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/time.h>
27 #include <linux/uaccess.h>
29 #include <trace/events/block.h>
31 #include "trace_output.h"
33 #ifdef CONFIG_BLK_DEV_IO_TRACE
35 static unsigned int blktrace_seq __read_mostly = 1;
37 static struct trace_array *blk_tr;
38 static bool blk_tracer_enabled __read_mostly;
40 /* Select an alternative, minimalistic output than the original one */
41 #define TRACE_BLK_OPT_CLASSIC 0x1
43 static struct tracer_opt blk_tracer_opts[] = {
44 /* Default disable the minimalistic output */
45 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
49 static struct tracer_flags blk_tracer_flags = {
51 .opts = blk_tracer_opts,
54 /* Global reference count of probes */
55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
57 static void blk_register_tracepoints(void);
58 static void blk_unregister_tracepoints(void);
61 * Send out a notify message.
63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
64 const void *data, size_t len)
66 struct blk_io_trace *t;
67 struct ring_buffer_event *event = NULL;
68 struct ring_buffer *buffer = NULL;
70 int cpu = smp_processor_id();
71 bool blk_tracer = blk_tracer_enabled;
74 buffer = blk_tr->buffer;
76 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
81 t = ring_buffer_event_data(event);
88 t = relay_reserve(bt->rchan, sizeof(*t) + len);
90 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
91 t->time = ktime_to_ns(ktime_get());
98 memcpy((void *) t + sizeof(*t), data, len);
101 trace_buffer_unlock_commit(buffer, event, 0, pc);
106 * Send out a notify for this process, if we haven't done so since a trace
109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
111 tsk->btrace_seq = blktrace_seq;
112 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
115 static void trace_note_time(struct blk_trace *bt)
121 getnstimeofday(&now);
122 words[0] = now.tv_sec;
123 words[1] = now.tv_nsec;
125 local_irq_save(flags);
126 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
127 local_irq_restore(flags);
130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
137 if (unlikely(bt->trace_state != Blktrace_running &&
138 !blk_tracer_enabled))
142 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
143 * message to the trace.
145 if (!(bt->act_mask & BLK_TC_NOTIFY))
148 local_irq_save(flags);
149 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
151 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
154 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
155 local_irq_restore(flags);
157 EXPORT_SYMBOL_GPL(__trace_note_message);
159 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
162 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
164 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
166 if (bt->pid && pid != bt->pid)
173 * Data direction bit lookup
175 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
176 BLK_TC_ACT(BLK_TC_WRITE) };
178 #define BLK_TC_RAHEAD BLK_TC_AHEAD
180 /* The ilog2() calls fall out because they're constant */
181 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
182 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
185 * The worker for the various blk_add_trace*() types. Fills out a
186 * blk_io_trace structure and places it in a per-cpu subbuffer.
188 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
189 int rw, u32 what, int error, int pdu_len, void *pdu_data)
191 struct task_struct *tsk = current;
192 struct ring_buffer_event *event = NULL;
193 struct ring_buffer *buffer = NULL;
194 struct blk_io_trace *t;
195 unsigned long flags = 0;
196 unsigned long *sequence;
199 bool blk_tracer = blk_tracer_enabled;
201 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
204 what |= ddir_act[rw & WRITE];
205 what |= MASK_TC_BIT(rw, SYNC);
206 what |= MASK_TC_BIT(rw, RAHEAD);
207 what |= MASK_TC_BIT(rw, META);
208 what |= MASK_TC_BIT(rw, DISCARD);
209 what |= MASK_TC_BIT(rw, FLUSH);
210 what |= MASK_TC_BIT(rw, FUA);
213 if (act_log_check(bt, what, sector, pid))
215 cpu = raw_smp_processor_id();
218 tracing_record_cmdline(current);
220 buffer = blk_tr->buffer;
221 pc = preempt_count();
222 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
223 sizeof(*t) + pdu_len,
227 t = ring_buffer_event_data(event);
232 * A word about the locking here - we disable interrupts to reserve
233 * some space in the relay per-cpu buffer, to prevent an irq
234 * from coming in and stepping on our toes.
