1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/blktrace_api.h>
12 #include <linux/percpu.h>
13 #include <linux/init.h>
14 #include <linux/mutex.h>
15 #include <linux/slab.h>
16 #include <linux/debugfs.h>
17 #include <linux/export.h>
18 #include <linux/time.h>
19 #include <linux/uaccess.h>
20 #include <linux/list.h>
21 #include <linux/blk-cgroup.h>
23 #include "../../block/blk.h"
25 #include <trace/events/block.h>
27 #include "trace_output.h"
29 #ifdef CONFIG_BLK_DEV_IO_TRACE
31 static unsigned int blktrace_seq __read_mostly = 1;
33 static struct trace_array *blk_tr;
34 static bool blk_tracer_enabled __read_mostly;
36 static LIST_HEAD(running_trace_list);
37 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
39 /* Select an alternative, minimalistic output than the original one */
40 #define TRACE_BLK_OPT_CLASSIC 0x1
41 #define TRACE_BLK_OPT_CGROUP 0x2
42 #define TRACE_BLK_OPT_CGNAME 0x4
44 static struct tracer_opt blk_tracer_opts[] = {
45 /* Default disable the minimalistic output */
46 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
47 #ifdef CONFIG_BLK_CGROUP
48 { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
49 { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
54 static struct tracer_flags blk_tracer_flags = {
56 .opts = blk_tracer_opts,
59 /* Global reference count of probes */
60 static DEFINE_MUTEX(blk_probe_mutex);
61 static int blk_probes_ref;
63 static void blk_register_tracepoints(void);
64 static void blk_unregister_tracepoints(void);
67 * Send out a notify message.
69 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
70 const void *data, size_t len, u64 cgid)
72 struct blk_io_trace *t;
73 struct ring_buffer_event *event = NULL;
74 struct trace_buffer *buffer = NULL;
76 int cpu = smp_processor_id();
77 bool blk_tracer = blk_tracer_enabled;
78 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
81 buffer = blk_tr->array_buffer.buffer;
83 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
84 sizeof(*t) + len + cgid_len,
88 t = ring_buffer_event_data(event);
95 t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
97 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
98 t->time = ktime_to_ns(ktime_get());
101 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
104 t->pdu_len = len + cgid_len;
106 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
107 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
110 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
115 * Send out a notify for this process, if we haven't done so since a trace
118 static void trace_note_tsk(struct task_struct *tsk)
121 struct blk_trace *bt;
123 tsk->btrace_seq = blktrace_seq;
124 spin_lock_irqsave(&running_trace_lock, flags);
125 list_for_each_entry(bt, &running_trace_list, running_list) {
126 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
127 sizeof(tsk->comm), 0);
129 spin_unlock_irqrestore(&running_trace_lock, flags);
132 static void trace_note_time(struct blk_trace *bt)
134 struct timespec64 now;
138 /* need to check user space to see if this breaks in y2038 or y2106 */
139 ktime_get_real_ts64(&now);
140 words[0] = (u32)now.tv_sec;
141 words[1] = now.tv_nsec;
143 local_irq_save(flags);
144 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
145 local_irq_restore(flags);
148 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
149 const char *fmt, ...)
156 if (unlikely(bt->trace_state != Blktrace_running &&
157 !blk_tracer_enabled))
161 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
162 * message to the trace.
164 if (!(bt->act_mask & BLK_TC_NOTIFY))
167 local_irq_save(flags);
168 buf = this_cpu_ptr(bt->msg_data);
170 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
173 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
175 #ifdef CONFIG_BLK_CGROUP
176 trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n,
177 blkcg ? cgroup_id(blkcg->css.cgroup) : 1);
179 trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, 0);
181 local_irq_restore(flags);
183 EXPORT_SYMBOL_GPL(__trace_note_message);
185 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
188 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
190 if (sector && (sector < bt->start_lba || sector > bt->end_lba))
192 if (bt->pid && pid != bt->pid)
199 * Data direction bit lookup
201 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
202 BLK_TC_ACT(BLK_TC_WRITE) };
204 #define BLK_TC_RAHEAD BLK_TC_AHEAD
205 #define BLK_TC_PREFLUSH BLK_TC_FLUSH
207 /* The ilog2() calls fall out because they're constant */
208 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
209 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
212 * The worker for the various blk_add_trace*() types. Fills out a
213 * blk_io_trace structure and places it in a per-cpu subbuffer.
