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/smp_lock.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))
141 local_irq_save(flags);
142 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
144 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
147 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
148 local_irq_restore(flags);
150 EXPORT_SYMBOL_GPL(__trace_note_message);
152 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
155 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
157 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
159 if (bt->pid && pid != bt->pid)
166 * Data direction bit lookup
168 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
169 BLK_TC_ACT(BLK_TC_WRITE) };
171 /* The ilog2() calls fall out because they're constant */
172 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
173 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
176 * The worker for the various blk_add_trace*() types. Fills out a
177 * blk_io_trace structure and places it in a per-cpu subbuffer.
179 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
180 int rw, u32 what, int error, int pdu_len, void *pdu_data)
182 struct task_struct *tsk = current;
183 struct ring_buffer_event *event = NULL;
184 struct ring_buffer *buffer = NULL;
185 struct blk_io_trace *t;
186 unsigned long flags = 0;
187 unsigned long *sequence;
190 bool blk_tracer = blk_tracer_enabled;
192 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
195 what |= ddir_act[rw & WRITE];
196 what |= MASK_TC_BIT(rw, BARRIER);
197 what |= MASK_TC_BIT(rw, SYNCIO);
198 what |= MASK_TC_BIT(rw, AHEAD);
199 what |= MASK_TC_BIT(rw, META);
200 what |= MASK_TC_BIT(rw, DISCARD);
203 if (act_log_check(bt, what, sector, pid))
205 cpu = raw_smp_processor_id();
208 tracing_record_cmdline(current);
210 buffer = blk_tr->buffer;
211 pc = preempt_count();
212 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
213 sizeof(*t) + pdu_len,
217 t = ring_buffer_event_data(event);
222 * A word about the locking here - we disable interrupts to reserve
223 * some space in the relay per-cpu buffer, to prevent an irq
224 * from coming in and stepping on our toes.
226 local_irq_save(flags);
228 if (unlikely(tsk->btrace_seq != blktrace_seq))
229 trace_note_tsk(bt, tsk);
231 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
233 sequence = per_cpu_ptr(bt->sequence, cpu);
235 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
236 t->sequence = ++(*sequence);
237 t->time = ktime_to_ns(ktime_get());
240 * These two are not needed in ftrace as they are in the
241 * generic trace_entry, filled by tracing_generic_entry_update,
242 * but for the trace_event->bin() synthesizer benefit we do it
253 t->pdu_len = pdu_len;
256 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
259 trace_buffer_unlock_commit(buffer, event, 0, pc);
264 local_irq_restore(flags);
267 static struct dentry *blk_tree_root;
268 static DEFINE_MUTEX(blk_tree_mutex);
270 static void blk_trace_free(struct blk_trace *bt)
272 debugfs_remove(bt->msg_file);
273 debugfs_remove(bt->dropped_file);
274 relay_close(bt->rchan);
275 debugfs_remove(bt->dir);
276 free_percpu(bt->sequence);
277 free_percpu(bt->msg_data);
281 static void blk_trace_cleanup(struct blk_trace *bt)
284 if (atomic_dec_and_test(&blk_probes_ref))
285 blk_unregister_tracepoints();
288 int blk_trace_remove(struct request_queue *q)
290 struct blk_trace *bt;
292 bt = xchg(&q->blk_trace, NULL);
296 if (bt->trace_state != Blktrace_running)
297 blk_trace_cleanup(bt);
301 EXPORT_SYMBOL_GPL(blk_trace_remove);
303 static int blk_dropped_open(struct inode *inode, struct file *filp)
305 filp->private_data = inode->i_private;
310 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
311 size_t count, loff_t *ppos)
313 struct blk_trace *bt = filp->private_data;
316 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
318 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
321 static const struct file_operations blk_dropped_fops = {
322 .owner = THIS_MODULE,
323 .open = blk_dropped_open,
324 .read = blk_dropped_read,
327 static int blk_msg_open(struct inode *inode, struct file *filp)
329 filp->private_data = inode->i_private;
334 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
335 size_t count, loff_t *ppos)
338 struct blk_trace *bt;
340 if (count >= BLK_TN_MAX_MSG)
343 msg = kmalloc(count + 1, GFP_KERNEL);
347 if (copy_from_user(msg, buffer, count)) {
353 bt = filp->private_data;
354 __trace_note_message(bt, "%s", msg);
360 static const struct file_operations blk_msg_fops = {
361 .owner = THIS_MODULE,
362 .open = blk_msg_open,
363 .write = blk_msg_write,
367 * Keep track of how many times we encountered a full subbuffer, to aid
368 * the user space app in telling how many lost events there were.
