2 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include <linux/kernel.h>
19 #include <linux/blkdev.h>
20 #include <linux/blktrace_api.h>
21 #include <linux/percpu.h>
22 #include <linux/init.h>
23 #include <linux/mutex.h>
24 #include <linux/slab.h>
25 #include <linux/debugfs.h>
26 #include <linux/smp_lock.h>
27 #include <linux/time.h>
28 #include <linux/uaccess.h>
30 #include <trace/events/block.h>
32 #include "trace_output.h"
34 #ifdef CONFIG_BLK_DEV_IO_TRACE
36 static unsigned int blktrace_seq __read_mostly = 1;
38 static struct trace_array *blk_tr;
39 static bool blk_tracer_enabled __read_mostly;
41 /* Select an alternative, minimalistic output than the original one */
42 #define TRACE_BLK_OPT_CLASSIC 0x1
44 static struct tracer_opt blk_tracer_opts[] = {
45 /* Default disable the minimalistic output */
46 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
50 static struct tracer_flags blk_tracer_flags = {
52 .opts = blk_tracer_opts,
55 /* Global reference count of probes */
56 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
58 static void blk_register_tracepoints(void);
59 static void blk_unregister_tracepoints(void);
62 * Send out a notify message.
64 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
65 const void *data, size_t len)
67 struct blk_io_trace *t;
68 struct ring_buffer_event *event = NULL;
69 struct ring_buffer *buffer = NULL;
71 int cpu = smp_processor_id();
72 bool blk_tracer = blk_tracer_enabled;
75 buffer = blk_tr->buffer;
77 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
82 t = ring_buffer_event_data(event);
89 t = relay_reserve(bt->rchan, sizeof(*t) + len);
91 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
92 t->time = ktime_to_ns(ktime_get());
99 memcpy((void *) t + sizeof(*t), data, len);
102 trace_buffer_unlock_commit(buffer, event, 0, pc);
107 * Send out a notify for this process, if we haven't done so since a trace
110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
112 tsk->btrace_seq = blktrace_seq;
113 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
116 static void trace_note_time(struct blk_trace *bt)
122 getnstimeofday(&now);
123 words[0] = now.tv_sec;
124 words[1] = now.tv_nsec;
126 local_irq_save(flags);
127 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
128 local_irq_restore(flags);
131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
138 if (unlikely(bt->trace_state != Blktrace_running &&
139 !blk_tracer_enabled))
142 local_irq_save(flags);
143 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
145 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
148 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
149 local_irq_restore(flags);
151 EXPORT_SYMBOL_GPL(__trace_note_message);
153 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
156 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
158 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
160 if (bt->pid && pid != bt->pid)
167 * Data direction bit lookup
169 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
170 BLK_TC_ACT(BLK_TC_WRITE) };
172 #define BLK_TC_HARDBARRIER BLK_TC_BARRIER
173 #define BLK_TC_RAHEAD BLK_TC_AHEAD
175 /* The ilog2() calls fall out because they're constant */
176 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
177 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
180 * The worker for the various blk_add_trace*() types. Fills out a
181 * blk_io_trace structure and places it in a per-cpu subbuffer.
183 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
184 int rw, u32 what, int error, int pdu_len, void *pdu_data)
186 struct task_struct *tsk = current;
187 struct ring_buffer_event *event = NULL;
188 struct ring_buffer *buffer = NULL;
189 struct blk_io_trace *t;
190 unsigned long flags = 0;
191 unsigned long *sequence;
194 bool blk_tracer = blk_tracer_enabled;
196 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
199 what |= ddir_act[rw & WRITE];
200 what |= MASK_TC_BIT(rw, HARDBARRIER);
201 what |= MASK_TC_BIT(rw, SYNC);
202 what |= MASK_TC_BIT(rw, RAHEAD);
203 what |= MASK_TC_BIT(rw, META);
204 what |= MASK_TC_BIT(rw, DISCARD);
207 if (act_log_check(bt, what, sector, pid))
209 cpu = raw_smp_processor_id();
212 tracing_record_cmdline(current);
214 buffer = blk_tr->buffer;
215 pc = preempt_count();
216 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
217 sizeof(*t) + pdu_len,
221 t = ring_buffer_event_data(event);
226 * A word about the locking here - we disable interrupts to reserve
227 * some space in the relay per-cpu buffer, to prevent an irq
228 * from coming in and stepping on our toes.
