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
)
461 struct hd_struct
*part
= NULL
;
464 part
= bdev
->bd_part
;
467 bt
->start_lba
= part
->start_sect
;
468 bt
->end_lba
= part
->start_sect
+ part
->nr_sects
;
476 * Setup everything required to start tracing
478 static int do_blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
479 struct block_device
*bdev
,
480 struct blk_user_trace_setup
*buts
)
482 struct blk_trace
*bt
= NULL
;
483 struct dentry
*dir
= NULL
;
486 lockdep_assert_held(&q
->debugfs_mutex
);
488 if (!buts
->buf_size
|| !buts
->buf_nr
)
491 strncpy(buts
->name
, name
, BLKTRACE_BDEV_SIZE
);
492 buts
->name
[BLKTRACE_BDEV_SIZE
- 1] = '\0';
495 * some device names have larger paths - convert the slashes
496 * to underscores for this to work as expected
498 strreplace(buts
->name
, '/', '_');
501 * bdev can be NULL, as with scsi-generic, this is a helpful as
504 if (rcu_dereference_protected(q
->blk_trace
,
505 lockdep_is_held(&q
->debugfs_mutex
))) {
506 pr_warn("Concurrent blktraces are not allowed on %s\n",
511 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
516 bt
->sequence
= alloc_percpu(unsigned long);
520 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
525 * When tracing the whole disk reuse the existing debugfs directory
526 * created by the block layer on init. For partitions block devices,
527 * and scsi-generic block devices we create a temporary new debugfs
528 * directory that will be removed once the trace ends.
530 if (bdev
&& !bdev_is_partition(bdev
))
531 dir
= q
->debugfs_dir
;
533 bt
->dir
= dir
= debugfs_create_dir(buts
->name
, blk_debugfs_root
);
536 * As blktrace relies on debugfs for its interface the debugfs directory
537 * is required, contrary to the usual mantra of not checking for debugfs
538 * files or directories.
540 if (IS_ERR_OR_NULL(dir
)) {
541 pr_warn("debugfs_dir not present for %s so skipping\n",
548 atomic_set(&bt
->dropped
, 0);
549 INIT_LIST_HEAD(&bt
->running_list
);
552 bt
->dropped_file
= debugfs_create_file("dropped", 0444, dir
, bt
,
555 bt
->msg_file
= debugfs_create_file("msg", 0222, dir
, bt
, &blk_msg_fops
);
557 bt
->rchan
= relay_open("trace", dir
, buts
->buf_size
,
558 buts
->buf_nr
, &blk_relay_callbacks
, bt
);
562 bt
->act_mask
= buts
->act_mask
;
564 bt
->act_mask
= (u16
) -1;
566 blk_trace_setup_lba(bt
, bdev
);
568 /* overwrite with user settings */
570 bt
->start_lba
= buts
->start_lba
;
572 bt
->end_lba
= buts
->end_lba
;
575 bt
->trace_state
= Blktrace_setup
;
577 rcu_assign_pointer(q
->blk_trace
, bt
);
587 static int __blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
588 struct block_device
*bdev
, char __user
*arg
)
590 struct blk_user_trace_setup buts
;
593 ret
= copy_from_user(&buts
, arg
, sizeof(buts
));
597 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
601 if (copy_to_user(arg
, &buts
, sizeof(buts
))) {
602 __blk_trace_remove(q
);
608 int blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
609 struct block_device
*bdev
,
614 mutex_lock(&q
->debugfs_mutex
);
615 ret
= __blk_trace_setup(q
, name
, dev
, bdev
, arg
);
616 mutex_unlock(&q
->debugfs_mutex
);
620 EXPORT_SYMBOL_GPL(blk_trace_setup
);
622 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
623 static int compat_blk_trace_setup(struct request_queue
*q
, char *name
,
624 dev_t dev
, struct block_device
*bdev
,
627 struct blk_user_trace_setup buts
;
628 struct compat_blk_user_trace_setup cbuts
;
631 if (copy_from_user(&cbuts
, arg
, sizeof(cbuts
)))
634 buts
= (struct blk_user_trace_setup
) {
635 .act_mask
= cbuts
.act_mask
,
636 .buf_size
= cbuts
.buf_size
,