215 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
216 int op, int op_flags, u32 what, int error, int pdu_len,
217 void *pdu_data, u64 cgid)
219 struct task_struct *tsk = current;
220 struct ring_buffer_event *event = NULL;
221 struct trace_buffer *buffer = NULL;
222 struct blk_io_trace *t;
223 unsigned long flags = 0;
224 unsigned long *sequence;
227 bool blk_tracer = blk_tracer_enabled;
228 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
230 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
233 what |= ddir_act[op_is_write(op) ? WRITE : READ];
234 what |= MASK_TC_BIT(op_flags, SYNC);
235 what |= MASK_TC_BIT(op_flags, RAHEAD);
236 what |= MASK_TC_BIT(op_flags, META);
237 what |= MASK_TC_BIT(op_flags, PREFLUSH);
238 what |= MASK_TC_BIT(op_flags, FUA);
239 if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
240 what |= BLK_TC_ACT(BLK_TC_DISCARD);
241 if (op == REQ_OP_FLUSH)
242 what |= BLK_TC_ACT(BLK_TC_FLUSH);
244 what |= __BLK_TA_CGROUP;
247 if (act_log_check(bt, what, sector, pid))
249 cpu = raw_smp_processor_id();
252 tracing_record_cmdline(current);
254 buffer = blk_tr->array_buffer.buffer;
255 pc = preempt_count();
256 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
257 sizeof(*t) + pdu_len + cgid_len,
261 t = ring_buffer_event_data(event);
265 if (unlikely(tsk->btrace_seq != blktrace_seq))
269 * A word about the locking here - we disable interrupts to reserve
270 * some space in the relay per-cpu buffer, to prevent an irq
271 * from coming in and stepping on our toes.
273 local_irq_save(flags);
274 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
276 sequence = per_cpu_ptr(bt->sequence, cpu);
278 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
279 t->sequence = ++(*sequence);
280 t->time = ktime_to_ns(ktime_get());
283 * These two are not needed in ftrace as they are in the
284 * generic trace_entry, filled by tracing_generic_entry_update,
285 * but for the trace_event->bin() synthesizer benefit we do it
296 t->pdu_len = pdu_len + cgid_len;
299 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
301 memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
304 trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
309 local_irq_restore(flags);
312 static void blk_trace_free(struct blk_trace *bt)
314 debugfs_remove(bt->msg_file);
315 debugfs_remove(bt->dropped_file);
316 relay_close(bt->rchan);
317 debugfs_remove(bt->dir);
318 free_percpu(bt->sequence);
319 free_percpu(bt->msg_data);
323 static void get_probe_ref(void)
325 mutex_lock(&blk_probe_mutex);
326 if (++blk_probes_ref == 1)
327 blk_register_tracepoints();
328 mutex_unlock(&blk_probe_mutex);
331 static void put_probe_ref(void)
333 mutex_lock(&blk_probe_mutex);
334 if (!--blk_probes_ref)
335 blk_unregister_tracepoints();
336 mutex_unlock(&blk_probe_mutex);
339 static void blk_trace_cleanup(struct blk_trace *bt)
346 static int __blk_trace_remove(struct request_queue *q)
348 struct blk_trace *bt;
350 bt = rcu_replace_pointer(q->blk_trace, NULL,
351 lockdep_is_held(&q->debugfs_mutex));
355 if (bt->trace_state != Blktrace_running)
356 blk_trace_cleanup(bt);
361 int blk_trace_remove(struct request_queue *q)
365 mutex_lock(&q->debugfs_mutex);
366 ret = __blk_trace_remove(q);
367 mutex_unlock(&q->debugfs_mutex);
371 EXPORT_SYMBOL_GPL(blk_trace_remove);
373 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
374 size_t count, loff_t *ppos)
376 struct blk_trace *bt = filp->private_data;
379 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
381 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
384 static const struct file_operations blk_dropped_fops = {
385 .owner = THIS_MODULE,
387 .read = blk_dropped_read,
388 .llseek = default_llseek,
391 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
392 size_t count, loff_t *ppos)
395 struct blk_trace *bt;
397 if (count >= BLK_TN_MAX_MSG)
400 msg = memdup_user_nul(buffer, count);
404 bt = filp->private_data;
405 __trace_note_message(bt, NULL, "%s", msg);
411 static const struct file_operations blk_msg_fops = {
412 .owner = THIS_MODULE,
414 .write = blk_msg_write,
415 .llseek = noop_llseek,
419 * Keep track of how many times we encountered a full subbuffer, to aid
420 * the user space app in telling how many lost events there were.
