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/debugfs.h>
25 #include <linux/time.h>
26 #include <trace/block.h>
27 #include <linux/uaccess.h>
28 #include "trace_output.h"
30 static unsigned int blktrace_seq __read_mostly = 1;
32 static struct trace_array *blk_tr;
33 static bool blk_tracer_enabled __read_mostly;
35 /* Select an alternative, minimalistic output than the original one */
36 #define TRACE_BLK_OPT_CLASSIC 0x1
38 static struct tracer_opt blk_tracer_opts[] = {
39 /* Default disable the minimalistic output */
40 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
44 static struct tracer_flags blk_tracer_flags = {
46 .opts = blk_tracer_opts,
49 /* Global reference count of probes */
50 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
52 static void blk_register_tracepoints(void);
53 static void blk_unregister_tracepoints(void);
56 * Send out a notify message.
58 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
59 const void *data, size_t len)
61 struct blk_io_trace *t;
66 t = relay_reserve(bt->rchan, sizeof(*t) + len);
68 const int cpu = smp_processor_id();
70 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
71 t->time = ktime_to_ns(ktime_get());
77 memcpy((void *) t + sizeof(*t), data, len);
82 * Send out a notify for this process, if we haven't done so since a trace
85 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
87 tsk->btrace_seq = blktrace_seq;
88 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
91 static void trace_note_time(struct blk_trace *bt)
98 words[0] = now.tv_sec;
99 words[1] = now.tv_nsec;
101 local_irq_save(flags);
102 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
103 local_irq_restore(flags);
106 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
115 ftrace_vprintk(fmt, args);
123 local_irq_save(flags);
124 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
126 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
129 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
130 local_irq_restore(flags);
132 EXPORT_SYMBOL_GPL(__trace_note_message);
134 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
137 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
139 if (sector < bt->start_lba || sector > bt->end_lba)
141 if (bt->pid && pid != bt->pid)
148 * Data direction bit lookup
150 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
151 BLK_TC_ACT(BLK_TC_WRITE) };
153 /* The ilog2() calls fall out because they're constant */
154 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
155 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
158 * The worker for the various blk_add_trace*() types. Fills out a
159 * blk_io_trace structure and places it in a per-cpu subbuffer.
161 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
162 int rw, u32 what, int error, int pdu_len, void *pdu_data)
164 struct task_struct *tsk = current;
165 struct ring_buffer_event *event = NULL;
166 struct blk_io_trace *t;
167 unsigned long flags = 0;
168 unsigned long *sequence;
172 if (unlikely(bt->trace_state != Blktrace_running ||
173 !blk_tracer_enabled))
176 what |= ddir_act[rw & WRITE];
177 what |= MASK_TC_BIT(rw, BARRIER);
178 what |= MASK_TC_BIT(rw, SYNCIO);
179 what |= MASK_TC_BIT(rw, AHEAD);
180 what |= MASK_TC_BIT(rw, META);
181 what |= MASK_TC_BIT(rw, DISCARD);
184 if (unlikely(act_log_check(bt, what, sector, pid)))
186 cpu = raw_smp_processor_id();
189 tracing_record_cmdline(current);
191 pc = preempt_count();
192 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
193 sizeof(*t) + pdu_len,
197 t = ring_buffer_event_data(event);
202 * A word about the locking here - we disable interrupts to reserve
203 * some space in the relay per-cpu buffer, to prevent an irq
204 * from coming in and stepping on our toes.
