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->trace_buffer.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 = this_cpu_ptr(bt->msg_data);
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->trace_buffer.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 ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
315 size_t count, loff_t *ppos)
317 struct blk_trace *bt = filp->private_data;
320 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
322 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
325 static const struct file_operations blk_dropped_fops = {
326 .owner = THIS_MODULE,
328 .read = blk_dropped_read,
329 .llseek = default_llseek,
332 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
333 size_t count, loff_t *ppos)
336 struct blk_trace *bt;
338 if (count >= BLK_TN_MAX_MSG)
341 msg = kmalloc(count + 1, GFP_KERNEL);
345 if (copy_from_user(msg, buffer, count)) {
351 bt = filp->private_data;
352 __trace_note_message(bt, "%s", msg);
358 static const struct file_operations blk_msg_fops = {
359 .owner = THIS_MODULE,
361 .write = blk_msg_write,
362 .llseek = noop_llseek,
366 * Keep track of how many times we encountered a full subbuffer, to aid
367 * the user space app in telling how many lost events there were.
369 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
370 void *prev_subbuf, size_t prev_padding)
372 struct blk_trace *bt;
374 if (!relay_buf_full(buf))
377 bt = buf->chan->private_data;
378 atomic_inc(&bt->dropped);
382 static int blk_remove_buf_file_callback(struct dentry *dentry)
384 debugfs_remove(dentry);
389 static struct dentry *blk_create_buf_file_callback(const char *filename,
390 struct dentry *parent,
392 struct rchan_buf *buf,
395 return debugfs_create_file(filename, mode, parent, buf,
396 &relay_file_operations);
399 static struct rchan_callbacks blk_relay_callbacks = {
400 .subbuf_start = blk_subbuf_start_callback,
401 .create_buf_file = blk_create_buf_file_callback,
402 .remove_buf_file = blk_remove_buf_file_callback,
405 static void blk_trace_setup_lba(struct blk_trace *bt,
406 struct block_device *bdev)
408 struct hd_struct *part = NULL;
411 part = bdev->bd_part;
414 bt->start_lba = part->start_sect;
415 bt->end_lba = part->start_sect + part->nr_sects;
423 * Setup everything required to start tracing
425 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
426 struct block_device *bdev,
427 struct blk_user_trace_setup *buts)
429 struct blk_trace *old_bt, *bt = NULL;
430 struct dentry *dir = NULL;
433 if (!buts->buf_size || !buts->buf_nr)
436 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
437 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
440 * some device names have larger paths - convert the slashes
441 * to underscores for this to work as expected
443 for (i = 0; i < strlen(buts->name); i++)
444 if (buts->name[i] == '/')
447 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
452 bt->sequence = alloc_percpu(unsigned long);
456 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
462 mutex_lock(&blk_tree_mutex);
463 if (!blk_tree_root) {
464 blk_tree_root = debugfs_create_dir("block", NULL);
465 if (!blk_tree_root) {
466 mutex_unlock(&blk_tree_mutex);
470 mutex_unlock(&blk_tree_mutex);
472 dir = debugfs_create_dir(buts->name, blk_tree_root);
479 atomic_set(&bt->dropped, 0);
482 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
484 if (!bt->dropped_file)
487 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
491 bt->rchan = relay_open("trace", dir, buts->buf_size,
492 buts->buf_nr, &blk_relay_callbacks, bt);
496 bt->act_mask = buts->act_mask;
498 bt->act_mask = (u16) -1;
500 blk_trace_setup_lba(bt, bdev);
502 /* overwrite with user settings */
504 bt->start_lba = buts->start_lba;
506 bt->end_lba = buts->end_lba;
509 bt->trace_state = Blktrace_setup;
512 old_bt = xchg(&q->blk_trace, bt);
514 (void) xchg(&q->blk_trace, old_bt);
518 if (atomic_inc_return(&blk_probes_ref) == 1)
519 blk_register_tracepoints();