422 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
423 void *prev_subbuf, size_t prev_padding)
425 struct blk_trace *bt;
427 if (!relay_buf_full(buf))
430 bt = buf->chan->private_data;
431 atomic_inc(&bt->dropped);
435 static int blk_remove_buf_file_callback(struct dentry *dentry)
437 debugfs_remove(dentry);
442 static struct dentry *blk_create_buf_file_callback(const char *filename,
443 struct dentry *parent,
445 struct rchan_buf *buf,
448 return debugfs_create_file(filename, mode, parent, buf,
449 &relay_file_operations);
452 static struct rchan_callbacks blk_relay_callbacks = {
453 .subbuf_start = blk_subbuf_start_callback,
454 .create_buf_file = blk_create_buf_file_callback,
455 .remove_buf_file = blk_remove_buf_file_callback,
458 static void blk_trace_setup_lba(struct blk_trace *bt,
459 struct block_device *bdev)
462 bt->start_lba = bdev->bd_start_sect;
463 bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
471 * Setup everything required to start tracing
473 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
474 struct block_device *bdev,
475 struct blk_user_trace_setup *buts)
477 struct blk_trace *bt = NULL;
478 struct dentry *dir = NULL;
481 lockdep_assert_held(&q->debugfs_mutex);
483 if (!buts->buf_size || !buts->buf_nr)
486 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
487 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
490 * some device names have larger paths - convert the slashes
491 * to underscores for this to work as expected
493 strreplace(buts->name, '/', '_');
496 * bdev can be NULL, as with scsi-generic, this is a helpful as
499 if (rcu_dereference_protected(q->blk_trace,
500 lockdep_is_held(&q->debugfs_mutex))) {
501 pr_warn("Concurrent blktraces are not allowed on %s\n",
506 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
511 bt->sequence = alloc_percpu(unsigned long);
515 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
520 * When tracing the whole disk reuse the existing debugfs directory
521 * created by the block layer on init. For partitions block devices,
522 * and scsi-generic block devices we create a temporary new debugfs
523 * directory that will be removed once the trace ends.
525 if (bdev && !bdev_is_partition(bdev))
526 dir = q->debugfs_dir;
528 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
531 * As blktrace relies on debugfs for its interface the debugfs directory
532 * is required, contrary to the usual mantra of not checking for debugfs
533 * files or directories.
535 if (IS_ERR_OR_NULL(dir)) {
536 pr_warn("debugfs_dir not present for %s so skipping\n",
543 atomic_set(&bt->dropped, 0);
544 INIT_LIST_HEAD(&bt->running_list);
547 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
550 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
552 bt->rchan = relay_open("trace", dir, buts->buf_size,
553 buts->buf_nr, &blk_relay_callbacks, bt);
557 bt->act_mask = buts->act_mask;
559 bt->act_mask = (u16) -1;
561 blk_trace_setup_lba(bt, bdev);
563 /* overwrite with user settings */
565 bt->start_lba = buts->start_lba;
567 bt->end_lba = buts->end_lba;
570 bt->trace_state = Blktrace_setup;
572 rcu_assign_pointer(q->blk_trace, bt);
582 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
583 struct block_device *bdev, char __user *arg)
585 struct blk_user_trace_setup buts;
588 ret = copy_from_user(&buts, arg, sizeof(buts));
592 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
596 if (copy_to_user(arg, &buts, sizeof(buts))) {
597 __blk_trace_remove(q);
603 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
604 struct block_device *bdev,
609 mutex_lock(&q->debugfs_mutex);
610 ret = __blk_trace_setup(q, name, dev, bdev, arg);
611 mutex_unlock(&q->debugfs_mutex);
615 EXPORT_SYMBOL_GPL(blk_trace_setup);
617 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
618 static int compat_blk_trace_setup(struct request_queue *q, char *name,
619 dev_t dev, struct block_device *bdev,
622 struct blk_user_trace_setup buts;
623 struct compat_blk_user_trace_setup cbuts;
626 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
629 buts = (struct blk_user_trace_setup) {
630 .act_mask = cbuts.act_mask,
631 .buf_size = cbuts.buf_size,
632 .buf_nr = cbuts.buf_nr,
633 .start_lba = cbuts.start_lba,
634 .end_lba = cbuts.end_lba,
638 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
642 if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
643 __blk_trace_remove(q);
651 static int __blk_trace_startstop(struct request_queue *q, int start)
654 struct blk_trace *bt;
656 bt = rcu_dereference_protected(q->blk_trace,
657 lockdep_is_held(&q->debugfs_mutex));
662 * For starting a trace, we can transition from a setup or stopped
663 * trace. For stopping a trace, the state must be running
667 if (bt->trace_state == Blktrace_setup ||
668 bt->trace_state == Blktrace_stopped) {
671 bt->trace_state = Blktrace_running;
672 spin_lock_irq(&running_trace_lock);
673 list_add(&bt->running_list, &running_trace_list);
674 spin_unlock_irq(&running_trace_lock);
680 if (bt->trace_state == Blktrace_running) {
681 bt->trace_state = Blktrace_stopped;
682 spin_lock_irq(&running_trace_lock);
683 list_del_init(&bt->running_list);
684 spin_unlock_irq(&running_trace_lock);
685 relay_flush(bt->rchan);
693 int blk_trace_startstop(struct request_queue *q, int start)
697 mutex_lock(&q->debugfs_mutex);
698 ret = __blk_trace_startstop(q, start);
699 mutex_unlock(&q->debugfs_mutex);
703 EXPORT_SYMBOL_GPL(blk_trace_startstop);
706 * When reading or writing the blktrace sysfs files, the references to the
707 * opened sysfs or device files should prevent the underlying block device
708 * from being removed. So no further delete protection is really needed.