236 local_irq_save(flags);
238 if (unlikely(tsk->btrace_seq != blktrace_seq))
239 trace_note_tsk(bt, tsk);
241 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
243 sequence = per_cpu_ptr(bt->sequence, cpu);
245 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
246 t->sequence = ++(*sequence);
247 t->time = ktime_to_ns(ktime_get());
250 * These two are not needed in ftrace as they are in the
251 * generic trace_entry, filled by tracing_generic_entry_update,
252 * but for the trace_event->bin() synthesizer benefit we do it
263 t->pdu_len = pdu_len;
266 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
269 trace_buffer_unlock_commit(buffer, event, 0, pc);
274 local_irq_restore(flags);
277 static struct dentry *blk_tree_root;
278 static DEFINE_MUTEX(blk_tree_mutex);
280 static void blk_trace_free(struct blk_trace *bt)
282 debugfs_remove(bt->msg_file);
283 debugfs_remove(bt->dropped_file);
284 relay_close(bt->rchan);
285 debugfs_remove(bt->dir);
286 free_percpu(bt->sequence);
287 free_percpu(bt->msg_data);
291 static void blk_trace_cleanup(struct blk_trace *bt)
294 if (atomic_dec_and_test(&blk_probes_ref))
295 blk_unregister_tracepoints();
298 int blk_trace_remove(struct request_queue *q)
300 struct blk_trace *bt;
302 bt = xchg(&q->blk_trace, NULL);
306 if (bt->trace_state != Blktrace_running)
307 blk_trace_cleanup(bt);
311 EXPORT_SYMBOL_GPL(blk_trace_remove);
313 static int blk_dropped_open(struct inode *inode, struct file *filp)
315 filp->private_data = inode->i_private;
320 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
321 size_t count, loff_t *ppos)
323 struct blk_trace *bt = filp->private_data;
326 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
328 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
331 static const struct file_operations blk_dropped_fops = {
332 .owner = THIS_MODULE,
333 .open = blk_dropped_open,
334 .read = blk_dropped_read,
335 .llseek = default_llseek,
338 static int blk_msg_open(struct inode *inode, struct file *filp)
340 filp->private_data = inode->i_private;
345 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
346 size_t count, loff_t *ppos)
349 struct blk_trace *bt;
351 if (count >= BLK_TN_MAX_MSG)
354 msg = kmalloc(count + 1, GFP_KERNEL);
358 if (copy_from_user(msg, buffer, count)) {
364 bt = filp->private_data;
365 __trace_note_message(bt, "%s", msg);
371 static const struct file_operations blk_msg_fops = {
372 .owner = THIS_MODULE,
373 .open = blk_msg_open,
374 .write = blk_msg_write,
375 .llseek = noop_llseek,
379 * Keep track of how many times we encountered a full subbuffer, to aid
380 * the user space app in telling how many lost events there were.
382 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
383 void *prev_subbuf, size_t prev_padding)
385 struct blk_trace *bt;
387 if (!relay_buf_full(buf))
390 bt = buf->chan->private_data;
391 atomic_inc(&bt->dropped);
395 static int blk_remove_buf_file_callback(struct dentry *dentry)
397 debugfs_remove(dentry);
402 static struct dentry *blk_create_buf_file_callback(const char *filename,
403 struct dentry *parent,
405 struct rchan_buf *buf,
408 return debugfs_create_file(filename, mode, parent, buf,
409 &relay_file_operations);
412 static struct rchan_callbacks blk_relay_callbacks = {
413 .subbuf_start = blk_subbuf_start_callback,
414 .create_buf_file = blk_create_buf_file_callback,
415 .remove_buf_file = blk_remove_buf_file_callback,
418 static void blk_trace_setup_lba(struct blk_trace *bt,
419 struct block_device *bdev)
421 struct hd_struct *part = NULL;
424 part = bdev->bd_part;
427 bt->start_lba = part->start_sect;
428 bt->end_lba = part->start_sect + part->nr_sects;
436 * Setup everything required to start tracing
438 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
439 struct block_device *bdev,
440 struct blk_user_trace_setup *buts)
442 struct blk_trace *old_bt, *bt = NULL;
443 struct dentry *dir = NULL;
446 if (!buts->buf_size || !buts->buf_nr)
449 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