527 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
528 struct block_device *bdev,
531 struct blk_user_trace_setup buts;
534 ret = copy_from_user(&buts, arg, sizeof(buts));
538 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
542 if (copy_to_user(arg, &buts, sizeof(buts))) {
548 EXPORT_SYMBOL_GPL(blk_trace_setup);
550 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
551 static int compat_blk_trace_setup(struct request_queue *q, char *name,
552 dev_t dev, struct block_device *bdev,
555 struct blk_user_trace_setup buts;
556 struct compat_blk_user_trace_setup cbuts;
559 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
562 buts = (struct blk_user_trace_setup) {
563 .act_mask = cbuts.act_mask,
564 .buf_size = cbuts.buf_size,
565 .buf_nr = cbuts.buf_nr,
566 .start_lba = cbuts.start_lba,
567 .end_lba = cbuts.end_lba,
570 memcpy(&buts.name, &cbuts.name, 32);
572 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
576 if (copy_to_user(arg, &buts.name, 32)) {
585 int blk_trace_startstop(struct request_queue *q, int start)
588 struct blk_trace *bt = q->blk_trace;
594 * For starting a trace, we can transition from a setup or stopped
595 * trace. For stopping a trace, the state must be running
599 if (bt->trace_state == Blktrace_setup ||
600 bt->trace_state == Blktrace_stopped) {
603 bt->trace_state = Blktrace_running;
609 if (bt->trace_state == Blktrace_running) {
610 bt->trace_state = Blktrace_stopped;
611 relay_flush(bt->rchan);
618 EXPORT_SYMBOL_GPL(blk_trace_startstop);
621 * blk_trace_ioctl: - handle the ioctls associated with tracing
622 * @bdev: the block device
623 * @cmd: the ioctl cmd
624 * @arg: the argument data, if any
627 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
629 struct request_queue *q;
631 char b[BDEVNAME_SIZE];
633 q = bdev_get_queue(bdev);
637 mutex_lock(&bdev->bd_mutex);
642 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
644 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
645 case BLKTRACESETUP32:
647 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
653 ret = blk_trace_startstop(q, start);
655 case BLKTRACETEARDOWN:
656 ret = blk_trace_remove(q);
663 mutex_unlock(&bdev->bd_mutex);
668 * blk_trace_shutdown: - stop and cleanup trace structures
669 * @q: the request queue associated with the device
672 void blk_trace_shutdown(struct request_queue *q)
675 blk_trace_startstop(q, 0);
685 * blk_add_trace_rq - Add a trace for a request oriented action
686 * @q: queue the io is for
687 * @rq: the source request
691 * Records an action against a request. Will log the bio offset + size.
694 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
697 struct blk_trace *bt = q->blk_trace;
702 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
703 what |= BLK_TC_ACT(BLK_TC_PC);
704 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rq->cmd_flags,
705 what, rq->errors, rq->cmd_len, rq->cmd);
707 what |= BLK_TC_ACT(BLK_TC_FS);
708 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
709 rq->cmd_flags, what, rq->errors, 0, NULL);
713 static void blk_add_trace_rq_abort(void *ignore,
714 struct request_queue *q, struct request *rq)
716 blk_add_trace_rq(q, rq, BLK_TA_ABORT);
719 static void blk_add_trace_rq_insert(void *ignore,
720 struct request_queue *q, struct request *rq)
722 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
725 static void blk_add_trace_rq_issue(void *ignore,
726 struct request_queue *q, struct request *rq)
728 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
731 static void blk_add_trace_rq_requeue(void *ignore,
732 struct request_queue *q,
735 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
738 static void blk_add_trace_rq_complete(void *ignore,
739 struct request_queue *q,
742 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
746 * blk_add_trace_bio - Add a trace for a bio oriented action
747 * @q: queue the io is for
748 * @bio: the source bio
750 * @error: error, if any
753 * Records an action against a bio. Will log the bio offset + size.