712 * blk_trace_ioctl: - handle the ioctls associated with tracing
713 * @bdev: the block device
714 * @cmd: the ioctl cmd
715 * @arg: the argument data, if any
718 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
720 struct request_queue *q;
722 char b[BDEVNAME_SIZE];
724 q = bdev_get_queue(bdev);
728 mutex_lock(&q->debugfs_mutex);
733 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
735 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
736 case BLKTRACESETUP32:
738 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
745 ret = __blk_trace_startstop(q, start);
747 case BLKTRACETEARDOWN:
748 ret = __blk_trace_remove(q);
755 mutex_unlock(&q->debugfs_mutex);
760 * blk_trace_shutdown: - stop and cleanup trace structures
761 * @q: the request queue associated with the device
764 void blk_trace_shutdown(struct request_queue *q)
766 mutex_lock(&q->debugfs_mutex);
767 if (rcu_dereference_protected(q->blk_trace,
768 lockdep_is_held(&q->debugfs_mutex))) {
769 __blk_trace_startstop(q, 0);
770 __blk_trace_remove(q);
773 mutex_unlock(&q->debugfs_mutex);
776 #ifdef CONFIG_BLK_CGROUP
777 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
779 struct blk_trace *bt;
781 /* We don't use the 'bt' value here except as an optimization... */
782 bt = rcu_dereference_protected(q->blk_trace, 1);
783 if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
788 return cgroup_id(bio_blkcg(bio)->css.cgroup);
791 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
798 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
802 /* Use the first bio */
803 return blk_trace_bio_get_cgid(q, rq->bio);
811 * blk_add_trace_rq - Add a trace for a request oriented action
812 * @rq: the source request
813 * @error: return status to log
814 * @nr_bytes: number of completed bytes
816 * @cgid: the cgroup info
819 * Records an action against a request. Will log the bio offset + size.
822 static void blk_add_trace_rq(struct request *rq, int error,
823 unsigned int nr_bytes, u32 what, u64 cgid)
825 struct blk_trace *bt;
828 bt = rcu_dereference(rq->q->blk_trace);
834 if (blk_rq_is_passthrough(rq))
835 what |= BLK_TC_ACT(BLK_TC_PC);
837 what |= BLK_TC_ACT(BLK_TC_FS);
839 __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
840 rq->cmd_flags, what, error, 0, NULL, cgid);
844 static void blk_add_trace_rq_insert(void *ignore,
845 struct request_queue *q, struct request *rq)
847 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
848 blk_trace_request_get_cgid(q, rq));
851 static void blk_add_trace_rq_issue(void *ignore,
852 struct request_queue *q, struct request *rq)
854 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
855 blk_trace_request_get_cgid(q, rq));
858 static void blk_add_trace_rq_merge(void *ignore,
859 struct request_queue *q, struct request *rq)
861 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
862 blk_trace_request_get_cgid(q, rq));
865 static void blk_add_trace_rq_requeue(void *ignore,
866 struct request_queue *q,
869 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
870 blk_trace_request_get_cgid(q, rq));
873 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
874 int error, unsigned int nr_bytes)
876 blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
877 blk_trace_request_get_cgid(rq->q, rq));
881 * blk_add_trace_bio - Add a trace for a bio oriented action
882 * @q: queue the io is for
883 * @bio: the source bio
885 * @error: error, if any
888 * Records an action against a bio. Will log the bio offset + size.
891 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
894 struct blk_trace *bt;
897 bt = rcu_dereference(q->blk_trace);
903 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
904 bio_op(bio), bio->bi_opf, what, error, 0, NULL,
905 blk_trace_bio_get_cgid(q, bio));
909 static void blk_add_trace_bio_bounce(void *ignore,
910 struct request_queue *q, struct bio *bio)
912 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
915 static void blk_add_trace_bio_complete(void *ignore,
916 struct request_queue *q, struct bio *bio)
918 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
919 blk_status_to_errno(bio->bi_status));
922 static void blk_add_trace_bio_backmerge(void *ignore,
923 struct request_queue *q,
927 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
930 static void blk_add_trace_bio_frontmerge(void *ignore,
931 struct request_queue *q,
935 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
938 static void blk_add_trace_bio_queue(void *ignore,
939 struct request_queue *q, struct bio *bio)
941 blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
944 static void blk_add_trace_getrq(void *ignore,
945 struct request_queue *q,
946 struct bio *bio, int rw)
949 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
951 struct blk_trace *bt;
954 bt = rcu_dereference(q->blk_trace);
956 __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
962 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
964 struct blk_trace *bt;
967 bt = rcu_dereference(q->blk_trace);
969 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
973 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
974 unsigned int depth, bool explicit)
976 struct blk_trace *bt;
979 bt = rcu_dereference(q->blk_trace);
981 __be64 rpdu = cpu_to_be64(depth);
985 what = BLK_TA_UNPLUG_IO;
987 what = BLK_TA_UNPLUG_TIMER;
989 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
994 static void blk_add_trace_split(void *ignore,
995 struct request_queue *q, struct bio *bio,
998 struct blk_trace *bt;
1001 bt = rcu_dereference(q->blk_trace);
1003 __be64 rpdu = cpu_to_be64(pdu);
1005 __blk_add_trace(bt, bio->bi_iter.bi_sector,
1006 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
1008 blk_status_to_errno(bio->bi_status),
1009 sizeof(rpdu), &rpdu,
1010 blk_trace_bio_get_cgid(q, bio));
1016 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1017 * @ignore: trace callback data parameter (not used)
1018 * @q: queue the io is for
1019 * @bio: the source bio
1020 * @dev: target device
1021 * @from: source sector
1024 * Device mapper or raid target sometimes need to split a bio because
1025 * it spans a stripe (or similar). Add a trace for that action.
