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