230 local_irq_save(flags);
232 if (unlikely(tsk->btrace_seq != blktrace_seq))
233 trace_note_tsk(bt, tsk);
235 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
237 sequence = per_cpu_ptr(bt->sequence, cpu);
239 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
240 t->sequence = ++(*sequence);
241 t->time = ktime_to_ns(ktime_get());
244 * These two are not needed in ftrace as they are in the
245 * generic trace_entry, filled by tracing_generic_entry_update,
246 * but for the trace_event->bin() synthesizer benefit we do it
257 t->pdu_len = pdu_len;
260 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
263 trace_buffer_unlock_commit(buffer, event, 0, pc);
268 local_irq_restore(flags);
271 static struct dentry *blk_tree_root;
272 static DEFINE_MUTEX(blk_tree_mutex);
274 static void blk_trace_free(struct blk_trace *bt)
276 debugfs_remove(bt->msg_file);
277 debugfs_remove(bt->dropped_file);
278 relay_close(bt->rchan);
279 debugfs_remove(bt->dir);
280 free_percpu(bt->sequence);
281 free_percpu(bt->msg_data);
285 static void blk_trace_cleanup(struct blk_trace *bt)
288 if (atomic_dec_and_test(&blk_probes_ref))
289 blk_unregister_tracepoints();
292 int blk_trace_remove(struct request_queue *q)
294 struct blk_trace *bt;
296 bt = xchg(&q->blk_trace, NULL);
300 if (bt->trace_state != Blktrace_running)
301 blk_trace_cleanup(bt);
305 EXPORT_SYMBOL_GPL(blk_trace_remove);
307 static int blk_dropped_open(struct inode *inode, struct file *filp)
309 filp->private_data = inode->i_private;
314 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
315 size_t count, loff_t *ppos)
317 struct blk_trace *bt = filp->private_data;
320 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
322 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
325 static const struct file_operations blk_dropped_fops = {
326 .owner = THIS_MODULE,
327 .open = blk_dropped_open,
328 .read = blk_dropped_read,
331 static int blk_msg_open(struct inode *inode, struct file *filp)
333 filp->private_data = inode->i_private;
338 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
339 size_t count, loff_t *ppos)
342 struct blk_trace *bt;
344 if (count >= BLK_TN_MAX_MSG)
347 msg = kmalloc(count + 1, GFP_KERNEL);
351 if (copy_from_user(msg, buffer, count)) {
357 bt = filp->private_data;
358 __trace_note_message(bt, "%s", msg);
364 static const struct file_operations blk_msg_fops = {
365 .owner = THIS_MODULE,
366 .open = blk_msg_open,
367 .write = blk_msg_write,
371 * Keep track of how many times we encountered a full subbuffer, to aid
372 * the user space app in telling how many lost events there were.
374 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
375 void *prev_subbuf, size_t prev_padding)
377 struct blk_trace *bt;
379 if (!relay_buf_full(buf))
382 bt = buf->chan->private_data;
383 atomic_inc(&bt->dropped);
387 static int blk_remove_buf_file_callback(struct dentry *dentry)
389 debugfs_remove(dentry);
394 static struct dentry *blk_create_buf_file_callback(const char *filename,
395 struct dentry *parent,
397 struct rchan_buf *buf,
400 return debugfs_create_file(filename, mode, parent, buf,
401 &relay_file_operations);
404 static struct rchan_callbacks blk_relay_callbacks = {
405 .subbuf_start = blk_subbuf_start_callback,
406 .create_buf_file = blk_create_buf_file_callback,
407 .remove_buf_file = blk_remove_buf_file_callback,
410 static void blk_trace_setup_lba(struct blk_trace *bt,
411 struct block_device *bdev)
413 struct hd_struct *part = NULL;
416 part = bdev->bd_part;
419 bt->start_lba = part->start_sect;
420 bt->end_lba = part->start_sect + part->nr_sects;
428 * Setup everything required to start tracing
430 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
431 struct block_device *bdev,
432 struct blk_user_trace_setup *buts)
434 struct blk_trace *old_bt, *bt = NULL;
435 struct dentry *dir = NULL;
438 if (!buts->buf_size || !buts->buf_nr)
441 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