1028 static void blk_add_trace_bio_remap(void *ignore,
1029 struct request_queue *q, struct bio *bio,
1030 dev_t dev, sector_t from)
1032 struct blk_trace *bt;
1033 struct blk_io_trace_remap r;
1036 bt = rcu_dereference(q->blk_trace);
1042 r.device_from = cpu_to_be32(dev);
1043 r.device_to = cpu_to_be32(bio_dev(bio));
1044 r.sector_from = cpu_to_be64(from);
1046 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1047 bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1048 blk_status_to_errno(bio->bi_status),
1049 sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1054 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1055 * @ignore: trace callback data parameter (not used)
1056 * @q: queue the io is for
1057 * @rq: the source request
1058 * @dev: target device
1059 * @from: source sector
1062 * Device mapper remaps request to other devices.
1063 * Add a trace for that action.
1066 static void blk_add_trace_rq_remap(void *ignore,
1067 struct request_queue *q,
1068 struct request *rq, dev_t dev,
1071 struct blk_trace *bt;
1072 struct blk_io_trace_remap r;
1075 bt = rcu_dereference(q->blk_trace);
1081 r.device_from = cpu_to_be32(dev);
1082 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk));
1083 r.sector_from = cpu_to_be64(from);
1085 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1086 rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1087 sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1092 * blk_add_driver_data - Add binary message with driver-specific data
1093 * @q: queue the io is for
1095 * @data: driver-specific data
1096 * @len: length of driver-specific data
1099 * Some drivers might want to write driver-specific data per request.
1102 void blk_add_driver_data(struct request_queue *q,
1104 void *data, size_t len)
1106 struct blk_trace *bt;
1109 bt = rcu_dereference(q->blk_trace);
1115 __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1116 BLK_TA_DRV_DATA, 0, len, data,
1117 blk_trace_request_get_cgid(q, rq));
1120 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1122 static void blk_register_tracepoints(void)
1126 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1128 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1130 ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1132 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1134 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1136 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1138 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1140 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1142 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1144 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1146 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1148 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1150 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1152 ret = register_trace_block_split(blk_add_trace_split, NULL);
1154 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1156 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1160 static void blk_unregister_tracepoints(void)
1162 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1163 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1164 unregister_trace_block_split(blk_add_trace_split, NULL);
1165 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1166 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1167 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1168 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1169 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1170 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1171 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1172 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1173 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1174 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1175 unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1176 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1177 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1179 tracepoint_synchronize_unregister();
1183 * struct blk_io_tracer formatting routines
1186 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1189 int tc = t->action >> BLK_TC_SHIFT;
1191 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1196 if (tc & BLK_TC_FLUSH)
1199 if (tc & BLK_TC_DISCARD)
1201 else if (tc & BLK_TC_WRITE)
1208 if (tc & BLK_TC_FUA)
1210 if (tc & BLK_TC_AHEAD)
1212 if (tc & BLK_TC_SYNC)
1214 if (tc & BLK_TC_META)
1221 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1223 return (const struct blk_io_trace *)ent;
1226 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1228 return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1231 static inline u64 t_cgid(const struct trace_entry *ent)
1233 return *(u64 *)(te_blk_io_trace(ent) + 1);
1236 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1238 return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1241 static inline u32 t_action(const struct trace_entry *ent)
1243 return te_blk_io_trace(ent)->action;
1246 static inline u32 t_bytes(const struct trace_entry *ent)
1248 return te_blk_io_trace(ent)->bytes;
1251 static inline u32 t_sec(const struct trace_entry *ent)
1253 return te_blk_io_trace(ent)->bytes >> 9;
1256 static inline unsigned long long t_sector(const struct trace_entry *ent)
1258 return te_blk_io_trace(ent)->sector;
1261 static inline __u16 t_error(const struct trace_entry *ent)
1263 return te_blk_io_trace(ent)->error;
1266 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1268 const __be64 *val = pdu_start(ent, has_cg);
1269 return be64_to_cpu(*val);
1272 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1275 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1278 char rwbs[RWBS_LEN];
1279 unsigned long long ts = iter->ts;
1280 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1281 unsigned secs = (unsigned long)ts;
1282 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1286 trace_seq_printf(&iter->seq,
1287 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1288 MAJOR(t->device), MINOR(t->device), iter->cpu,
1289 secs, nsec_rem, iter->ent->pid, act, rwbs);
1292 static void blk_log_action(struct trace_iterator *iter, const char *act,
1295 char rwbs[RWBS_LEN];
1296 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1300 u64 id = t_cgid(iter->ent);
1302 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1303 char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1305 cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1306 sizeof(blkcg_name_buf));
1307 trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1308 MAJOR(t->device), MINOR(t->device),
1309 blkcg_name_buf, act, rwbs);
1312 * The cgid portion used to be "INO,GEN". Userland
1313 * builds a FILEID_INO32_GEN fid out of them and
1314 * opens the cgroup using open_by_handle_at(2).
