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_RAW_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;
75 unsigned int trace_ctx = 0;
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;
82 trace_ctx = tracing_gen_ctx_flags(0);
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, trace_ctx);
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 raw_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 raw_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 __blk_trace_note_message(struct blk_trace *bt,
149 struct cgroup_subsys_state *css, const char *fmt, ...)
157 if (unlikely(bt->trace_state != Blktrace_running &&
158 !blk_tracer_enabled))
162 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
163 * message to the trace.
165 if (!(bt->act_mask & BLK_TC_NOTIFY))
168 local_irq_save(flags);
169 buf = this_cpu_ptr(bt->msg_data);
171 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
174 #ifdef CONFIG_BLK_CGROUP
175 if (css && (blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
176 cgid = cgroup_id(css->cgroup);
180 trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, cgid);
181 local_irq_restore(flags);
183 EXPORT_SYMBOL_GPL(__blk_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) ((__force u32)(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 const blk_opf_t opf, u32 what, int error,
217 int pdu_len, 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;
225 unsigned int trace_ctx = 0;
228 bool blk_tracer = blk_tracer_enabled;
229 ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
230 const enum req_op op = opf & REQ_OP_MASK;
232 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
235 what |= ddir_act[op_is_write(op) ? WRITE : READ];
236 what |= MASK_TC_BIT(opf, SYNC);
237 what |= MASK_TC_BIT(opf, RAHEAD);
238 what |= MASK_TC_BIT(opf, META);
239 what |= MASK_TC_BIT(opf, PREFLUSH);
240 what |= MASK_TC_BIT(opf, FUA);
241 if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
242 what |= BLK_TC_ACT(BLK_TC_DISCARD);
243 if (op == REQ_OP_FLUSH)
244 what |= BLK_TC_ACT(BLK_TC_FLUSH);
246 what |= __BLK_TA_CGROUP;
249 if (act_log_check(bt, what, sector, pid))
251 cpu = raw_smp_processor_id();
254 tracing_record_cmdline(current);
256 buffer = blk_tr->array_buffer.buffer;
257 trace_ctx = tracing_gen_ctx_flags(0);
258 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
259 sizeof(*t) + pdu_len + cgid_len,
263 t = ring_buffer_event_data(event);
267 if (unlikely(tsk->btrace_seq != blktrace_seq))
271 * A word about the locking here - we disable interrupts to reserve
272 * some space in the relay per-cpu buffer, to prevent an irq
273 * from coming in and stepping on our toes.
275 local_irq_save(flags);
276 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
278 sequence = per_cpu_ptr(bt->sequence, cpu);
280 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
281 t->sequence = ++(*sequence);
282 t->time = ktime_to_ns(ktime_get());
285 * These two are not needed in ftrace as they are in the
286 * generic trace_entry, filled by tracing_generic_entry_update,
287 * but for the trace_event->bin() synthesizer benefit we do it
298 t->pdu_len = pdu_len + cgid_len;
301 memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
303 memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
306 trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx);
311 local_irq_restore(flags);
314 static void blk_trace_free(struct request_queue *q, struct blk_trace *bt)
316 relay_close(bt->rchan);
319 * If 'bt->dir' is not set, then both 'dropped' and 'msg' are created
320 * under 'q->debugfs_dir', thus lookup and remove them.