206 local_irq_save(flags);
208 if (unlikely(tsk->btrace_seq != blktrace_seq))
209 trace_note_tsk(bt, tsk);
211 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
213 sequence = per_cpu_ptr(bt->sequence, cpu);
215 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
216 t->sequence = ++(*sequence);
217 t->time = ktime_to_ns(ktime_get());
220 * These two are not needed in ftrace as they are in the
221 * generic trace_entry, filled by tracing_generic_entry_update,
222 * but for the trace_event->bin() synthesizer benefit we do it
233 t->pdu_len = pdu_len;
236 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
239 trace_buffer_unlock_commit(blk_tr, event, 0, pc);
244 local_irq_restore(flags);
247 static struct dentry *blk_tree_root;
248 static DEFINE_MUTEX(blk_tree_mutex);
250 static void blk_trace_cleanup(struct blk_trace *bt)
252 debugfs_remove(bt->msg_file);
253 debugfs_remove(bt->dropped_file);
254 relay_close(bt->rchan);
255 free_percpu(bt->sequence);
256 free_percpu(bt->msg_data);
258 if (atomic_dec_and_test(&blk_probes_ref))
259 blk_unregister_tracepoints();
262 int blk_trace_remove(struct request_queue *q)
264 struct blk_trace *bt;
266 bt = xchg(&q->blk_trace, NULL);
270 if (bt->trace_state == Blktrace_setup ||
271 bt->trace_state == Blktrace_stopped)
272 blk_trace_cleanup(bt);
276 EXPORT_SYMBOL_GPL(blk_trace_remove);
278 static int blk_dropped_open(struct inode *inode, struct file *filp)
280 filp->private_data = inode->i_private;
285 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
286 size_t count, loff_t *ppos)
288 struct blk_trace *bt = filp->private_data;
291 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
293 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
296 static const struct file_operations blk_dropped_fops = {
297 .owner = THIS_MODULE,
298 .open = blk_dropped_open,
299 .read = blk_dropped_read,
302 static int blk_msg_open(struct inode *inode, struct file *filp)
304 filp->private_data = inode->i_private;
309 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
310 size_t count, loff_t *ppos)
313 struct blk_trace *bt;
315 if (count > BLK_TN_MAX_MSG)
318 msg = kmalloc(count, GFP_KERNEL);
322 if (copy_from_user(msg, buffer, count)) {
327 bt = filp->private_data;
328 __trace_note_message(bt, "%s", msg);
334 static const struct file_operations blk_msg_fops = {
335 .owner = THIS_MODULE,
336 .open = blk_msg_open,
337 .write = blk_msg_write,
341 * Keep track of how many times we encountered a full subbuffer, to aid
342 * the user space app in telling how many lost events there were.
344 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
345 void *prev_subbuf, size_t prev_padding)
347 struct blk_trace *bt;
349 if (!relay_buf_full(buf))
352 bt = buf->chan->private_data;
353 atomic_inc(&bt->dropped);
357 static int blk_remove_buf_file_callback(struct dentry *dentry)
359 struct dentry *parent = dentry->d_parent;
360 debugfs_remove(dentry);
363 * this will fail for all but the last file, but that is ok. what we
364 * care about is the top level buts->name directory going away, when
365 * the last trace file is gone. Then we don't have to rmdir() that
366 * manually on trace stop, so it nicely solves the issue with
367 * force killing of running traces.
370 debugfs_remove(parent);
374 static struct dentry *blk_create_buf_file_callback(const char *filename,
375 struct dentry *parent,
377 struct rchan_buf *buf,
380 return debugfs_create_file(filename, mode, parent, buf,
381 &relay_file_operations);
384 static struct rchan_callbacks blk_relay_callbacks = {
385 .subbuf_start = blk_subbuf_start_callback,
386 .create_buf_file = blk_create_buf_file_callback,
387 .remove_buf_file = blk_remove_buf_file_callback,
391 * Setup everything required to start tracing
393 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
394 struct blk_user_trace_setup *buts)
396 struct blk_trace *old_bt, *bt = NULL;
397 struct dentry *dir = NULL;
400 if (!buts->buf_size || !buts->buf_nr)
403 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
404 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
407 * some device names have larger paths - convert the slashes
408 * to underscores for this to work as expected