637 .buf_nr
= cbuts
.buf_nr
,
638 .start_lba
= cbuts
.start_lba
,
639 .end_lba
= cbuts
.end_lba
,
643 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
647 if (copy_to_user(arg
, &buts
.name
, ARRAY_SIZE(buts
.name
))) {
648 __blk_trace_remove(q
);
656 static int __blk_trace_startstop(struct request_queue
*q
, int start
)
659 struct blk_trace
*bt
;
661 bt
= rcu_dereference_protected(q
->blk_trace
,
662 lockdep_is_held(&q
->debugfs_mutex
));
667 * For starting a trace, we can transition from a setup or stopped
668 * trace. For stopping a trace, the state must be running
672 if (bt
->trace_state
== Blktrace_setup
||
673 bt
->trace_state
== Blktrace_stopped
) {
676 bt
->trace_state
= Blktrace_running
;
677 spin_lock_irq(&running_trace_lock
);
678 list_add(&bt
->running_list
, &running_trace_list
);
679 spin_unlock_irq(&running_trace_lock
);
685 if (bt
->trace_state
== Blktrace_running
) {
686 bt
->trace_state
= Blktrace_stopped
;
687 spin_lock_irq(&running_trace_lock
);
688 list_del_init(&bt
->running_list
);
689 spin_unlock_irq(&running_trace_lock
);
690 relay_flush(bt
->rchan
);
698 int blk_trace_startstop(struct request_queue
*q
, int start
)
702 mutex_lock(&q
->debugfs_mutex
);
703 ret
= __blk_trace_startstop(q
, start
);
704 mutex_unlock(&q
->debugfs_mutex
);
708 EXPORT_SYMBOL_GPL(blk_trace_startstop
);
711 * When reading or writing the blktrace sysfs files, the references to the
712 * opened sysfs or device files should prevent the underlying block device
713 * from being removed. So no further delete protection is really needed.
717 * blk_trace_ioctl: - handle the ioctls associated with tracing
718 * @bdev: the block device
719 * @cmd: the ioctl cmd
720 * @arg: the argument data, if any
723 int blk_trace_ioctl(struct block_device
*bdev
, unsigned cmd
, char __user
*arg
)
725 struct request_queue
*q
;
727 char b
[BDEVNAME_SIZE
];
729 q
= bdev_get_queue(bdev
);
733 mutex_lock(&q
->debugfs_mutex
);
738 ret
= __blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
740 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
741 case BLKTRACESETUP32
:
743 ret
= compat_blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
750 ret
= __blk_trace_startstop(q
, start
);
752 case BLKTRACETEARDOWN
:
753 ret
= __blk_trace_remove(q
);
760 mutex_unlock(&q
->debugfs_mutex
);
765 * blk_trace_shutdown: - stop and cleanup trace structures
766 * @q: the request queue associated with the device
769 void blk_trace_shutdown(struct request_queue
*q
)
771 mutex_lock(&q
->debugfs_mutex
);
772 if (rcu_dereference_protected(q
->blk_trace
,
773 lockdep_is_held(&q
->debugfs_mutex
))) {
774 __blk_trace_startstop(q
, 0);
775 __blk_trace_remove(q
);
778 mutex_unlock(&q
->debugfs_mutex
);
781 #ifdef CONFIG_BLK_CGROUP
782 static u64
blk_trace_bio_get_cgid(struct request_queue
*q
, struct bio
*bio
)
784 struct blk_trace
*bt
;
786 /* We don't use the 'bt' value here except as an optimization... */
787 bt
= rcu_dereference_protected(q
->blk_trace
, 1);
788 if (!bt
|| !(blk_tracer_flags
.val
& TRACE_BLK_OPT_CGROUP
))
793 return cgroup_id(bio_blkcg(bio
)->css
.cgroup
);
796 static u64
blk_trace_bio_get_cgid(struct request_queue
*q
, struct bio
*bio
)
803 blk_trace_request_get_cgid(struct request_queue
*q
, struct request
*rq
)
807 /* Use the first bio */
808 return blk_trace_bio_get_cgid(q
, rq
->bio
);
816 * blk_add_trace_rq - Add a trace for a request oriented action
817 * @rq: the source request
818 * @error: return status to log
819 * @nr_bytes: number of completed bytes
821 * @cgid: the cgroup info
824 * Records an action against a request. Will log the bio offset + size.