370 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
371 void *prev_subbuf, size_t prev_padding)
373 struct blk_trace *bt;
375 if (!relay_buf_full(buf))
378 bt = buf->chan->private_data;
379 atomic_inc(&bt->dropped);
383 static int blk_remove_buf_file_callback(struct dentry *dentry)
385 debugfs_remove(dentry);
390 static struct dentry *blk_create_buf_file_callback(const char *filename,
391 struct dentry *parent,
393 struct rchan_buf *buf,
396 return debugfs_create_file(filename, mode, parent, buf,
397 &relay_file_operations);
400 static struct rchan_callbacks blk_relay_callbacks = {
401 .subbuf_start = blk_subbuf_start_callback,
402 .create_buf_file = blk_create_buf_file_callback,
403 .remove_buf_file = blk_remove_buf_file_callback,
406 static void blk_trace_setup_lba(struct blk_trace *bt,
407 struct block_device *bdev)
409 struct hd_struct *part = NULL;
412 part = bdev->bd_part;
415 bt->start_lba = part->start_sect;
416 bt->end_lba = part->start_sect + part->nr_sects;
424 * Setup everything required to start tracing
426 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
427 struct block_device *bdev,
428 struct blk_user_trace_setup *buts)
430 struct blk_trace *old_bt, *bt = NULL;
431 struct dentry *dir = NULL;
434 if (!buts->buf_size || !buts->buf_nr)
437 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
438 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
441 * some device names have larger paths - convert the slashes
442 * to underscores for this to work as expected
444 for (i = 0; i < strlen(buts->name); i++)
445 if (buts->name[i] == '/')
448 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
453 bt->sequence = alloc_percpu(unsigned long);
457 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
463 mutex_lock(&blk_tree_mutex);
464 if (!blk_tree_root) {
465 blk_tree_root = debugfs_create_dir("block", NULL);
466 if (!blk_tree_root) {
467 mutex_unlock(&blk_tree_mutex);
471 mutex_unlock(&blk_tree_mutex);
473 dir = debugfs_create_dir(buts->name, blk_tree_root);
480 atomic_set(&bt->dropped, 0);
483 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
485 if (!bt->dropped_file)
488 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
492 bt->rchan = relay_open("trace", dir, buts->buf_size,
493 buts->buf_nr, &blk_relay_callbacks, bt);
497 bt->act_mask = buts->act_mask;
499 bt->act_mask = (u16) -1;
501 blk_trace_setup_lba(bt, bdev);
503 /* overwrite with user settings */
505 bt->start_lba = buts->start_lba;
507 bt->end_lba = buts->end_lba;
510 bt->trace_state = Blktrace_setup;
513 old_bt = xchg(&q->blk_trace, bt);
515 (void) xchg(&q->blk_trace, old_bt);
519 if (atomic_inc_return(&blk_probes_ref) == 1)
520 blk_register_tracepoints();
528 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
529 struct block_device *bdev,
532 struct blk_user_trace_setup buts;
535 ret = copy_from_user(&buts, arg, sizeof(buts));
539 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
543 if (copy_to_user(arg, &buts, sizeof(buts))) {
549 EXPORT_SYMBOL_GPL(blk_trace_setup);
551 int blk_trace_startstop(struct request_queue *q, int start)
554 struct blk_trace *bt = q->blk_trace;
560 * For starting a trace, we can transition from a setup or stopped
561 * trace. For stopping a trace, the state must be running
565 if (bt->trace_state == Blktrace_setup ||
566 bt->trace_state == Blktrace_stopped) {
569 bt->trace_state = Blktrace_running;
575 if (bt->trace_state == Blktrace_running) {
576 bt->trace_state = Blktrace_stopped;
577 relay_flush(bt->rchan);
584 EXPORT_SYMBOL_GPL(blk_trace_startstop);
587 * blk_trace_ioctl: - handle the ioctls associated with tracing
588 * @bdev: the block device
589 * @cmd: the ioctl cmd
590 * @arg: the argument data, if any
593 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
595 struct request_queue *q;
597 char b[BDEVNAME_SIZE];
599 q = bdev_get_queue(bdev);
603 mutex_lock(&bdev->bd_mutex);
608 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
613 ret = blk_trace_startstop(q, start);
615 case BLKTRACETEARDOWN:
616 ret = blk_trace_remove(q);
623 mutex_unlock(&bdev->bd_mutex);
628 * blk_trace_shutdown: - stop and cleanup trace structures
629 * @q: the request queue associated with the device
632 void blk_trace_shutdown(struct request_queue *q)
635 blk_trace_startstop(q, 0);
645 * blk_add_trace_rq - Add a trace for a request oriented action
646 * @q: queue the io is for
647 * @rq: the source request
651 * Records an action against a request. Will log the bio offset + size.