323 debugfs_lookup_and_remove("dropped", q->debugfs_dir);
324 debugfs_lookup_and_remove("msg", q->debugfs_dir);
326 debugfs_remove(bt->dir);
328 free_percpu(bt->sequence);
329 free_percpu(bt->msg_data);
333 static void get_probe_ref(void)
335 mutex_lock(&blk_probe_mutex);
336 if (++blk_probes_ref == 1)
337 blk_register_tracepoints();
338 mutex_unlock(&blk_probe_mutex);
341 static void put_probe_ref(void)
343 mutex_lock(&blk_probe_mutex);
344 if (!--blk_probes_ref)
345 blk_unregister_tracepoints();
346 mutex_unlock(&blk_probe_mutex);
349 static int blk_trace_start(struct blk_trace *bt)
351 if (bt->trace_state != Blktrace_setup &&
352 bt->trace_state != Blktrace_stopped)
357 bt->trace_state = Blktrace_running;
358 raw_spin_lock_irq(&running_trace_lock);
359 list_add(&bt->running_list, &running_trace_list);
360 raw_spin_unlock_irq(&running_trace_lock);
366 static int blk_trace_stop(struct blk_trace *bt)
368 if (bt->trace_state != Blktrace_running)
371 bt->trace_state = Blktrace_stopped;
372 raw_spin_lock_irq(&running_trace_lock);
373 list_del_init(&bt->running_list);
374 raw_spin_unlock_irq(&running_trace_lock);
375 relay_flush(bt->rchan);
380 static void blk_trace_cleanup(struct request_queue *q, struct blk_trace *bt)
384 blk_trace_free(q, bt);
388 static int __blk_trace_remove(struct request_queue *q)
390 struct blk_trace *bt;
392 bt = rcu_replace_pointer(q->blk_trace, NULL,
393 lockdep_is_held(&q->debugfs_mutex));
397 blk_trace_cleanup(q, bt);
402 int blk_trace_remove(struct request_queue *q)
406 mutex_lock(&q->debugfs_mutex);
407 ret = __blk_trace_remove(q);
408 mutex_unlock(&q->debugfs_mutex);
412 EXPORT_SYMBOL_GPL(blk_trace_remove);
414 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
415 size_t count, loff_t *ppos)
417 struct blk_trace *bt = filp->private_data;
420 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
422 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
425 static const struct file_operations blk_dropped_fops = {
426 .owner = THIS_MODULE,
428 .read = blk_dropped_read,
429 .llseek = default_llseek,
432 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
433 size_t count, loff_t *ppos)
436 struct blk_trace *bt;
438 if (count >= BLK_TN_MAX_MSG)
441 msg = memdup_user_nul(buffer, count);
445 bt = filp->private_data;
446 __blk_trace_note_message(bt, NULL, "%s", msg);
452 static const struct file_operations blk_msg_fops = {
453 .owner = THIS_MODULE,
455 .write = blk_msg_write,
456 .llseek = noop_llseek,
460 * Keep track of how many times we encountered a full subbuffer, to aid
461 * the user space app in telling how many lost events there were.
463 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
464 void *prev_subbuf, size_t prev_padding)
466 struct blk_trace *bt;
468 if (!relay_buf_full(buf))
471 bt = buf->chan->private_data;
472 atomic_inc(&bt->dropped);
476 static int blk_remove_buf_file_callback(struct dentry *dentry)
478 debugfs_remove(dentry);
483 static struct dentry *blk_create_buf_file_callback(const char *filename,
484 struct dentry *parent,
486 struct rchan_buf *buf,
489 return debugfs_create_file(filename, mode, parent, buf,
490 &relay_file_operations);
493 static const struct rchan_callbacks blk_relay_callbacks = {
494 .subbuf_start = blk_subbuf_start_callback,
495 .create_buf_file = blk_create_buf_file_callback,
496 .remove_buf_file = blk_remove_buf_file_callback,
499 static void blk_trace_setup_lba(struct blk_trace *bt,
500 struct block_device *bdev)
503 bt->start_lba = bdev->bd_start_sect;
504 bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
512 * Setup everything required to start tracing
514 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
515 struct block_device *bdev,
516 struct blk_user_trace_setup *buts)
518 struct blk_trace *bt = NULL;
519 struct dentry *dir = NULL;
522 lockdep_assert_held(&q->debugfs_mutex);
524 if (!buts->buf_size || !buts->buf_nr)
527 strscpy_pad(buts->name, name, BLKTRACE_BDEV_SIZE);
530 * some device names have larger paths - convert the slashes
531 * to underscores for this to work as expected
533 strreplace(buts->name, '/', '_');
536 * bdev can be NULL, as with scsi-generic, this is a helpful as
539 if (rcu_dereference_protected(q->blk_trace,
540 lockdep_is_held(&q->debugfs_mutex))) {
541 pr_warn("Concurrent blktraces are not allowed on %s\n",
546 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
551 bt->sequence = alloc_percpu(unsigned long);
555 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
560 * When tracing the whole disk reuse the existing debugfs directory
561 * created by the block layer on init. For partitions block devices,
562 * and scsi-generic block devices we create a temporary new debugfs
563 * directory that will be removed once the trace ends.
565 if (bdev && !bdev_is_partition(bdev))
566 dir = q->debugfs_dir;
568 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
571 * As blktrace relies on debugfs for its interface the debugfs directory
572 * is required, contrary to the usual mantra of not checking for debugfs
573 * files or directories.
