
1275typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1276 bool has_cg);
1277
1278static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1279 bool has_cg)
1280{
1281 char rwbs[RWBS_LEN];
1282 unsigned long long ts = iter->ts;
1283 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1284 unsigned secs = (unsigned long)ts;
1285 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1286
1287 fill_rwbs(rwbs, t);
1288
1289 trace_seq_printf(&iter->seq,
1290 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1291 MAJOR(t->device), MINOR(t->device), iter->cpu,
1292 secs, nsec_rem, iter->ent->pid, act, rwbs);
1293}
1294
1295static void blk_log_action(struct trace_iterator *iter, const char *act,
1296 bool has_cg)
1297{
1298 char rwbs[RWBS_LEN];
1299 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1300
1301 fill_rwbs(rwbs, t);
1302 if (has_cg) {
1303 u64 id = t_cgid(iter->ent);
1304
1305 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1306 char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1307
1308 cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1309 sizeof(blkcg_name_buf));
1310 trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1311 MAJOR(t->device), MINOR(t->device),
1312 blkcg_name_buf, act, rwbs);
1313 } else {
1314 /*
1315 * The cgid portion used to be "INO,GEN". Userland
1316 * builds a FILEID_INO32_GEN fid out of them and
1317 * opens the cgroup using open_by_handle_at(2).
1318 * While 32bit ino setups are still the same, 64bit
1319 * ones now use the 64bit ino as the whole ID and
1320 * no longer use generation.
1321 *
1322 * Regarldess of the content, always output
1323 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1324 * be mapped back to @id on both 64 and 32bit ino
1325 * setups. See __kernfs_fh_to_dentry().
1326 */
1327 trace_seq_printf(&iter->seq,
1328 "%3d,%-3d %llx,%-llx %2s %3s ",
1329 MAJOR(t->device), MINOR(t->device),
1330 id & U32_MAX, id >> 32, act, rwbs);
1331 }
1332 } else
1333 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1334 MAJOR(t->device), MINOR(t->device), act, rwbs);
1335}
1336
1337static void blk_log_dump_pdu(struct trace_seq *s,
1338 const struct trace_entry *ent, bool has_cg)
1339{
1340 const unsigned char *pdu_buf;
1341 int pdu_len;
1342 int i, end;
1343
1344 pdu_buf = pdu_start(ent, has_cg);
1345 pdu_len = pdu_real_len(ent, has_cg);
1346
1347 if (!pdu_len)
1348 return;
1349
1350 /* find the last zero that needs to be printed */
1351 for (end = pdu_len - 1; end >= 0; end--)
1352 if (pdu_buf[end])
1353 break;
1354 end++;
1355
1356 trace_seq_putc(s, '(');
1357
1358 for (i = 0; i < pdu_len; i++) {
1359
1360 trace_seq_printf(s, "%s%02x",
1361 i == 0 ? "" : " ", pdu_buf[i]);
1362
1363 /*
1364 * stop when the rest is just zeroes and indicate so
1365 * with a ".." appended
1366 */
1367 if (i == end && end != pdu_len - 1) {
1368 trace_seq_puts(s, " ..) ");
1369 return;
1370 }
1371 }
1372
1373 trace_seq_puts(s, ") ");
1374}
1375
1376static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1377{
1378 char cmd[TASK_COMM_LEN];
1379
1380 trace_find_cmdline(ent->pid, cmd);
1381
1382 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1383 trace_seq_printf(s, "%u ", t_bytes(ent));
1384 blk_log_dump_pdu(s, ent, has_cg);
1385 trace_seq_printf(s, "[%s]\n", cmd);
1386 } else {
1387 if (t_sec(ent))
1388 trace_seq_printf(s, "%llu + %u [%s]\n",
1389 t_sector(ent), t_sec(ent), cmd);
1390 else
1391 trace_seq_printf(s, "[%s]\n", cmd);
1392 }
1393}
1394
1395static void blk_log_with_error(struct trace_seq *s,
1396 const struct trace_entry *ent, bool has_cg)
1397{
1398 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1399 blk_log_dump_pdu(s, ent, has_cg);
1400 trace_seq_printf(s, "[%d]\n", t_error(ent));
1401 } else {
1402 if (t_sec(ent))