450 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
453 * some device names have larger paths - convert the slashes
454 * to underscores for this to work as expected
456 for (i = 0; i < strlen(buts->name); i++)
457 if (buts->name[i] == '/')
460 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
465 bt->sequence = alloc_percpu(unsigned long);
469 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
475 mutex_lock(&blk_tree_mutex);
476 if (!blk_tree_root) {
477 blk_tree_root = debugfs_create_dir("block", NULL);
478 if (!blk_tree_root) {
479 mutex_unlock(&blk_tree_mutex);
483 mutex_unlock(&blk_tree_mutex);
485 dir = debugfs_create_dir(buts->name, blk_tree_root);
492 atomic_set(&bt->dropped, 0);
495 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
497 if (!bt->dropped_file)
500 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
504 bt->rchan = relay_open("trace", dir, buts->buf_size,
505 buts->buf_nr, &blk_relay_callbacks, bt);
509 bt->act_mask = buts->act_mask;
511 bt->act_mask = (u16) -1;
513 blk_trace_setup_lba(bt, bdev);
515 /* overwrite with user settings */
517 bt->start_lba = buts->start_lba;
519 bt->end_lba = buts->end_lba;
522 bt->trace_state = Blktrace_setup;
525 old_bt = xchg(&q->blk_trace, bt);
527 (void) xchg(&q->blk_trace, old_bt);
531 if (atomic_inc_return(&blk_probes_ref) == 1)
532 blk_register_tracepoints();
540 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
541 struct block_device *bdev,
544 struct blk_user_trace_setup buts;
547 ret = copy_from_user(&buts, arg, sizeof(buts));
551 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
555 if (copy_to_user(arg, &buts, sizeof(buts))) {
561 EXPORT_SYMBOL_GPL(blk_trace_setup);
563 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
564 static int compat_blk_trace_setup(struct request_queue *q, char *name,
565 dev_t dev, struct block_device *bdev,
568 struct blk_user_trace_setup buts;
569 struct compat_blk_user_trace_setup cbuts;
572 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
575 buts = (struct blk_user_trace_setup) {
576 .act_mask = cbuts.act_mask,
577 .buf_size = cbuts.buf_size,
578 .buf_nr = cbuts.buf_nr,
579 .start_lba = cbuts.start_lba,
580 .end_lba = cbuts.end_lba,
583 memcpy(&buts.name, &cbuts.name, 32);
585 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
589 if (copy_to_user(arg, &buts.name, 32)) {
598 int blk_trace_startstop(struct request_queue *q, int start)
601 struct blk_trace *bt = q->blk_trace;
607 * For starting a trace, we can transition from a setup or stopped
608 * trace. For stopping a trace, the state must be running
612 if (bt->trace_state == Blktrace_setup ||
613 bt->trace_state == Blktrace_stopped) {
616 bt->trace_state = Blktrace_running;
622 if (bt->trace_state == Blktrace_running) {
623 bt->trace_state = Blktrace_stopped;
624 relay_flush(bt->rchan);
631 EXPORT_SYMBOL_GPL(blk_trace_startstop);
634 * blk_trace_ioctl: - handle the ioctls associated with tracing
635 * @bdev: the block device
636 * @cmd: the ioctl cmd
637 * @arg: the argument data, if any
640 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
642 struct request_queue *q;
644 char b[BDEVNAME_SIZE];
646 q = bdev_get_queue(bdev);
650 mutex_lock(&bdev->bd_mutex);
655 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
657 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
658 case BLKTRACESETUP32:
660 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
666 ret = blk_trace_startstop(q, start);
668 case BLKTRACETEARDOWN:
669 ret = blk_trace_remove(q);
676 mutex_unlock(&bdev->bd_mutex);
681 * blk_trace_shutdown: - stop and cleanup trace structures
682 * @q: the request queue associated with the device
685 void blk_trace_shutdown(struct request_queue *q)
688 blk_trace_startstop(q, 0);
698 * blk_add_trace_rq - Add a trace for a request oriented action
699 * @q: queue the io is for
700 * @rq: the source request
704 * Records an action against a request. Will log the bio offset + size.
