?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.9.199
[linux/fpc-iii.git] / kernel / trace / blktrace.c
blobbfa8bb3a6e1966937037c91b2baa7a689014dd80
1 /*
2 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
16 *
17 */
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/slab.h>
25 #include <linux/debugfs.h>
26 #include <linux/export.h>
27 #include <linux/time.h>
28 #include <linux/uaccess.h>
29 #include <linux/list.h>
30
31 #include <trace/events/block.h>
32
33 #include "trace_output.h"
34
35 #ifdef CONFIG_BLK_DEV_IO_TRACE
36
37 static unsigned int blktrace_seq __read_mostly = 1;
38
39 static struct trace_array *blk_tr;
40 static bool blk_tracer_enabled __read_mostly;
41
42 static LIST_HEAD(running_trace_list);
43 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
44
45 /* Select an alternative, minimalistic output than the original one */
46 #define TRACE_BLK_OPT_CLASSIC 0x1
47
48 static struct tracer_opt blk_tracer_opts[] = {
49 /* Default disable the minimalistic output */
50 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
51 { }
52 };
53
54 static struct tracer_flags blk_tracer_flags = {
55 .val = 0,
56 .opts = blk_tracer_opts,
57 };
58
59 /* Global reference count of probes */
60 static DEFINE_MUTEX(blk_probe_mutex);
61 static int blk_probes_ref;
62
63 static void blk_register_tracepoints(void);
64 static void blk_unregister_tracepoints(void);
65
66 /*
67 * Send out a notify message.
68 */
69 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
70 const void *data, size_t len)
71 {
72 struct blk_io_trace *t;
73 struct ring_buffer_event *event = NULL;
74 struct ring_buffer *buffer = NULL;
75 int pc = 0;
76 int cpu = smp_processor_id();
77 bool blk_tracer = blk_tracer_enabled;
78
79 if (blk_tracer) {
80 buffer = blk_tr->trace_buffer.buffer;
81 pc = preempt_count();
82 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
83 sizeof(*t) + len,
84 0, pc);
85 if (!event)
86 return;
87 t = ring_buffer_event_data(event);
88 goto record_it;
89 }
90
91 if (!bt->rchan)
92 return;
93
94 t = relay_reserve(bt->rchan, sizeof(*t) + len);
95 if (t) {
96 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
97 t->time = ktime_to_ns(ktime_get());
98 record_it:
99 t->device = bt->dev;
100 t->action = action;
101 t->pid = pid;
102 t->cpu = cpu;
103 t->pdu_len = len;
104 memcpy((void *) t + sizeof(*t), 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 */
115 static 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));
125 }
126 spin_unlock_irqrestore(&running_trace_lock, flags);
127 }
128
129 static 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));
142 local_irq_restore(flags);
143 }
144
145 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
146 {
147 int n;
148 va_list args;
149 unsigned long flags;
150 char *buf;
151
152 if (unlikely(bt->trace_state != Blktrace_running &&
153 !blk_tracer_enabled))
154 return;
155
156 /*
157 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
158 * message to the trace.
159 */
160 if (!(bt->act_mask & BLK_TC_NOTIFY))
161 return;
162
163 local_irq_save(flags);
164 buf = this_cpu_ptr(bt->msg_data);
165 va_start(args, fmt);
166 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
167 va_end(args);
168
169 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
170 local_irq_restore(flags);
171 }
172 EXPORT_SYMBOL_GPL(__trace_note_message);
173
174 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
175 pid_t pid)
176 {
177 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
178 return 1;
179 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
180 return 1;
181 if (bt->pid && pid != bt->pid)
182 return 1;
183
184 return 0;
185 }
186
187 /*
188 * Data direction bit lookup
189 */
190 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
191 BLK_TC_ACT(BLK_TC_WRITE) };
192
193 #define BLK_TC_RAHEAD BLK_TC_AHEAD
194 #define BLK_TC_PREFLUSH BLK_TC_FLUSH
195
196 /* The ilog2() calls fall out because they're constant */
197 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
198 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
199
200 /*
201 * The worker for the various blk_add_trace*() types. Fills out a
202 * blk_io_trace structure and places it in a per-cpu subbuffer.
203 */
204 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
205 int op, int op_flags, u32 what, int error, int pdu_len,
206 void *pdu_data)
207 {
208 struct task_struct *tsk = current;
209 struct ring_buffer_event *event = NULL;
210 struct ring_buffer *buffer = NULL;
211 struct blk_io_trace *t;
212 unsigned long flags = 0;
213 unsigned long *sequence;
214 pid_t pid;
215 int cpu, pc = 0;
216 bool blk_tracer = blk_tracer_enabled;
217
218 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
219 return;
220
221 what |= ddir_act[op_is_write(op) ? WRITE : READ];
222 what |= MASK_TC_BIT(op_flags, SYNC);
223 what |= MASK_TC_BIT(op_flags, RAHEAD);
224 what |= MASK_TC_BIT(op_flags, META);
225 what |= MASK_TC_BIT(op_flags, PREFLUSH);
226 what |= MASK_TC_BIT(op_flags, FUA);
227 if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
228 what |= BLK_TC_ACT(BLK_TC_DISCARD);
229 if (op == REQ_OP_FLUSH)
230 what |= BLK_TC_ACT(BLK_TC_FLUSH);
231
232 pid = tsk->pid;
233 if (act_log_check(bt, what, sector, pid))
234 return;
235 cpu = raw_smp_processor_id();
236
237 if (blk_tracer) {
238 tracing_record_cmdline(current);
239
240 buffer = blk_tr->trace_buffer.buffer;
241 pc = preempt_count();
242 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
243 sizeof(*t) + pdu_len,
244 0, pc);
245 if (!event)
246 return;
247 t = ring_buffer_event_data(event);
248 goto record_it;
249 }
250
251 if (unlikely(tsk->btrace_seq != blktrace_seq))
252 trace_note_tsk(tsk);
253
254 /*
255 * A word about the locking here - we disable interrupts to reserve
256 * some space in the relay per-cpu buffer, to prevent an irq
257 * from coming in and stepping on our toes.