410 for (i = 0; i < strlen(buts->name); i++)
411 if (buts->name[i] == '/')
415 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
419 bt->sequence = alloc_percpu(unsigned long);
423 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
429 if (!blk_tree_root) {
430 blk_tree_root = debugfs_create_dir("block", NULL);
435 dir = debugfs_create_dir(buts->name, blk_tree_root);
442 atomic_set(&bt->dropped, 0);
445 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
447 if (!bt->dropped_file)
450 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
454 bt->rchan = relay_open("trace", dir, buts->buf_size,
455 buts->buf_nr, &blk_relay_callbacks, bt);
459 bt->act_mask = buts->act_mask;
461 bt->act_mask = (u16) -1;
463 bt->start_lba = buts->start_lba;
464 bt->end_lba = buts->end_lba;
469 bt->trace_state = Blktrace_setup;
472 old_bt = xchg(&q->blk_trace, bt);
474 (void) xchg(&q->blk_trace, old_bt);
478 if (atomic_add_return(1, &blk_probes_ref) == 1)
479 blk_register_tracepoints();
485 debugfs_remove(bt->msg_file);
486 if (bt->dropped_file)
487 debugfs_remove(bt->dropped_file);
488 free_percpu(bt->sequence);
489 free_percpu(bt->msg_data);
491 relay_close(bt->rchan);
497 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
500 struct blk_user_trace_setup buts;
503 ret = copy_from_user(&buts, arg, sizeof(buts));
507 ret = do_blk_trace_setup(q, name, dev, &buts);
511 if (copy_to_user(arg, &buts, sizeof(buts)))
516 EXPORT_SYMBOL_GPL(blk_trace_setup);
518 int blk_trace_startstop(struct request_queue *q, int start)
521 struct blk_trace *bt = q->blk_trace;
527 * For starting a trace, we can transition from a setup or stopped
528 * trace. For stopping a trace, the state must be running
532 if (bt->trace_state == Blktrace_setup ||
533 bt->trace_state == Blktrace_stopped) {
536 bt->trace_state = Blktrace_running;
542 if (bt->trace_state == Blktrace_running) {
543 bt->trace_state = Blktrace_stopped;
544 relay_flush(bt->rchan);
551 EXPORT_SYMBOL_GPL(blk_trace_startstop);
554 * blk_trace_ioctl: - handle the ioctls associated with tracing
555 * @bdev: the block device
556 * @cmd: the ioctl cmd
557 * @arg: the argument data, if any
560 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
562 struct request_queue *q;
564 char b[BDEVNAME_SIZE];
566 q = bdev_get_queue(bdev);
570 mutex_lock(&bdev->bd_mutex);
575 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
580 ret = blk_trace_startstop(q, start);
582 case BLKTRACETEARDOWN:
583 ret = blk_trace_remove(q);
590 mutex_unlock(&bdev->bd_mutex);
595 * blk_trace_shutdown: - stop and cleanup trace structures
596 * @q: the request queue associated with the device
599 void blk_trace_shutdown(struct request_queue *q)
602 blk_trace_startstop(q, 0);
612 * blk_add_trace_rq - Add a trace for a request oriented action
613 * @q: queue the io is for
614 * @rq: the source request
618 * Records an action against a request. Will log the bio offset + size.
621 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
624 struct blk_trace *bt = q->blk_trace;
625 int rw = rq->cmd_flags & 0x03;
630 if (blk_discard_rq(rq))
631 rw |= (1 << BIO_RW_DISCARD);
633 if (blk_pc_request(rq)) {
634 what |= BLK_TC_ACT(BLK_TC_PC);
635 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
636 sizeof(rq->cmd), rq->cmd);
638 what |= BLK_TC_ACT(BLK_TC_FS);
639 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
640 rw, what, rq->errors, 0, NULL);
644 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
646 blk_add_trace_rq(q, rq, BLK_TA_ABORT);
649 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
651 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
654 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
656 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
659 static void blk_add_trace_rq_requeue(struct request_queue *q,
662 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
665 static void blk_add_trace_rq_complete(struct request_queue *q,
668 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
672 * blk_add_trace_bio - Add a trace for a bio oriented action
673 * @q: queue the io is for
674 * @bio: the source bio
678 * Records an action against a bio. Will log the bio offset + size.