442 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
445 * some device names have larger paths - convert the slashes
446 * to underscores for this to work as expected
448 for (i = 0; i < strlen(buts->name); i++)
449 if (buts->name[i] == '/')
452 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
457 bt->sequence = alloc_percpu(unsigned long);
461 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
467 mutex_lock(&blk_tree_mutex);
468 if (!blk_tree_root) {
469 blk_tree_root = debugfs_create_dir("block", NULL);
470 if (!blk_tree_root) {
471 mutex_unlock(&blk_tree_mutex);
475 mutex_unlock(&blk_tree_mutex);
477 dir = debugfs_create_dir(buts->name, blk_tree_root);
484 atomic_set(&bt->dropped, 0);
487 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
489 if (!bt->dropped_file)
492 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
496 bt->rchan = relay_open("trace", dir, buts->buf_size,
497 buts->buf_nr, &blk_relay_callbacks, bt);
501 bt->act_mask = buts->act_mask;
503 bt->act_mask = (u16) -1;
505 blk_trace_setup_lba(bt, bdev);
507 /* overwrite with user settings */
509 bt->start_lba = buts->start_lba;
511 bt->end_lba = buts->end_lba;
514 bt->trace_state = Blktrace_setup;
517 old_bt = xchg(&q->blk_trace, bt);
519 (void) xchg(&q->blk_trace, old_bt);
523 if (atomic_inc_return(&blk_probes_ref) == 1)
524 blk_register_tracepoints();
532 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
533 struct block_device *bdev,
536 struct blk_user_trace_setup buts;
539 ret = copy_from_user(&buts, arg, sizeof(buts));
543 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
547 if (copy_to_user(arg, &buts, sizeof(buts))) {
553 EXPORT_SYMBOL_GPL(blk_trace_setup);
555 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
556 static int compat_blk_trace_setup(struct request_queue *q, char *name,
557 dev_t dev, struct block_device *bdev,
560 struct blk_user_trace_setup buts;
561 struct compat_blk_user_trace_setup cbuts;
564 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
567 buts = (struct blk_user_trace_setup) {
568 .act_mask = cbuts.act_mask,
569 .buf_size = cbuts.buf_size,
570 .buf_nr = cbuts.buf_nr,
571 .start_lba = cbuts.start_lba,
572 .end_lba = cbuts.end_lba,
575 memcpy(&buts.name, &cbuts.name, 32);
577 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
581 if (copy_to_user(arg, &buts.name, 32)) {
590 int blk_trace_startstop(struct request_queue *q, int start)
593 struct blk_trace *bt = q->blk_trace;
599 * For starting a trace, we can transition from a setup or stopped
600 * trace. For stopping a trace, the state must be running
604 if (bt->trace_state == Blktrace_setup ||
605 bt->trace_state == Blktrace_stopped) {
608 bt->trace_state = Blktrace_running;
614 if (bt->trace_state == Blktrace_running) {
615 bt->trace_state = Blktrace_stopped;
616 relay_flush(bt->rchan);
623 EXPORT_SYMBOL_GPL(blk_trace_startstop);
626 * blk_trace_ioctl: - handle the ioctls associated with tracing
627 * @bdev: the block device
628 * @cmd: the ioctl cmd
629 * @arg: the argument data, if any
632 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
634 struct request_queue *q;
636 char b[BDEVNAME_SIZE];
638 q = bdev_get_queue(bdev);
643 mutex_lock(&bdev->bd_mutex);
648 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
650 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
651 case BLKTRACESETUP32:
653 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
659 ret = blk_trace_startstop(q, start);
661 case BLKTRACETEARDOWN:
662 ret = blk_trace_remove(q);
669 mutex_unlock(&bdev->bd_mutex);
675 * blk_trace_shutdown: - stop and cleanup trace structures
676 * @q: the request queue associated with the device
679 void blk_trace_shutdown(struct request_queue *q)
682 blk_trace_startstop(q, 0);
692 * blk_add_trace_rq - Add a trace for a request oriented action
693 * @q: queue the io is for
694 * @rq: the source request
698 * Records an action against a request. Will log the bio offset + size.