258 */
259 local_irq_save(flags);
260 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
261 if (t) {
262 sequence = per_cpu_ptr(bt->sequence, cpu);
263
264 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
265 t->sequence = ++(*sequence);
266 t->time = ktime_to_ns(ktime_get());
267 record_it:
268 /*
269 * These two are not needed in ftrace as they are in the
270 * generic trace_entry, filled by tracing_generic_entry_update,
271 * but for the trace_event->bin() synthesizer benefit we do it
272 * here too.
273 */
274 t->cpu = cpu;
275 t->pid = pid;
276
277 t->sector = sector;
278 t->bytes = bytes;
279 t->action = what;
280 t->device = bt->dev;
281 t->error = error;
282 t->pdu_len = pdu_len;
283
284 if (pdu_len)
285 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
286
287 if (blk_tracer) {
288 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
289 return;
290 }
291 }
292
293 local_irq_restore(flags);
294 }
295
296 static struct dentry *blk_tree_root;
297 static DEFINE_MUTEX(blk_tree_mutex);
298
299 static void blk_trace_free(struct blk_trace *bt)
300 {
301 debugfs_remove(bt->msg_file);
302 debugfs_remove(bt->dropped_file);
303 relay_close(bt->rchan);
304 debugfs_remove(bt->dir);
305 free_percpu(bt->sequence);
306 free_percpu(bt->msg_data);
307 kfree(bt);
308 }
309
310 static void get_probe_ref(void)
311 {
312 mutex_lock(&blk_probe_mutex);
313 if (++blk_probes_ref == 1)
314 blk_register_tracepoints();
315 mutex_unlock(&blk_probe_mutex);
316 }
317
318 static void put_probe_ref(void)
319 {
320 mutex_lock(&blk_probe_mutex);
321 if (!--blk_probes_ref)
322 blk_unregister_tracepoints();
323 mutex_unlock(&blk_probe_mutex);
324 }
325
326 static void blk_trace_cleanup(struct blk_trace *bt)
327 {
328 blk_trace_free(bt);
329 put_probe_ref();
330 }
331
332 int blk_trace_remove(struct request_queue *q)
333 {
334 struct blk_trace *bt;
335
336 bt = xchg(&q->blk_trace, NULL);
337 if (!bt)
338 return -EINVAL;
339
340 if (bt->trace_state != Blktrace_running)
341 blk_trace_cleanup(bt);
342
343 return 0;
344 }
345 EXPORT_SYMBOL_GPL(blk_trace_remove);
346
347 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
348 size_t count, loff_t *ppos)
349 {
350 struct blk_trace *bt = filp->private_data;
351 char buf[16];
352
353 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
354
355 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
356 }
357
358 static const struct file_operations blk_dropped_fops = {
359 .owner = THIS_MODULE,
360 .open = simple_open,
361 .read = blk_dropped_read,
362 .llseek = default_llseek,
363 };
364
365 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
366 size_t count, loff_t *ppos)
367 {
368 char *msg;
369 struct blk_trace *bt;
370
371 if (count >= BLK_TN_MAX_MSG)
372 return -EINVAL;
373
374 msg = memdup_user_nul(buffer, count);
375 if (IS_ERR(msg))
376 return PTR_ERR(msg);
377
378 bt = filp->private_data;
379 __trace_note_message(bt, "%s", msg);
380 kfree(msg);
381
382 return count;
383 }
384
385 static const struct file_operations blk_msg_fops = {
386 .owner = THIS_MODULE,
387 .open = simple_open,
388 .write = blk_msg_write,
389 .llseek = noop_llseek,
390 };
391
392 /*
393 * Keep track of how many times we encountered a full subbuffer, to aid
394 * the user space app in telling how many lost events there were.
395 */
396 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
397 void *prev_subbuf, size_t prev_padding)
398 {
399 struct blk_trace *bt;
400
401 if (!relay_buf_full(buf))
402 return 1;
403
404 bt = buf->chan->private_data;
405 atomic_inc(&bt->dropped);
406 return 0;
407 }
408
409 static int blk_remove_buf_file_callback(struct dentry *dentry)
410 {
411 debugfs_remove(dentry);
412
413 return 0;
414 }
415
416 static struct dentry *blk_create_buf_file_callback(const char *filename,
417 struct dentry *parent,
418 umode_t mode,
419 struct rchan_buf *buf,
420 int *is_global)
421 {
422 return debugfs_create_file(filename, mode, parent, buf,
423 &relay_file_operations);
424 }
425
426 static struct rchan_callbacks blk_relay_callbacks = {
427 .subbuf_start = blk_subbuf_start_callback,
428 .create_buf_file = blk_create_buf_file_callback,
429 .remove_buf_file = blk_remove_buf_file_callback,
430 };
431
432 static void blk_trace_setup_lba(struct blk_trace *bt,
433 struct block_device *bdev)
434 {
435 struct hd_struct *part = NULL;
436
437 if (bdev)
438 part = bdev->bd_part;
439
440 if (part) {
441 bt->start_lba = part->start_sect;
442 bt->end_lba = part->start_sect + part->nr_sects;
443 } else {
444 bt->start_lba = 0;
445 bt->end_lba = -1ULL;
446 }
447 }
448
449 /*
450 * Setup everything required to start tracing
451 */
452 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
453 struct block_device *bdev,
454 struct blk_user_trace_setup *buts)
455 {
456 struct blk_trace *bt = NULL;
457 struct dentry *dir = NULL;
458 int ret;
459
460 if (!buts->buf_size || !buts->buf_nr)
461 return -EINVAL;
462
463 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
464 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
465
466 /*
467 * some device names have larger paths - convert the slashes
468 * to underscores for this to work as expected
469 */
470 strreplace(buts->name, '/', '_');
471
472 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
473 if (!bt)
474 return -ENOMEM;
475
476 ret = -ENOMEM;
477 bt->sequence = alloc_percpu(unsigned long);
478 if (!bt->sequence)
479 goto err;
480
481 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
482 if (!bt->msg_data)
483 goto err;
484