575 if (IS_ERR_OR_NULL(dir)) {
576 pr_warn("debugfs_dir not present for %s so skipping\n",
583 atomic_set(&bt->dropped, 0);
584 INIT_LIST_HEAD(&bt->running_list);
587 debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
588 debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
590 bt->rchan = relay_open("trace", dir, buts->buf_size,
591 buts->buf_nr, &blk_relay_callbacks, bt);
595 bt->act_mask = buts->act_mask;
597 bt->act_mask = (u16) -1;
599 blk_trace_setup_lba(bt, bdev);
601 /* overwrite with user settings */
603 bt->start_lba = buts->start_lba;
605 bt->end_lba = buts->end_lba;
608 bt->trace_state = Blktrace_setup;
610 rcu_assign_pointer(q->blk_trace, bt);
616 blk_trace_free(q, bt);
620 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
621 struct block_device *bdev, char __user *arg)
623 struct blk_user_trace_setup buts;
626 ret = copy_from_user(&buts, arg, sizeof(buts));
630 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
634 if (copy_to_user(arg, &buts, sizeof(buts))) {
635 __blk_trace_remove(q);
641 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
642 struct block_device *bdev,
647 mutex_lock(&q->debugfs_mutex);
648 ret = __blk_trace_setup(q, name, dev, bdev, arg);
649 mutex_unlock(&q->debugfs_mutex);
653 EXPORT_SYMBOL_GPL(blk_trace_setup);
655 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
656 static int compat_blk_trace_setup(struct request_queue *q, char *name,
657 dev_t dev, struct block_device *bdev,
660 struct blk_user_trace_setup buts;
661 struct compat_blk_user_trace_setup cbuts;
664 if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
667 buts = (struct blk_user_trace_setup) {
668 .act_mask = cbuts.act_mask,
669 .buf_size = cbuts.buf_size,
670 .buf_nr = cbuts.buf_nr,
671 .start_lba = cbuts.start_lba,
672 .end_lba = cbuts.end_lba,
676 ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
680 if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
681 __blk_trace_remove(q);
689 static int __blk_trace_startstop(struct request_queue *q, int start)
691 struct blk_trace *bt;
693 bt = rcu_dereference_protected(q->blk_trace,
694 lockdep_is_held(&q->debugfs_mutex));
699 return blk_trace_start(bt);
701 return blk_trace_stop(bt);
704 int blk_trace_startstop(struct request_queue *q, int start)
708 mutex_lock(&q->debugfs_mutex);
709 ret = __blk_trace_startstop(q, start);
710 mutex_unlock(&q->debugfs_mutex);
714 EXPORT_SYMBOL_GPL(blk_trace_startstop);
717 * When reading or writing the blktrace sysfs files, the references to the
718 * opened sysfs or device files should prevent the underlying block device
719 * from being removed. So no further delete protection is really needed.
723 * blk_trace_ioctl - handle the ioctls associated with tracing
724 * @bdev: the block device
725 * @cmd: the ioctl cmd
726 * @arg: the argument data, if any
729 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
731 struct request_queue *q = bdev_get_queue(bdev);
733 char b[BDEVNAME_SIZE];
735 mutex_lock(&q->debugfs_mutex);
739 snprintf(b, sizeof(b), "%pg", bdev);
740 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
742 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
743 case BLKTRACESETUP32:
744 snprintf(b, sizeof(b), "%pg", bdev);
745 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
752 ret = __blk_trace_startstop(q, start);
754 case BLKTRACETEARDOWN:
755 ret = __blk_trace_remove(q);
762 mutex_unlock(&q->debugfs_mutex);
767 * blk_trace_shutdown - stop and cleanup trace structures
768 * @q: the request queue associated with the device
771 void blk_trace_shutdown(struct request_queue *q)
773 if (rcu_dereference_protected(q->blk_trace,
774 lockdep_is_held(&q->debugfs_mutex)))
775 __blk_trace_remove(q);
778 #ifdef CONFIG_BLK_CGROUP
779 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
781 struct cgroup_subsys_state *blkcg_css;
782 struct blk_trace *bt;
784 /* We don't use the 'bt' value here except as an optimization... */
785 bt = rcu_dereference_protected(q->blk_trace, 1);
786 if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
789 blkcg_css = bio_blkcg_css(bio);
792 return cgroup_id(blkcg_css->cgroup);
795 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
802 blk_trace_request_get_cgid(struct request *rq)
806 /* Use the first bio */
807 return blk_trace_bio_get_cgid(rq->q, rq->bio);
815 * blk_add_trace_rq - Add a trace for a request oriented action
816 * @rq: the source request
817 * @error: return status to log
818 * @nr_bytes: number of completed bytes
820 * @cgid: the cgroup info
823 * Records an action against a request. Will log the bio offset + size.