756 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
759 struct blk_trace *bt = q->blk_trace;
764 if (!error && !bio_flagged(bio, BIO_UPTODATE))
767 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
771 static void blk_add_trace_bio_bounce(void *ignore,
772 struct request_queue *q, struct bio *bio)
774 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
777 static void blk_add_trace_bio_complete(void *ignore,
778 struct request_queue *q, struct bio *bio,
781 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
784 static void blk_add_trace_bio_backmerge(void *ignore,
785 struct request_queue *q,
789 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
792 static void blk_add_trace_bio_frontmerge(void *ignore,
793 struct request_queue *q,
797 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
800 static void blk_add_trace_bio_queue(void *ignore,
801 struct request_queue *q, struct bio *bio)
803 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
806 static void blk_add_trace_getrq(void *ignore,
807 struct request_queue *q,
808 struct bio *bio, int rw)
811 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
813 struct blk_trace *bt = q->blk_trace;
816 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
821 static void blk_add_trace_sleeprq(void *ignore,
822 struct request_queue *q,
823 struct bio *bio, int rw)
826 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
828 struct blk_trace *bt = q->blk_trace;
831 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
836 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
838 struct blk_trace *bt = q->blk_trace;
841 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
844 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
845 unsigned int depth, bool explicit)
847 struct blk_trace *bt = q->blk_trace;
850 __be64 rpdu = cpu_to_be64(depth);
854 what = BLK_TA_UNPLUG_IO;
856 what = BLK_TA_UNPLUG_TIMER;
858 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
862 static void blk_add_trace_split(void *ignore,
863 struct request_queue *q, struct bio *bio,
866 struct blk_trace *bt = q->blk_trace;
869 __be64 rpdu = cpu_to_be64(pdu);
871 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
872 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
873 sizeof(rpdu), &rpdu);
878 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
879 * @ignore: trace callback data parameter (not used)
880 * @q: queue the io is for
881 * @bio: the source bio
882 * @dev: target device
883 * @from: source sector
886 * Device mapper or raid target sometimes need to split a bio because
887 * it spans a stripe (or similar). Add a trace for that action.
890 static void blk_add_trace_bio_remap(void *ignore,
891 struct request_queue *q, struct bio *bio,
892 dev_t dev, sector_t from)
894 struct blk_trace *bt = q->blk_trace;
895 struct blk_io_trace_remap r;
900 r.device_from = cpu_to_be32(dev);
901 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev);
902 r.sector_from = cpu_to_be64(from);
904 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
905 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
910 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
911 * @ignore: trace callback data parameter (not used)
912 * @q: queue the io is for
913 * @rq: the source request
914 * @dev: target device
915 * @from: source sector
918 * Device mapper remaps request to other devices.
919 * Add a trace for that action.
922 static void blk_add_trace_rq_remap(void *ignore,
923 struct request_queue *q,
924 struct request *rq, dev_t dev,
927 struct blk_trace *bt = q->blk_trace;
928 struct blk_io_trace_remap r;
933 r.device_from = cpu_to_be32(dev);
934 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
935 r.sector_from = cpu_to_be64(from);
937 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
938 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
943 * blk_add_driver_data - Add binary message with driver-specific data
944 * @q: queue the io is for
946 * @data: driver-specific data
947 * @len: length of driver-specific data
950 * Some drivers might want to write driver-specific data per request.