485 ret = -ENOENT;
486
487 mutex_lock(&blk_tree_mutex);
488 if (!blk_tree_root) {
489 blk_tree_root = debugfs_create_dir("block", NULL);
490 if (!blk_tree_root) {
491 mutex_unlock(&blk_tree_mutex);
492 goto err;
493 }
494 }
495 mutex_unlock(&blk_tree_mutex);
496
497 dir = debugfs_create_dir(buts->name, blk_tree_root);
498
499 if (!dir)
500 goto err;
501
502 bt->dir = dir;
503 bt->dev = dev;
504 atomic_set(&bt->dropped, 0);
505 INIT_LIST_HEAD(&bt->running_list);
506
507 ret = -EIO;
508 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
509 &blk_dropped_fops);
510 if (!bt->dropped_file)
511 goto err;
512
513 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
514 if (!bt->msg_file)
515 goto err;
516
517 bt->rchan = relay_open("trace", dir, buts->buf_size,
518 buts->buf_nr, &blk_relay_callbacks, bt);
519 if (!bt->rchan)
520 goto err;
521
522 bt->act_mask = buts->act_mask;
523 if (!bt->act_mask)
524 bt->act_mask = (u16) -1;
525
526 blk_trace_setup_lba(bt, bdev);
527
528 /* overwrite with user settings */
529 if (buts->start_lba)
530 bt->start_lba = buts->start_lba;
531 if (buts->end_lba)
532 bt->end_lba = buts->end_lba;
533
534 bt->pid = buts->pid;
535 bt->trace_state = Blktrace_setup;
536
537 ret = -EBUSY;
538 if (cmpxchg(&q->blk_trace, NULL, bt))
539 goto err;
540
541 get_probe_ref();
542
543 return 0;
544 err:
545 blk_trace_free(bt);
546 return ret;
547 }
548
549 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
550 struct block_device *bdev,
551 char __user *arg)
552 {
553 struct blk_user_trace_setup buts;
554 int ret;
555
556 ret = copy_from_user(&buts, arg, sizeof(buts));
557 if (ret)
558 return -EFAULT;
559
560 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
561 if (ret)
562 return ret;
563
564 if (copy_to_user(arg, &buts, sizeof(buts))) {
565 blk_trace_remove(q);
566 return -EFAULT;
567 }
568 return 0;
569 }
570 EXPORT_SYMBOL_GPL(blk_trace_setup);
571
572 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
573 static int compat_blk_trace_setup(struct request_queue *q, char *name,
574 dev_t dev, struct block_device *bdev,
575 char __user *arg)
576 {
577 struct blk_user_trace_setup buts;
578 struct compat_blk_user_trace_setup cbuts;
579 int ret;
580
581 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
582 return -EFAULT;
583
584 buts = (struct blk_user_trace_setup) {
585 .act_mask = cbuts.act_mask,
586 .buf_size = cbuts.buf_size,
587 .buf_nr = cbuts.buf_nr,
588 .start_lba = cbuts.start_lba,
589 .end_lba = cbuts.end_lba,
590 .pid = cbuts.pid,
591 };
592
593 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
594 if (ret)
595 return ret;
596
597 if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
598 blk_trace_remove(q);
599 return -EFAULT;
600 }
601
602 return 0;
603 }
604 #endif
605
606 int blk_trace_startstop(struct request_queue *q, int start)
607 {
608 int ret;
609 struct blk_trace *bt = q->blk_trace;
610
611 if (bt == NULL)
612 return -EINVAL;
613
614 /*
615 * For starting a trace, we can transition from a setup or stopped
616 * trace. For stopping a trace, the state must be running
617 */
618 ret = -EINVAL;
619 if (start) {
620 if (bt->trace_state == Blktrace_setup ||
621 bt->trace_state == Blktrace_stopped) {
622 blktrace_seq++;
623 smp_mb();
624 bt->trace_state = Blktrace_running;
625 spin_lock_irq(&running_trace_lock);
626 list_add(&bt->running_list, &running_trace_list);
627 spin_unlock_irq(&running_trace_lock);
628
629 trace_note_time(bt);
630 ret = 0;
631 }
632 } else {
633 if (bt->trace_state == Blktrace_running) {
634 bt->trace_state = Blktrace_stopped;
635 spin_lock_irq(&running_trace_lock);
636 list_del_init(&bt->running_list);
637 spin_unlock_irq(&running_trace_lock);
638 relay_flush(bt->rchan);
639 ret = 0;
640 }
641 }
642
643 return ret;
644 }
645 EXPORT_SYMBOL_GPL(blk_trace_startstop);
646
647 /**
648 * blk_trace_ioctl: - handle the ioctls associated with tracing
649 * @bdev: the block device
650 * @cmd: the ioctl cmd
651 * @arg: the argument data, if any
652 *
653 **/
654 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
655 {
656 struct request_queue *q;
657 int ret, start = 0;
658 char b[BDEVNAME_SIZE];
659
660 q = bdev_get_queue(bdev);
661 if (!q)
662 return -ENXIO;
663
664 mutex_lock(&bdev->bd_mutex);
665
666 switch (cmd) {
667 case BLKTRACESETUP:
668 bdevname(bdev, b);
669 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
670 break;
671 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
672 case BLKTRACESETUP32:
673 bdevname(bdev, b);
674 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
675 break;
676 #endif
677 case BLKTRACESTART:
678 start = 1;
679 case BLKTRACESTOP:
680 ret = blk_trace_startstop(q, start);
681 break;
682 case BLKTRACETEARDOWN:
683 ret = blk_trace_remove(q);
684 break;
685 default:
686 ret = -ENOTTY;
687 break;
688 }
689
690 mutex_unlock(&bdev->bd_mutex);
691 return ret;
692 }
693
694 /**
695 * blk_trace_shutdown: - stop and cleanup trace structures
696 * @q: the request queue associated with the device
697 *
698 **/
699 void blk_trace_shutdown(struct request_queue *q)
700 {
701 if (q->blk_trace) {
702 blk_trace_startstop(q, 0);
703 blk_trace_remove(q);
704 }
705 }
706
707 /*
708 * blktrace probes
709 */
710
711 /**
712 * blk_add_trace_rq - Add a trace for a request oriented action
713 * @q: queue the io is for
714 * @rq: the source request
715 * @nr_bytes: number of completed bytes
716 * @what: the action
717 *
718 * Description:
719 * Records an action against a request. Will log the bio offset + size.