707 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
710 struct blk_trace *bt = q->blk_trace;
715 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
716 what |= BLK_TC_ACT(BLK_TC_PC);
717 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rq->cmd_flags,
718 what, rq->errors, rq->cmd_len, rq->cmd);
720 what |= BLK_TC_ACT(BLK_TC_FS);
721 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
722 rq->cmd_flags, what, rq->errors, 0, NULL);
726 static void blk_add_trace_rq_abort(void *ignore,
727 struct request_queue *q, struct request *rq)
729 blk_add_trace_rq(q, rq, BLK_TA_ABORT);
732 static void blk_add_trace_rq_insert(void *ignore,
733 struct request_queue *q, struct request *rq)
735 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
738 static void blk_add_trace_rq_issue(void *ignore,
739 struct request_queue *q, struct request *rq)
741 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
744 static void blk_add_trace_rq_requeue(void *ignore,
745 struct request_queue *q,
748 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
751 static void blk_add_trace_rq_complete(void *ignore,
752 struct request_queue *q,
755 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
759 * blk_add_trace_bio - Add a trace for a bio oriented action
760 * @q: queue the io is for
761 * @bio: the source bio
763 * @error: error, if any
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 if (!error && !bio_flagged(bio, BIO_UPTODATE))
780 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
784 static void blk_add_trace_bio_bounce(void *ignore,
785 struct request_queue *q, struct bio *bio)
787 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
790 static void blk_add_trace_bio_complete(void *ignore,
791 struct request_queue *q, struct bio *bio,
794 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
797 static void blk_add_trace_bio_backmerge(void *ignore,
798 struct request_queue *q,
801 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
804 static void blk_add_trace_bio_frontmerge(void *ignore,
805 struct request_queue *q,
808 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
811 static void blk_add_trace_bio_queue(void *ignore,
812 struct request_queue *q, struct bio *bio)
814 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
817 static void blk_add_trace_getrq(void *ignore,
818 struct request_queue *q,
819 struct bio *bio, int rw)
822 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
824 struct blk_trace *bt = q->blk_trace;
827 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
832 static void blk_add_trace_sleeprq(void *ignore,
833 struct request_queue *q,
834 struct bio *bio, int rw)
837 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
839 struct blk_trace *bt = q->blk_trace;
842 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
847 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
849 struct blk_trace *bt = q->blk_trace;
852 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
855 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
856 unsigned int depth, bool explicit)
858 struct blk_trace *bt = q->blk_trace;
861 __be64 rpdu = cpu_to_be64(depth);
865 what = BLK_TA_UNPLUG_IO;
867 what = BLK_TA_UNPLUG_TIMER;
869 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
873 static void blk_add_trace_split(void *ignore,
874 struct request_queue *q, struct bio *bio,
877 struct blk_trace *bt = q->blk_trace;
880 __be64 rpdu = cpu_to_be64(pdu);
882 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
883 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
884 sizeof(rpdu), &rpdu);
889 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
890 * @ignore: trace callback data parameter (not used)
891 * @q: queue the io is for
892 * @bio: the source bio
893 * @dev: target device
894 * @from: source sector
897 * Device mapper or raid target sometimes need to split a bio because
898 * it spans a stripe (or similar). Add a trace for that action.
901 static void blk_add_trace_bio_remap(void *ignore,
902 struct request_queue *q, struct bio *bio,
903 dev_t dev, sector_t from)
905 struct blk_trace *bt = q->blk_trace;
906 struct blk_io_trace_remap r;
911 r.device_from = cpu_to_be32(dev);
912 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
913 r.sector_from = cpu_to_be64(from);
915 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
916 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
921 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
922 * @ignore: trace callback data parameter (not used)
923 * @q: queue the io is for
924 * @rq: the source request
925 * @dev: target device
926 * @from: source sector
929 * Device mapper remaps request to other devices.
930 * Add a trace for that action.
933 static void blk_add_trace_rq_remap(void *ignore,
934 struct request_queue *q,
935 struct request *rq, dev_t dev,
938 struct blk_trace *bt = q->blk_trace;
939 struct blk_io_trace_remap r;
944 r.device_from = cpu_to_be32(dev);
945 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
946 r.sector_from = cpu_to_be64(from);
948 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
949 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
954 * blk_add_driver_data - Add binary message with driver-specific data
955 * @q: queue the io is for
957 * @data: driver-specific data
958 * @len: length of driver-specific data
961 * Some drivers might want to write driver-specific data per request.