681 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
684 struct blk_trace *bt = q->blk_trace;
689 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
690 !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
693 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
695 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
698 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
700 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
703 static void blk_add_trace_bio_backmerge(struct request_queue *q,
706 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
709 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
712 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
715 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
717 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
720 static void blk_add_trace_getrq(struct request_queue *q,
721 struct bio *bio, int rw)
724 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
726 struct blk_trace *bt = q->blk_trace;
729 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
734 static void blk_add_trace_sleeprq(struct request_queue *q,
735 struct bio *bio, int rw)
738 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
740 struct blk_trace *bt = q->blk_trace;
743 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
748 static void blk_add_trace_plug(struct request_queue *q)
750 struct blk_trace *bt = q->blk_trace;
753 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
756 static void blk_add_trace_unplug_io(struct request_queue *q)
758 struct blk_trace *bt = q->blk_trace;
761 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
762 __be64 rpdu = cpu_to_be64(pdu);
764 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
765 sizeof(rpdu), &rpdu);
769 static void blk_add_trace_unplug_timer(struct request_queue *q)
771 struct blk_trace *bt = q->blk_trace;
774 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
775 __be64 rpdu = cpu_to_be64(pdu);
777 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
778 sizeof(rpdu), &rpdu);
782 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
785 struct blk_trace *bt = q->blk_trace;
788 __be64 rpdu = cpu_to_be64(pdu);
790 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
791 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
792 sizeof(rpdu), &rpdu);
797 * blk_add_trace_remap - Add a trace for a remap operation
798 * @q: queue the io is for
799 * @bio: the source bio
800 * @dev: target device
801 * @from: source sector
805 * Device mapper or raid target sometimes need to split a bio because
806 * it spans a stripe (or similar). Add a trace for that action.
809 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
810 dev_t dev, sector_t from, sector_t to)
812 struct blk_trace *bt = q->blk_trace;
813 struct blk_io_trace_remap r;
818 r.device = cpu_to_be32(dev);
819 r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
820 r.sector = cpu_to_be64(to);
822 __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
823 !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
827 * blk_add_driver_data - Add binary message with driver-specific data
828 * @q: queue the io is for
830 * @data: driver-specific data
831 * @len: length of driver-specific data
834 * Some drivers might want to write driver-specific data per request.
837 void blk_add_driver_data(struct request_queue *q,
839 void *data, size_t len)
841 struct blk_trace *bt = q->blk_trace;
846 if (blk_pc_request(rq))
847 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
848 rq->errors, len, data);
850 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
851 0, BLK_TA_DRV_DATA, rq->errors, len, data);
853 EXPORT_SYMBOL_GPL(blk_add_driver_data);
855 static void blk_register_tracepoints(void)
859 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
861 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
863 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
865 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
867 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
869 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
871 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
873 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
875 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
877 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
879 ret = register_trace_block_getrq(blk_add_trace_getrq);
881 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
883 ret = register_trace_block_plug(blk_add_trace_plug);
885 ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
887 ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
889 ret = register_trace_block_split(blk_add_trace_split);
891 ret = register_trace_block_remap(blk_add_trace_remap);
895 static void blk_unregister_tracepoints(void)
897 unregister_trace_block_remap(blk_add_trace_remap);
898 unregister_trace_block_split(blk_add_trace_split);
899 unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
900 unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
901 unregister_trace_block_plug(blk_add_trace_plug);
902 unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
903 unregister_trace_block_getrq(blk_add_trace_getrq);
904 unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
905 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
906 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
907 unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
908 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
909 unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
910 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
911 unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
912 unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
913 unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
915 tracepoint_synchronize_unregister();
919 * struct blk_io_tracer formatting routines
922 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
925 int tc = t->action >> BLK_TC_SHIFT;
927 if (tc & BLK_TC_DISCARD)
929 else if (tc & BLK_TC_WRITE)
936 if (tc & BLK_TC_AHEAD)
938 if (tc & BLK_TC_BARRIER)
940 if (tc & BLK_TC_SYNC)
942 if (tc & BLK_TC_META)
949 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