720 *
721 **/
722 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
723 unsigned int nr_bytes, u32 what)
724 {
725 struct blk_trace *bt = q->blk_trace;
726
727 if (likely(!bt))
728 return;
729
730 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
731 what |= BLK_TC_ACT(BLK_TC_PC);
732 __blk_add_trace(bt, 0, nr_bytes, req_op(rq), rq->cmd_flags,
733 what, rq->errors, rq->cmd_len, rq->cmd);
734 } else {
735 what |= BLK_TC_ACT(BLK_TC_FS);
736 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes, req_op(rq),
737 rq->cmd_flags, what, rq->errors, 0, NULL);
738 }
739 }
740
741 static void blk_add_trace_rq_abort(void *ignore,
742 struct request_queue *q, struct request *rq)
743 {
744 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
745 }
746
747 static void blk_add_trace_rq_insert(void *ignore,
748 struct request_queue *q, struct request *rq)
749 {
750 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
751 }
752
753 static void blk_add_trace_rq_issue(void *ignore,
754 struct request_queue *q, struct request *rq)
755 {
756 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
757 }
758
759 static void blk_add_trace_rq_requeue(void *ignore,
760 struct request_queue *q,
761 struct request *rq)
762 {
763 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
764 }
765
766 static void blk_add_trace_rq_complete(void *ignore,
767 struct request_queue *q,
768 struct request *rq,
769 unsigned int nr_bytes)
770 {
771 blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
772 }
773
774 /**
775 * blk_add_trace_bio - Add a trace for a bio oriented action
776 * @q: queue the io is for
777 * @bio: the source bio
778 * @what: the action
779 * @error: error, if any
780 *
781 * Description:
782 * Records an action against a bio. Will log the bio offset + size.
783 *
784 **/
785 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
786 u32 what, int error)
787 {
788 struct blk_trace *bt = q->blk_trace;
789
790 if (likely(!bt))
791 return;
792
793 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
794 bio_op(bio), bio->bi_opf, what, error, 0, NULL);
795 }
796
797 static void blk_add_trace_bio_bounce(void *ignore,
798 struct request_queue *q, struct bio *bio)
799 {
800 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
801 }
802
803 static void blk_add_trace_bio_complete(void *ignore,
804 struct request_queue *q, struct bio *bio,
805 int error)
806 {
807 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
808 }
809
810 static void blk_add_trace_bio_backmerge(void *ignore,
811 struct request_queue *q,
812 struct request *rq,
813 struct bio *bio)
814 {
815 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
816 }
817
818 static void blk_add_trace_bio_frontmerge(void *ignore,
819 struct request_queue *q,
820 struct request *rq,
821 struct bio *bio)
822 {
823 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
824 }
825
826 static void blk_add_trace_bio_queue(void *ignore,
827 struct request_queue *q, struct bio *bio)
828 {
829 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
830 }
831
832 static void blk_add_trace_getrq(void *ignore,
833 struct request_queue *q,
834 struct bio *bio, int rw)
835 {
836 if (bio)
837 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
838 else {
839 struct blk_trace *bt = q->blk_trace;
840
841 if (bt)
842 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
843 NULL);
844 }
845 }
846
847
848 static void blk_add_trace_sleeprq(void *ignore,
849 struct request_queue *q,
850 struct bio *bio, int rw)
851 {
852 if (bio)
853 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
854 else {
855 struct blk_trace *bt = q->blk_trace;
856
857 if (bt)
858 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
859 0, 0, NULL);
860 }
861 }
862
863 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
864 {
865 struct blk_trace *bt = q->blk_trace;
866
867 if (bt)
868 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
869 }
870
871 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
872 unsigned int depth, bool explicit)
873 {
874 struct blk_trace *bt = q->blk_trace;
875
876 if (bt) {
877 __be64 rpdu = cpu_to_be64(depth);
878 u32 what;
879
880 if (explicit)
881 what = BLK_TA_UNPLUG_IO;
882 else
883 what = BLK_TA_UNPLUG_TIMER;
884
885 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
886 }
887 }
888
889 static void blk_add_trace_split(void *ignore,
890 struct request_queue *q, struct bio *bio,
891 unsigned int pdu)
892 {
893 struct blk_trace *bt = q->blk_trace;
894
895 if (bt) {
896 __be64 rpdu = cpu_to_be64(pdu);
897
898 __blk_add_trace(bt, bio->bi_iter.bi_sector,
899 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
900 BLK_TA_SPLIT, bio->bi_error, sizeof(rpdu),
901 &rpdu);
902 }
903 }
904
905 /**
906 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
907 * @ignore: trace callback data parameter (not used)
908 * @q: queue the io is for
909 * @bio: the source bio
910 * @dev: target device
911 * @from: source sector
912 *
913 * Description:
914 * Device mapper or raid target sometimes need to split a bio because
915 * it spans a stripe (or similar). Add a trace for that action.
916 *
917 **/
918 static void blk_add_trace_bio_remap(void *ignore,
919 struct request_queue *q, struct bio *bio,
920 dev_t dev, sector_t from)
921 {
922 struct blk_trace *bt = q->blk_trace;
923 struct blk_io_trace_remap r;
924
925 if (likely(!bt))
926 return;
927
928 r.device_from = cpu_to_be32(dev);
929 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
930 r.sector_from = cpu_to_be64(from);
931
932 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
933 bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_error,
934 sizeof(r), &r);
935 }
936
937 /**
938 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
939 * @ignore: trace callback data parameter (not used)
940 * @q: queue the io is for
941 * @rq: the source request
942 * @dev: target device
943 * @from: source sector
944 *
945 * Description:
946 * Device mapper remaps request to other devices.
947 * Add a trace for that action.
948 *
949 **/
950 static void blk_add_trace_rq_remap(void *ignore,
951 struct request_queue *q,
952 struct request *rq, dev_t dev,
953 sector_t from)
954 {
955 struct blk_trace *bt = q->blk_trace;
956 struct blk_io_trace_remap r;
957
958 if (likely(!bt))
959 return;
960
961 r.device_from = cpu_to_be32(dev);
962 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
963 r.sector_from = cpu_to_be64(from);
964
965 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
966 rq_data_dir(rq), 0, BLK_TA_REMAP, !!rq->errors,
967 sizeof(r), &r);
968 }
969
970 /**
971 * blk_add_driver_data - Add binary message with driver-specific data
972 * @q: queue the io is for
973 * @rq: io request
974 * @data: driver-specific data
975 * @len: length of driver-specific data
976 *
977 * Description:
978 * Some drivers might want to write driver-specific data per request.