953 void blk_add_driver_data(struct request_queue *q,
955 void *data, size_t len)
957 struct blk_trace *bt = q->blk_trace;
962 if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
963 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
964 BLK_TA_DRV_DATA, rq->errors, len, data);
966 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
967 BLK_TA_DRV_DATA, rq->errors, len, data);
969 EXPORT_SYMBOL_GPL(blk_add_driver_data);
971 static void blk_register_tracepoints(void)
975 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
977 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
979 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
981 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
983 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
985 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
987 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
989 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
991 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
993 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
995 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
997 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
999 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1001 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1003 ret = register_trace_block_split(blk_add_trace_split, NULL);
1005 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1007 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1011 static void blk_unregister_tracepoints(void)
1013 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1014 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1015 unregister_trace_block_split(blk_add_trace_split, NULL);
1016 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1017 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1018 unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1019 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1020 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1021 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1022 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1023 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1024 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1025 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1026 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1027 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1028 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1029 unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1031 tracepoint_synchronize_unregister();
1035 * struct blk_io_tracer formatting routines
1038 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1041 int tc = t->action >> BLK_TC_SHIFT;
1043 if (t->action == BLK_TN_MESSAGE) {
1048 if (tc & BLK_TC_FLUSH)
1051 if (tc & BLK_TC_DISCARD)
1053 else if (tc & BLK_TC_WRITE)
1060 if (tc & BLK_TC_FUA)
1062 if (tc & BLK_TC_AHEAD)
1064 if (tc & BLK_TC_SYNC)
1066 if (tc & BLK_TC_META)
1073 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1075 return (const struct blk_io_trace *)ent;
1078 static inline const void *pdu_start(const struct trace_entry *ent)
1080 return te_blk_io_trace(ent) + 1;
1083 static inline u32 t_action(const struct trace_entry *ent)
1085 return te_blk_io_trace(ent)->action;
1088 static inline u32 t_bytes(const struct trace_entry *ent)
1090 return te_blk_io_trace(ent)->bytes;
1093 static inline u32 t_sec(const struct trace_entry *ent)
1095 return te_blk_io_trace(ent)->bytes >> 9;
1098 static inline unsigned long long t_sector(const struct trace_entry *ent)
1100 return te_blk_io_trace(ent)->sector;
1103 static inline __u16 t_error(const struct trace_entry *ent)
1105 return te_blk_io_trace(ent)->error;
1108 static __u64 get_pdu_int(const struct trace_entry *ent)
1110 const __u64 *val = pdu_start(ent);
1111 return be64_to_cpu(*val);
1114 static void get_pdu_remap(const struct trace_entry *ent,
1115 struct blk_io_trace_remap *r)
1117 const struct blk_io_trace_remap *__r = pdu_start(ent);
1118 __u64 sector_from = __r->sector_from;
1120 r->device_from = be32_to_cpu(__r->device_from);
1121 r->device_to = be32_to_cpu(__r->device_to);
1122 r->sector_from = be64_to_cpu(sector_from);
1125 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1127 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1129 char rwbs[RWBS_LEN];
1130 unsigned long long ts = iter->ts;
1131 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1132 unsigned secs = (unsigned long)ts;
1133 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1137 return trace_seq_printf(&iter->seq,
1138 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1139 MAJOR(t->device), MINOR(t->device), iter->cpu,
1140 secs, nsec_rem, iter->ent->pid, act, rwbs);
1143 static int blk_log_action(struct trace_iterator *iter, const char *act)
1145 char rwbs[RWBS_LEN];
1146 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1149 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1150 MAJOR(t->device), MINOR(t->device), act, rwbs);
1153 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1155 const unsigned char *pdu_buf;
1159 pdu_buf = pdu_start(ent);
1160 pdu_len = te_blk_io_trace(ent)->pdu_len;
1165 /* find the last zero that needs to be printed */
1166 for (end = pdu_len - 1; end >= 0; end--)
1171 if (!trace_seq_putc(s, '('))
1174 for (i = 0; i < pdu_len; i++) {