701 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
704 struct blk_trace *bt = q->blk_trace;
705 int rw = rq->cmd_flags & 0x03;
710 if (rq->cmd_flags & REQ_DISCARD)
713 if (rq->cmd_flags & REQ_SECURE)
716 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
717 what |= BLK_TC_ACT(BLK_TC_PC);
718 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
719 what, rq->errors, rq->cmd_len, rq->cmd);
721 what |= BLK_TC_ACT(BLK_TC_FS);
722 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
723 what, rq->errors, 0, NULL);
727 static void blk_add_trace_rq_abort(void *ignore,
728 struct request_queue *q, struct request *rq)
730 blk_add_trace_rq(q, rq, BLK_TA_ABORT);
733 static void blk_add_trace_rq_insert(void *ignore,
734 struct request_queue *q, struct request *rq)
736 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
739 static void blk_add_trace_rq_issue(void *ignore,
740 struct request_queue *q, struct request *rq)
742 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
745 static void blk_add_trace_rq_requeue(void *ignore,
746 struct request_queue *q,
749 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
752 static void blk_add_trace_rq_complete(void *ignore,
753 struct request_queue *q,
756 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
760 * blk_add_trace_bio - Add a trace for a bio oriented action
761 * @q: queue the io is for
762 * @bio: the source bio
766 * Records an action against a bio. Will log the bio offset + size.
769 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
772 struct blk_trace *bt = q->blk_trace;
777 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
778 !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
781 static void blk_add_trace_bio_bounce(void *ignore,
782 struct request_queue *q, struct bio *bio)
784 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
787 static void blk_add_trace_bio_complete(void *ignore,
788 struct request_queue *q, struct bio *bio)
790 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
793 static void blk_add_trace_bio_backmerge(void *ignore,
794 struct request_queue *q,
797 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
800 static void blk_add_trace_bio_frontmerge(void *ignore,
801 struct request_queue *q,
804 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
807 static void blk_add_trace_bio_queue(void *ignore,
808 struct request_queue *q, struct bio *bio)
810 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
813 static void blk_add_trace_getrq(void *ignore,
814 struct request_queue *q,
815 struct bio *bio, int rw)
818 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
820 struct blk_trace *bt = q->blk_trace;
823 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
828 static void blk_add_trace_sleeprq(void *ignore,
829 struct request_queue *q,
830 struct bio *bio, int rw)
833 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
835 struct blk_trace *bt = q->blk_trace;
838 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
843 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
845 struct blk_trace *bt = q->blk_trace;
848 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
851 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
853 struct blk_trace *bt = q->blk_trace;
856 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
857 __be64 rpdu = cpu_to_be64(pdu);
859 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
860 sizeof(rpdu), &rpdu);
864 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q)
866 struct blk_trace *bt = q->blk_trace;
869 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
870 __be64 rpdu = cpu_to_be64(pdu);
872 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
873 sizeof(rpdu), &rpdu);
877 static void blk_add_trace_split(void *ignore,
878 struct request_queue *q, struct bio *bio,
881 struct blk_trace *bt = q->blk_trace;
884 __be64 rpdu = cpu_to_be64(pdu);
886 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
887 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
888 sizeof(rpdu), &rpdu);
893 * blk_add_trace_remap - Add a trace for a remap operation
894 * @ignore: trace callback data parameter (not used)
895 * @q: queue the io is for
896 * @bio: the source bio
897 * @dev: target device
898 * @from: source sector
901 * Device mapper or raid target sometimes need to split a bio because
902 * it spans a stripe (or similar). Add a trace for that action.
905 static void blk_add_trace_remap(void *ignore,
906 struct request_queue *q, struct bio *bio,
907 dev_t dev, sector_t from)
909 struct blk_trace *bt = q->blk_trace;
910 struct blk_io_trace_remap r;
915 r.device_from = cpu_to_be32(dev);
916 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
917 r.sector_from = cpu_to_be64(from);
919 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
920 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
925 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
926 * @ignore: trace callback data parameter (not used)
927 * @q: queue the io is for
928 * @rq: the source request
929 * @dev: target device
930 * @from: source sector
933 * Device mapper remaps request to other devices.
934 * Add a trace for that action.
937 static void blk_add_trace_rq_remap(void *ignore,
938 struct request_queue *q,
939 struct request *rq, dev_t dev,
942 struct blk_trace *bt = q->blk_trace;
943 struct blk_io_trace_remap r;
948 r.device_from = cpu_to_be32(dev);
949 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
950 r.sector_from = cpu_to_be64(from);
952 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
953 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
958 * blk_add_driver_data - Add binary message with driver-specific data
959 * @q: queue the io is for
961 * @data: driver-specific data
962 * @len: length of driver-specific data
965 * Some drivers might want to write driver-specific data per request.