979 *
980 **/
981 void blk_add_driver_data(struct request_queue *q,
982 struct request *rq,
983 void *data, size_t len)
984 {
985 struct blk_trace *bt = q->blk_trace;
986
987 if (likely(!bt))
988 return;
989
990 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
991 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, 0,
992 BLK_TA_DRV_DATA, rq->errors, len, data);
993 else
994 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, 0,
995 BLK_TA_DRV_DATA, rq->errors, len, data);
996 }
997 EXPORT_SYMBOL_GPL(blk_add_driver_data);
998
999 static void blk_register_tracepoints(void)
1000 {
1001 int ret;
1002
1003 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1004 WARN_ON(ret);
1005 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1006 WARN_ON(ret);
1007 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1008 WARN_ON(ret);
1009 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1010 WARN_ON(ret);
1011 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1012 WARN_ON(ret);
1013 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1014 WARN_ON(ret);
1015 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1016 WARN_ON(ret);
1017 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1018 WARN_ON(ret);
1019 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1020 WARN_ON(ret);
1021 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1022 WARN_ON(ret);
1023 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1024 WARN_ON(ret);
1025 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1026 WARN_ON(ret);
1027 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1028 WARN_ON(ret);
1029 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1030 WARN_ON(ret);
1031 ret = register_trace_block_split(blk_add_trace_split, NULL);
1032 WARN_ON(ret);
1033 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1034 WARN_ON(ret);
1035 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1036 WARN_ON(ret);
1037 }
1038
1039 static void blk_unregister_tracepoints(void)
1040 {
1041 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1042 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1043 unregister_trace_block_split(blk_add_trace_split, NULL);
1044 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1045 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1046 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1047 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1048 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1049 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1050 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1051 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1052 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1053 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1054 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1055 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1056 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1057 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1058
1059 tracepoint_synchronize_unregister();
1060 }
1061
1062 /*
1063 * struct blk_io_tracer formatting routines
1064 */
1065
1066 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1067 {
1068 int i = 0;
1069 int tc = t->action >> BLK_TC_SHIFT;
1070
1071 if (t->action == BLK_TN_MESSAGE) {
1072 rwbs[i++] = 'N';
1073 goto out;
1074 }
1075
1076 if (tc & BLK_TC_FLUSH)
1077 rwbs[i++] = 'F';
1078
1079 if (tc & BLK_TC_DISCARD)
1080 rwbs[i++] = 'D';
1081 else if (tc & BLK_TC_WRITE)
1082 rwbs[i++] = 'W';
1083 else if (t->bytes)
1084 rwbs[i++] = 'R';
1085 else
1086 rwbs[i++] = 'N';
1087
1088 if (tc & BLK_TC_FUA)
1089 rwbs[i++] = 'F';
1090 if (tc & BLK_TC_AHEAD)
1091 rwbs[i++] = 'A';
1092 if (tc & BLK_TC_SYNC)
1093 rwbs[i++] = 'S';
1094 if (tc & BLK_TC_META)
1095 rwbs[i++] = 'M';
1096 out:
1097 rwbs[i] = '\0';
1098 }
1099
1100 static inline
1101 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1102 {
1103 return (const struct blk_io_trace *)ent;
1104 }
1105
1106 static inline const void *pdu_start(const struct trace_entry *ent)
1107 {
1108 return te_blk_io_trace(ent) + 1;
1109 }
1110
1111 static inline u32 t_action(const struct trace_entry *ent)
1112 {
1113 return te_blk_io_trace(ent)->action;
1114 }
1115
1116 static inline u32 t_bytes(const struct trace_entry *ent)
1117 {
1118 return te_blk_io_trace(ent)->bytes;
1119 }
1120
1121 static inline u32 t_sec(const struct trace_entry *ent)
1122 {
1123 return te_blk_io_trace(ent)->bytes >> 9;
1124 }
1125
1126 static inline unsigned long long t_sector(const struct trace_entry *ent)
1127 {
1128 return te_blk_io_trace(ent)->sector;
1129 }
1130
1131 static inline __u16 t_error(const struct trace_entry *ent)
1132 {
1133 return te_blk_io_trace(ent)->error;
1134 }
1135
1136 static __u64 get_pdu_int(const struct trace_entry *ent)
1137 {
1138 const __u64 *val = pdu_start(ent);
1139 return be64_to_cpu(*val);
1140 }
1141
1142 static void get_pdu_remap(const struct trace_entry *ent,
1143 struct blk_io_trace_remap *r)
1144 {
1145 const struct blk_io_trace_remap *__r = pdu_start(ent);
1146 __u64 sector_from = __r->sector_from;
1147
1148 r->device_from = be32_to_cpu(__r->device_from);
1149 r->device_to = be32_to_cpu(__r->device_to);
1150 r->sector_from = be64_to_cpu(sector_from);
1151 }
1152
1153 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1154
1155 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1156 {
1157 char rwbs[RWBS_LEN];
1158 unsigned long long ts = iter->ts;
1159 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1160 unsigned secs = (unsigned long)ts;
1161 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1162
1163 fill_rwbs(rwbs, t);
1164
1165 trace_seq_printf(&iter->seq,
1166 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1167 MAJOR(t->device), MINOR(t->device), iter->cpu,