1176 ret = trace_seq_printf(s, "%s%02x",
1177 i == 0 ? "" : " ", pdu_buf[i]);
1182 * stop when the rest is just zeroes and indicate so
1183 * with a ".." appended
1185 if (i == end && end != pdu_len - 1)
1186 return trace_seq_puts(s, " ..) ");
1189 return trace_seq_puts(s, ") ");
1192 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1194 char cmd[TASK_COMM_LEN];
1196 trace_find_cmdline(ent->pid, cmd);
1198 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1201 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1204 ret = blk_log_dump_pdu(s, ent);
1207 return trace_seq_printf(s, "[%s]\n", cmd);
1210 return trace_seq_printf(s, "%llu + %u [%s]\n",
1211 t_sector(ent), t_sec(ent), cmd);
1212 return trace_seq_printf(s, "[%s]\n", cmd);
1216 static int blk_log_with_error(struct trace_seq *s,
1217 const struct trace_entry *ent)
1219 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1222 ret = blk_log_dump_pdu(s, ent);
1224 return trace_seq_printf(s, "[%d]\n", t_error(ent));
1228 return trace_seq_printf(s, "%llu + %u [%d]\n",
1230 t_sec(ent), t_error(ent));
1231 return trace_seq_printf(s, "%llu [%d]\n",
1232 t_sector(ent), t_error(ent));
1236 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1238 struct blk_io_trace_remap r = { .device_from = 0, };
1240 get_pdu_remap(ent, &r);
1241 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1242 t_sector(ent), t_sec(ent),
1243 MAJOR(r.device_from), MINOR(r.device_from),
1244 (unsigned long long)r.sector_from);
1247 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1249 char cmd[TASK_COMM_LEN];
1251 trace_find_cmdline(ent->pid, cmd);
1253 return trace_seq_printf(s, "[%s]\n", cmd);
1256 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1258 char cmd[TASK_COMM_LEN];
1260 trace_find_cmdline(ent->pid, cmd);
1262 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1265 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1267 char cmd[TASK_COMM_LEN];
1269 trace_find_cmdline(ent->pid, cmd);
1271 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1272 get_pdu_int(ent), cmd);
1275 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1278 const struct blk_io_trace *t = te_blk_io_trace(ent);
1280 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1282 return trace_seq_putc(s, '\n');
1287 * struct tracer operations
1290 static void blk_tracer_print_header(struct seq_file *m)
1292 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1294 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1298 static void blk_tracer_start(struct trace_array *tr)
1300 blk_tracer_enabled = true;
1303 static int blk_tracer_init(struct trace_array *tr)
1306 blk_tracer_start(tr);
1310 static void blk_tracer_stop(struct trace_array *tr)
1312 blk_tracer_enabled = false;
1315 static void blk_tracer_reset(struct trace_array *tr)
1317 blk_tracer_stop(tr);
1320 static const struct {
1322 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1324 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1325 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1326 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1327 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1328 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1329 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1330 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1331 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1332 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1333 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1334 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1335 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1336 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1337 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1338 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1341 static enum print_line_t print_one_line(struct trace_iterator *iter,
1344 struct trace_seq *s = &iter->seq;
1345 const struct blk_io_trace *t;
1349 blk_log_action_t *log_action;
1351 t = te_blk_io_trace(iter->ent);
1352 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1353 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1354 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1356 if (t->action == BLK_TN_MESSAGE) {
1357 ret = log_action(iter, long_act ? "message" : "m");
1359 ret = blk_log_msg(s, iter->ent);
1363 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1364 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1366 ret = log_action(iter, what2act[what].act[long_act]);
1368 ret = what2act[what].print(s, iter->ent);
1371 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1374 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1375 int flags, struct trace_event *event)
1377 return print_one_line(iter, false);
1380 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1382 struct trace_seq *s = &iter->seq;
1383 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1384 const int offset = offsetof(struct blk_io_trace, sector);
1385 struct blk_io_trace old = {
1386 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1390 if (!trace_seq_putmem(s, &old, offset))
1392 return trace_seq_putmem(s, &t->sector,
1393 sizeof(old) - offset + t->pdu_len);
1396 static enum print_line_t
1397 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1398 struct trace_event *event)
1400 return blk_trace_synthesize_old_trace(iter) ?