1403 trace_seq_printf(s, "%llu + %u [%d]\n",
1404 t_sector(ent),
1405 t_sec(ent), t_error(ent));
1406 else
1407 trace_seq_printf(s, "%llu [%d]\n",
1408 t_sector(ent), t_error(ent));
1409 }
1410}
1411
1412static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1413{
1414 const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1415
1416 trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1417 t_sector(ent), t_sec(ent),
1418 MAJOR(be32_to_cpu(__r->device_from)),
1419 MINOR(be32_to_cpu(__r->device_from)),
1420 be64_to_cpu(__r->sector_from));
1421}
1422
1423static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1424{
1425 char cmd[TASK_COMM_LEN];
1426
1427 trace_find_cmdline(ent->pid, cmd);
1428
1429 trace_seq_printf(s, "[%s]\n", cmd);
1430}
1431
1432static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1433{
1434 char cmd[TASK_COMM_LEN];
1435
1436 trace_find_cmdline(ent->pid, cmd);
1437
1438 trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1439}
1440
1441static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1442{
1443 char cmd[TASK_COMM_LEN];
1444
1445 trace_find_cmdline(ent->pid, cmd);
1446
1447 trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1448 get_pdu_int(ent, has_cg), cmd);
1449}
1450
1451static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1452 bool has_cg)
1453{
1454
1455 trace_seq_putmem(s, pdu_start(ent, has_cg),
1456 pdu_real_len(ent, has_cg));
1457 trace_seq_putc(s, '\n');
1458}
1459
1460/*
1461 * struct tracer operations
1462 */
1463
1464static void blk_tracer_print_header(struct seq_file *m)
1465{
1466 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1467 return;
1468 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1469 "# | | | | | |\n");
1470}
1471
1472static void blk_tracer_start(struct trace_array *tr)
1473{
1474 blk_tracer_enabled = true;
1475}
1476
1477static int blk_tracer_init(struct trace_array *tr)
1478{
1479 blk_tr = tr;
1480 blk_tracer_start(tr);
1481 return 0;
1482}
1483
1484static void blk_tracer_stop(struct trace_array *tr)
1485{
1486 blk_tracer_enabled = false;
1487}
1488
1489static void blk_tracer_reset(struct trace_array *tr)
1490{
1491 blk_tracer_stop(tr);
1492}
1493
1494static const struct {
1495 const char *act[2];
1496 void (*print)(struct trace_seq *s, const struct trace_entry *ent,
1497 bool has_cg);
1498} what2act[] = {
1499 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1500 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1501 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1502 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1503 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1504 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1505 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1506 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1507 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1508 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1509 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1510 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1511 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1512 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1513 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1514};
1515
1516static enum print_line_t print_one_line(struct trace_iterator *iter,
1517 bool classic)
1518{
1519 struct trace_array *tr = iter->tr;
1520 struct trace_seq *s = &iter->seq;
1521 const struct blk_io_trace *t;
1522 u16 what;
1523 bool long_act;
1524 blk_log_action_t *log_action;
1525 bool has_cg;
1526
1527 t = te_blk_io_trace(iter->ent);
1528 what = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1529 long_act = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1530 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1531 has_cg = t->action & __BLK_TA_CGROUP;