951 return (const struct blk_io_trace *)ent;
954 static inline const void *pdu_start(const struct trace_entry *ent)
956 return te_blk_io_trace(ent) + 1;
959 static inline u32 t_sec(const struct trace_entry *ent)
961 return te_blk_io_trace(ent)->bytes >> 9;
964 static inline unsigned long long t_sector(const struct trace_entry *ent)
966 return te_blk_io_trace(ent)->sector;
969 static inline __u16 t_error(const struct trace_entry *ent)
971 return te_blk_io_trace(ent)->error;
974 static __u64 get_pdu_int(const struct trace_entry *ent)
976 const __u64 *val = pdu_start(ent);
977 return be64_to_cpu(*val);
980 static void get_pdu_remap(const struct trace_entry *ent,
981 struct blk_io_trace_remap *r)
983 const struct blk_io_trace_remap *__r = pdu_start(ent);
984 __u64 sector = __r->sector;
986 r->device = be32_to_cpu(__r->device);
987 r->device_from = be32_to_cpu(__r->device_from);
988 r->sector = be64_to_cpu(sector);
991 static int blk_log_action_iter(struct trace_iterator *iter, const char *act)
994 unsigned long long ts = ns2usecs(iter->ts);
995 unsigned long usec_rem = do_div(ts, USEC_PER_SEC);
996 unsigned secs = (unsigned long)ts;
997 const struct trace_entry *ent = iter->ent;
998 const struct blk_io_trace *t = (const struct blk_io_trace *)ent;
1002 return trace_seq_printf(&iter->seq,
1003 "%3d,%-3d %2d %5d.%06lu %5u %2s %3s ",
1004 MAJOR(t->device), MINOR(t->device), iter->cpu,
1005 secs, usec_rem, ent->pid, act, rwbs);
1008 static int blk_log_action_seq(struct trace_seq *s, const struct blk_io_trace *t,
1013 return trace_seq_printf(s, "%3d,%-3d %2s %3s ",
1014 MAJOR(t->device), MINOR(t->device), act, rwbs);
1017 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1019 char cmd[TASK_COMM_LEN];
1021 trace_find_cmdline(ent->pid, cmd);
1024 return trace_seq_printf(s, "%llu + %u [%s]\n",
1025 t_sector(ent), t_sec(ent), cmd);
1026 return trace_seq_printf(s, "[%s]\n", cmd);
1029 static int blk_log_with_error(struct trace_seq *s,
1030 const struct trace_entry *ent)
1033 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1034 t_sec(ent), t_error(ent));
1035 return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1038 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1040 struct blk_io_trace_remap r = { .device = 0, };
1042 get_pdu_remap(ent, &r);
1043 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1045 t_sec(ent), MAJOR(r.device), MINOR(r.device),
1046 (unsigned long long)r.sector);
1049 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1051 char cmd[TASK_COMM_LEN];
1053 trace_find_cmdline(ent->pid, cmd);
1055 return trace_seq_printf(s, "[%s]\n", cmd);
1058 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1060 char cmd[TASK_COMM_LEN];
1062 trace_find_cmdline(ent->pid, cmd);
1064 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1067 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1069 char cmd[TASK_COMM_LEN];
1071 trace_find_cmdline(ent->pid, cmd);
1073 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1074 get_pdu_int(ent), cmd);
1078 * struct tracer operations
1081 static void blk_tracer_print_header(struct seq_file *m)
1083 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1085 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1089 static void blk_tracer_start(struct trace_array *tr)
1091 if (atomic_add_return(1, &blk_probes_ref) == 1)
1092 blk_register_tracepoints();
1093 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1096 static int blk_tracer_init(struct trace_array *tr)
1099 blk_tracer_start(tr);
1100 blk_tracer_enabled = true;
1104 static void blk_tracer_stop(struct trace_array *tr)
1106 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1107 if (atomic_dec_and_test(&blk_probes_ref))
1108 blk_unregister_tracepoints();
1111 static void blk_tracer_reset(struct trace_array *tr)
1113 if (!atomic_read(&blk_probes_ref))
1116 blk_tracer_enabled = false;
1117 blk_tracer_stop(tr);
1120 static const struct {
1122 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1124 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1125 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1126 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1127 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1128 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1129 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1130 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1131 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1132 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1133 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1134 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1135 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1136 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1137 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1138 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1141 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1144 struct trace_seq *s = &iter->seq;
1145 const struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1146 const u16 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1149 if (!trace_print_context(iter))
1150 return TRACE_TYPE_PARTIAL_LINE;
1152 if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1153 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1155 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1156 ret = blk_log_action_seq(s, t, what2act[what].act[long_act]);
1158 ret = what2act[what].print(s, iter->ent);
1161 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1164 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1166 struct trace_seq *s = &iter->seq;
1167 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1168 const int offset = offsetof(struct blk_io_trace, sector);
1169 struct blk_io_trace old = {
1170 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1171 .time = ns2usecs(iter->ts),
1174 if (!trace_seq_putmem(s, &old, offset))
1176 return trace_seq_putmem(s, &t->sector,
1177 sizeof(old) - offset + t->pdu_len);
1180 static enum print_line_t
1181 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1183 return blk_trace_synthesize_old_trace(iter) ?