827 static void blk_add_trace_rq(struct request
*rq
, int error
,
828 unsigned int nr_bytes
, u32 what
, u64 cgid
)
830 struct blk_trace
*bt
;
833 bt
= rcu_dereference(rq
->q
->blk_trace
);
839 if (blk_rq_is_passthrough(rq
))
840 what
|= BLK_TC_ACT(BLK_TC_PC
);
842 what
|= BLK_TC_ACT(BLK_TC_FS
);
844 __blk_add_trace(bt
, blk_rq_trace_sector(rq
), nr_bytes
, req_op(rq
),
845 rq
->cmd_flags
, what
, error
, 0, NULL
, cgid
);
849 static void blk_add_trace_rq_insert(void *ignore
,
850 struct request_queue
*q
, struct request
*rq
)
852 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_INSERT
,
853 blk_trace_request_get_cgid(q
, rq
));
856 static void blk_add_trace_rq_issue(void *ignore
,
857 struct request_queue
*q
, struct request
*rq
)
859 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_ISSUE
,
860 blk_trace_request_get_cgid(q
, rq
));
863 static void blk_add_trace_rq_merge(void *ignore
,
864 struct request_queue
*q
, struct request
*rq
)
866 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_BACKMERGE
,
867 blk_trace_request_get_cgid(q
, rq
));
870 static void blk_add_trace_rq_requeue(void *ignore
,
871 struct request_queue
*q
,
874 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_REQUEUE
,
875 blk_trace_request_get_cgid(q
, rq
));
878 static void blk_add_trace_rq_complete(void *ignore
, struct request
*rq
,
879 int error
, unsigned int nr_bytes
)
881 blk_add_trace_rq(rq
, error
, nr_bytes
, BLK_TA_COMPLETE
,
882 blk_trace_request_get_cgid(rq
->q
, rq
));
886 * blk_add_trace_bio - Add a trace for a bio oriented action
887 * @q: queue the io is for
888 * @bio: the source bio
890 * @error: error, if any
893 * Records an action against a bio. Will log the bio offset + size.