964 void blk_add_driver_data(struct request_queue *q,
966 void *data, size_t len)
968 struct blk_trace *bt = q->blk_trace;
973 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
974 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
975 BLK_TA_DRV_DATA, rq->errors, len, data);
977 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
978 BLK_TA_DRV_DATA, rq->errors, len, data);
980 EXPORT_SYMBOL_GPL(blk_add_driver_data);
982 static void blk_register_tracepoints(void)
986 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
988 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
990 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
992 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
994 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
996 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
998 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1000 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1002 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1004 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1006 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1008 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1010 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1012 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1014 ret = register_trace_block_split(blk_add_trace_split, NULL);
1016 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1018 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1022 static void blk_unregister_tracepoints(void)
1024 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1025 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1026 unregister_trace_block_split(blk_add_trace_split, NULL);
1027 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1028 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1029 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1030 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1031 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1032 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1033 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1034 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1035 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1036 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1037 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1038 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1039 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1040 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1042 tracepoint_synchronize_unregister();
1046 * struct blk_io_tracer formatting routines
1049 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1052 int tc = t->action >> BLK_TC_SHIFT;
1054 if (t->action == BLK_TN_MESSAGE) {
1059 if (tc & BLK_TC_FLUSH)
1062 if (tc & BLK_TC_DISCARD)
1064 else if (tc & BLK_TC_WRITE)
1071 if (tc & BLK_TC_FUA)
1073 if (tc & BLK_TC_AHEAD)
1075 if (tc & BLK_TC_SYNC)
1077 if (tc & BLK_TC_META)
1084 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1086 return (const struct blk_io_trace *)ent;
1089 static inline const void *pdu_start(const struct trace_entry *ent)
1091 return te_blk_io_trace(ent) + 1;
1094 static inline u32 t_action(const struct trace_entry *ent)
1096 return te_blk_io_trace(ent)->action;
1099 static inline u32 t_bytes(const struct trace_entry *ent)
1101 return te_blk_io_trace(ent)->bytes;
1104 static inline u32 t_sec(const struct trace_entry *ent)
1106 return te_blk_io_trace(ent)->bytes >> 9;
1109 static inline unsigned long long t_sector(const struct trace_entry *ent)
1111 return te_blk_io_trace(ent)->sector;
1114 static inline __u16 t_error(const struct trace_entry *ent)
1116 return te_blk_io_trace(ent)->error;
1119 static __u64 get_pdu_int(const struct trace_entry *ent)
1121 const __u64 *val = pdu_start(ent);
1122 return be64_to_cpu(*val);
1125 static void get_pdu_remap(const struct trace_entry *ent,
1126 struct blk_io_trace_remap *r)
1128 const struct blk_io_trace_remap *__r = pdu_start(ent);
1129 __u64 sector_from = __r->sector_from;
1131 r->device_from = be32_to_cpu(__r->device_from);
1132 r->device_to = be32_to_cpu(__r->device_to);
1133 r->sector_from = be64_to_cpu(sector_from);
1136 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1138 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1140 char rwbs[RWBS_LEN];
1141 unsigned long long ts = iter->ts;
1142 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1143 unsigned secs = (unsigned long)ts;
1144 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1148 return trace_seq_printf(&iter->seq,
1149 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1150 MAJOR(t->device), MINOR(t->device), iter->cpu,
1151 secs, nsec_rem, iter->ent->pid, act, rwbs);
1154 static int blk_log_action(struct trace_iterator *iter, const char *act)
1156 char rwbs[RWBS_LEN];
1157 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1160 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1161 MAJOR(t->device), MINOR(t->device), act, rwbs);
1164 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1166 const unsigned char *pdu_buf;
1170 pdu_buf = pdu_start(ent);
1171 pdu_len = te_blk_io_trace(ent)->pdu_len;
1176 /* find the last zero that needs to be printed */
1177 for (end = pdu_len - 1; end >= 0; end--)
1182 if (!trace_seq_putc(s, '('))
1185 for (i = 0; i < pdu_len; i++) {