1168 secs, nsec_rem, iter->ent->pid, act, rwbs);
1169 }
1170
1171 static void blk_log_action(struct trace_iterator *iter, const char *act)
1172 {
1173 char rwbs[RWBS_LEN];
1174 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1175
1176 fill_rwbs(rwbs, t);
1177 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1178 MAJOR(t->device), MINOR(t->device), act, rwbs);
1179 }
1180
1181 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1182 {
1183 const unsigned char *pdu_buf;
1184 int pdu_len;
1185 int i, end;
1186
1187 pdu_buf = pdu_start(ent);
1188 pdu_len = te_blk_io_trace(ent)->pdu_len;
1189
1190 if (!pdu_len)
1191 return;
1192
1193 /* find the last zero that needs to be printed */
1194 for (end = pdu_len - 1; end >= 0; end--)
1195 if (pdu_buf[end])
1196 break;
1197 end++;
1198
1199 trace_seq_putc(s, '(');
1200
1201 for (i = 0; i < pdu_len; i++) {
1202
1203 trace_seq_printf(s, "%s%02x",
1204 i == 0 ? "" : " ", pdu_buf[i]);
1205
1206 /*
1207 * stop when the rest is just zeroes and indicate so
1208 * with a ".." appended
1209 */
1210 if (i == end && end != pdu_len - 1) {
1211 trace_seq_puts(s, " ..) ");
1212 return;
1213 }
1214 }
1215
1216 trace_seq_puts(s, ") ");
1217 }
1218
1219 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1220 {
1221 char cmd[TASK_COMM_LEN];
1222
1223 trace_find_cmdline(ent->pid, cmd);
1224
1225 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1226 trace_seq_printf(s, "%u ", t_bytes(ent));
1227 blk_log_dump_pdu(s, ent);
1228 trace_seq_printf(s, "[%s]\n", cmd);
1229 } else {
1230 if (t_sec(ent))
1231 trace_seq_printf(s, "%llu + %u [%s]\n",
1232 t_sector(ent), t_sec(ent), cmd);
1233 else
1234 trace_seq_printf(s, "[%s]\n", cmd);
1235 }
1236 }
1237
1238 static void blk_log_with_error(struct trace_seq *s,
1239 const struct trace_entry *ent)
1240 {
1241 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1242 blk_log_dump_pdu(s, ent);
1243 trace_seq_printf(s, "[%d]\n", t_error(ent));
1244 } else {
1245 if (t_sec(ent))
1246 trace_seq_printf(s, "%llu + %u [%d]\n",
1247 t_sector(ent),
1248 t_sec(ent), t_error(ent));
1249 else
1250 trace_seq_printf(s, "%llu [%d]\n",
1251 t_sector(ent), t_error(ent));
1252 }
1253 }
1254
1255 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1256 {
1257 struct blk_io_trace_remap r = { .device_from = 0, };
1258
1259 get_pdu_remap(ent, &r);
1260 trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1261 t_sector(ent), t_sec(ent),
1262 MAJOR(r.device_from), MINOR(r.device_from),
1263 (unsigned long long)r.sector_from);
1264 }
1265
1266 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1267 {
1268 char cmd[TASK_COMM_LEN];
1269
1270 trace_find_cmdline(ent->pid, cmd);
1271
1272 trace_seq_printf(s, "[%s]\n", cmd);
1273 }
1274
1275 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1276 {
1277 char cmd[TASK_COMM_LEN];
1278
1279 trace_find_cmdline(ent->pid, cmd);
1280
1281 trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1282 }
1283
1284 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1285 {
1286 char cmd[TASK_COMM_LEN];
1287
1288 trace_find_cmdline(ent->pid, cmd);
1289
1290 trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1291 get_pdu_int(ent), cmd);
1292 }
1293
1294 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1295 {
1296 const struct blk_io_trace *t = te_blk_io_trace(ent);
1297
1298 trace_seq_putmem(s, t + 1, t->pdu_len);
1299 trace_seq_putc(s, '\n');
1300 }
1301
1302 /*
1303 * struct tracer operations
1304 */
1305
1306 static void blk_tracer_print_header(struct seq_file *m)
1307 {
1308 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1309 return;
1310 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1311 "# | | | | | |\n");
1312 }
1313
1314 static void blk_tracer_start(struct trace_array *tr)
1315 {
1316 blk_tracer_enabled = true;
1317 }
1318
1319 static int blk_tracer_init(struct trace_array *tr)
1320 {
1321 blk_tr = tr;
1322 blk_tracer_start(tr);
1323 return 0;
1324 }
1325
1326 static void blk_tracer_stop(struct trace_array *tr)
1327 {
1328 blk_tracer_enabled = false;
1329 }
1330
1331 static void blk_tracer_reset(struct trace_array *tr)
1332 {
1333 blk_tracer_stop(tr);
1334 }
1335
1336 static const struct {
1337 const char *act[2];
1338 void (*print)(struct trace_seq *s, const struct trace_entry *ent);
1339 } what2act[] = {
1340 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1341 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1342 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1343 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1344 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1345 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1346 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1347 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1348 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1349 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1350 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1351 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1352 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1353 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1354 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1355 };
1356
1357 static enum print_line_t print_one_line(struct trace_iterator *iter,
1358 bool classic)
1359 {
1360 struct trace_array *tr = iter->tr;
1361 struct trace_seq *s = &iter->seq;
1362 const struct blk_io_trace *t;
1363 u16 what;
1364 bool long_act;
1365 blk_log_action_t *log_action;
1366
1367 t = te_blk_io_trace(iter->ent);
1368 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1369 long_act = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1370 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1371
1372 if (t->action == BLK_TN_MESSAGE) {