896 static void blk_add_trace_bio(struct request_queue
*q
, struct bio
*bio
,
899 struct blk_trace
*bt
;
902 bt
= rcu_dereference(q
->blk_trace
);
908 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
909 bio_op(bio
), bio
->bi_opf
, what
, error
, 0, NULL
,
910 blk_trace_bio_get_cgid(q
, bio
));
914 static void blk_add_trace_bio_bounce(void *ignore
,
915 struct request_queue
*q
, struct bio
*bio
)
917 blk_add_trace_bio(q
, bio
, BLK_TA_BOUNCE
, 0);
920 static void blk_add_trace_bio_complete(void *ignore
,
921 struct request_queue
*q
, struct bio
*bio
)
923 blk_add_trace_bio(q
, bio
, BLK_TA_COMPLETE
,
924 blk_status_to_errno(bio
->bi_status
));
927 static void blk_add_trace_bio_backmerge(void *ignore
,
928 struct request_queue
*q
,
932 blk_add_trace_bio(q
, bio
, BLK_TA_BACKMERGE
, 0);
935 static void blk_add_trace_bio_frontmerge(void *ignore
,
936 struct request_queue
*q
,
940 blk_add_trace_bio(q
, bio
, BLK_TA_FRONTMERGE
, 0);
943 static void blk_add_trace_bio_queue(void *ignore
,
944 struct request_queue
*q
, struct bio
*bio
)
946 blk_add_trace_bio(q
, bio
, BLK_TA_QUEUE
, 0);
949 static void blk_add_trace_getrq(void *ignore
,
950 struct request_queue
*q
,
951 struct bio
*bio
, int rw
)
954 blk_add_trace_bio(q
, bio
, BLK_TA_GETRQ
, 0);
956 struct blk_trace
*bt
;
959 bt
= rcu_dereference(q
->blk_trace
);
961 __blk_add_trace(bt
, 0, 0, rw
, 0, BLK_TA_GETRQ
, 0, 0,
968 static void blk_add_trace_sleeprq(void *ignore
,
969 struct request_queue
*q
,
970 struct bio
*bio
, int rw
)
973 blk_add_trace_bio(q
, bio
, BLK_TA_SLEEPRQ
, 0);
975 struct blk_trace
*bt
;
978 bt
= rcu_dereference(q
->blk_trace
);
980 __blk_add_trace(bt
, 0, 0, rw
, 0, BLK_TA_SLEEPRQ
,
986 static void blk_add_trace_plug(void *ignore
, struct request_queue
*q
)
988 struct blk_trace
*bt
;
991 bt
= rcu_dereference(q
->blk_trace
);
993 __blk_add_trace(bt
, 0, 0, 0, 0, BLK_TA_PLUG
, 0, 0, NULL
, 0);
997 static void blk_add_trace_unplug(void *ignore
, struct request_queue
*q
,
998 unsigned int depth
, bool explicit)
1000 struct blk_trace
*bt
;
1003 bt
= rcu_dereference(q
->blk_trace
);
1005 __be64 rpdu
= cpu_to_be64(depth
);
1009 what
= BLK_TA_UNPLUG_IO
;
1011 what
= BLK_TA_UNPLUG_TIMER
;
1013 __blk_add_trace(bt
, 0, 0, 0, 0, what
, 0, sizeof(rpdu
), &rpdu
, 0);
1018 static void blk_add_trace_split(void *ignore
,
1019 struct request_queue
*q
, struct bio
*bio
,
1022 struct blk_trace
*bt
;
1025 bt
= rcu_dereference(q
->blk_trace
);
1027 __be64 rpdu
= cpu_to_be64(pdu
);
1029 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
,
1030 bio
->bi_iter
.bi_size
, bio_op(bio
), bio
->bi_opf
,
1032 blk_status_to_errno(bio
->bi_status
),
1033 sizeof(rpdu
), &rpdu
,
1034 blk_trace_bio_get_cgid(q
, bio
));
1040 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1041 * @ignore: trace callback data parameter (not used)
1042 * @q: queue the io is for
1043 * @bio: the source bio
1044 * @dev: target device
1045 * @from: source sector
1048 * Device mapper or raid target sometimes need to split a bio because
1049 * it spans a stripe (or similar). Add a trace for that action.
1052 static void blk_add_trace_bio_remap(void *ignore
,
1053 struct request_queue
*q
, struct bio
*bio
,
1054 dev_t dev
, sector_t from
)
1056 struct blk_trace
*bt
;
1057 struct blk_io_trace_remap r
;
1060 bt
= rcu_dereference(q
->blk_trace
);
1066 r
.device_from
= cpu_to_be32(dev
);
1067 r
.device_to
= cpu_to_be32(bio_dev(bio
));
1068 r
.sector_from
= cpu_to_be64(from
);
1070 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
1071 bio_op(bio
), bio
->bi_opf
, BLK_TA_REMAP
,
1072 blk_status_to_errno(bio
->bi_status
),
1073 sizeof(r
), &r
, blk_trace_bio_get_cgid(q
, bio
));
1078 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1079 * @ignore: trace callback data parameter (not used)
1080 * @q: queue the io is for
1081 * @rq: the source request
1082 * @dev: target device
1083 * @from: source sector
1086 * Device mapper remaps request to other devices.