1401 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1404 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1406 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1407 return TRACE_TYPE_UNHANDLED;
1409 return print_one_line(iter, true);
1412 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1414 /* don't output context-info for blk_classic output */
1415 if (bit == TRACE_BLK_OPT_CLASSIC) {
1417 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1419 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1424 static struct tracer blk_tracer __read_mostly = {
1426 .init = blk_tracer_init,
1427 .reset = blk_tracer_reset,
1428 .start = blk_tracer_start,
1429 .stop = blk_tracer_stop,
1430 .print_header = blk_tracer_print_header,
1431 .print_line = blk_tracer_print_line,
1432 .flags = &blk_tracer_flags,
1433 .set_flag = blk_tracer_set_flag,
1436 static struct trace_event_functions trace_blk_event_funcs = {
1437 .trace = blk_trace_event_print,
1438 .binary = blk_trace_event_print_binary,
1441 static struct trace_event trace_blk_event = {
1443 .funcs = &trace_blk_event_funcs,
1446 static int __init init_blk_tracer(void)
1448 if (!register_ftrace_event(&trace_blk_event)) {
1449 pr_warning("Warning: could not register block events\n");
1453 if (register_tracer(&blk_tracer) != 0) {
1454 pr_warning("Warning: could not register the block tracer\n");
1455 unregister_ftrace_event(&trace_blk_event);
1462 device_initcall(init_blk_tracer);
1464 static int blk_trace_remove_queue(struct request_queue *q)
1466 struct blk_trace *bt;
1468 bt = xchg(&q->blk_trace, NULL);
1472 if (atomic_dec_and_test(&blk_probes_ref))
1473 blk_unregister_tracepoints();
1480 * Setup everything required to start tracing
1482 static int blk_trace_setup_queue(struct request_queue *q,
1483 struct block_device *bdev)
1485 struct blk_trace *old_bt, *bt = NULL;
1488 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1492 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1496 bt->dev = bdev->bd_dev;
1497 bt->act_mask = (u16)-1;
1499 blk_trace_setup_lba(bt, bdev);
1501 old_bt = xchg(&q->blk_trace, bt);
1502 if (old_bt != NULL) {
1503 (void)xchg(&q->blk_trace, old_bt);
1508 if (atomic_inc_return(&blk_probes_ref) == 1)
1509 blk_register_tracepoints();
1518 * sysfs interface to enable and configure tracing
1521 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1522 struct device_attribute *attr,
1524 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1525 struct device_attribute *attr,
1526 const char *buf, size_t count);
1527 #define BLK_TRACE_DEVICE_ATTR(_name) \
1528 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1529 sysfs_blk_trace_attr_show, \
1530 sysfs_blk_trace_attr_store)
1532 static BLK_TRACE_DEVICE_ATTR(enable);
1533 static BLK_TRACE_DEVICE_ATTR(act_mask);
1534 static BLK_TRACE_DEVICE_ATTR(pid);
1535 static BLK_TRACE_DEVICE_ATTR(start_lba);
1536 static BLK_TRACE_DEVICE_ATTR(end_lba);
1538 static struct attribute *blk_trace_attrs[] = {
1539 &dev_attr_enable.attr,
1540 &dev_attr_act_mask.attr,
1542 &dev_attr_start_lba.attr,
1543 &dev_attr_end_lba.attr,
1547 struct attribute_group blk_trace_attr_group = {
1549 .attrs = blk_trace_attrs,
1552 static const struct {
1556 { BLK_TC_READ, "read" },
1557 { BLK_TC_WRITE, "write" },
1558 { BLK_TC_FLUSH, "flush" },
1559 { BLK_TC_SYNC, "sync" },
1560 { BLK_TC_QUEUE, "queue" },
1561 { BLK_TC_REQUEUE, "requeue" },
1562 { BLK_TC_ISSUE, "issue" },
1563 { BLK_TC_COMPLETE, "complete" },
1564 { BLK_TC_FS, "fs" },
1565 { BLK_TC_PC, "pc" },
1566 { BLK_TC_AHEAD, "ahead" },
1567 { BLK_TC_META, "meta" },
1568 { BLK_TC_DISCARD, "discard" },
1569 { BLK_TC_DRV_DATA, "drv_data" },
1570 { BLK_TC_FUA, "fua" },
1573 static int blk_trace_str2mask(const char *str)