654 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
657 struct blk_trace *bt = q->blk_trace;
658 int rw = rq->cmd_flags & 0x03;
663 if (blk_discard_rq(rq))
664 rw |= (1 << BIO_RW_DISCARD);
666 if (blk_pc_request(rq)) {
667 what |= BLK_TC_ACT(BLK_TC_PC);
668 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
669 what, rq->errors, rq->cmd_len, rq->cmd);
671 what |= BLK_TC_ACT(BLK_TC_FS);
672 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
673 what, rq->errors, 0, NULL);
677 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
679 blk_add_trace_rq(q, rq, BLK_TA_ABORT);
682 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
684 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
687 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
689 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
692 static void blk_add_trace_rq_requeue(struct request_queue *q,
695 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
698 static void blk_add_trace_rq_complete(struct request_queue *q,
701 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
705 * blk_add_trace_bio - Add a trace for a bio oriented action
706 * @q: queue the io is for
707 * @bio: the source bio
711 * Records an action against a bio. Will log the bio offset + size.
714 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
717 struct blk_trace *bt = q->blk_trace;
722 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
723 !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
726 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
728 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
731 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
733 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
736 static void blk_add_trace_bio_backmerge(struct request_queue *q,
739 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
742 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
745 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
748 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
750 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
753 static void blk_add_trace_getrq(struct request_queue *q,
754 struct bio *bio, int rw)
757 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
759 struct blk_trace *bt = q->blk_trace;
762 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
767 static void blk_add_trace_sleeprq(struct request_queue *q,
768 struct bio *bio, int rw)
771 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
773 struct blk_trace *bt = q->blk_trace;
776 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
781 static void blk_add_trace_plug(struct request_queue *q)
783 struct blk_trace *bt = q->blk_trace;
786 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
789 static void blk_add_trace_unplug_io(struct request_queue *q)
791 struct blk_trace *bt = q->blk_trace;
794 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
795 __be64 rpdu = cpu_to_be64(pdu);
797 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
798 sizeof(rpdu), &rpdu);
802 static void blk_add_trace_unplug_timer(struct request_queue *q)
804 struct blk_trace *bt = q->blk_trace;
807 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
808 __be64 rpdu = cpu_to_be64(pdu);
810 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
811 sizeof(rpdu), &rpdu);
815 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
818 struct blk_trace *bt = q->blk_trace;
821 __be64 rpdu = cpu_to_be64(pdu);
823 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
824 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
825 sizeof(rpdu), &rpdu);
830 * blk_add_trace_remap - Add a trace for a remap operation
831 * @q: queue the io is for
832 * @bio: the source bio
833 * @dev: target device
834 * @from: source sector
837 * Device mapper or raid target sometimes need to split a bio because
838 * it spans a stripe (or similar). Add a trace for that action.
841 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
842 dev_t dev, sector_t from)
844 struct blk_trace *bt = q->blk_trace;
845 struct blk_io_trace_remap r;
850 r.device_from = cpu_to_be32(dev);
851 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
852 r.sector_from = cpu_to_be64(from);
854 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
855 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
860 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
861 * @q: queue the io is for
862 * @rq: the source request
863 * @dev: target device
864 * @from: source sector
867 * Device mapper remaps request to other devices.
868 * Add a trace for that action.
871 static void blk_add_trace_rq_remap(struct request_queue *q,
872 struct request *rq, dev_t dev,
875 struct blk_trace *bt = q->blk_trace;
876 struct blk_io_trace_remap r;
881 r.device_from = cpu_to_be32(dev);
882 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
883 r.sector_from = cpu_to_be64(from);
885 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
886 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
891 * blk_add_driver_data - Add binary message with driver-specific data
892 * @q: queue the io is for
894 * @data: driver-specific data
895 * @len: length of driver-specific data
898 * Some drivers might want to write driver-specific data per request.