968 void blk_add_driver_data(struct request_queue *q,
970 void *data, size_t len)
972 struct blk_trace *bt = q->blk_trace;
977 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
978 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
979 BLK_TA_DRV_DATA, rq->errors, len, data);
981 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
982 BLK_TA_DRV_DATA, rq->errors, len, data);
984 EXPORT_SYMBOL_GPL(blk_add_driver_data);
986 static void blk_register_tracepoints(void)
990 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
992 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
994 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
996 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
998 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1000 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1002 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1004 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1006 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1008 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1010 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1012 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1014 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1016 ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1018 ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1020 ret = register_trace_block_split(blk_add_trace_split, NULL);
1022 ret = register_trace_block_remap(blk_add_trace_remap, NULL);
1024 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1028 static void blk_unregister_tracepoints(void)
1030 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1031 unregister_trace_block_remap(blk_add_trace_remap, NULL);
1032 unregister_trace_block_split(blk_add_trace_split, NULL);
1033 unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1034 unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1035 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1036 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1037 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1038 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1039 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1040 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1041 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1042 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1043 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1044 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1045 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1046 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1047 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1049 tracepoint_synchronize_unregister();
1053 * struct blk_io_tracer formatting routines
1056 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1059 int tc = t->action >> BLK_TC_SHIFT;
1061 if (t->action == BLK_TN_MESSAGE) {
1066 if (tc & BLK_TC_DISCARD)
1068 else if (tc & BLK_TC_WRITE)
1075 if (tc & BLK_TC_AHEAD)
1077 if (tc & BLK_TC_BARRIER)
1079 if (tc & BLK_TC_SYNC)
1081 if (tc & BLK_TC_META)
1088 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1090 return (const struct blk_io_trace *)ent;
1093 static inline const void *pdu_start(const struct trace_entry *ent)
1095 return te_blk_io_trace(ent) + 1;
1098 static inline u32 t_action(const struct trace_entry *ent)
1100 return te_blk_io_trace(ent)->action;
1103 static inline u32 t_bytes(const struct trace_entry *ent)
1105 return te_blk_io_trace(ent)->bytes;
1108 static inline u32 t_sec(const struct trace_entry *ent)
1110 return te_blk_io_trace(ent)->bytes >> 9;
1113 static inline unsigned long long t_sector(const struct trace_entry *ent)
1115 return te_blk_io_trace(ent)->sector;
1118 static inline __u16 t_error(const struct trace_entry *ent)
1120 return te_blk_io_trace(ent)->error;
1123 static __u64 get_pdu_int(const struct trace_entry *ent)
1125 const __u64 *val = pdu_start(ent);
1126 return be64_to_cpu(*val);
1129 static void get_pdu_remap(const struct trace_entry *ent,
1130 struct blk_io_trace_remap *r)
1132 const struct blk_io_trace_remap *__r = pdu_start(ent);
1133 __u64 sector_from = __r->sector_from;
1135 r->device_from = be32_to_cpu(__r->device_from);
1136 r->device_to = be32_to_cpu(__r->device_to);
1137 r->sector_from = be64_to_cpu(sector_from);
1140 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1142 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1145 unsigned long long ts = iter->ts;
1146 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1147 unsigned secs = (unsigned long)ts;
1148 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1152 return trace_seq_printf(&iter->seq,
1153 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1154 MAJOR(t->device), MINOR(t->device), iter->cpu,
1155 secs, nsec_rem, iter->ent->pid, act, rwbs);
1158 static int blk_log_action(struct trace_iterator *iter, const char *act)
1161 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1164 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1165 MAJOR(t->device), MINOR(t->device), act, rwbs);
1168 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1170 const unsigned char *pdu_buf;
1174 pdu_buf = pdu_start(ent);
1175 pdu_len = te_blk_io_trace(ent)->pdu_len;
1180 /* find the last zero that needs to be printed */
1181 for (end = pdu_len - 1; end >= 0; end--)
1186 if (!trace_seq_putc(s, '('))