826 static void blk_add_trace_rq(struct request *rq, blk_status_t error,
827 unsigned int nr_bytes, u32 what, u64 cgid)
829 struct blk_trace *bt;
832 bt = rcu_dereference(rq->q->blk_trace);
838 if (blk_rq_is_passthrough(rq))
839 what |= BLK_TC_ACT(BLK_TC_PC);
841 what |= BLK_TC_ACT(BLK_TC_FS);
843 __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, rq->cmd_flags,
844 what, blk_status_to_errno(error), 0, NULL, cgid);
848 static void blk_add_trace_rq_insert(void *ignore, struct request *rq)
850 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
851 blk_trace_request_get_cgid(rq));
854 static void blk_add_trace_rq_issue(void *ignore, struct request *rq)
856 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
857 blk_trace_request_get_cgid(rq));
860 static void blk_add_trace_rq_merge(void *ignore, struct request *rq)
862 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
863 blk_trace_request_get_cgid(rq));
866 static void blk_add_trace_rq_requeue(void *ignore, struct request *rq)
868 blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
869 blk_trace_request_get_cgid(rq));
872 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
873 blk_status_t error, unsigned int nr_bytes)
875 blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
876 blk_trace_request_get_cgid(rq));
880 * blk_add_trace_bio - Add a trace for a bio oriented action
881 * @q: queue the io is for
882 * @bio: the source bio
884 * @error: error, if any
887 * Records an action against a bio. Will log the bio offset + size.
890 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
893 struct blk_trace *bt;
896 bt = rcu_dereference(q->blk_trace);
902 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
903 bio->bi_opf, what, error, 0, NULL,
904 blk_trace_bio_get_cgid(q, bio));
908 static void blk_add_trace_bio_bounce(void *ignore, struct bio *bio)
910 blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BOUNCE, 0);
913 static void blk_add_trace_bio_complete(void *ignore,
914 struct request_queue *q, struct bio *bio)
916 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
917 blk_status_to_errno(bio->bi_status));
920 static void blk_add_trace_bio_backmerge(void *ignore, struct bio *bio)
922 blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BACKMERGE,
926 static void blk_add_trace_bio_frontmerge(void *ignore, struct bio *bio)
928 blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_FRONTMERGE,
932 static void blk_add_trace_bio_queue(void *ignore, struct bio *bio)
934 blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_QUEUE, 0);
937 static void blk_add_trace_getrq(void *ignore, struct bio *bio)
939 blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_GETRQ, 0);
942 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
944 struct blk_trace *bt;
947 bt = rcu_dereference(q->blk_trace);
949 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
953 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
954 unsigned int depth, bool explicit)
956 struct blk_trace *bt;
959 bt = rcu_dereference(q->blk_trace);
961 __be64 rpdu = cpu_to_be64(depth);
965 what = BLK_TA_UNPLUG_IO;
967 what = BLK_TA_UNPLUG_TIMER;
969 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
974 static void blk_add_trace_split(void *ignore, struct bio *bio, unsigned int pdu)
976 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
977 struct blk_trace *bt;
980 bt = rcu_dereference(q->blk_trace);
982 __be64 rpdu = cpu_to_be64(pdu);
984 __blk_add_trace(bt, bio->bi_iter.bi_sector,
985 bio->bi_iter.bi_size, bio->bi_opf, BLK_TA_SPLIT,
986 blk_status_to_errno(bio->bi_status),
988 blk_trace_bio_get_cgid(q, bio));
994 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
995 * @ignore: trace callback data parameter (not used)
996 * @bio: the source bio
997 * @dev: source device
998 * @from: source sector
1000 * Called after a bio is remapped to a different device and/or sector.
1002 static void blk_add_trace_bio_remap(void *ignore, struct bio *bio, dev_t dev,
1005 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1006 struct blk_trace *bt;
1007 struct blk_io_trace_remap r;
1010 bt = rcu_dereference(q->blk_trace);
1016 r.device_from = cpu_to_be32(dev);
1017 r.device_to = cpu_to_be32(bio_dev(bio));
1018 r.sector_from = cpu_to_be64(from);
1020 __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1021 bio->bi_opf, BLK_TA_REMAP,
1022 blk_status_to_errno(bio->bi_status),
1023 sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1028 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1029 * @ignore: trace callback data parameter (not used)
1030 * @rq: the source request
1031 * @dev: target device
1032 * @from: source sector
1035 * Device mapper remaps request to other devices.
1036 * Add a trace for that action.