1373 log_action(iter, long_act ? "message" : "m");
1374 blk_log_msg(s, iter->ent);
1375 return trace_handle_return(s);
1376 }
1377
1378 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1379 trace_seq_printf(s, "Unknown action %x\n", what);
1380 else {
1381 log_action(iter, what2act[what].act[long_act]);
1382 what2act[what].print(s, iter->ent);
1383 }
1384
1385 return trace_handle_return(s);
1386 }
1387
1388 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1389 int flags, struct trace_event *event)
1390 {
1391 return print_one_line(iter, false);
1392 }
1393
1394 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1395 {
1396 struct trace_seq *s = &iter->seq;
1397 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1398 const int offset = offsetof(struct blk_io_trace, sector);
1399 struct blk_io_trace old = {
1400 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1401 .time = iter->ts,
1402 };
1403
1404 trace_seq_putmem(s, &old, offset);
1405 trace_seq_putmem(s, &t->sector,
1406 sizeof(old) - offset + t->pdu_len);
1407 }
1408
1409 static enum print_line_t
1410 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1411 struct trace_event *event)
1412 {
1413 blk_trace_synthesize_old_trace(iter);
1414
1415 return trace_handle_return(&iter->seq);
1416 }
1417
1418 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1419 {
1420 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1421 return TRACE_TYPE_UNHANDLED;
1422
1423 return print_one_line(iter, true);
1424 }
1425
1426 static int
1427 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1428 {
1429 /* don't output context-info for blk_classic output */
1430 if (bit == TRACE_BLK_OPT_CLASSIC) {
1431 if (set)
1432 tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1433 else
1434 tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1435 }
1436 return 0;
1437 }
1438
1439 static struct tracer blk_tracer __read_mostly = {
1440 .name = "blk",
1441 .init = blk_tracer_init,
1442 .reset = blk_tracer_reset,
1443 .start = blk_tracer_start,
1444 .stop = blk_tracer_stop,
1445 .print_header = blk_tracer_print_header,
1446 .print_line = blk_tracer_print_line,
1447 .flags = &blk_tracer_flags,
1448 .set_flag = blk_tracer_set_flag,
1449 };
1450
1451 static struct trace_event_functions trace_blk_event_funcs = {
1452 .trace = blk_trace_event_print,
1453 .binary = blk_trace_event_print_binary,
1454 };
1455
1456 static struct trace_event trace_blk_event = {
1457 .type = TRACE_BLK,
1458 .funcs = &trace_blk_event_funcs,
1459 };
1460
1461 static int __init init_blk_tracer(void)
1462 {
1463 if (!register_trace_event(&trace_blk_event)) {
1464 pr_warn("Warning: could not register block events\n");
1465 return 1;
1466 }
1467
1468 if (register_tracer(&blk_tracer) != 0) {
1469 pr_warn("Warning: could not register the block tracer\n");
1470 unregister_trace_event(&trace_blk_event);
1471 return 1;
1472 }
1473
1474 return 0;
1475 }
1476
1477 device_initcall(init_blk_tracer);
1478
1479 static int blk_trace_remove_queue(struct request_queue *q)
1480 {
1481 struct blk_trace *bt;
1482
1483 bt = xchg(&q->blk_trace, NULL);
1484 if (bt == NULL)
1485 return -EINVAL;
1486
1487 put_probe_ref();
1488 blk_trace_free(bt);
1489 return 0;
1490 }
1491
1492 /*
1493 * Setup everything required to start tracing
1494 */
1495 static int blk_trace_setup_queue(struct request_queue *q,
1496 struct block_device *bdev)
1497 {
1498 struct blk_trace *bt = NULL;
1499 int ret = -ENOMEM;
1500
1501 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1502 if (!bt)
1503 return -ENOMEM;
1504
1505 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1506 if (!bt->msg_data)
1507 goto free_bt;
1508
1509 bt->dev = bdev->bd_dev;
1510 bt->act_mask = (u16)-1;
1511
1512 blk_trace_setup_lba(bt, bdev);
1513
1514 ret = -EBUSY;
1515 if (cmpxchg(&q->blk_trace, NULL, bt))
1516 goto free_bt;
1517
1518 get_probe_ref();
1519 return 0;
1520
1521 free_bt:
1522 blk_trace_free(bt);
1523 return ret;
1524 }
1525
1526 /*
1527 * sysfs interface to enable and configure tracing
1528 */
1529
1530 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1531 struct device_attribute *attr,
1532 char *buf);
1533 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1534 struct device_attribute *attr,
1535 const char *buf, size_t count);
1536 #define BLK_TRACE_DEVICE_ATTR(_name) \
1537 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1538 sysfs_blk_trace_attr_show, \
1539 sysfs_blk_trace_attr_store)
1540
1541 static BLK_TRACE_DEVICE_ATTR(enable);
1542 static BLK_TRACE_DEVICE_ATTR(act_mask);
1543 static BLK_TRACE_DEVICE_ATTR(pid);
1544 static BLK_TRACE_DEVICE_ATTR(start_lba);
1545 static BLK_TRACE_DEVICE_ATTR(end_lba);
1546
1547 static struct attribute *blk_trace_attrs[] = {
1548 &dev_attr_enable.attr,
1549 &dev_attr_act_mask.attr,
1550 &dev_attr_pid.attr,
1551 &dev_attr_start_lba.attr,
1552 &dev_attr_end_lba.attr,
1553 NULL
1554 };
1555
1556 struct attribute_group blk_trace_attr_group = {
1557 .name = "trace",
1558 .attrs = blk_trace_attrs,
1559 };
1560
1561 static const struct {
1562 int mask;
1563 const char *str;
1564 } mask_maps[] = {
1565 { BLK_TC_READ, "read" },
1566 { BLK_TC_WRITE, "write" },
1567 { BLK_TC_FLUSH, "flush" },
1568 { BLK_TC_SYNC, "sync" },
1569 { BLK_TC_QUEUE, "queue" },
1570 { BLK_TC_REQUEUE, "requeue" },
1571 { BLK_TC_ISSUE, "issue" },
1572 { BLK_TC_COMPLETE, "complete" },
1573 { BLK_TC_FS, "fs" },
1574 { BLK_TC_PC, "pc" },
1575 { BLK_TC_NOTIFY, "notify" },
1576 { BLK_TC_AHEAD, "ahead" },
1577 { BLK_TC_META, "meta" },
1578 { BLK_TC_DISCARD, "discard" },
1579 { BLK_TC_DRV_DATA, "drv_data" },
1580 { BLK_TC_FUA, "fua" },
1581 };
1582
1583 static int blk_trace_str2mask(const char *str)
1584 {
1585 int i;
1586 int mask = 0;
1587 char *buf, *s, *token;
1588