1189 for (i = 0; i < pdu_len; i++) {
1191 ret = trace_seq_printf(s, "%s%02x",
1192 i == 0 ? "" : " ", pdu_buf[i]);
1197 * stop when the rest is just zeroes and indicate so
1198 * with a ".." appended
1200 if (i == end && end != pdu_len - 1)
1201 return trace_seq_puts(s, " ..) ");
1204 return trace_seq_puts(s, ") ");
1207 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1209 char cmd[TASK_COMM_LEN];
1211 trace_find_cmdline(ent->pid, cmd);
1213 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1216 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1219 ret = blk_log_dump_pdu(s, ent);
1222 return trace_seq_printf(s, "[%s]\n", cmd);
1225 return trace_seq_printf(s, "%llu + %u [%s]\n",
1226 t_sector(ent), t_sec(ent), cmd);
1227 return trace_seq_printf(s, "[%s]\n", cmd);
1231 static int blk_log_with_error(struct trace_seq *s,
1232 const struct trace_entry *ent)
1234 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1237 ret = blk_log_dump_pdu(s, ent);
1239 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1243 return trace_seq_printf(s, "%llu + %u [%d]\n",
1245 t_sec(ent), t_error(ent));
1246 return trace_seq_printf(s, "%llu [%d]\n",
1247 t_sector(ent), t_error(ent));
1251 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1253 struct blk_io_trace_remap r = { .device_from = 0, };
1255 get_pdu_remap(ent, &r);
1256 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1257 t_sector(ent), t_sec(ent),
1258 MAJOR(r.device_from), MINOR(r.device_from),
1259 (unsigned long long)r.sector_from);
1262 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1264 char cmd[TASK_COMM_LEN];
1266 trace_find_cmdline(ent->pid, cmd);
1268 return trace_seq_printf(s, "[%s]\n", cmd);
1271 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1273 char cmd[TASK_COMM_LEN];
1275 trace_find_cmdline(ent->pid, cmd);
1277 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1280 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1282 char cmd[TASK_COMM_LEN];
1284 trace_find_cmdline(ent->pid, cmd);
1286 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1287 get_pdu_int(ent), cmd);
1290 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1293 const struct blk_io_trace *t = te_blk_io_trace(ent);
1295 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1297 return trace_seq_putc(s, '\n');
1302 * struct tracer operations
1305 static void blk_tracer_print_header(struct seq_file *m)
1307 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1309 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1313 static void blk_tracer_start(struct trace_array *tr)
1315 blk_tracer_enabled = true;
1318 static int blk_tracer_init(struct trace_array *tr)
1321 blk_tracer_start(tr);
1325 static void blk_tracer_stop(struct trace_array *tr)
1327 blk_tracer_enabled = false;
1330 static void blk_tracer_reset(struct trace_array *tr)
1332 blk_tracer_stop(tr);
1335 static const struct {
1337 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1339 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1340 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1341 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1342 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1343 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1344 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1345 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1346 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1347 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1348 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1349 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1350 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1351 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1352 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1353 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1356 static enum print_line_t print_one_line(struct trace_iterator *iter,
1359 struct trace_seq *s = &iter->seq;
1360 const struct blk_io_trace *t;
1364 blk_log_action_t *log_action;
1366 t = te_blk_io_trace(iter->ent);
1367 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1368 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1369 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1371 if (t->action == BLK_TN_MESSAGE) {
1372 ret = log_action(iter, long_act ? "message" : "m");
1374 ret = blk_log_msg(s, iter->ent);
1378 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1379 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1381 ret = log_action(iter, what2act[what].act[long_act]);
1383 ret = what2act[what].print(s, iter->ent);
1386 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1389 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1390 int flags, struct trace_event *event)
1392 return print_one_line(iter, false);
1395 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1397 struct trace_seq *s = &iter->seq;
1398 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1399 const int offset = offsetof(struct blk_io_trace, sector);
1400 struct blk_io_trace old = {
1401 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1405 if (!trace_seq_putmem(s, &old, offset))
1407 return trace_seq_putmem(s, &t->sector,
1408 sizeof(old) - offset + t->pdu_len);
1411 static enum print_line_t
1412 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1413 struct trace_event *event)
1415 return blk_trace_synthesize_old_trace(iter) ?
