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/export.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))
143 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
144 * message to the trace.
146 if (!(bt->act_mask & BLK_TC_NOTIFY))
149 local_irq_save(flags);
150 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
152 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
155 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
156 local_irq_restore(flags);
158 EXPORT_SYMBOL_GPL(__trace_note_message);
160 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
163 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
165 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
167 if (bt->pid && pid != bt->pid)
174 * Data direction bit lookup
176 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
177 BLK_TC_ACT(BLK_TC_WRITE) };
179 #define BLK_TC_RAHEAD BLK_TC_AHEAD
181 /* The ilog2() calls fall out because they're constant */
182 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
183 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
186 * The worker for the various blk_add_trace*() types. Fills out a
187 * blk_io_trace structure and places it in a per-cpu subbuffer.
189 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
190 int rw, u32 what, int error, int pdu_len, void *pdu_data)
192 struct task_struct *tsk = current;
193 struct ring_buffer_event *event = NULL;
194 struct ring_buffer *buffer = NULL;
195 struct blk_io_trace *t;
196 unsigned long flags = 0;
197 unsigned long *sequence;
200 bool blk_tracer = blk_tracer_enabled;
202 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
205 what |= ddir_act[rw & WRITE];
206 what |= MASK_TC_BIT(rw, SYNC);
207 what |= MASK_TC_BIT(rw, RAHEAD);
208 what |= MASK_TC_BIT(rw, META);
209 what |= MASK_TC_BIT(rw, DISCARD);
210 what |= MASK_TC_BIT(rw, FLUSH);
211 what |= MASK_TC_BIT(rw, FUA);
214 if (act_log_check(bt, what, sector, pid))
216 cpu = raw_smp_processor_id();
219 tracing_record_cmdline(current);
221 buffer = blk_tr->buffer;
222 pc = preempt_count();
223 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
224 sizeof(*t) + pdu_len,
228 t = ring_buffer_event_data(event);
233 * A word about the locking here - we disable interrupts to reserve
234 * some space in the relay per-cpu buffer, to prevent an irq
235 * from coming in and stepping on our toes.
237 local_irq_save(flags);
239 if (unlikely(tsk->btrace_seq != blktrace_seq))
240 trace_note_tsk(bt, tsk);
242 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
244 sequence = per_cpu_ptr(bt->sequence, cpu);
246 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
247 t->sequence = ++(*sequence);
248 t->time = ktime_to_ns(ktime_get());
251 * These two are not needed in ftrace as they are in the
252 * generic trace_entry, filled by tracing_generic_entry_update,
253 * but for the trace_event->bin() synthesizer benefit we do it
264 t->pdu_len = pdu_len;
267 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
270 trace_buffer_unlock_commit(buffer, event, 0, pc);
275 local_irq_restore(flags);
278 static struct dentry *blk_tree_root;
279 static DEFINE_MUTEX(blk_tree_mutex);
281 static void blk_trace_free(struct blk_trace *bt)
283 debugfs_remove(bt->msg_file);
284 debugfs_remove(bt->dropped_file);
285 relay_close(bt->rchan);
286 debugfs_remove(bt->dir);
287 free_percpu(bt->sequence);
288 free_percpu(bt->msg_data);
292 static void blk_trace_cleanup(struct blk_trace *bt)
295 if (atomic_dec_and_test(&blk_probes_ref))
296 blk_unregister_tracepoints();
299 int blk_trace_remove(struct request_queue *q)
301 struct blk_trace *bt;
303 bt = xchg(&q->blk_trace, NULL);
307 if (bt->trace_state != Blktrace_running)
308 blk_trace_cleanup(bt);
312 EXPORT_SYMBOL_GPL(blk_trace_remove);
314 static int blk_dropped_open(struct inode *inode, struct file *filp)
316 filp->private_data = inode->i_private;
321 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
322 size_t count, loff_t *ppos)
324 struct blk_trace *bt = filp->private_data;
327 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
329 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
332 static const struct file_operations blk_dropped_fops = {
333 .owner = THIS_MODULE,
334 .open = blk_dropped_open,
335 .read = blk_dropped_read,
336 .llseek = default_llseek,
339 static int blk_msg_open(struct inode *inode, struct file *filp)
341 filp->private_data = inode->i_private;
346 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
347 size_t count, loff_t *ppos)
350 struct blk_trace *bt;
352 if (count >= BLK_TN_MAX_MSG)
355 msg = kmalloc(count + 1, GFP_KERNEL);
359 if (copy_from_user(msg, buffer, count)) {
365 bt = filp->private_data;
366 __trace_note_message(bt, "%s", msg);
372 static const struct file_operations blk_msg_fops = {
373 .owner = THIS_MODULE,
374 .open = blk_msg_open,
375 .write = blk_msg_write,
376 .llseek = noop_llseek,
380 * Keep track of how many times we encountered a full subbuffer, to aid
381 * the user space app in telling how many lost events there were.