901 void blk_add_driver_data(struct request_queue *q,
903 void *data, size_t len)
905 struct blk_trace *bt = q->blk_trace;
910 if (blk_pc_request(rq))
911 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
912 BLK_TA_DRV_DATA, rq->errors, len, data);
914 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
915 BLK_TA_DRV_DATA, rq->errors, len, data);
917 EXPORT_SYMBOL_GPL(blk_add_driver_data);
919 static void blk_register_tracepoints(void)
923 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
925 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
927 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
929 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
931 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
933 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
935 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
937 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
939 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
941 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
943 ret = register_trace_block_getrq(blk_add_trace_getrq);
945 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
947 ret = register_trace_block_plug(blk_add_trace_plug);
949 ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
951 ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
953 ret = register_trace_block_split(blk_add_trace_split);
955 ret = register_trace_block_remap(blk_add_trace_remap);
957 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap);
961 static void blk_unregister_tracepoints(void)
963 unregister_trace_block_rq_remap(blk_add_trace_rq_remap);
964 unregister_trace_block_remap(blk_add_trace_remap);
965 unregister_trace_block_split(blk_add_trace_split);
966 unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
967 unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
968 unregister_trace_block_plug(blk_add_trace_plug);
969 unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
970 unregister_trace_block_getrq(blk_add_trace_getrq);
971 unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
972 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
973 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
974 unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
975 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
976 unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
977 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
978 unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
979 unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
980 unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
982 tracepoint_synchronize_unregister();
986 * struct blk_io_tracer formatting routines
989 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
992 int tc = t->action >> BLK_TC_SHIFT;
994 if (t->action == BLK_TN_MESSAGE) {
999 if (tc & BLK_TC_DISCARD)
1001 else if (tc & BLK_TC_WRITE)
1008 if (tc & BLK_TC_AHEAD)
1010 if (tc & BLK_TC_BARRIER)
1012 if (tc & BLK_TC_SYNC)
1014 if (tc & BLK_TC_META)
1021 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1023 return (const struct blk_io_trace *)ent;
1026 static inline const void *pdu_start(const struct trace_entry *ent)
1028 return te_blk_io_trace(ent) + 1;
1031 static inline u32 t_action(const struct trace_entry *ent)
1033 return te_blk_io_trace(ent)->action;
1036 static inline u32 t_bytes(const struct trace_entry *ent)
1038 return te_blk_io_trace(ent)->bytes;
1041 static inline u32 t_sec(const struct trace_entry *ent)
1043 return te_blk_io_trace(ent)->bytes >> 9;
1046 static inline unsigned long long t_sector(const struct trace_entry *ent)
1048 return te_blk_io_trace(ent)->sector;
1051 static inline __u16 t_error(const struct trace_entry *ent)
1053 return te_blk_io_trace(ent)->error;
1056 static __u64 get_pdu_int(const struct trace_entry *ent)
1058 const __u64 *val = pdu_start(ent);
1059 return be64_to_cpu(*val);
1062 static void get_pdu_remap(const struct trace_entry *ent,
1063 struct blk_io_trace_remap *r)
1065 const struct blk_io_trace_remap *__r = pdu_start(ent);
1066 __u64 sector_from = __r->sector_from;
1068 r->device_from = be32_to_cpu(__r->device_from);
1069 r->device_to = be32_to_cpu(__r->device_to);
1070 r->sector_from = be64_to_cpu(sector_from);
1073 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1075 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1078 unsigned long long ts = iter->ts;
1079 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1080 unsigned secs = (unsigned long)ts;
1081 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1085 return trace_seq_printf(&iter->seq,
1086 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1087 MAJOR(t->device), MINOR(t->device), iter->cpu,
1088 secs, nsec_rem, iter->ent->pid, act, rwbs);
1091 static int blk_log_action(struct trace_iterator *iter, const char *act)
1094 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1097 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1098 MAJOR(t->device), MINOR(t->device), act, rwbs);
1101 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1103 const unsigned char *pdu_buf;
1107 pdu_buf = pdu_start(ent);
1108 pdu_len = te_blk_io_trace(ent)->pdu_len;
1113 /* find the last zero that needs to be printed */
1114 for (end = pdu_len - 1; end >= 0; end--)
1119 if (!trace_seq_putc(s, '('))