1087 * Add a trace for that action.
1090 static void blk_add_trace_rq_remap(void *ignore
,
1091 struct request_queue
*q
,
1092 struct request
*rq
, dev_t dev
,
1095 struct blk_trace
*bt
;
1096 struct blk_io_trace_remap r
;
1099 bt
= rcu_dereference(q
->blk_trace
);
1105 r
.device_from
= cpu_to_be32(dev
);
1106 r
.device_to
= cpu_to_be32(disk_devt(rq
->rq_disk
));
1107 r
.sector_from
= cpu_to_be64(from
);
1109 __blk_add_trace(bt
, blk_rq_pos(rq
), blk_rq_bytes(rq
),
1110 rq_data_dir(rq
), 0, BLK_TA_REMAP
, 0,
1111 sizeof(r
), &r
, blk_trace_request_get_cgid(q
, rq
));
1116 * blk_add_driver_data - Add binary message with driver-specific data
1117 * @q: queue the io is for
1119 * @data: driver-specific data
1120 * @len: length of driver-specific data
1123 * Some drivers might want to write driver-specific data per request.
1126 void blk_add_driver_data(struct request_queue
*q
,
1128 void *data
, size_t len
)
1130 struct blk_trace
*bt
;
1133 bt
= rcu_dereference(q
->blk_trace
);
1139 __blk_add_trace(bt
, blk_rq_trace_sector(rq
), blk_rq_bytes(rq
), 0, 0,
1140 BLK_TA_DRV_DATA
, 0, len
, data
,
1141 blk_trace_request_get_cgid(q
, rq
));
1144 EXPORT_SYMBOL_GPL(blk_add_driver_data
);
1146 static void blk_register_tracepoints(void)
1150 ret
= register_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1152 ret
= register_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1154 ret
= register_trace_block_rq_merge(blk_add_trace_rq_merge
, NULL
);
1156 ret
= register_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1158 ret
= register_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1160 ret
= register_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1162 ret
= register_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1164 ret
= register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1166 ret
= register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1168 ret
= register_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1170 ret
= register_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1172 ret
= register_trace_block_sleeprq(blk_add_trace_sleeprq
, NULL
);
1174 ret
= register_trace_block_plug(blk_add_trace_plug
, NULL
);
1176 ret
= register_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1178 ret
= register_trace_block_split(blk_add_trace_split
, NULL
);
1180 ret
= register_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1182 ret
= register_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1186 static void blk_unregister_tracepoints(void)
1188 unregister_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1189 unregister_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1190 unregister_trace_block_split(blk_add_trace_split
, NULL
);
1191 unregister_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1192 unregister_trace_block_plug(blk_add_trace_plug
, NULL
);
1193 unregister_trace_block_sleeprq(blk_add_trace_sleeprq
, NULL
);
1194 unregister_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1195 unregister_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1196 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1197 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1198 unregister_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1199 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1200 unregister_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1201 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1202 unregister_trace_block_rq_merge(blk_add_trace_rq_merge
, NULL
);