383 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
384 void *prev_subbuf, size_t prev_padding)
386 struct blk_trace *bt;
388 if (!relay_buf_full(buf))
391 bt = buf->chan->private_data;
392 atomic_inc(&bt->dropped);
396 static int blk_remove_buf_file_callback(struct dentry *dentry)
398 debugfs_remove(dentry);
403 static struct dentry *blk_create_buf_file_callback(const char *filename,
404 struct dentry *parent,
406 struct rchan_buf *buf,
409 return debugfs_create_file(filename, mode, parent, buf,
410 &relay_file_operations);
413 static struct rchan_callbacks blk_relay_callbacks = {
414 .subbuf_start = blk_subbuf_start_callback,
415 .create_buf_file = blk_create_buf_file_callback,
416 .remove_buf_file = blk_remove_buf_file_callback,
419 static void blk_trace_setup_lba(struct blk_trace *bt,
420 struct block_device *bdev)
422 struct hd_struct *part = NULL;
425 part = bdev->bd_part;
428 bt->start_lba = part->start_sect;
429 bt->end_lba = part->start_sect + part->nr_sects;
437 * Setup everything required to start tracing
439 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
440 struct block_device *bdev,
441 struct blk_user_trace_setup *buts)
443 struct blk_trace *old_bt, *bt = NULL;
444 struct dentry *dir = NULL;
447 if (!buts->buf_size || !buts->buf_nr)
450 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
451 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
454 * some device names have larger paths - convert the slashes
455 * to underscores for this to work as expected
457 for (i = 0; i < strlen(buts->name); i++)
458 if (buts->name[i] == '/')
461 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
466 bt->sequence = alloc_percpu(unsigned long);
470 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
476 mutex_lock(&blk_tree_mutex);
477 if (!blk_tree_root) {
478 blk_tree_root = debugfs_create_dir("block", NULL);
479 if (!blk_tree_root) {
480 mutex_unlock(&blk_tree_mutex);
484 mutex_unlock(&blk_tree_mutex);
486 dir = debugfs_create_dir(buts->name, blk_tree_root);
493 atomic_set(&bt->dropped, 0);
496 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
498 if (!bt->dropped_file)
501 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
505 bt->rchan = relay_open("trace", dir, buts->buf_size,
506 buts->buf_nr, &blk_relay_callbacks, bt);
510 bt->act_mask = buts->act_mask;
512 bt->act_mask = (u16) -1;
514 blk_trace_setup_lba(bt, bdev);
516 /* overwrite with user settings */
518 bt->start_lba = buts->start_lba;
520 bt->end_lba = buts->end_lba;
523 bt->trace_state = Blktrace_setup;
526 old_bt = xchg(&q->blk_trace, bt);
528 (void) xchg(&q->blk_trace, old_bt);
532 if (atomic_inc_return(&blk_probes_ref) == 1)
533 blk_register_tracepoints();
541 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
542 struct block_device *bdev,
545 struct blk_user_trace_setup buts;
548 ret = copy_from_user(&buts, arg, sizeof(buts));
552 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
556 if (copy_to_user(arg, &buts, sizeof(buts))) {
562 EXPORT_SYMBOL_GPL(blk_trace_setup);
564 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
565 static int compat_blk_trace_setup(struct request_queue *q, char *name,
566 dev_t dev, struct block_device *bdev,
569 struct blk_user_trace_setup buts;
570 struct compat_blk_user_trace_setup cbuts;
573 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
576 buts = (struct blk_user_trace_setup) {
577 .act_mask = cbuts.act_mask,
578 .buf_size = cbuts.buf_size,
579 .buf_nr = cbuts.buf_nr,
580 .start_lba = cbuts.start_lba,
581 .end_lba = cbuts.end_lba,
584 memcpy(&buts.name, &cbuts.name, 32);
586 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
590 if (copy_to_user(arg, &buts.name, 32)) {
599 int blk_trace_startstop(struct request_queue *q, int start)
602 struct blk_trace *bt = q->blk_trace;
608 * For starting a trace, we can transition from a setup or stopped
609 * trace. For stopping a trace, the state must be running
613 if (bt->trace_state == Blktrace_setup ||
614 bt->trace_state == Blktrace_stopped) {
617 bt->trace_state = Blktrace_running;
623 if (bt->trace_state == Blktrace_running) {
624 bt->trace_state = Blktrace_stopped;
625 relay_flush(bt->rchan);
632 EXPORT_SYMBOL_GPL(blk_trace_startstop);
635 * blk_trace_ioctl: - handle the ioctls associated with tracing
636 * @bdev: the block device
637 * @cmd: the ioctl cmd
638 * @arg: the argument data, if any
641 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
643 struct request_queue *q;
645 char b[BDEVNAME_SIZE];
647 q = bdev_get_queue(bdev);
651 mutex_lock(&bdev->bd_mutex);
656 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
658 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
659 case BLKTRACESETUP32:
661 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