1184 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1187 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1189 const struct blk_io_trace *t;
1193 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1194 return TRACE_TYPE_UNHANDLED;
1196 t = (const struct blk_io_trace *)iter->ent;
1197 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1199 if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1200 ret = trace_seq_printf(&iter->seq, "Bad pc action %x\n", what);
1202 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1203 ret = blk_log_action_iter(iter, what2act[what].act[long_act]);
1205 ret = what2act[what].print(&iter->seq, iter->ent);
1208 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1211 static struct tracer blk_tracer __read_mostly = {
1213 .init = blk_tracer_init,
1214 .reset = blk_tracer_reset,
1215 .start = blk_tracer_start,
1216 .stop = blk_tracer_stop,
1217 .print_header = blk_tracer_print_header,
1218 .print_line = blk_tracer_print_line,
1219 .flags = &blk_tracer_flags,
1222 static struct trace_event trace_blk_event = {
1224 .trace = blk_trace_event_print,
1225 .binary = blk_trace_event_print_binary,
1228 static int __init init_blk_tracer(void)
1230 if (!register_ftrace_event(&trace_blk_event)) {
1231 pr_warning("Warning: could not register block events\n");
1235 if (register_tracer(&blk_tracer) != 0) {
1236 pr_warning("Warning: could not register the block tracer\n");
1237 unregister_ftrace_event(&trace_blk_event);
1244 device_initcall(init_blk_tracer);
1246 static int blk_trace_remove_queue(struct request_queue *q)
1248 struct blk_trace *bt;
1250 bt = xchg(&q->blk_trace, NULL);
1259 * Setup everything required to start tracing
1261 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1263 struct blk_trace *old_bt, *bt = NULL;
1265 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1270 bt->act_mask = (u16)-1;
1271 bt->end_lba = -1ULL;
1272 bt->trace_state = Blktrace_running;
1274 old_bt = xchg(&q->blk_trace, bt);
1275 if (old_bt != NULL) {
1276 (void)xchg(&q->blk_trace, old_bt);
1285 * sysfs interface to enable and configure tracing
1288 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1289 struct device_attribute *attr,
1291 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1292 struct device_attribute *attr,
1293 const char *buf, size_t count);
1294 #define BLK_TRACE_DEVICE_ATTR(_name) \
1295 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1296 sysfs_blk_trace_attr_show, \
1297 sysfs_blk_trace_attr_store)
1299 static BLK_TRACE_DEVICE_ATTR(enable);
1300 static BLK_TRACE_DEVICE_ATTR(act_mask);
1301 static BLK_TRACE_DEVICE_ATTR(pid);
1302 static BLK_TRACE_DEVICE_ATTR(start_lba);
1303 static BLK_TRACE_DEVICE_ATTR(end_lba);
1305 static struct attribute *blk_trace_attrs[] = {
1306 &dev_attr_enable.attr,
1307 &dev_attr_act_mask.attr,
1309 &dev_attr_start_lba.attr,
1310 &dev_attr_end_lba.attr,
1314 struct attribute_group blk_trace_attr_group = {
1316 .attrs = blk_trace_attrs,
1319 static const struct {
1323 { BLK_TC_READ, "read" },
1324 { BLK_TC_WRITE, "write" },
1325 { BLK_TC_BARRIER, "barrier" },
1326 { BLK_TC_SYNC, "sync" },
1327 { BLK_TC_QUEUE, "queue" },