1039 static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev,
1042 struct blk_trace *bt;
1043 struct blk_io_trace_remap r;
1046 bt = rcu_dereference(rq->q->blk_trace);
1052 r.device_from = cpu_to_be32(dev);
1053 r.device_to = cpu_to_be32(disk_devt(rq->q->disk));
1054 r.sector_from = cpu_to_be64(from);
1056 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1057 rq->cmd_flags, BLK_TA_REMAP, 0,
1058 sizeof(r), &r, blk_trace_request_get_cgid(rq));
1063 * blk_add_driver_data - Add binary message with driver-specific data
1065 * @data: driver-specific data
1066 * @len: length of driver-specific data
1069 * Some drivers might want to write driver-specific data per request.
1072 void blk_add_driver_data(struct request *rq, void *data, size_t len)
1074 struct blk_trace *bt;
1077 bt = rcu_dereference(rq->q->blk_trace);
1083 __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0,
1084 BLK_TA_DRV_DATA, 0, len, data,
1085 blk_trace_request_get_cgid(rq));
1088 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1090 static void blk_register_tracepoints(void)
1094 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1096 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1098 ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1100 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1102 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1104 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1106 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1108 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1110 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1112 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1114 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1116 ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1118 ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1120 ret = register_trace_block_split(blk_add_trace_split, NULL);
1122 ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1124 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1128 static void blk_unregister_tracepoints(void)
1130 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1131 unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1132 unregister_trace_block_split(blk_add_trace_split, NULL);
1133 unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1134 unregister_trace_block_plug(blk_add_trace_plug, NULL);
1135 unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1136 unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1137 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1138 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1139 unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1140 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1141 unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1142 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1143 unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1144 unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1145 unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1147 tracepoint_synchronize_unregister();
1151 * struct blk_io_tracer formatting routines
1154 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1157 int tc = t->action >> BLK_TC_SHIFT;
1159 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1164 if (tc & BLK_TC_FLUSH)
1167 if (tc & BLK_TC_DISCARD)
1169 else if (tc & BLK_TC_WRITE)
1176 if (tc & BLK_TC_FUA)
1178 if (tc & BLK_TC_AHEAD)
1180 if (tc & BLK_TC_SYNC)
1182 if (tc & BLK_TC_META)
1189 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1191 return (const struct blk_io_trace *)ent;
1194 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1196 return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1199 static inline u64 t_cgid(const struct trace_entry *ent)
1201 return *(u64 *)(te_blk_io_trace(ent) + 1);
1204 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1206 return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1209 static inline u32 t_action(const struct trace_entry *ent)
1211 return te_blk_io_trace(ent)->action;
1214 static inline u32 t_bytes(const struct trace_entry *ent)
1216 return te_blk_io_trace(ent)->bytes;
1219 static inline u32 t_sec(const struct trace_entry *ent)
1221 return te_blk_io_trace(ent)->bytes >> 9;
1224 static inline unsigned long long t_sector(const struct trace_entry *ent)
1226 return te_blk_io_trace(ent)->sector;
1229 static inline __u16 t_error(const struct trace_entry *ent)
1231 return te_blk_io_trace(ent)->error;
1234 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1236 const __be64 *val = pdu_start(ent, has_cg);
1237 return be64_to_cpu(*val);
1240 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1243 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1246 char rwbs[RWBS_LEN];
1247 unsigned long long ts = iter->ts;
1248 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1249 unsigned secs = (unsigned long)ts;
1250 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1254 trace_seq_printf(&iter->seq,
1255 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1256 MAJOR(t->device), MINOR(t->device), iter->cpu,
1257 secs, nsec_rem, iter->ent->pid, act, rwbs);
1260 static void blk_log_action(struct trace_iterator *iter, const char *act,
1263 char rwbs[RWBS_LEN];
1264 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1268 u64 id = t_cgid(iter->ent);
1270 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1271 char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1273 cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1274 sizeof(blkcg_name_buf));
1275 trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1276 MAJOR(t->device), MINOR(t->device),
1277 blkcg_name_buf, act, rwbs);
1280 * The cgid portion used to be "INO,GEN". Userland
1281 * builds a FILEID_INO32_GEN fid out of them and
1282 * opens the cgroup using open_by_handle_at(2).
1283 * While 32bit ino setups are still the same, 64bit
1284 * ones now use the 64bit ino as the whole ID and
1285 * no longer use generation.
1287 * Regardless of the content, always output
1288 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1289 * be mapped back to @id on both 64 and 32bit ino
1290 * setups. See __kernfs_fh_to_dentry().