667 ret = blk_trace_startstop(q, start);
669 case BLKTRACETEARDOWN:
670 ret = blk_trace_remove(q);
677 mutex_unlock(&bdev->bd_mutex);
682 * blk_trace_shutdown: - stop and cleanup trace structures
683 * @q: the request queue associated with the device
686 void blk_trace_shutdown(struct request_queue *q)
689 blk_trace_startstop(q, 0);
699 * blk_add_trace_rq - Add a trace for a request oriented action
700 * @q: queue the io is for
701 * @rq: the source request
702 * @nr_bytes: number of completed bytes
706 * Records an action against a request. Will log the bio offset + size.
709 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
710 unsigned int nr_bytes, u32 what)
712 struct blk_trace *bt = q->blk_trace;
717 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
718 what |= BLK_TC_ACT(BLK_TC_PC);
719 __blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
720 what, rq->errors, rq->cmd_len, rq->cmd);
722 what |= BLK_TC_ACT(BLK_TC_FS);
723 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
724 rq->cmd_flags, what, rq->errors, 0, NULL);
728 static void blk_add_trace_rq_abort(void *ignore,
729 struct request_queue *q, struct request *rq)
731 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
734 static void blk_add_trace_rq_insert(void *ignore,
735 struct request_queue *q, struct request *rq)
737 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
740 static void blk_add_trace_rq_issue(void *ignore,
741 struct request_queue *q, struct request *rq)
743 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
746 static void blk_add_trace_rq_requeue(void *ignore,
747 struct request_queue *q,
750 blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
753 static void blk_add_trace_rq_complete(void *ignore,
754 struct request_queue *q,
756 unsigned int nr_bytes)
758 blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
762 * blk_add_trace_bio - Add a trace for a bio oriented action
763 * @q: queue the io is for
764 * @bio: the source bio
766 * @error: error, if any
769 * Records an action against a bio. Will log the bio offset + size.
772 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
775 struct blk_trace *bt = q->blk_trace;
780 if (!error && !bio_flagged(bio, BIO_UPTODATE))
783 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
787 static void blk_add_trace_bio_bounce(void *ignore,
788 struct request_queue *q, struct bio *bio)
790 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
793 static void blk_add_trace_bio_complete(void *ignore,
794 struct request_queue *q, struct bio *bio,
797 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
800 static void blk_add_trace_bio_backmerge(void *ignore,
801 struct request_queue *q,
804 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
807 static void blk_add_trace_bio_frontmerge(void *ignore,
808 struct request_queue *q,
811 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
814 static void blk_add_trace_bio_queue(void *ignore,
815 struct request_queue *q, struct bio *bio)
817 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
820 static void blk_add_trace_getrq(void *ignore,
821 struct request_queue *q,
822 struct bio *bio, int rw)
825 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
827 struct blk_trace *bt = q->blk_trace;
830 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
835 static void blk_add_trace_sleeprq(void *ignore,
836 struct request_queue *q,
837 struct bio *bio, int rw)
840 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
842 struct blk_trace *bt = q->blk_trace;
845 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
850 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
852 struct blk_trace *bt = q->blk_trace;
855 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
858 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
859 unsigned int depth, bool explicit)
861 struct blk_trace *bt = q->blk_trace;
864 __be64 rpdu = cpu_to_be64(depth);
868 what = BLK_TA_UNPLUG_IO;
870 what = BLK_TA_UNPLUG_TIMER;
872 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
876 static void blk_add_trace_split(void *ignore,
877 struct request_queue *q, struct bio *bio,
880 struct blk_trace *bt = q->blk_trace;
883 __be64 rpdu = cpu_to_be64(pdu);
885 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
886 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
887 sizeof(rpdu), &rpdu);
892 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
893 * @ignore: trace callback data parameter (not used)
894 * @q: queue the io is for
895 * @bio: the source bio
896 * @dev: target device
897 * @from: source sector
900 * Device mapper or raid target sometimes need to split a bio because
901 * it spans a stripe (or similar). Add a trace for that action.