1532
1533 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1534 log_action(iter, long_act ? "message" : "m", has_cg);
1535 blk_log_msg(s, iter->ent, has_cg);
1536 return trace_handle_return(s);
1537 }
1538
1539 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1540 trace_seq_printf(s, "Unknown action %x\n", what);
1541 else {
1542 log_action(iter, what2act[what].act[long_act], has_cg);
1543 what2act[what].print(s, iter->ent, has_cg);
1544 }
1545
1546 return trace_handle_return(s);
1547}
1548
1549static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1550 int flags, struct trace_event *event)
1551{
1552 return print_one_line(iter, false);
1553}
1554
1555static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1556{
1557 struct trace_seq *s = &iter->seq;
1558 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1559 const int offset = offsetof(struct blk_io_trace, sector);
1560 struct blk_io_trace old = {
1561 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1562 .time = iter->ts,
1563 };
1564
1565 trace_seq_putmem(s, &old, offset);
1566 trace_seq_putmem(s, &t->sector,
1567 sizeof(old) - offset + t->pdu_len);
1568}
1569
1570static enum print_line_t
1571blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1572 struct trace_event *event)
1573{
1574 blk_trace_synthesize_old_trace(iter);
1575
1576 return trace_handle_return(&iter->seq);
1577}
1578
1579static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1580{
1581 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1582 return TRACE_TYPE_UNHANDLED;
1583
1584 return print_one_line(iter, true);
1585}
1586
1587static int
1588blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1589{
1590 /* don't output context-info for blk_classic output */
1591 if (bit == TRACE_BLK_OPT_CLASSIC) {
1592 if (set)
1593 tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1594 else
1595 tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1596 }
1597 return 0;
1598}
1599
1600static struct tracer blk_tracer __read_mostly = {
1601 .name = "blk",
1602 .init = blk_tracer_init,
1603 .reset = blk_tracer_reset,
1604 .start = blk_tracer_start,
1605 .stop = blk_tracer_stop,
1606 .print_header = blk_tracer_print_header,
1607 .print_line = blk_tracer_print_line,
1608 .flags = &blk_tracer_flags,
1609 .set_flag = blk_tracer_set_flag,
1610};
1611
1612static struct trace_event_functions trace_blk_event_funcs = {
1613 .trace = blk_trace_event_print,
1614 .binary = blk_trace_event_print_binary,
1615};
1616
1617static struct trace_event trace_blk_event = {
1618 .type = TRACE_BLK,
1619 .funcs = &trace_blk_event_funcs,
1620};
1621
1622static int __init init_blk_tracer(void)
1623{
1624 if (!register_trace_event(&trace_blk_event)) {
1625 pr_warn("Warning: could not register block events\n");
1626 return 1;
1627 }
1628
1629 if (register_tracer(&blk_tracer) != 0) {
1630 pr_warn("Warning: could not register the block tracer\n");
1631 unregister_trace_event(&trace_blk_event);
1632 return 1;
1633 }
1634
1635 return 0;
1636}
1637
1638device_initcall(init_blk_tracer);
1639
1640static int blk_trace_remove_queue(struct request_queue *q)
1641{
1642 struct blk_trace *bt;
1643
1644 bt = rcu_replace_pointer(q->blk_trace, NULL,
1645 lockdep_is_held(&q->blk_trace_mutex));
1646 if (bt == NULL)
1647 return -EINVAL;
1648
1649 put_probe_ref();
1650 synchronize_rcu();
1651 blk_trace_free(bt);
1652 return 0;
1653}
1654
1655/*
1656 * Setup everything required to start tracing
1657 */
1658static int blk_trace_setup_queue(struct request_queue *q,
1659 struct block_device *bdev)
1660{
1661 struct blk_trace *bt = NULL;
1662 int ret = -ENOMEM;
1663
1664 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1665 if (!bt)
1666 return -ENOMEM;
1667