1416 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1419 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1421 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1422 return TRACE_TYPE_UNHANDLED;
1424 return print_one_line(iter, true);
1427 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1429 /* don't output context-info for blk_classic output */
1430 if (bit == TRACE_BLK_OPT_CLASSIC) {
1432 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1434 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1439 static struct tracer blk_tracer __read_mostly = {
1441 .init = blk_tracer_init,
1442 .reset = blk_tracer_reset,
1443 .start = blk_tracer_start,
1444 .stop = blk_tracer_stop,
1445 .print_header = blk_tracer_print_header,
1446 .print_line = blk_tracer_print_line,
1447 .flags = &blk_tracer_flags,
1448 .set_flag = blk_tracer_set_flag,
1451 static struct trace_event_functions trace_blk_event_funcs = {
1452 .trace = blk_trace_event_print,
1453 .binary = blk_trace_event_print_binary,
1456 static struct trace_event trace_blk_event = {
1458 .funcs = &trace_blk_event_funcs,
1461 static int __init init_blk_tracer(void)
1463 if (!register_ftrace_event(&trace_blk_event)) {
1464 pr_warning("Warning: could not register block events\n");
1468 if (register_tracer(&blk_tracer) != 0) {
1469 pr_warning("Warning: could not register the block tracer\n");
1470 unregister_ftrace_event(&trace_blk_event);
1477 device_initcall(init_blk_tracer);
1479 static int blk_trace_remove_queue(struct request_queue *q)
1481 struct blk_trace *bt;
1483 bt = xchg(&q->blk_trace, NULL);
1487 if (atomic_dec_and_test(&blk_probes_ref))
1488 blk_unregister_tracepoints();
1495 * Setup everything required to start tracing
1497 static int blk_trace_setup_queue(struct request_queue *q,
1498 struct block_device *bdev)
1500 struct blk_trace *old_bt, *bt = NULL;
1503 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1507 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1511 bt->dev = bdev->bd_dev;
1512 bt->act_mask = (u16)-1;
1514 blk_trace_setup_lba(bt, bdev);
1516 old_bt = xchg(&q->blk_trace, bt);
1517 if (old_bt != NULL) {
1518 (void)xchg(&q->blk_trace, old_bt);
1523 if (atomic_inc_return(&blk_probes_ref) == 1)
1524 blk_register_tracepoints();
1533 * sysfs interface to enable and configure tracing
1536 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1537 struct device_attribute *attr,
1539 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1540 struct device_attribute *attr,
1541 const char *buf, size_t count);
1542 #define BLK_TRACE_DEVICE_ATTR(_name) \
1543 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1544 sysfs_blk_trace_attr_show, \
1545 sysfs_blk_trace_attr_store)
1547 static BLK_TRACE_DEVICE_ATTR(enable);
1548 static BLK_TRACE_DEVICE_ATTR(act_mask);
1549 static BLK_TRACE_DEVICE_ATTR(pid);
1550 static BLK_TRACE_DEVICE_ATTR(start_lba);
1551 static BLK_TRACE_DEVICE_ATTR(end_lba);
1553 static struct attribute *blk_trace_attrs[] = {
1554 &dev_attr_enable.attr,
1555 &dev_attr_act_mask.attr,
1557 &dev_attr_start_lba.attr,
1558 &dev_attr_end_lba.attr,
1562 struct attribute_group blk_trace_attr_group = {
1564 .attrs = blk_trace_attrs,
1567 static const struct {
1571 { BLK_TC_READ, "read" },
1572 { BLK_TC_WRITE, "write" },
1573 { BLK_TC_BARRIER, "barrier" },
1574 { BLK_TC_SYNC, "sync" },
1575 { BLK_TC_QUEUE, "queue" },
1576 { BLK_TC_REQUEUE, "requeue" },
1577 { BLK_TC_ISSUE, "issue" },
1578 { BLK_TC_COMPLETE, "complete" },
1579 { BLK_TC_FS, "fs" },
1580 { BLK_TC_PC, "pc" },
1581 { BLK_TC_AHEAD, "ahead" },
1582 { BLK_TC_META, "meta" },
1583 { BLK_TC_DISCARD, "discard" },
1584 { BLK_TC_DRV_DATA, "drv_data" },
1587 static int blk_trace_str2mask(const char *str)
1591 char *buf, *s, *token;
1593 buf = kstrdup(str, GFP_KERNEL);
1599 token = strsep(&s, ",");