1315 * While 32bit ino setups are still the same, 64bit
1316 * ones now use the 64bit ino as the whole ID and
1317 * no longer use generation.
1319 * Regarldess of the content, always output
1320 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1321 * be mapped back to @id on both 64 and 32bit ino
1322 * setups. See __kernfs_fh_to_dentry().
1324 trace_seq_printf(&iter->seq,
1325 "%3d,%-3d %llx,%-llx %2s %3s ",
1326 MAJOR(t->device), MINOR(t->device),
1327 id & U32_MAX, id >> 32, act, rwbs);
1330 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1331 MAJOR(t->device), MINOR(t->device), act, rwbs);
1334 static void blk_log_dump_pdu(struct trace_seq *s,
1335 const struct trace_entry *ent, bool has_cg)
1337 const unsigned char *pdu_buf;
1341 pdu_buf = pdu_start(ent, has_cg);
1342 pdu_len = pdu_real_len(ent, has_cg);
1347 /* find the last zero that needs to be printed */
1348 for (end = pdu_len - 1; end >= 0; end--)
1353 trace_seq_putc(s, '(');
1355 for (i = 0; i < pdu_len; i++) {
1357 trace_seq_printf(s, "%s%02x",
1358 i == 0 ? "" : " ", pdu_buf[i]);
1361 * stop when the rest is just zeroes and indicate so
1362 * with a ".." appended
1364 if (i == end && end != pdu_len - 1) {
1365 trace_seq_puts(s, " ..) ");
1370 trace_seq_puts(s, ") ");
1373 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1375 char cmd[TASK_COMM_LEN];
1377 trace_find_cmdline(ent->pid, cmd);
1379 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1380 trace_seq_printf(s, "%u ", t_bytes(ent));
1381 blk_log_dump_pdu(s, ent, has_cg);
1382 trace_seq_printf(s, "[%s]\n", cmd);
1385 trace_seq_printf(s, "%llu + %u [%s]\n",
1386 t_sector(ent), t_sec(ent), cmd);
1388 trace_seq_printf(s, "[%s]\n", cmd);
1392 static void blk_log_with_error(struct trace_seq *s,
1393 const struct trace_entry *ent, bool has_cg)
1395 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1396 blk_log_dump_pdu(s, ent, has_cg);
1397 trace_seq_printf(s, "[%d]\n", t_error(ent));
1400 trace_seq_printf(s, "%llu + %u [%d]\n",
1402 t_sec(ent), t_error(ent));
1404 trace_seq_printf(s, "%llu [%d]\n",
1405 t_sector(ent), t_error(ent));
1409 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1411 const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1413 trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1414 t_sector(ent), t_sec(ent),
1415 MAJOR(be32_to_cpu(__r->device_from)),
1416 MINOR(be32_to_cpu(__r->device_from)),
1417 be64_to_cpu(__r->sector_from));
1420 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1422 char cmd[TASK_COMM_LEN];
1424 trace_find_cmdline(ent->pid, cmd);
1426 trace_seq_printf(s, "[%s]\n", cmd);
1429 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1431 char cmd[TASK_COMM_LEN];
1433 trace_find_cmdline(ent->pid, cmd);
1435 trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1438 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1440 char cmd[TASK_COMM_LEN];
1442 trace_find_cmdline(ent->pid, cmd);
1444 trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1445 get_pdu_int(ent, has_cg), cmd);
1448 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1452 trace_seq_putmem(s, pdu_start(ent, has_cg),
1453 pdu_real_len(ent, has_cg));
1454 trace_seq_putc(s, '\n');
1458 * struct tracer operations
1461 static void blk_tracer_print_header(struct seq_file *m)
1463 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1465 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1469 static void blk_tracer_start(struct trace_array *tr)
1471 blk_tracer_enabled = true;
1474 static int blk_tracer_init(struct trace_array *tr)
1477 blk_tracer_start(tr);
1481 static void blk_tracer_stop(struct trace_array *tr)
1483 blk_tracer_enabled = false;
1486 static void blk_tracer_reset(struct trace_array *tr)
1488 blk_tracer_stop(tr);
1491 static const struct {
1493 void (*print)(struct trace_seq *s, const struct trace_entry *ent,
1496 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1497 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1498 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1499 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1500 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1501 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1502 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1503 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1504 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1505 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1506 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1507 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1508 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1509 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1510 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1513 static enum print_line_t print_one_line(struct trace_iterator *iter,