1187 ret = trace_seq_printf(s, "%s%02x",
1188 i == 0 ? "" : " ", pdu_buf[i]);
1193 * stop when the rest is just zeroes and indicate so
1194 * with a ".." appended
1196 if (i == end && end != pdu_len - 1)
1197 return trace_seq_puts(s, " ..) ");
1200 return trace_seq_puts(s, ") ");
1203 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1205 char cmd[TASK_COMM_LEN];
1207 trace_find_cmdline(ent->pid, cmd);
1209 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1212 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1215 ret = blk_log_dump_pdu(s, ent);
1218 return trace_seq_printf(s, "[%s]\n", cmd);
1221 return trace_seq_printf(s, "%llu + %u [%s]\n",
1222 t_sector(ent), t_sec(ent), cmd);
1223 return trace_seq_printf(s, "[%s]\n", cmd);
1227 static int blk_log_with_error(struct trace_seq *s,
1228 const struct trace_entry *ent)
1230 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1233 ret = blk_log_dump_pdu(s, ent);
1235 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1239 return trace_seq_printf(s, "%llu + %u [%d]\n",
1241 t_sec(ent), t_error(ent));
1242 return trace_seq_printf(s, "%llu [%d]\n",
1243 t_sector(ent), t_error(ent));
1247 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1249 struct blk_io_trace_remap r = { .device_from = 0, };
1251 get_pdu_remap(ent, &r);
1252 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1253 t_sector(ent), t_sec(ent),
1254 MAJOR(r.device_from), MINOR(r.device_from),
1255 (unsigned long long)r.sector_from);
1258 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1260 char cmd[TASK_COMM_LEN];
1262 trace_find_cmdline(ent->pid, cmd);
1264 return trace_seq_printf(s, "[%s]\n", cmd);
1267 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1269 char cmd[TASK_COMM_LEN];
1271 trace_find_cmdline(ent->pid, cmd);
1273 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1276 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1278 char cmd[TASK_COMM_LEN];
1280 trace_find_cmdline(ent->pid, cmd);
1282 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1283 get_pdu_int(ent), cmd);
1286 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1289 const struct blk_io_trace *t = te_blk_io_trace(ent);
1291 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1293 return trace_seq_putc(s, '\n');
1298 * struct tracer operations
1301 static void blk_tracer_print_header(struct seq_file *m)
1303 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1305 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1309 static void blk_tracer_start(struct trace_array *tr)
1311 blk_tracer_enabled = true;
1314 static int blk_tracer_init(struct trace_array *tr)
1317 blk_tracer_start(tr);
1321 static void blk_tracer_stop(struct trace_array *tr)
1323 blk_tracer_enabled = false;
1326 static void blk_tracer_reset(struct trace_array *tr)
1328 blk_tracer_stop(tr);
1331 static const struct {
1333 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1335 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1336 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1337 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1338 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1339 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1340 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1341 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1342 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1343 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1344 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1345 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1346 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1347 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1348 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1349 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1352 static enum print_line_t print_one_line(struct trace_iterator *iter,
1355 struct trace_seq *s = &iter->seq;
1356 const struct blk_io_trace *t;
1360 blk_log_action_t *log_action;
1362 t = te_blk_io_trace(iter->ent);
1363 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1364 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1365 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1367 if (t->action == BLK_TN_MESSAGE) {
1368 ret = log_action(iter, long_act ? "message" : "m");
1370 ret = blk_log_msg(s, iter->ent);
1374 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1375 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1377 ret = log_action(iter, what2act[what].act[long_act]);
1379 ret = what2act[what].print(s, iter->ent);
1382 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1385 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1386 int flags, struct trace_event *event)
1388 return print_one_line(iter, false);
1391 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1393 struct trace_seq *s = &iter->seq;
1394 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1395 const int offset = offsetof(struct blk_io_trace, sector);
1396 struct blk_io_trace old = {
1397 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1401 if (!trace_seq_putmem(s, &old, offset))
1403 return trace_seq_putmem(s, &t->sector,
1404 sizeof(old) - offset + t->pdu_len);
1407 static enum print_line_t
1408 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1409 struct trace_event *event)
1411 return blk_trace_synthesize_old_trace(iter) ?