1122 for (i = 0; i < pdu_len; i++) {
1124 ret = trace_seq_printf(s, "%s%02x",
1125 i == 0 ? "" : " ", pdu_buf[i]);
1130 * stop when the rest is just zeroes and indicate so
1131 * with a ".." appended
1133 if (i == end && end != pdu_len - 1)
1134 return trace_seq_puts(s, " ..) ");
1137 return trace_seq_puts(s, ") ");
1140 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1142 char cmd[TASK_COMM_LEN];
1144 trace_find_cmdline(ent->pid, cmd);
1146 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1149 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1152 ret = blk_log_dump_pdu(s, ent);
1155 return trace_seq_printf(s, "[%s]\n", cmd);
1158 return trace_seq_printf(s, "%llu + %u [%s]\n",
1159 t_sector(ent), t_sec(ent), cmd);
1160 return trace_seq_printf(s, "[%s]\n", cmd);
1164 static int blk_log_with_error(struct trace_seq *s,
1165 const struct trace_entry *ent)
1167 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1170 ret = blk_log_dump_pdu(s, ent);
1172 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1176 return trace_seq_printf(s, "%llu + %u [%d]\n",
1178 t_sec(ent), t_error(ent));
1179 return trace_seq_printf(s, "%llu [%d]\n",
1180 t_sector(ent), t_error(ent));
1184 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1186 struct blk_io_trace_remap r = { .device_from = 0, };
1188 get_pdu_remap(ent, &r);
1189 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1190 t_sector(ent), t_sec(ent),
1191 MAJOR(r.device_from), MINOR(r.device_from),
1192 (unsigned long long)r.sector_from);
1195 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1197 char cmd[TASK_COMM_LEN];
1199 trace_find_cmdline(ent->pid, cmd);
1201 return trace_seq_printf(s, "[%s]\n", cmd);
1204 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1206 char cmd[TASK_COMM_LEN];
1208 trace_find_cmdline(ent->pid, cmd);
1210 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1213 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1215 char cmd[TASK_COMM_LEN];
1217 trace_find_cmdline(ent->pid, cmd);
1219 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1220 get_pdu_int(ent), cmd);
1223 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1226 const struct blk_io_trace *t = te_blk_io_trace(ent);
1228 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1230 return trace_seq_putc(s, '\n');
1235 * struct tracer operations
1238 static void blk_tracer_print_header(struct seq_file *m)
1240 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1242 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1246 static void blk_tracer_start(struct trace_array *tr)
1248 blk_tracer_enabled = true;
1251 static int blk_tracer_init(struct trace_array *tr)
1254 blk_tracer_start(tr);
1258 static void blk_tracer_stop(struct trace_array *tr)
1260 blk_tracer_enabled = false;
1263 static void blk_tracer_reset(struct trace_array *tr)
1265 blk_tracer_stop(tr);
1268 static const struct {
1270 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1272 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1273 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1274 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1275 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1276 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1277 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1278 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1279 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1280 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1281 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1282 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1283 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1284 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1285 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1286 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1289 static enum print_line_t print_one_line(struct trace_iterator *iter,
1292 struct trace_seq *s = &iter->seq;
1293 const struct blk_io_trace *t;
1297 blk_log_action_t *log_action;
1299 t = te_blk_io_trace(iter->ent);
1300 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1301 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1302 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1304 if (t->action == BLK_TN_MESSAGE) {
1305 ret = log_action(iter, long_act ? "message" : "m");
1307 ret = blk_log_msg(s, iter->ent);
1311 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1312 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1314 ret = log_action(iter, what2act[what].act[long_act]);
1316 ret = what2act[what].print(s, iter->ent);
1319 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1322 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1325 return print_one_line(iter, false);
1328 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1330 struct trace_seq *s = &iter->seq;
1331 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1332 const int offset = offsetof(struct blk_io_trace, sector);
1333 struct blk_io_trace old = {
1334 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1338 if (!trace_seq_putmem(s, &old, offset))
1340 return trace_seq_putmem(s, &t->sector,
1341 sizeof(old) - offset + t->pdu_len);
1344 static enum print_line_t
1345 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1347 return blk_trace_synthesize_old_trace(iter) ?