1516 struct trace_array *tr = iter->tr;
1517 struct trace_seq *s = &iter->seq;
1518 const struct blk_io_trace *t;
1521 blk_log_action_t *log_action;
1524 t = te_blk_io_trace(iter->ent);
1525 what = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1526 long_act = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1527 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1528 has_cg = t->action & __BLK_TA_CGROUP;
1530 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1531 log_action(iter, long_act ? "message" : "m", has_cg);
1532 blk_log_msg(s, iter->ent, has_cg);
1533 return trace_handle_return(s);
1536 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1537 trace_seq_printf(s, "Unknown action %x\n", what);
1539 log_action(iter, what2act[what].act[long_act], has_cg);
1540 what2act[what].print(s, iter->ent, has_cg);
1543 return trace_handle_return(s);
1546 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1547 int flags, struct trace_event *event)
1549 return print_one_line(iter, false);
1552 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1554 struct trace_seq *s = &iter->seq;
1555 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1556 const int offset = offsetof(struct blk_io_trace, sector);
1557 struct blk_io_trace old = {
1558 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1562 trace_seq_putmem(s, &old, offset);
1563 trace_seq_putmem(s, &t->sector,
1564 sizeof(old) - offset + t->pdu_len);
1567 static enum print_line_t
1568 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1569 struct trace_event *event)
1571 blk_trace_synthesize_old_trace(iter);
1573 return trace_handle_return(&iter->seq);
1576 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1578 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1579 return TRACE_TYPE_UNHANDLED;
1581 return print_one_line(iter, true);
1585 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1587 /* don't output context-info for blk_classic output */
1588 if (bit == TRACE_BLK_OPT_CLASSIC) {
1590 tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1592 tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1597 static struct tracer blk_tracer __read_mostly = {
1599 .init = blk_tracer_init,
1600 .reset = blk_tracer_reset,
1601 .start = blk_tracer_start,
1602 .stop = blk_tracer_stop,
1603 .print_header = blk_tracer_print_header,
1604 .print_line = blk_tracer_print_line,
1605 .flags = &blk_tracer_flags,
1606 .set_flag = blk_tracer_set_flag,
1609 static struct trace_event_functions trace_blk_event_funcs = {
1610 .trace = blk_trace_event_print,
1611 .binary = blk_trace_event_print_binary,
1614 static struct trace_event trace_blk_event = {
1616 .funcs = &trace_blk_event_funcs,
1619 static int __init init_blk_tracer(void)
1621 if (!register_trace_event(&trace_blk_event)) {
1622 pr_warn("Warning: could not register block events\n");
1626 if (register_tracer(&blk_tracer) != 0) {
1627 pr_warn("Warning: could not register the block tracer\n");
1628 unregister_trace_event(&trace_blk_event);
1635 device_initcall(init_blk_tracer);
1637 static int blk_trace_remove_queue(struct request_queue *q)
1639 struct blk_trace *bt;
1641 bt = rcu_replace_pointer(q->blk_trace, NULL,
1642 lockdep_is_held(&q->debugfs_mutex));
1653 * Setup everything required to start tracing
1655 static int blk_trace_setup_queue(struct request_queue *q,
1656 struct block_device *bdev)
1658 struct blk_trace *bt = NULL;
1661 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1665 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1669 bt->dev = bdev->bd_dev;
1670 bt->act_mask = (u16)-1;
1672 blk_trace_setup_lba(bt, bdev);
1674 rcu_assign_pointer(q->blk_trace, bt);
1684 * sysfs interface to enable and configure tracing
1687 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1688 struct device_attribute *attr,
1690 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1691 struct device_attribute *attr,
1692 const char *buf, size_t count);
1693 #define BLK_TRACE_DEVICE_ATTR(_name) \
1694 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1695 sysfs_blk_trace_attr_show, \