1589 buf = kstrdup(str, GFP_KERNEL);
1590 if (buf == NULL)
1591 return -ENOMEM;
1592 s = strstrip(buf);
1593
1594 while (1) {
1595 token = strsep(&s, ",");
1596 if (token == NULL)
1597 break;
1598
1599 if (*token == '\0')
1600 continue;
1601
1602 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1603 if (strcasecmp(token, mask_maps[i].str) == 0) {
1604 mask |= mask_maps[i].mask;
1605 break;
1606 }
1607 }
1608 if (i == ARRAY_SIZE(mask_maps)) {
1609 mask = -EINVAL;
1610 break;
1611 }
1612 }
1613 kfree(buf);
1614
1615 return mask;
1616 }
1617
1618 static ssize_t blk_trace_mask2str(char *buf, int mask)
1619 {
1620 int i;
1621 char *p = buf;
1622
1623 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1624 if (mask & mask_maps[i].mask) {
1625 p += sprintf(p, "%s%s",
1626 (p == buf) ? "" : ",", mask_maps[i].str);
1627 }
1628 }
1629 *p++ = '\n';
1630
1631 return p - buf;
1632 }
1633
1634 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1635 {
1636 if (bdev->bd_disk == NULL)
1637 return NULL;
1638
1639 return bdev_get_queue(bdev);
1640 }
1641
1642 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1643 struct device_attribute *attr,
1644 char *buf)
1645 {
1646 struct hd_struct *p = dev_to_part(dev);
1647 struct request_queue *q;
1648 struct block_device *bdev;
1649 ssize_t ret = -ENXIO;
1650
1651 bdev = bdget(part_devt(p));
1652 if (bdev == NULL)
1653 goto out;
1654
1655 q = blk_trace_get_queue(bdev);
1656 if (q == NULL)
1657 goto out_bdput;
1658
1659 mutex_lock(&bdev->bd_mutex);
1660
1661 if (attr == &dev_attr_enable) {
1662 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1663 goto out_unlock_bdev;
1664 }
1665
1666 if (q->blk_trace == NULL)
1667 ret = sprintf(buf, "disabled\n");
1668 else if (attr == &dev_attr_act_mask)
1669 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1670 else if (attr == &dev_attr_pid)
1671 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1672 else if (attr == &dev_attr_start_lba)
1673 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1674 else if (attr == &dev_attr_end_lba)
1675 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1676
1677 out_unlock_bdev:
1678 mutex_unlock(&bdev->bd_mutex);
1679 out_bdput:
1680 bdput(bdev);
1681 out:
1682 return ret;
1683 }
1684
1685 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1686 struct device_attribute *attr,
1687 const char *buf, size_t count)
1688 {
1689 struct block_device *bdev;
1690 struct request_queue *q;
1691 struct hd_struct *p;
1692 u64 value;
1693 ssize_t ret = -EINVAL;
1694
1695 if (count == 0)
1696 goto out;
1697
1698 if (attr == &dev_attr_act_mask) {
1699 if (sscanf(buf, "%llx", &value) != 1) {
1700 /* Assume it is a list of trace category names */
1701 ret = blk_trace_str2mask(buf);
1702 if (ret < 0)
1703 goto out;
1704 value = ret;
1705 }
1706 } else if (sscanf(buf, "%llu", &value) != 1)
1707 goto out;
1708
1709 ret = -ENXIO;
1710
1711 p = dev_to_part(dev);
1712 bdev = bdget(part_devt(p));
1713 if (bdev == NULL)
1714 goto out;
1715
1716 q = blk_trace_get_queue(bdev);
1717 if (q == NULL)
1718 goto out_bdput;
1719
1720 mutex_lock(&bdev->bd_mutex);
1721
1722 if (attr == &dev_attr_enable) {
1723 if (!!value == !!q->blk_trace) {
1724 ret = 0;
1725 goto out_unlock_bdev;
1726 }
1727 if (value)
1728 ret = blk_trace_setup_queue(q, bdev);
1729 else
1730 ret = blk_trace_remove_queue(q);
1731 goto out_unlock_bdev;
1732 }
1733
1734 ret = 0;
1735 if (q->blk_trace == NULL)
1736 ret = blk_trace_setup_queue(q, bdev);
1737
1738 if (ret == 0) {
1739 if (attr == &dev_attr_act_mask)
1740 q->blk_trace->act_mask = value;
1741 else if (attr == &dev_attr_pid)
1742 q->blk_trace->pid = value;
1743 else if (attr == &dev_attr_start_lba)
1744 q->blk_trace->start_lba = value;
1745 else if (attr == &dev_attr_end_lba)
1746 q->blk_trace->end_lba = value;
1747 }
1748
1749 out_unlock_bdev:
1750 mutex_unlock(&bdev->bd_mutex);
1751 out_bdput:
1752 bdput(bdev);
1753 out:
1754 return ret ? ret : count;
1755 }
1756
1757 int blk_trace_init_sysfs(struct device *dev)
1758 {
1759 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1760 }
1761
1762 void blk_trace_remove_sysfs(struct device *dev)
1763 {
1764 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1765 }
1766
1767 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1768
1769 #ifdef CONFIG_EVENT_TRACING
1770
1771 void blk_dump_cmd(char *buf, struct request *rq)
1772 {
1773 int i, end;
1774 int len = rq->cmd_len;
1775 unsigned char *cmd = rq->cmd;
1776
1777 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1778 buf[0] = '\0';
1779 return;
1780 }
1781
1782 for (end = len - 1; end >= 0; end--)
1783 if (cmd[end])
1784 break;
1785 end++;
1786
1787 for (i = 0; i < len; i++) {
1788 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1789 if (i == end && end != len - 1) {
1790 sprintf(buf, " ..");
1791 break;
1792 }
1793 }
1794 }
1795
1796 void blk_fill_rwbs(char *rwbs, int op, u32 rw, int bytes)
1797 {
1798 int i = 0;
1799
1800 if (rw & REQ_PREFLUSH)
1801 rwbs[i++] = 'F';
1802
1803 switch (op) {
1804 case REQ_OP_WRITE:
1805 case REQ_OP_WRITE_SAME:
1806 rwbs[i++] = 'W';
1807 break;
1808 case REQ_OP_DISCARD:
1809 rwbs[i++] = 'D';
1810 break;
1811 case REQ_OP_SECURE_ERASE:
1812 rwbs[i++] = 'D';
1813 rwbs[i++] = 'E';
1814 break;
1815 case REQ_OP_FLUSH:
1816 rwbs[i++] = 'F';
1817 break;
1818 case REQ_OP_READ:
1819 rwbs[i++] = 'R';
1820 break;
1821 default:
1822 rwbs[i++] = 'N';
1823 }
1824
1825 if (rw & REQ_FUA)
1826 rwbs[i++] = 'F';
1827 if (rw & REQ_RAHEAD)
1828 rwbs[i++] = 'A';
1829 if (rw & REQ_SYNC)
1830 rwbs[i++] = 'S';
1831 if (rw & REQ_META)
1832 rwbs[i++] = 'M';
1833
1834 rwbs[i] = '\0';
1835 }
1836 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1837
1838 #endif /* CONFIG_EVENT_TRACING */
1839
Linux with added FPC-III support