1348 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1351 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1353 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1354 return TRACE_TYPE_UNHANDLED;
1356 return print_one_line(iter, true);
1359 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1361 /* don't output context-info for blk_classic output */
1362 if (bit == TRACE_BLK_OPT_CLASSIC) {
1364 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1366 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1371 static struct tracer blk_tracer __read_mostly = {
1373 .init = blk_tracer_init,
1374 .reset = blk_tracer_reset,
1375 .start = blk_tracer_start,
1376 .stop = blk_tracer_stop,
1377 .print_header = blk_tracer_print_header,
1378 .print_line = blk_tracer_print_line,
1379 .flags = &blk_tracer_flags,
1380 .set_flag = blk_tracer_set_flag,
1383 static struct trace_event trace_blk_event = {
1385 .trace = blk_trace_event_print,
1386 .binary = blk_trace_event_print_binary,
1389 static int __init init_blk_tracer(void)
1391 if (!register_ftrace_event(&trace_blk_event)) {
1392 pr_warning("Warning: could not register block events\n");
1396 if (register_tracer(&blk_tracer) != 0) {
1397 pr_warning("Warning: could not register the block tracer\n");
1398 unregister_ftrace_event(&trace_blk_event);
1405 device_initcall(init_blk_tracer);
1407 static int blk_trace_remove_queue(struct request_queue *q)
1409 struct blk_trace *bt;
1411 bt = xchg(&q->blk_trace, NULL);
1415 if (atomic_dec_and_test(&blk_probes_ref))
1416 blk_unregister_tracepoints();
1423 * Setup everything required to start tracing
1425 static int blk_trace_setup_queue(struct request_queue *q,
1426 struct block_device *bdev)
1428 struct blk_trace *old_bt, *bt = NULL;
1431 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1435 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1439 bt->dev = bdev->bd_dev;
1440 bt->act_mask = (u16)-1;
1442 blk_trace_setup_lba(bt, bdev);
1444 old_bt = xchg(&q->blk_trace, bt);
1445 if (old_bt != NULL) {
1446 (void)xchg(&q->blk_trace, old_bt);
1451 if (atomic_inc_return(&blk_probes_ref) == 1)
1452 blk_register_tracepoints();
1461 * sysfs interface to enable and configure tracing
1464 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1465 struct device_attribute *attr,
1467 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1468 struct device_attribute *attr,
1469 const char *buf, size_t count);
1470 #define BLK_TRACE_DEVICE_ATTR(_name) \
1471 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1472 sysfs_blk_trace_attr_show, \
1473 sysfs_blk_trace_attr_store)
1475 static BLK_TRACE_DEVICE_ATTR(enable);
1476 static BLK_TRACE_DEVICE_ATTR(act_mask);
1477 static BLK_TRACE_DEVICE_ATTR(pid);
1478 static BLK_TRACE_DEVICE_ATTR(start_lba);
1479 static BLK_TRACE_DEVICE_ATTR(end_lba);
1481 static struct attribute *blk_trace_attrs[] = {
1482 &dev_attr_enable.attr,
1483 &dev_attr_act_mask.attr,
1485 &dev_attr_start_lba.attr,
1486 &dev_attr_end_lba.attr,
1490 struct attribute_group blk_trace_attr_group = {
1492 .attrs = blk_trace_attrs,
1495 static const struct {
1499 { BLK_TC_READ, "read" },
1500 { BLK_TC_WRITE, "write" },
1501 { BLK_TC_BARRIER, "barrier" },
1502 { BLK_TC_SYNC, "sync" },
1503 { BLK_TC_QUEUE, "queue" },
1504 { BLK_TC_REQUEUE, "requeue" },
1505 { BLK_TC_ISSUE, "issue" },
1506 { BLK_TC_COMPLETE, "complete" },
1507 { BLK_TC_FS, "fs" },
1508 { BLK_TC_PC, "pc" },
1509 { BLK_TC_AHEAD, "ahead" },
1510 { BLK_TC_META, "meta" },
1511 { BLK_TC_DISCARD, "discard" },
1512 { BLK_TC_DRV_DATA, "drv_data" },
1515 static int blk_trace_str2mask(const char *str)
1519 char *buf, *s, *token;
1521 buf = kstrdup(str, GFP_KERNEL);
1527 token = strsep(&s, ",");
1534 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1535 if (strcasecmp(token, mask_maps[i].str) == 0) {