904 static void blk_add_trace_bio_remap(void *ignore,
905 struct request_queue *q, struct bio *bio,
906 dev_t dev, sector_t from)
908 struct blk_trace *bt = q->blk_trace;
909 struct blk_io_trace_remap r;
914 r.device_from = cpu_to_be32(dev);
915 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
916 r.sector_from = cpu_to_be64(from);
918 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
919 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
924 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
925 * @ignore: trace callback data parameter (not used)
926 * @q: queue the io is for
927 * @rq: the source request
928 * @dev: target device
929 * @from: source sector
932 * Device mapper remaps request to other devices.
933 * Add a trace for that action.
936 static void blk_add_trace_rq_remap(void *ignore,
937 struct request_queue *q,
938 struct request *rq, dev_t dev,
941 struct blk_trace *bt = q->blk_trace;
942 struct blk_io_trace_remap r;
947 r.device_from = cpu_to_be32(dev);
948 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
949 r.sector_from = cpu_to_be64(from);
951 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
952 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
957 * blk_add_driver_data - Add binary message with driver-specific data
958 * @q: queue the io is for
960 * @data: driver-specific data
961 * @len: length of driver-specific data
964 * Some drivers might want to write driver-specific data per request.
967 void blk_add_driver_data(struct request_queue *q,
969 void *data, size_t len)
971 struct blk_trace *bt = q->blk_trace;
976 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
977 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
978 BLK_TA_DRV_DATA, rq->errors, len, data);
980 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
981 BLK_TA_DRV_DATA, rq->errors, len, data);
983 EXPORT_SYMBOL_GPL(blk_add_driver_data);
985 static void blk_register_tracepoints(void)
989 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
991 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
993 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
995 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
997 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
999 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1001 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1003 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1005 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1007 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1009 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1011 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1013 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1015 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1017 ret = register_trace_block_split(blk_add_trace_split, NULL);
1019 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1021 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1025 static void blk_unregister_tracepoints(void)
1027 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1028 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1029 unregister_trace_block_split(blk_add_trace_split, NULL);
1030 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1031 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1032 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1033 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1034 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1035 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1036 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1037 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1038 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1039 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1040 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1041 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1042 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1043 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1045 tracepoint_synchronize_unregister();
1049 * struct blk_io_tracer formatting routines
1052 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1055 int tc = t->action >> BLK_TC_SHIFT;
1057 if (t->action == BLK_TN_MESSAGE) {
1062 if (tc & BLK_TC_FLUSH)
1065 if (tc & BLK_TC_DISCARD)
1067 else if (tc & BLK_TC_WRITE)
1074 if (tc & BLK_TC_FUA)
1076 if (tc & BLK_TC_AHEAD)
1078 if (tc & BLK_TC_SYNC)
1080 if (tc & BLK_TC_META)
1087 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1089 return (const struct blk_io_trace *)ent;
1092 static inline const void *pdu_start(const struct trace_entry *ent)
1094 return te_blk_io_trace(ent) + 1;
1097 static inline u32 t_action(const struct trace_entry *ent)