1203 unregister_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1204 unregister_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1206 tracepoint_synchronize_unregister();
1210 * struct blk_io_tracer formatting routines
1213 static void fill_rwbs(char *rwbs
, const struct blk_io_trace
*t
)
1216 int tc
= t
->action
>> BLK_TC_SHIFT
;
1218 if ((t
->action
& ~__BLK_TN_CGROUP
) == BLK_TN_MESSAGE
) {
1223 if (tc
& BLK_TC_FLUSH
)
1226 if (tc
& BLK_TC_DISCARD
)
1228 else if (tc
& BLK_TC_WRITE
)
1235 if (tc
& BLK_TC_FUA
)
1237 if (tc
& BLK_TC_AHEAD
)
1239 if (tc
& BLK_TC_SYNC
)
1241 if (tc
& BLK_TC_META
)
1248 const struct blk_io_trace
*te_blk_io_trace(const struct trace_entry
*ent
)
1250 return (const struct blk_io_trace
*)ent
;
1253 static inline const void *pdu_start(const struct trace_entry
*ent
, bool has_cg
)
1255 return (void *)(te_blk_io_trace(ent
) + 1) + (has_cg
? sizeof(u64
) : 0);
1258 static inline u64
t_cgid(const struct trace_entry
*ent
)
1260 return *(u64
*)(te_blk_io_trace(ent
) + 1);
1263 static inline int pdu_real_len(const struct trace_entry
*ent
, bool has_cg
)
1265 return te_blk_io_trace(ent
)->pdu_len
- (has_cg
? sizeof(u64
) : 0);
1268 static inline u32
t_action(const struct trace_entry
*ent
)
1270 return te_blk_io_trace(ent
)->action
;
1273 static inline u32
t_bytes(const struct trace_entry
*ent
)
1275 return te_blk_io_trace(ent
)->bytes
;
1278 static inline u32
t_sec(const struct trace_entry
*ent
)
1280 return te_blk_io_trace(ent
)->bytes
>> 9;
1283 static inline unsigned long long t_sector(const struct trace_entry
*ent
)
1285 return te_blk_io_trace(ent
)->sector
;
1288 static inline __u16
t_error(const struct trace_entry
*ent
)
1290 return te_blk_io_trace(ent
)->error
;
1293 static __u64
get_pdu_int(const struct trace_entry
*ent
, bool has_cg
)
1295 const __be64
*val
= pdu_start(ent
, has_cg
);
1296 return be64_to_cpu(*val
);
1299 typedef void (blk_log_action_t
) (struct trace_iterator
*iter
, const char *act
,
1302 static void blk_log_action_classic(struct trace_iterator
*iter
, const char *act
,
1305 char rwbs
[RWBS_LEN
];
1306 unsigned long long ts
= iter
->ts
;
1307 unsigned long nsec_rem
= do_div(ts
, NSEC_PER_SEC
);
1308 unsigned secs
= (unsigned long)ts
;
1309 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1313 trace_seq_printf(&iter
->seq
,
1314 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1315 MAJOR(t
->device
), MINOR(t
->device
), iter
->cpu
,
1316 secs
, nsec_rem
, iter
->ent
->pid
, act
, rwbs
);
1319 static void blk_log_action(struct trace_iterator
*iter
, const char *act
,
1322 char rwbs
[RWBS_LEN
];
1323 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1327 u64 id
= t_cgid(iter
->ent
);
1329 if (blk_tracer_flags
.val
& TRACE_BLK_OPT_CGNAME
) {
1330 char blkcg_name_buf
[NAME_MAX
+ 1] = "<...>";
1332 cgroup_path_from_kernfs_id(id
, blkcg_name_buf
,
1333 sizeof(blkcg_name_buf
));
1334 trace_seq_printf(&iter
->seq
, "%3d,%-3d %s %2s %3s ",
1335 MAJOR(t
->device
), MINOR(t
->device
),
1336 blkcg_name_buf
, act
, rwbs
);
1339 * The cgid portion used to be "INO,GEN". Userland
1340 * builds a FILEID_INO32_GEN fid out of them and
1341 * opens the cgroup using open_by_handle_at(2).
1342 * While 32bit ino setups are still the same, 64bit
1343 * ones now use the 64bit ino as the whole ID and
1344 * no longer use generation.
1346 * Regarldess of the content, always output
1347 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1348 * be mapped back to @id on both 64 and 32bit ino
1349 * setups. See __kernfs_fh_to_dentry().