1412 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1415 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1417 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1418 return TRACE_TYPE_UNHANDLED;
1420 return print_one_line(iter, true);
1423 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1425 /* don't output context-info for blk_classic output */
1426 if (bit == TRACE_BLK_OPT_CLASSIC) {
1428 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1430 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1435 static struct tracer blk_tracer __read_mostly = {
1437 .init = blk_tracer_init,
1438 .reset = blk_tracer_reset,
1439 .start = blk_tracer_start,
1440 .stop = blk_tracer_stop,
1441 .print_header = blk_tracer_print_header,
1442 .print_line = blk_tracer_print_line,
1443 .flags = &blk_tracer_flags,
1444 .set_flag = blk_tracer_set_flag,
1447 static struct trace_event_functions trace_blk_event_funcs = {
1448 .trace = blk_trace_event_print,
1449 .binary = blk_trace_event_print_binary,
1452 static struct trace_event trace_blk_event = {
1454 .funcs = &trace_blk_event_funcs,
1457 static int __init init_blk_tracer(void)
1459 if (!register_ftrace_event(&trace_blk_event)) {
1460 pr_warning("Warning: could not register block events\n");
1464 if (register_tracer(&blk_tracer) != 0) {
1465 pr_warning("Warning: could not register the block tracer\n");
1466 unregister_ftrace_event(&trace_blk_event);
1473 device_initcall(init_blk_tracer);
1475 static int blk_trace_remove_queue(struct request_queue *q)
1477 struct blk_trace *bt;
1479 bt = xchg(&q->blk_trace, NULL);
1483 if (atomic_dec_and_test(&blk_probes_ref))
1484 blk_unregister_tracepoints();
1491 * Setup everything required to start tracing
1493 static int blk_trace_setup_queue(struct request_queue *q,
1494 struct block_device *bdev)
1496 struct blk_trace *old_bt, *bt = NULL;
1499 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1503 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1507 bt->dev = bdev->bd_dev;
1508 bt->act_mask = (u16)-1;
1510 blk_trace_setup_lba(bt, bdev);
1512 old_bt = xchg(&q->blk_trace, bt);
1513 if (old_bt != NULL) {
1514 (void)xchg(&q->blk_trace, old_bt);
1519 if (atomic_inc_return(&blk_probes_ref) == 1)
1520 blk_register_tracepoints();
1529 * sysfs interface to enable and configure tracing
1532 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1533 struct device_attribute *attr,
1535 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1536 struct device_attribute *attr,
1537 const char *buf, size_t count);
1538 #define BLK_TRACE_DEVICE_ATTR(_name) \
1539 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1540 sysfs_blk_trace_attr_show, \
1541 sysfs_blk_trace_attr_store)
1543 static BLK_TRACE_DEVICE_ATTR(enable);
1544 static BLK_TRACE_DEVICE_ATTR(act_mask);
1545 static BLK_TRACE_DEVICE_ATTR(pid);
1546 static BLK_TRACE_DEVICE_ATTR(start_lba);
1547 static BLK_TRACE_DEVICE_ATTR(end_lba);
1549 static struct attribute *blk_trace_attrs[] = {
1550 &dev_attr_enable.attr,
1551 &dev_attr_act_mask.attr,
1553 &dev_attr_start_lba.attr,
1554 &dev_attr_end_lba.attr,
1558 struct attribute_group blk_trace_attr_group = {
1560 .attrs = blk_trace_attrs,
1563 static const struct {
1567 { BLK_TC_READ, "read" },
1568 { BLK_TC_WRITE, "write" },
1569 { BLK_TC_FLUSH, "flush" },
1570 { BLK_TC_SYNC, "sync" },
1571 { BLK_TC_QUEUE, "queue" },
1572 { BLK_TC_REQUEUE, "requeue" },
1573 { BLK_TC_ISSUE, "issue" },
1574 { BLK_TC_COMPLETE, "complete" },
1575 { BLK_TC_FS, "fs" },
1576 { BLK_TC_PC, "pc" },
1577 { BLK_TC_AHEAD, "ahead" },
1578 { BLK_TC_META, "meta" },
1579 { BLK_TC_DISCARD, "discard" },
1580 { BLK_TC_DRV_DATA, "drv_data" },
1581 { BLK_TC_FUA, "fua" },