1292 trace_seq_printf(&iter->seq,
1293 "%3d,%-3d %llx,%-llx %2s %3s ",
1294 MAJOR(t->device), MINOR(t->device),
1295 id & U32_MAX, id >> 32, act, rwbs);
1298 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1299 MAJOR(t->device), MINOR(t->device), act, rwbs);
1302 static void blk_log_dump_pdu(struct trace_seq *s,
1303 const struct trace_entry *ent, bool has_cg)
1305 const unsigned char *pdu_buf;
1309 pdu_buf = pdu_start(ent, has_cg);
1310 pdu_len = pdu_real_len(ent, has_cg);
1315 /* find the last zero that needs to be printed */
1316 for (end = pdu_len - 1; end >= 0; end--)
1321 trace_seq_putc(s, '(');
1323 for (i = 0; i < pdu_len; i++) {
1325 trace_seq_printf(s, "%s%02x",
1326 i == 0 ? "" : " ", pdu_buf[i]);
1329 * stop when the rest is just zeros and indicate so
1330 * with a ".." appended
1332 if (i == end && end != pdu_len - 1) {
1333 trace_seq_puts(s, " ..) ");
1338 trace_seq_puts(s, ") ");
1341 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1343 char cmd[TASK_COMM_LEN];
1345 trace_find_cmdline(ent->pid, cmd);
1347 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1348 trace_seq_printf(s, "%u ", t_bytes(ent));
1349 blk_log_dump_pdu(s, ent, has_cg);
1350 trace_seq_printf(s, "[%s]\n", cmd);
1353 trace_seq_printf(s, "%llu + %u [%s]\n",
1354 t_sector(ent), t_sec(ent), cmd);
1356 trace_seq_printf(s, "[%s]\n", cmd);
1360 static void blk_log_with_error(struct trace_seq *s,
1361 const struct trace_entry *ent, bool has_cg)
1363 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1364 blk_log_dump_pdu(s, ent, has_cg);
1365 trace_seq_printf(s, "[%d]\n", t_error(ent));
1368 trace_seq_printf(s, "%llu + %u [%d]\n",
1370 t_sec(ent), t_error(ent));
1372 trace_seq_printf(s, "%llu [%d]\n",
1373 t_sector(ent), t_error(ent));
1377 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1379 const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1381 trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1382 t_sector(ent), t_sec(ent),
1383 MAJOR(be32_to_cpu(__r->device_from)),
1384 MINOR(be32_to_cpu(__r->device_from)),
1385 be64_to_cpu(__r->sector_from));
1388 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1390 char cmd[TASK_COMM_LEN];
1392 trace_find_cmdline(ent->pid, cmd);
1394 trace_seq_printf(s, "[%s]\n", cmd);
1397 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1399 char cmd[TASK_COMM_LEN];
1401 trace_find_cmdline(ent->pid, cmd);
1403 trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1406 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1408 char cmd[TASK_COMM_LEN];
1410 trace_find_cmdline(ent->pid, cmd);
1412 trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1413 get_pdu_int(ent, has_cg), cmd);
1416 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1420 trace_seq_putmem(s, pdu_start(ent, has_cg),
1421 pdu_real_len(ent, has_cg));
1422 trace_seq_putc(s, '\n');
1426 * struct tracer operations
1429 static void blk_tracer_print_header(struct seq_file *m)
1431 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1433 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1437 static void blk_tracer_start(struct trace_array *tr)
1439 blk_tracer_enabled = true;
1442 static int blk_tracer_init(struct trace_array *tr)
1445 blk_tracer_start(tr);
1449 static void blk_tracer_stop(struct trace_array *tr)
1451 blk_tracer_enabled = false;
1454 static void blk_tracer_reset(struct trace_array *tr)
1456 blk_tracer_stop(tr);
1459 static const struct {
1461 void (*print)(struct trace_seq *s, const struct trace_entry *ent,
1464 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1465 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1466 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1467 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1468 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1469 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1470 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1471 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1472 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1473 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1474 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1475 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1476 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1477 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1478 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1481 static enum print_line_t print_one_line(struct trace_iterator *iter,