1099 return te_blk_io_trace(ent)->action;
1102 static inline u32 t_bytes(const struct trace_entry *ent)
1104 return te_blk_io_trace(ent)->bytes;
1107 static inline u32 t_sec(const struct trace_entry *ent)
1109 return te_blk_io_trace(ent)->bytes >> 9;
1112 static inline unsigned long long t_sector(const struct trace_entry *ent)
1114 return te_blk_io_trace(ent)->sector;
1117 static inline __u16 t_error(const struct trace_entry *ent)
1119 return te_blk_io_trace(ent)->error;
1122 static __u64 get_pdu_int(const struct trace_entry *ent)
1124 const __u64 *val = pdu_start(ent);
1125 return be64_to_cpu(*val);
1128 static void get_pdu_remap(const struct trace_entry *ent,
1129 struct blk_io_trace_remap *r)
1131 const struct blk_io_trace_remap *__r = pdu_start(ent);
1132 __u64 sector_from = __r->sector_from;
1134 r->device_from = be32_to_cpu(__r->device_from);
1135 r->device_to = be32_to_cpu(__r->device_to);
1136 r->sector_from = be64_to_cpu(sector_from);
1139 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1141 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1143 char rwbs[RWBS_LEN];
1144 unsigned long long ts = iter->ts;
1145 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1146 unsigned secs = (unsigned long)ts;
1147 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1151 return trace_seq_printf(&iter->seq,
1152 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1153 MAJOR(t->device), MINOR(t->device), iter->cpu,
1154 secs, nsec_rem, iter->ent->pid, act, rwbs);
1157 static int blk_log_action(struct trace_iterator *iter, const char *act)
1159 char rwbs[RWBS_LEN];
1160 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1163 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1164 MAJOR(t->device), MINOR(t->device), act, rwbs);
1167 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1169 const unsigned char *pdu_buf;
1173 pdu_buf = pdu_start(ent);
1174 pdu_len = te_blk_io_trace(ent)->pdu_len;
1179 /* find the last zero that needs to be printed */
1180 for (end = pdu_len - 1; end >= 0; end--)
1185 if (!trace_seq_putc(s, '('))
1188 for (i = 0; i < pdu_len; i++) {
1190 ret = trace_seq_printf(s, "%s%02x",
1191 i == 0 ? "" : " ", pdu_buf[i]);
1196 * stop when the rest is just zeroes and indicate so
1197 * with a ".." appended
1199 if (i == end && end != pdu_len - 1)
1200 return trace_seq_puts(s, " ..) ");
1203 return trace_seq_puts(s, ") ");
1206 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1208 char cmd[TASK_COMM_LEN];
1210 trace_find_cmdline(ent->pid, cmd);
1212 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1215 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1218 ret = blk_log_dump_pdu(s, ent);
1221 return trace_seq_printf(s, "[%s]\n", cmd);
1224 return trace_seq_printf(s, "%llu + %u [%s]\n",
1225 t_sector(ent), t_sec(ent), cmd);
1226 return trace_seq_printf(s, "[%s]\n", cmd);
1230 static int blk_log_with_error(struct trace_seq *s,
1231 const struct trace_entry *ent)
1233 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1236 ret = blk_log_dump_pdu(s, ent);
1238 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1242 return trace_seq_printf(s, "%llu + %u [%d]\n",
1244 t_sec(ent), t_error(ent));
1245 return trace_seq_printf(s, "%llu [%d]\n",
1246 t_sector(ent), t_error(ent));
1250 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1252 struct blk_io_trace_remap r = { .device_from = 0, };
1254 get_pdu_remap(ent, &r);
1255 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1256 t_sector(ent), t_sec(ent),
1257 MAJOR(r.device_from), MINOR(r.device_from),
1258 (unsigned long long)r.sector_from);
1261 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1263 char cmd[TASK_COMM_LEN];
1265 trace_find_cmdline(ent->pid, cmd);
1267 return trace_seq_printf(s, "[%s]\n", cmd);
1270 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1272 char cmd[TASK_COMM_LEN];
1274 trace_find_cmdline(ent->pid, cmd);
1276 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1279 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1281 char cmd[TASK_COMM_LEN];
1283 trace_find_cmdline(ent->pid, cmd);
1285 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1286 get_pdu_int(ent), cmd);
1289 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1292 const struct blk_io_trace *t = te_blk_io_trace(ent);
1294 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1296 return trace_seq_putc(s, '\n');
1301 * struct tracer operations
1304 static void blk_tracer_print_header(struct seq_file *m)
1306 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1308 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1312 static void blk_tracer_start(struct trace_array *tr)