1328 { BLK_TC_REQUEUE, "requeue" },
1329 { BLK_TC_ISSUE, "issue" },
1330 { BLK_TC_COMPLETE, "complete" },
1331 { BLK_TC_FS, "fs" },
1332 { BLK_TC_PC, "pc" },
1333 { BLK_TC_AHEAD, "ahead" },
1334 { BLK_TC_META, "meta" },
1335 { BLK_TC_DISCARD, "discard" },
1336 { BLK_TC_DRV_DATA, "drv_data" },
1339 static int blk_trace_str2mask(const char *str)
1345 s = kstrdup(str, GFP_KERNEL);
1351 token = strsep(&s, ",");
1358 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1359 if (strcasecmp(token, mask_maps[i].str) == 0) {
1360 mask |= mask_maps[i].mask;
1364 if (i == ARRAY_SIZE(mask_maps)) {
1374 static ssize_t blk_trace_mask2str(char *buf, int mask)
1379 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1380 if (mask & mask_maps[i].mask) {
1381 p += sprintf(p, "%s%s",
1382 (p == buf) ? "" : ",", mask_maps[i].str);
1390 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1392 if (bdev->bd_disk == NULL)
1395 return bdev_get_queue(bdev);
1398 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1399 struct device_attribute *attr,
1402 struct hd_struct *p = dev_to_part(dev);
1403 struct request_queue *q;
1404 struct block_device *bdev;
1405 ssize_t ret = -ENXIO;
1408 bdev = bdget(part_devt(p));
1410 goto out_unlock_kernel;
1412 q = blk_trace_get_queue(bdev);
1416 mutex_lock(&bdev->bd_mutex);
1418 if (attr == &dev_attr_enable) {
1419 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1420 goto out_unlock_bdev;
1423 if (q->blk_trace == NULL)
1424 ret = sprintf(buf, "disabled\n");
1425 else if (attr == &dev_attr_act_mask)
1426 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1427 else if (attr == &dev_attr_pid)
1428 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1429 else if (attr == &dev_attr_start_lba)
1430 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1431 else if (attr == &dev_attr_end_lba)
1432 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1435 mutex_unlock(&bdev->bd_mutex);
1443 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1444 struct device_attribute *attr,
1445 const char *buf, size_t count)
1447 struct block_device *bdev;
1448 struct request_queue *q;
1449 struct hd_struct *p;
1451 ssize_t ret = -EINVAL;
1456 if (attr == &dev_attr_act_mask) {
1457 if (sscanf(buf, "%llx", &value) != 1) {
1458 /* Assume it is a list of trace category names */
1459 ret = blk_trace_str2mask(buf);
1464 } else if (sscanf(buf, "%llu", &value) != 1)
1470 p = dev_to_part(dev);
1471 bdev = bdget(part_devt(p));
1473 goto out_unlock_kernel;
1475 q = blk_trace_get_queue(bdev);
1479 mutex_lock(&bdev->bd_mutex);
1481 if (attr == &dev_attr_enable) {
1483 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1485 ret = blk_trace_remove_queue(q);
1486 goto out_unlock_bdev;
1490 if (q->blk_trace == NULL)
1491 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1494 if (attr == &dev_attr_act_mask)
1495 q->blk_trace->act_mask = value;
1496 else if (attr == &dev_attr_pid)
1497 q->blk_trace->pid = value;
1498 else if (attr == &dev_attr_start_lba)
1499 q->blk_trace->start_lba = value;
1500 else if (attr == &dev_attr_end_lba)
1501 q->blk_trace->end_lba = value;
1505 mutex_unlock(&bdev->bd_mutex);
1511 return ret ? ret : count;