1484 struct trace_array *tr = iter->tr;
1485 struct trace_seq *s = &iter->seq;
1486 const struct blk_io_trace *t;
1489 blk_log_action_t *log_action;
1492 t = te_blk_io_trace(iter->ent);
1493 what = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1494 long_act = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1495 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1496 has_cg = t->action & __BLK_TA_CGROUP;
1498 if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1499 log_action(iter, long_act ? "message" : "m", has_cg);
1500 blk_log_msg(s, iter->ent, has_cg);
1501 return trace_handle_return(s);
1504 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1505 trace_seq_printf(s, "Unknown action %x\n", what);
1507 log_action(iter, what2act[what].act[long_act], has_cg);
1508 what2act[what].print(s, iter->ent, has_cg);
1511 return trace_handle_return(s);
1514 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1515 int flags, struct trace_event *event)
1517 return print_one_line(iter, false);
1520 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1522 struct trace_seq *s = &iter->seq;
1523 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1524 const int offset = offsetof(struct blk_io_trace, sector);
1525 struct blk_io_trace old = {
1526 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1530 trace_seq_putmem(s, &old, offset);
1531 trace_seq_putmem(s, &t->sector,
1532 sizeof(old) - offset + t->pdu_len);
1535 static enum print_line_t
1536 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1537 struct trace_event *event)
1539 blk_trace_synthesize_old_trace(iter);
1541 return trace_handle_return(&iter->seq);
1544 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1546 if ((iter->ent->type != TRACE_BLK) ||
1547 !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1548 return TRACE_TYPE_UNHANDLED;
1550 return print_one_line(iter, true);
1554 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1556 /* don't output context-info for blk_classic output */
1557 if (bit == TRACE_BLK_OPT_CLASSIC) {
1559 tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1561 tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1566 static struct tracer blk_tracer __read_mostly = {
1568 .init = blk_tracer_init,
1569 .reset = blk_tracer_reset,
1570 .start = blk_tracer_start,
1571 .stop = blk_tracer_stop,
1572 .print_header = blk_tracer_print_header,
1573 .print_line = blk_tracer_print_line,
1574 .flags = &blk_tracer_flags,
1575 .set_flag = blk_tracer_set_flag,
1578 static struct trace_event_functions trace_blk_event_funcs = {
1579 .trace = blk_trace_event_print,
1580 .binary = blk_trace_event_print_binary,
1583 static struct trace_event trace_blk_event = {
1585 .funcs = &trace_blk_event_funcs,
1588 static int __init init_blk_tracer(void)
1590 if (!register_trace_event(&trace_blk_event)) {
1591 pr_warn("Warning: could not register block events\n");
1595 if (register_tracer(&blk_tracer) != 0) {
1596 pr_warn("Warning: could not register the block tracer\n");
1597 unregister_trace_event(&trace_blk_event);
1604 device_initcall(init_blk_tracer);
1606 static int blk_trace_remove_queue(struct request_queue *q)
1608 struct blk_trace *bt;
1610 bt = rcu_replace_pointer(q->blk_trace, NULL,
1611 lockdep_is_held(&q->debugfs_mutex));
1619 blk_trace_free(q, bt);
1624 * Setup everything required to start tracing
1626 static int blk_trace_setup_queue(struct request_queue *q,
1627 struct block_device *bdev)
1629 struct blk_trace *bt = NULL;
1632 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1636 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1640 bt->dev = bdev->bd_dev;
1641 bt->act_mask = (u16)-1;
1643 blk_trace_setup_lba(bt, bdev);
1645 rcu_assign_pointer(q->blk_trace, bt);
1650 blk_trace_free(q, bt);
1655 * sysfs interface to enable and configure tracing
1658 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1659 struct device_attribute *attr,
1661 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1662 struct device_attribute *attr,
1663 const char *buf, size_t count);
1664 #define BLK_TRACE_DEVICE_ATTR(_name) \
1665 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1666 sysfs_blk_trace_attr_show, \
1667 sysfs_blk_trace_attr_store)
1669 static BLK_TRACE_DEVICE_ATTR(enable);
1670 static BLK_TRACE_DEVICE_ATTR(act_mask);