1536 mask |= mask_maps[i].mask;
1540 if (i == ARRAY_SIZE(mask_maps)) {
1550 static ssize_t blk_trace_mask2str(char *buf, int mask)
1555 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1556 if (mask & mask_maps[i].mask) {
1557 p += sprintf(p, "%s%s",
1558 (p == buf) ? "" : ",", mask_maps[i].str);
1566 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1568 if (bdev->bd_disk == NULL)
1571 return bdev_get_queue(bdev);
1574 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1575 struct device_attribute *attr,
1578 struct hd_struct *p = dev_to_part(dev);
1579 struct request_queue *q;
1580 struct block_device *bdev;
1581 ssize_t ret = -ENXIO;
1584 bdev = bdget(part_devt(p));
1586 goto out_unlock_kernel;
1588 q = blk_trace_get_queue(bdev);
1592 mutex_lock(&bdev->bd_mutex);
1594 if (attr == &dev_attr_enable) {
1595 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1596 goto out_unlock_bdev;
1599 if (q->blk_trace == NULL)
1600 ret = sprintf(buf, "disabled\n");
1601 else if (attr == &dev_attr_act_mask)
1602 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1603 else if (attr == &dev_attr_pid)
1604 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1605 else if (attr == &dev_attr_start_lba)
1606 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1607 else if (attr == &dev_attr_end_lba)
1608 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1611 mutex_unlock(&bdev->bd_mutex);
1619 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1620 struct device_attribute *attr,
1621 const char *buf, size_t count)
1623 struct block_device *bdev;
1624 struct request_queue *q;
1625 struct hd_struct *p;
1627 ssize_t ret = -EINVAL;
1632 if (attr == &dev_attr_act_mask) {
1633 if (sscanf(buf, "%llx", &value) != 1) {
1634 /* Assume it is a list of trace category names */
1635 ret = blk_trace_str2mask(buf);
1640 } else if (sscanf(buf, "%llu", &value) != 1)
1646 p = dev_to_part(dev);
1647 bdev = bdget(part_devt(p));
1649 goto out_unlock_kernel;
1651 q = blk_trace_get_queue(bdev);
1655 mutex_lock(&bdev->bd_mutex);
1657 if (attr == &dev_attr_enable) {
1659 ret = blk_trace_setup_queue(q, bdev);
1661 ret = blk_trace_remove_queue(q);
1662 goto out_unlock_bdev;
1666 if (q->blk_trace == NULL)
1667 ret = blk_trace_setup_queue(q, bdev);
1670 if (attr == &dev_attr_act_mask)
1671 q->blk_trace->act_mask = value;
1672 else if (attr == &dev_attr_pid)
1673 q->blk_trace->pid = value;
1674 else if (attr == &dev_attr_start_lba)
1675 q->blk_trace->start_lba = value;
1676 else if (attr == &dev_attr_end_lba)
1677 q->blk_trace->end_lba = value;
1681 mutex_unlock(&bdev->bd_mutex);
1687 return ret ? ret : count;
1690 int blk_trace_init_sysfs(struct device *dev)
1692 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1695 void blk_trace_remove_sysfs(struct device *dev)
1697 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1700 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1702 #ifdef CONFIG_EVENT_TRACING
1704 void blk_dump_cmd(char *buf, struct request *rq)
1707 int len = rq->cmd_len;
1708 unsigned char *cmd = rq->cmd;
1710 if (!blk_pc_request(rq)) {
1715 for (end = len - 1; end >= 0; end--)
1720 for (i = 0; i < len; i++) {
1721 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1722 if (i == end && end != len - 1) {
1723 sprintf(buf, " ..");
1729 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1735 else if (rw & 1 << BIO_RW_DISCARD)
1742 if (rw & 1 << BIO_RW_AHEAD)
1744 if (rw & 1 << BIO_RW_BARRIER)
1746 if (rw & 1 << BIO_RW_SYNCIO)
1748 if (rw & 1 << BIO_RW_META)
1754 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1756 int rw = rq->cmd_flags & 0x03;
1759 if (blk_discard_rq(rq))
1760 rw |= (1 << BIO_RW_DISCARD);
1762 bytes = blk_rq_bytes(rq);
1764 blk_fill_rwbs(rwbs, rw, bytes);
1767 #endif /* CONFIG_EVENT_TRACING */