1668 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1669 if (!bt->msg_data)
1670 goto free_bt;
1671
1672 bt->dev = bdev->bd_dev;
1673 bt->act_mask = (u16)-1;
1674
1675 blk_trace_setup_lba(bt, bdev);
1676
1677 rcu_assign_pointer(q->blk_trace, bt);
1678 get_probe_ref();
1679 return 0;
1680
1681free_bt:
1682 blk_trace_free(bt);
1683 return ret;
1684}
1685
1686/*
1687 * sysfs interface to enable and configure tracing
1688 */
1689
1690static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1691 struct device_attribute *attr,
1692 char *buf);
1693static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1694 struct device_attribute *attr,
1695 const char *buf, size_t count);
1696#define BLK_TRACE_DEVICE_ATTR(_name) \
1697 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1698 sysfs_blk_trace_attr_show, \
1699 sysfs_blk_trace_attr_store)
1700
1701static BLK_TRACE_DEVICE_ATTR(enable);
1702static BLK_TRACE_DEVICE_ATTR(act_mask);
1703static BLK_TRACE_DEVICE_ATTR(pid);
1704static BLK_TRACE_DEVICE_ATTR(start_lba);
1705static BLK_TRACE_DEVICE_ATTR(end_lba);
1706
1707static struct attribute *blk_trace_attrs[] = {
1708 &dev_attr_enable.attr,
1709 &dev_attr_act_mask.attr,
1710 &dev_attr_pid.attr,
1711 &dev_attr_start_lba.attr,
1712 &dev_attr_end_lba.attr,
1713 NULL
1714};
1715
1716struct attribute_group blk_trace_attr_group = {
1717 .name = "trace",
1718 .attrs = blk_trace_attrs,
1719};
1720
1721static const struct {
1722 int mask;
1723 const char *str;
1724} mask_maps[] = {
1725 { BLK_TC_READ, "read" },
1726 { BLK_TC_WRITE, "write" },
1727 { BLK_TC_FLUSH, "flush" },
1728 { BLK_TC_SYNC, "sync" },
1729 { BLK_TC_QUEUE, "queue" },
1730 { BLK_TC_REQUEUE, "requeue" },
1731 { BLK_TC_ISSUE, "issue" },
1732 { BLK_TC_COMPLETE, "complete" },
1733 { BLK_TC_FS, "fs" },
1734 { BLK_TC_PC, "pc" },
1735 { BLK_TC_NOTIFY, "notify" },
1736 { BLK_TC_AHEAD, "ahead" },
1737 { BLK_TC_META, "meta" },
1738 { BLK_TC_DISCARD, "discard" },
1739 { BLK_TC_DRV_DATA, "drv_data" },
1740 { BLK_TC_FUA, "fua" },
1741};
1742
1743static int blk_trace_str2mask(const char *str)
1744{
1745 int i;
1746 int mask = 0;
1747 char *buf, *s, *token;
1748
1749 buf = kstrdup(str, GFP_KERNEL);
1750 if (buf == NULL)
1751 return -ENOMEM;
1752 s = strstrip(buf);
1753
1754 while (1) {
1755 token = strsep(&s, ",");
1756 if (token == NULL)
1757 break;
1758
1759 if (*token == '\0')
1760 continue;
1761
1762 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1763 if (strcasecmp(token, mask_maps[i].str) == 0) {
1764 mask |= mask_maps[i].mask;
1765 break;
1766 }
1767 }
1768 if (i == ARRAY_SIZE(mask_maps)) {
1769 mask = -EINVAL;
1770 break;
1771 }
1772 }
1773 kfree(buf);
1774
1775 return mask;
1776}
1777
1778static ssize_t blk_trace_mask2str(char *buf, int mask)
1779{
1780 int i;
1781 char *p = buf;
1782
1783 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1784 if (mask & mask_maps[i].mask) {
1785 p += sprintf(p, "%s%s",
1786 (p == buf) ? "" : ",", mask_maps[i].str);
1787 }
1788 }
1789 *p++ = '\n';
1790
1791 return p - buf;
1792}
1793
1794static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1795{
1796 if (bdev->bd_disk == NULL)
1797 return NULL;
1798
1799 return bdev_get_queue(bdev);
1800}
1801
1802static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1803 struct device_attribute *attr,
1804 char *buf)
1805{
1806 struct hd_struct *p = dev_to_part(dev);
1807 struct request_queue *q;
1808 struct block_device *bdev;
1809 struct blk_trace *bt;
1810 ssize_t ret = -ENXIO;
1811
1812 bdev = bdget(part_devt(p));
1813 if (bdev == NULL)
1814 goto out;
1815
1816 q = blk_trace_get_queue(bdev);
1817 if (q == NULL)
1818 goto out_bdput;
1819
1820 mutex_lock(&q->blk_trace_mutex);
1821
1822 bt = rcu_dereference_protected(q->blk_trace,
1823 lockdep_is_held(&q->blk_trace_mutex));