1577 char *buf, *s, *token;
1579 buf = kstrdup(str, GFP_KERNEL);
1585 token = strsep(&s, ",");
1592 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1593 if (strcasecmp(token, mask_maps[i].str) == 0) {
1594 mask |= mask_maps[i].mask;
1598 if (i == ARRAY_SIZE(mask_maps)) {
1608 static ssize_t blk_trace_mask2str(char *buf, int mask)
1613 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1614 if (mask & mask_maps[i].mask) {
1615 p += sprintf(p, "%s%s",
1616 (p == buf) ? "" : ",", mask_maps[i].str);
1624 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1626 if (bdev->bd_disk == NULL)
1629 return bdev_get_queue(bdev);
1632 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1633 struct device_attribute *attr,
1636 struct hd_struct *p = dev_to_part(dev);
1637 struct request_queue *q;
1638 struct block_device *bdev;
1639 ssize_t ret = -ENXIO;
1641 bdev = bdget(part_devt(p));
1645 q = blk_trace_get_queue(bdev);
1649 mutex_lock(&bdev->bd_mutex);
1651 if (attr == &dev_attr_enable) {
1652 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1653 goto out_unlock_bdev;
1656 if (q->blk_trace == NULL)
1657 ret = sprintf(buf, "disabled\n");
1658 else if (attr == &dev_attr_act_mask)
1659 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1660 else if (attr == &dev_attr_pid)
1661 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1662 else if (attr == &dev_attr_start_lba)
1663 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1664 else if (attr == &dev_attr_end_lba)
1665 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1668 mutex_unlock(&bdev->bd_mutex);
1675 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1676 struct device_attribute *attr,
1677 const char *buf, size_t count)
1679 struct block_device *bdev;
1680 struct request_queue *q;
1681 struct hd_struct *p;
1683 ssize_t ret = -EINVAL;
1688 if (attr == &dev_attr_act_mask) {
1689 if (sscanf(buf, "%llx", &value) != 1) {
1690 /* Assume it is a list of trace category names */
1691 ret = blk_trace_str2mask(buf);
1696 } else if (sscanf(buf, "%llu", &value) != 1)
1701 p = dev_to_part(dev);
1702 bdev = bdget(part_devt(p));
1706 q = blk_trace_get_queue(bdev);
1710 mutex_lock(&bdev->bd_mutex);
1712 if (attr == &dev_attr_enable) {
1714 ret = blk_trace_setup_queue(q, bdev);
1716 ret = blk_trace_remove_queue(q);
1717 goto out_unlock_bdev;
1721 if (q->blk_trace == NULL)
1722 ret = blk_trace_setup_queue(q, bdev);
1725 if (attr == &dev_attr_act_mask)
1726 q->blk_trace->act_mask = value;
1727 else if (attr == &dev_attr_pid)
1728 q->blk_trace->pid = value;
1729 else if (attr == &dev_attr_start_lba)
1730 q->blk_trace->start_lba = value;
1731 else if (attr == &dev_attr_end_lba)
1732 q->blk_trace->end_lba = value;
1736 mutex_unlock(&bdev->bd_mutex);
1740 return ret ? ret : count;
1743 int blk_trace_init_sysfs(struct device *dev)
1745 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1748 void blk_trace_remove_sysfs(struct device *dev)
1750 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1753 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1755 #ifdef CONFIG_EVENT_TRACING
1757 void blk_dump_cmd(char *buf, struct request *rq)
1760 int len = rq->cmd_len;
1761 unsigned char *cmd = rq->cmd;
1763 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1768 for (end = len - 1; end >= 0; end--)
1773 for (i = 0; i < len; i++) {
1774 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1775 if (i == end && end != len - 1) {
1776 sprintf(buf, " ..");
1782 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1791 else if (rw & REQ_DISCARD)
1800 if (rw & REQ_RAHEAD)
1806 if (rw & REQ_SECURE)
1811 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1813 #endif /* CONFIG_EVENT_TRACING */