1314 blk_tracer_enabled = true;
1317 static int blk_tracer_init(struct trace_array *tr)
1320 blk_tracer_start(tr);
1324 static void blk_tracer_stop(struct trace_array *tr)
1326 blk_tracer_enabled = false;
1329 static void blk_tracer_reset(struct trace_array *tr)
1331 blk_tracer_stop(tr);
1334 static const struct {
1336 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1338 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1339 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1340 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1341 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1342 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1343 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1344 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1345 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1346 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1347 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1348 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1349 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1350 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1351 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1352 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1355 static enum print_line_t print_one_line(struct trace_iterator *iter,
1358 struct trace_seq *s = &iter->seq;
1359 const struct blk_io_trace *t;
1363 blk_log_action_t *log_action;
1365 t = te_blk_io_trace(iter->ent);
1366 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1367 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1368 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1370 if (t->action == BLK_TN_MESSAGE) {
1371 ret = log_action(iter, long_act ? "message" : "m");
1373 ret = blk_log_msg(s, iter->ent);
1377 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1378 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1380 ret = log_action(iter, what2act[what].act[long_act]);
1382 ret = what2act[what].print(s, iter->ent);
1385 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1388 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1389 int flags, struct trace_event *event)
1391 return print_one_line(iter, false);
1394 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1396 struct trace_seq *s = &iter->seq;
1397 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1398 const int offset = offsetof(struct blk_io_trace, sector);
1399 struct blk_io_trace old = {
1400 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1404 if (!trace_seq_putmem(s, &old, offset))
1406 return trace_seq_putmem(s, &t->sector,
1407 sizeof(old) - offset + t->pdu_len);
1410 static enum print_line_t
1411 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1412 struct trace_event *event)
1414 return blk_trace_synthesize_old_trace(iter) ?
1415 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1418 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1420 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1421 return TRACE_TYPE_UNHANDLED;
1423 return print_one_line(iter, true);
1426 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1428 /* don't output context-info for blk_classic output */
1429 if (bit == TRACE_BLK_OPT_CLASSIC) {
1431 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1433 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1438 static struct tracer blk_tracer __read_mostly = {
1440 .init = blk_tracer_init,
1441 .reset = blk_tracer_reset,
1442 .start = blk_tracer_start,
1443 .stop = blk_tracer_stop,
1444 .print_header = blk_tracer_print_header,
1445 .print_line = blk_tracer_print_line,
1446 .flags = &blk_tracer_flags,
1447 .set_flag = blk_tracer_set_flag,
1450 static struct trace_event_functions trace_blk_event_funcs = {
1451 .trace = blk_trace_event_print,
1452 .binary = blk_trace_event_print_binary,
1455 static struct trace_event trace_blk_event = {
1457 .funcs = &trace_blk_event_funcs,
1460 static int __init init_blk_tracer(void)
1462 if (!register_ftrace_event(&trace_blk_event)) {
1463 pr_warning("Warning: could not register block events\n");
1467 if (register_tracer(&blk_tracer) != 0) {
1468 pr_warning("Warning: could not register the block tracer\n");
1469 unregister_ftrace_event(&trace_blk_event);
1476 device_initcall(init_blk_tracer);
1478 static int blk_trace_remove_queue(struct request_queue *q)
1480 struct blk_trace *bt;
1482 bt = xchg(&q->blk_trace, NULL);
1486 if (atomic_dec_and_test(&blk_probes_ref))
1487 blk_unregister_tracepoints();
1494 * Setup everything required to start tracing
1496 static int blk_trace_setup_queue(struct request_queue *q,
1497 struct block_device *bdev)
1499 struct blk_trace *old_bt, *bt = NULL;