1606 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1607 if (strcasecmp(token, mask_maps[i].str) == 0) {
1608 mask |= mask_maps[i].mask;
1612 if (i == ARRAY_SIZE(mask_maps)) {
1622 static ssize_t blk_trace_mask2str(char *buf, int mask)
1627 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1628 if (mask & mask_maps[i].mask) {
1629 p += sprintf(p, "%s%s",
1630 (p == buf) ? "" : ",", mask_maps[i].str);
1638 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1640 if (bdev->bd_disk == NULL)
1643 return bdev_get_queue(bdev);
1646 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1647 struct device_attribute *attr,
1650 struct hd_struct *p = dev_to_part(dev);
1651 struct request_queue *q;
1652 struct block_device *bdev;
1653 ssize_t ret = -ENXIO;
1656 bdev = bdget(part_devt(p));
1658 goto out_unlock_kernel;
1660 q = blk_trace_get_queue(bdev);
1664 mutex_lock(&bdev->bd_mutex);
1666 if (attr == &dev_attr_enable) {
1667 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1668 goto out_unlock_bdev;
1671 if (q->blk_trace == NULL)
1672 ret = sprintf(buf, "disabled\n");
1673 else if (attr == &dev_attr_act_mask)
1674 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1675 else if (attr == &dev_attr_pid)
1676 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1677 else if (attr == &dev_attr_start_lba)
1678 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1679 else if (attr == &dev_attr_end_lba)
1680 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1683 mutex_unlock(&bdev->bd_mutex);
1691 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1692 struct device_attribute *attr,
1693 const char *buf, size_t count)
1695 struct block_device *bdev;
1696 struct request_queue *q;
1697 struct hd_struct *p;
1699 ssize_t ret = -EINVAL;
1704 if (attr == &dev_attr_act_mask) {
1705 if (sscanf(buf, "%llx", &value) != 1) {
1706 /* Assume it is a list of trace category names */
1707 ret = blk_trace_str2mask(buf);
1712 } else if (sscanf(buf, "%llu", &value) != 1)
1718 p = dev_to_part(dev);
1719 bdev = bdget(part_devt(p));
1721 goto out_unlock_kernel;
1723 q = blk_trace_get_queue(bdev);
1727 mutex_lock(&bdev->bd_mutex);
1729 if (attr == &dev_attr_enable) {
1731 ret = blk_trace_setup_queue(q, bdev);
1733 ret = blk_trace_remove_queue(q);
1734 goto out_unlock_bdev;
1738 if (q->blk_trace == NULL)
1739 ret = blk_trace_setup_queue(q, bdev);
1742 if (attr == &dev_attr_act_mask)
1743 q->blk_trace->act_mask = value;
1744 else if (attr == &dev_attr_pid)
1745 q->blk_trace->pid = value;
1746 else if (attr == &dev_attr_start_lba)
1747 q->blk_trace->start_lba = value;
1748 else if (attr == &dev_attr_end_lba)
1749 q->blk_trace->end_lba = value;
1753 mutex_unlock(&bdev->bd_mutex);
1759 return ret ? ret : count;
1762 int blk_trace_init_sysfs(struct device *dev)
1764 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1767 void blk_trace_remove_sysfs(struct device *dev)
1769 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1772 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1774 #ifdef CONFIG_EVENT_TRACING
1776 void blk_dump_cmd(char *buf, struct request *rq)
1779 int len = rq->cmd_len;
1780 unsigned char *cmd = rq->cmd;
1782 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1787 for (end = len - 1; end >= 0; end--)
1792 for (i = 0; i < len; i++) {
1793 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1794 if (i == end && end != len - 1) {
1795 sprintf(buf, " ..");
1801 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1807 else if (rw & REQ_DISCARD)
1814 if (rw & REQ_RAHEAD)
1816 if (rw & REQ_HARDBARRIER)
1822 if (rw & REQ_SECURE)
1828 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1830 int rw = rq->cmd_flags & 0x03;
1833 if (rq->cmd_flags & REQ_DISCARD)
1836 if (rq->cmd_flags & REQ_SECURE)
1839 bytes = blk_rq_bytes(rq);
1841 blk_fill_rwbs(rwbs, rw, bytes);
1844 #endif /* CONFIG_EVENT_TRACING */