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