1502 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1506 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1510 bt->dev = bdev->bd_dev;
1511 bt->act_mask = (u16)-1;
1513 blk_trace_setup_lba(bt, bdev);
1515 old_bt = xchg(&q->blk_trace, bt);
1516 if (old_bt != NULL) {
1517 (void)xchg(&q->blk_trace, old_bt);
1522 if (atomic_inc_return(&blk_probes_ref) == 1)
1523 blk_register_tracepoints();
1532 * sysfs interface to enable and configure tracing
1535 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1536 struct device_attribute *attr,
1538 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1539 struct device_attribute *attr,
1540 const char *buf, size_t count);
1541 #define BLK_TRACE_DEVICE_ATTR(_name) \
1542 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1543 sysfs_blk_trace_attr_show, \
1544 sysfs_blk_trace_attr_store)
1546 static BLK_TRACE_DEVICE_ATTR(enable);
1547 static BLK_TRACE_DEVICE_ATTR(act_mask);
1548 static BLK_TRACE_DEVICE_ATTR(pid);
1549 static BLK_TRACE_DEVICE_ATTR(start_lba);
1550 static BLK_TRACE_DEVICE_ATTR(end_lba);
1552 static struct attribute *blk_trace_attrs[] = {
1553 &dev_attr_enable.attr,
1554 &dev_attr_act_mask.attr,
1556 &dev_attr_start_lba.attr,
1557 &dev_attr_end_lba.attr,
1561 struct attribute_group blk_trace_attr_group = {
1563 .attrs = blk_trace_attrs,
1566 static const struct {
1570 { BLK_TC_READ, "read" },
1571 { BLK_TC_WRITE, "write" },
1572 { BLK_TC_FLUSH, "flush" },
1573 { BLK_TC_SYNC, "sync" },
1574 { BLK_TC_QUEUE, "queue" },
1575 { BLK_TC_REQUEUE, "requeue" },
1576 { BLK_TC_ISSUE, "issue" },
1577 { BLK_TC_COMPLETE, "complete" },
1578 { BLK_TC_FS, "fs" },
1579 { BLK_TC_PC, "pc" },
1580 { BLK_TC_AHEAD, "ahead" },
1581 { BLK_TC_META, "meta" },
1582 { BLK_TC_DISCARD, "discard" },
1583 { BLK_TC_DRV_DATA, "drv_data" },
1584 { BLK_TC_FUA, "fua" },
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;
1655 bdev = bdget(part_devt(p));
1659 q = blk_trace_get_queue(bdev);
1663 mutex_lock(&bdev->bd_mutex);
1665 if (attr == &dev_attr_enable) {
1666 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1667 goto out_unlock_bdev;
1670 if (q->blk_trace == NULL)
1671 ret = sprintf(buf, "disabled\n");
1672 else if (attr == &dev_attr_act_mask)
1673 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1674 else if (attr == &dev_attr_pid)
1675 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1676 else if (attr == &dev_attr_start_lba)
1677 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1678 else if (attr == &dev_attr_end_lba)
1679 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1682 mutex_unlock(&bdev->bd_mutex);
1689 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1690 struct device_attribute *attr,
1691 const char *buf, size_t count)
1693 struct block_device *bdev;
1694 struct request_queue *q;
1695 struct hd_struct *p;
1697 ssize_t ret = -EINVAL;
1702 if (attr == &dev_attr_act_mask) {
1703 if (sscanf(buf, "%llx", &value) != 1) {
1704 /* Assume it is a list of trace category names */
1705 ret = blk_trace_str2mask(buf);
1710 } else if (sscanf(buf, "%llu", &value) != 1)
1715 p = dev_to_part(dev);
1716 bdev = bdget(part_devt(p));
1720 q = blk_trace_get_queue(bdev);
1724 mutex_lock(&bdev->bd_mutex);
1726 if (attr == &dev_attr_enable) {
1728 ret = blk_trace_setup_queue(q, bdev);
1730 ret = blk_trace_remove_queue(q);
1731 goto out_unlock_bdev;
1735 if (q->blk_trace == NULL)
1736 ret = blk_trace_setup_queue(q, bdev);
1739 if (attr == &dev_attr_act_mask)
1740 q->blk_trace->act_mask = value;
1741 else if (attr == &dev_attr_pid)
1742 q->blk_trace->pid = value;
1743 else if (attr == &dev_attr_start_lba)
1744 q->blk_trace->start_lba = value;
1745 else if (attr == &dev_attr_end_lba)
1746 q->blk_trace->end_lba = value;
1750 mutex_unlock(&bdev->bd_mutex);
1754 return ret ? ret : count;
1757 int blk_trace_init_sysfs(struct device *dev)
1759 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1762 void blk_trace_remove_sysfs(struct device *dev)
1764 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1767 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1769 #ifdef CONFIG_EVENT_TRACING
1771 void blk_dump_cmd(char *buf, struct request *rq)
1774 int len = rq->cmd_len;
1775 unsigned char *cmd = rq->cmd;
1777 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1782 for (end = len - 1; end >= 0; end--)
1787 for (i = 0; i < len; i++) {
1788 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1789 if (i == end && end != len - 1) {
1790 sprintf(buf, " ..");
1796 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1805 else if (rw & REQ_DISCARD)
1814 if (rw & REQ_RAHEAD)
1820 if (rw & REQ_SECURE)
1826 #endif /* CONFIG_EVENT_TRACING */