1696 sysfs_blk_trace_attr_store)
1698 static BLK_TRACE_DEVICE_ATTR(enable);
1699 static BLK_TRACE_DEVICE_ATTR(act_mask);
1700 static BLK_TRACE_DEVICE_ATTR(pid);
1701 static BLK_TRACE_DEVICE_ATTR(start_lba);
1702 static BLK_TRACE_DEVICE_ATTR(end_lba);
1704 static struct attribute *blk_trace_attrs[] = {
1705 &dev_attr_enable.attr,
1706 &dev_attr_act_mask.attr,
1708 &dev_attr_start_lba.attr,
1709 &dev_attr_end_lba.attr,
1713 struct attribute_group blk_trace_attr_group = {
1715 .attrs = blk_trace_attrs,
1718 static const struct {
1722 { BLK_TC_READ, "read" },
1723 { BLK_TC_WRITE, "write" },
1724 { BLK_TC_FLUSH, "flush" },
1725 { BLK_TC_SYNC, "sync" },
1726 { BLK_TC_QUEUE, "queue" },
1727 { BLK_TC_REQUEUE, "requeue" },
1728 { BLK_TC_ISSUE, "issue" },
1729 { BLK_TC_COMPLETE, "complete" },
1730 { BLK_TC_FS, "fs" },
1731 { BLK_TC_PC, "pc" },
1732 { BLK_TC_NOTIFY, "notify" },
1733 { BLK_TC_AHEAD, "ahead" },
1734 { BLK_TC_META, "meta" },
1735 { BLK_TC_DISCARD, "discard" },
1736 { BLK_TC_DRV_DATA, "drv_data" },
1737 { BLK_TC_FUA, "fua" },
1740 static int blk_trace_str2mask(const char *str)
1744 char *buf, *s, *token;
1746 buf = kstrdup(str, GFP_KERNEL);
1752 token = strsep(&s, ",");
1759 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1760 if (strcasecmp(token, mask_maps[i].str) == 0) {
1761 mask |= mask_maps[i].mask;
1765 if (i == ARRAY_SIZE(mask_maps)) {
1775 static ssize_t blk_trace_mask2str(char *buf, int mask)
1780 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1781 if (mask & mask_maps[i].mask) {
1782 p += sprintf(p, "%s%s",
1783 (p == buf) ? "" : ",", mask_maps[i].str);
1791 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1792 struct device_attribute *attr,
1795 struct block_device *bdev = dev_to_bdev(dev);
1796 struct request_queue *q = bdev_get_queue(bdev);
1797 struct blk_trace *bt;
1798 ssize_t ret = -ENXIO;
1800 mutex_lock(&q->debugfs_mutex);
1802 bt = rcu_dereference_protected(q->blk_trace,
1803 lockdep_is_held(&q->debugfs_mutex));
1804 if (attr == &dev_attr_enable) {
1805 ret = sprintf(buf, "%u\n", !!bt);
1806 goto out_unlock_bdev;
1810 ret = sprintf(buf, "disabled\n");
1811 else if (attr == &dev_attr_act_mask)
1812 ret = blk_trace_mask2str(buf, bt->act_mask);
1813 else if (attr == &dev_attr_pid)
1814 ret = sprintf(buf, "%u\n", bt->pid);
1815 else if (attr == &dev_attr_start_lba)
1816 ret = sprintf(buf, "%llu\n", bt->start_lba);
1817 else if (attr == &dev_attr_end_lba)
1818 ret = sprintf(buf, "%llu\n", bt->end_lba);
1821 mutex_unlock(&q->debugfs_mutex);
1825 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1826 struct device_attribute *attr,
1827 const char *buf, size_t count)
1829 struct block_device *bdev = dev_to_bdev(dev);
1830 struct request_queue *q = bdev_get_queue(bdev);
1831 struct blk_trace *bt;
1833 ssize_t ret = -EINVAL;
1838 if (attr == &dev_attr_act_mask) {
1839 if (kstrtoull(buf, 0, &value)) {
1840 /* Assume it is a list of trace category names */
1841 ret = blk_trace_str2mask(buf);
1847 if (kstrtoull(buf, 0, &value))
1851 mutex_lock(&q->debugfs_mutex);
1853 bt = rcu_dereference_protected(q->blk_trace,
1854 lockdep_is_held(&q->debugfs_mutex));
1855 if (attr == &dev_attr_enable) {
1856 if (!!value == !!bt) {
1858 goto out_unlock_bdev;
1861 ret = blk_trace_setup_queue(q, bdev);
1863 ret = blk_trace_remove_queue(q);
1864 goto out_unlock_bdev;
1869 ret = blk_trace_setup_queue(q, bdev);
1870 bt = rcu_dereference_protected(q->blk_trace,
1871 lockdep_is_held(&q->debugfs_mutex));
1875 if (attr == &dev_attr_act_mask)
1876 bt->act_mask = value;
1877 else if (attr == &dev_attr_pid)
1879 else if (attr == &dev_attr_start_lba)
1880 bt->start_lba = value;
1881 else if (attr == &dev_attr_end_lba)
1882 bt->end_lba = value;
1886 mutex_unlock(&q->debugfs_mutex);
1888 return ret ? ret : count;
1891 int blk_trace_init_sysfs(struct device *dev)
1893 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1896 void blk_trace_remove_sysfs(struct device *dev)
1898 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1901 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1903 #ifdef CONFIG_EVENT_TRACING
1905 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1909 if (op & REQ_PREFLUSH)
1912 switch (op & REQ_OP_MASK) {
1914 case REQ_OP_WRITE_SAME:
1917 case REQ_OP_DISCARD:
1920 case REQ_OP_SECURE_ERASE:
1936 if (op & REQ_RAHEAD)
1945 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1947 #endif /* CONFIG_EVENT_TRACING */