1824 if (attr == &dev_attr_enable) {
1825 ret = sprintf(buf, "%u\n", !!bt);
1826 goto out_unlock_bdev;
1827 }
1828
1829 if (bt == NULL)
1830 ret = sprintf(buf, "disabled\n");
1831 else if (attr == &dev_attr_act_mask)
1832 ret = blk_trace_mask2str(buf, bt->act_mask);
1833 else if (attr == &dev_attr_pid)
1834 ret = sprintf(buf, "%u\n", bt->pid);
1835 else if (attr == &dev_attr_start_lba)
1836 ret = sprintf(buf, "%llu\n", bt->start_lba);
1837 else if (attr == &dev_attr_end_lba)
1838 ret = sprintf(buf, "%llu\n", bt->end_lba);
1839
1840out_unlock_bdev:
1841 mutex_unlock(&q->blk_trace_mutex);
1842out_bdput:
1843 bdput(bdev);
1844out:
1845 return ret;
1846}
1847
1848static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1849 struct device_attribute *attr,
1850 const char *buf, size_t count)
1851{
1852 struct block_device *bdev;
1853 struct request_queue *q;
1854 struct hd_struct *p;
1855 struct blk_trace *bt;
1856 u64 value;
1857 ssize_t ret = -EINVAL;
1858
1859 if (count == 0)
1860 goto out;
1861
1862 if (attr == &dev_attr_act_mask) {
1863 if (kstrtoull(buf, 0, &value)) {
1864 /* Assume it is a list of trace category names */
1865 ret = blk_trace_str2mask(buf);
1866 if (ret < 0)
1867 goto out;
1868 value = ret;
1869 }
1870 } else if (kstrtoull(buf, 0, &value))
1871 goto out;
1872
1873 ret = -ENXIO;
1874
1875 p = dev_to_part(dev);
1876 bdev = bdget(part_devt(p));
1877 if (bdev == NULL)
1878 goto out;
1879
1880 q = blk_trace_get_queue(bdev);
1881 if (q == NULL)
1882 goto out_bdput;
1883
1884 mutex_lock(&q->blk_trace_mutex);
1885
1886 bt = rcu_dereference_protected(q->blk_trace,
1887 lockdep_is_held(&q->blk_trace_mutex));
1888 if (attr == &dev_attr_enable) {
1889 if (!!value == !!bt) {
1890 ret = 0;
1891 goto out_unlock_bdev;
1892 }
1893 if (value)
1894 ret = blk_trace_setup_queue(q, bdev);
1895 else
1896 ret = blk_trace_remove_queue(q);
1897 goto out_unlock_bdev;
1898 }
1899
1900 ret = 0;
1901 if (bt == NULL) {
1902 ret = blk_trace_setup_queue(q, bdev);
1903 bt = rcu_dereference_protected(q->blk_trace,
1904 lockdep_is_held(&q->blk_trace_mutex));
1905 }
1906
1907 if (ret == 0) {
1908 if (attr == &dev_attr_act_mask)
1909 bt->act_mask = value;
1910 else if (attr == &dev_attr_pid)
1911 bt->pid = value;
1912 else if (attr == &dev_attr_start_lba)
1913 bt->start_lba = value;
1914 else if (attr == &dev_attr_end_lba)
1915 bt->end_lba = value;
1916 }
1917
1918out_unlock_bdev:
1919 mutex_unlock(&q->blk_trace_mutex);
1920out_bdput:
1921 bdput(bdev);
1922out:
1923 return ret ? ret : count;
1924}
1925
1926int blk_trace_init_sysfs(struct device *dev)
1927{
1928 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1929}
1930
1931void blk_trace_remove_sysfs(struct device *dev)
1932{
1933 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1934}
1935
1936#endif /* CONFIG_BLK_DEV_IO_TRACE */
1937
1938#ifdef CONFIG_EVENT_TRACING
1939
1940void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1941{
1942 int i = 0;
1943
1944 if (op & REQ_PREFLUSH)
1945 rwbs[i++] = 'F';
1946
1947 switch (op & REQ_OP_MASK) {
1948 case REQ_OP_WRITE:
1949 case REQ_OP_WRITE_SAME:
1950 rwbs[i++] = 'W';
1951 break;
1952 case REQ_OP_DISCARD:
1953 rwbs[i++] = 'D';
1954 break;
1955 case REQ_OP_SECURE_ERASE:
1956 rwbs[i++] = 'D';
1957 rwbs[i++] = 'E';
1958 break;
1959 case REQ_OP_FLUSH:
1960 rwbs[i++] = 'F';
1961 break;
1962 case REQ_OP_READ:
1963 rwbs[i++] = 'R';
1964 break;
1965 default:
1966 rwbs[i++] = 'N';
1967 }
1968
1969 if (op & REQ_FUA)
1970 rwbs[i++] = 'F';
1971 if (op & REQ_RAHEAD)
1972 rwbs[i++] = 'A';
1973 if (op & REQ_SYNC)
1974 rwbs[i++] = 'S';
1975 if (op & REQ_META)
1976 rwbs[i++] = 'M';
1977
1978 rwbs[i] = '\0';
1979}
1980EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1981
1982#endif /* CONFIG_EVENT_TRACING */
1983