1584 static int blk_trace_str2mask(const char *str)
1588 char *buf, *s, *token;
1590 buf = kstrdup(str, GFP_KERNEL);
1596 token = strsep(&s, ",");
1603 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1604 if (strcasecmp(token, mask_maps[i].str) == 0) {
1605 mask |= mask_maps[i].mask;
1609 if (i == ARRAY_SIZE(mask_maps)) {
1619 static ssize_t blk_trace_mask2str(char *buf, int mask)
1624 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1625 if (mask & mask_maps[i].mask) {
1626 p += sprintf(p, "%s%s",
1627 (p == buf) ? "" : ",", mask_maps[i].str);
1635 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1637 if (bdev->bd_disk == NULL)
1640 return bdev_get_queue(bdev);
1643 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1644 struct device_attribute *attr,
1647 struct hd_struct *p = dev_to_part(dev);
1648 struct request_queue *q;
1649 struct block_device *bdev;
1650 ssize_t ret = -ENXIO;
1652 bdev = bdget(part_devt(p));
1656 q = blk_trace_get_queue(bdev);
1660 mutex_lock(&bdev->bd_mutex);
1662 if (attr == &dev_attr_enable) {
1663 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1664 goto out_unlock_bdev;
1667 if (q->blk_trace == NULL)
1668 ret = sprintf(buf, "disabled\n");
1669 else if (attr == &dev_attr_act_mask)
1670 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1671 else if (attr == &dev_attr_pid)
1672 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1673 else if (attr == &dev_attr_start_lba)
1674 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1675 else if (attr == &dev_attr_end_lba)
1676 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1679 mutex_unlock(&bdev->bd_mutex);
1686 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1687 struct device_attribute *attr,
1688 const char *buf, size_t count)
1690 struct block_device *bdev;
1691 struct request_queue *q;
1692 struct hd_struct *p;
1694 ssize_t ret = -EINVAL;
1699 if (attr == &dev_attr_act_mask) {
1700 if (sscanf(buf, "%llx", &value) != 1) {
1701 /* Assume it is a list of trace category names */
1702 ret = blk_trace_str2mask(buf);
1707 } else if (sscanf(buf, "%llu", &value) != 1)
1712 p = dev_to_part(dev);
1713 bdev = bdget(part_devt(p));
1717 q = blk_trace_get_queue(bdev);
1721 mutex_lock(&bdev->bd_mutex);
1723 if (attr == &dev_attr_enable) {
1725 ret = blk_trace_setup_queue(q, bdev);
1727 ret = blk_trace_remove_queue(q);
1728 goto out_unlock_bdev;
1732 if (q->blk_trace == NULL)
1733 ret = blk_trace_setup_queue(q, bdev);
1736 if (attr == &dev_attr_act_mask)
1737 q->blk_trace->act_mask = value;
1738 else if (attr == &dev_attr_pid)
1739 q->blk_trace->pid = value;
1740 else if (attr == &dev_attr_start_lba)
1741 q->blk_trace->start_lba = value;
1742 else if (attr == &dev_attr_end_lba)
1743 q->blk_trace->end_lba = value;
1747 mutex_unlock(&bdev->bd_mutex);
1751 return ret ? ret : count;
1754 int blk_trace_init_sysfs(struct device *dev)
1756 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1759 void blk_trace_remove_sysfs(struct device *dev)
1761 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1764 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1766 #ifdef CONFIG_EVENT_TRACING
1768 void blk_dump_cmd(char *buf, struct request *rq)
1771 int len = rq->cmd_len;
1772 unsigned char *cmd = rq->cmd;
1774 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1779 for (end = len - 1; end >= 0; end--)
1784 for (i = 0; i < len; i++) {
1785 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1786 if (i == end && end != len - 1) {
1787 sprintf(buf, " ..");
1793 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1802 else if (rw & REQ_DISCARD)
1811 if (rw & REQ_RAHEAD)
1817 if (rw & REQ_SECURE)
1823 #endif /* CONFIG_EVENT_TRACING */