1671 static BLK_TRACE_DEVICE_ATTR(pid);
1672 static BLK_TRACE_DEVICE_ATTR(start_lba);
1673 static BLK_TRACE_DEVICE_ATTR(end_lba);
1675 static struct attribute *blk_trace_attrs[] = {
1676 &dev_attr_enable.attr,
1677 &dev_attr_act_mask.attr,
1679 &dev_attr_start_lba.attr,
1680 &dev_attr_end_lba.attr,
1684 struct attribute_group blk_trace_attr_group = {
1686 .attrs = blk_trace_attrs,
1689 static const struct {
1693 { BLK_TC_READ, "read" },
1694 { BLK_TC_WRITE, "write" },
1695 { BLK_TC_FLUSH, "flush" },
1696 { BLK_TC_SYNC, "sync" },
1697 { BLK_TC_QUEUE, "queue" },
1698 { BLK_TC_REQUEUE, "requeue" },
1699 { BLK_TC_ISSUE, "issue" },
1700 { BLK_TC_COMPLETE, "complete" },
1701 { BLK_TC_FS, "fs" },
1702 { BLK_TC_PC, "pc" },
1703 { BLK_TC_NOTIFY, "notify" },
1704 { BLK_TC_AHEAD, "ahead" },
1705 { BLK_TC_META, "meta" },
1706 { BLK_TC_DISCARD, "discard" },
1707 { BLK_TC_DRV_DATA, "drv_data" },
1708 { BLK_TC_FUA, "fua" },
1711 static int blk_trace_str2mask(const char *str)
1715 char *buf, *s, *token;
1717 buf = kstrdup(str, GFP_KERNEL);
1723 token = strsep(&s, ",");
1730 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1731 if (strcasecmp(token, mask_maps[i].str) == 0) {
1732 mask |= mask_maps[i].mask;
1736 if (i == ARRAY_SIZE(mask_maps)) {
1746 static ssize_t blk_trace_mask2str(char *buf, int mask)
1751 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1752 if (mask & mask_maps[i].mask) {
1753 p += sprintf(p, "%s%s",
1754 (p == buf) ? "" : ",", mask_maps[i].str);
1762 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1763 struct device_attribute *attr,
1766 struct block_device *bdev = dev_to_bdev(dev);
1767 struct request_queue *q = bdev_get_queue(bdev);
1768 struct blk_trace *bt;
1769 ssize_t ret = -ENXIO;
1771 mutex_lock(&q->debugfs_mutex);
1773 bt = rcu_dereference_protected(q->blk_trace,
1774 lockdep_is_held(&q->debugfs_mutex));
1775 if (attr == &dev_attr_enable) {
1776 ret = sprintf(buf, "%u\n", !!bt);
1777 goto out_unlock_bdev;
1781 ret = sprintf(buf, "disabled\n");
1782 else if (attr == &dev_attr_act_mask)
1783 ret = blk_trace_mask2str(buf, bt->act_mask);
1784 else if (attr == &dev_attr_pid)
1785 ret = sprintf(buf, "%u\n", bt->pid);
1786 else if (attr == &dev_attr_start_lba)
1787 ret = sprintf(buf, "%llu\n", bt->start_lba);
1788 else if (attr == &dev_attr_end_lba)
1789 ret = sprintf(buf, "%llu\n", bt->end_lba);
1792 mutex_unlock(&q->debugfs_mutex);
1796 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1797 struct device_attribute *attr,
1798 const char *buf, size_t count)
1800 struct block_device *bdev = dev_to_bdev(dev);
1801 struct request_queue *q = bdev_get_queue(bdev);
1802 struct blk_trace *bt;
1804 ssize_t ret = -EINVAL;
1809 if (attr == &dev_attr_act_mask) {
1810 if (kstrtoull(buf, 0, &value)) {
1811 /* Assume it is a list of trace category names */
1812 ret = blk_trace_str2mask(buf);
1818 if (kstrtoull(buf, 0, &value))
1822 mutex_lock(&q->debugfs_mutex);
1824 bt = rcu_dereference_protected(q->blk_trace,
1825 lockdep_is_held(&q->debugfs_mutex));
1826 if (attr == &dev_attr_enable) {
1827 if (!!value == !!bt) {
1829 goto out_unlock_bdev;
1832 ret = blk_trace_setup_queue(q, bdev);
1834 ret = blk_trace_remove_queue(q);
1835 goto out_unlock_bdev;
1840 ret = blk_trace_setup_queue(q, bdev);
1841 bt = rcu_dereference_protected(q->blk_trace,
1842 lockdep_is_held(&q->debugfs_mutex));
1846 if (attr == &dev_attr_act_mask)
1847 bt->act_mask = value;
1848 else if (attr == &dev_attr_pid)
1850 else if (attr == &dev_attr_start_lba)
1851 bt->start_lba = value;
1852 else if (attr == &dev_attr_end_lba)
1853 bt->end_lba = value;
1857 mutex_unlock(&q->debugfs_mutex);
1859 return ret ? ret : count;
1861 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1863 #ifdef CONFIG_EVENT_TRACING
1866 * blk_fill_rwbs - Fill the buffer rwbs by mapping op to character string.
1867 * @rwbs: buffer to be filled
1868 * @opf: request operation type (REQ_OP_XXX) and flags for the tracepoint
1871 * Maps each request operation and flag to a single character and fills the
1872 * buffer provided by the caller with resulting string.
1875 void blk_fill_rwbs(char *rwbs, blk_opf_t opf)
1879 if (opf & REQ_PREFLUSH)
1882 switch (opf & REQ_OP_MASK) {
1886 case REQ_OP_DISCARD:
1889 case REQ_OP_SECURE_ERASE:
1905 if (opf & REQ_RAHEAD)
1914 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1916 #endif /* CONFIG_EVENT_TRACING */