1351 trace_seq_printf(&iter
->seq
,
1352 "%3d,%-3d %llx,%-llx %2s %3s ",
1353 MAJOR(t
->device
), MINOR(t
->device
),
1354 id
& U32_MAX
, id
>> 32, act
, rwbs
);
1357 trace_seq_printf(&iter
->seq
, "%3d,%-3d %2s %3s ",
1358 MAJOR(t
->device
), MINOR(t
->device
), act
, rwbs
);
1361 static void blk_log_dump_pdu(struct trace_seq
*s
,
1362 const struct trace_entry
*ent
, bool has_cg
)
1364 const unsigned char *pdu_buf
;
1368 pdu_buf
= pdu_start(ent
, has_cg
);
1369 pdu_len
= pdu_real_len(ent
, has_cg
);
1374 /* find the last zero that needs to be printed */
1375 for (end
= pdu_len
- 1; end
>= 0; end
--)
1380 trace_seq_putc(s
, '(');
1382 for (i
= 0; i
< pdu_len
; i
++) {
1384 trace_seq_printf(s
, "%s%02x",
1385 i
== 0 ? "" : " ", pdu_buf
[i
]);
1388 * stop when the rest is just zeroes and indicate so
1389 * with a ".." appended
1391 if (i
== end
&& end
!= pdu_len
- 1) {
1392 trace_seq_puts(s
, " ..) ");
1397 trace_seq_puts(s
, ") ");
1400 static void blk_log_generic(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1402 char cmd
[TASK_COMM_LEN
];
1404 trace_find_cmdline(ent
->pid
, cmd
);
1406 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1407 trace_seq_printf(s
, "%u ", t_bytes(ent
));
1408 blk_log_dump_pdu(s
, ent
, has_cg
);
1409 trace_seq_printf(s
, "[%s]\n", cmd
);
1412 trace_seq_printf(s
, "%llu + %u [%s]\n",
1413 t_sector(ent
), t_sec(ent
), cmd
);
1415 trace_seq_printf(s
, "[%s]\n", cmd
);
1419 static void blk_log_with_error(struct trace_seq
*s
,
1420 const struct trace_entry
*ent
, bool has_cg
)
1422 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1423 blk_log_dump_pdu(s
, ent
, has_cg
);
1424 trace_seq_printf(s
, "[%d]\n", t_error(ent
));
1427 trace_seq_printf(s
, "%llu + %u [%d]\n",
1429 t_sec(ent
), t_error(ent
));
1431 trace_seq_printf(s
, "%llu [%d]\n",
1432 t_sector(ent
), t_error(ent
));
1436 static void blk_log_remap(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1438 const struct blk_io_trace_remap
*__r
= pdu_start(ent
, has_cg
);
1440 trace_seq_printf(s
, "%llu + %u <- (%d,%d) %llu\n",
1441 t_sector(ent
), t_sec(ent
),
1442 MAJOR(be32_to_cpu(__r
->device_from
)),
1443 MINOR(be32_to_cpu(__r
->device_from
)),
1444 be64_to_cpu(__r
->sector_from
));
1447 static void blk_log_plug(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1449 char cmd
[TASK_COMM_LEN
];
1451 trace_find_cmdline(ent
->pid
, cmd
);
1453 trace_seq_printf(s
, "[%s]\n", cmd
);
1456 static void blk_log_unplug(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1458 char cmd
[TASK_COMM_LEN
];
1460 trace_find_cmdline(ent
->pid
, cmd
);
1462 trace_seq_printf(s
, "[%s] %llu\n", cmd
, get_pdu_int(ent
, has_cg
));
1465 static void blk_log_split(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1467 char cmd
[TASK_COMM_LEN
];
1469 trace_find_cmdline(ent
->pid
, cmd
);
1471 trace_seq_printf(s
, "%llu / %llu [%s]\n", t_sector(ent
),
1472 get_pdu_int(ent
, has_cg
), cmd
);
1475 static void blk_log_msg(struct trace_seq
*s
, const struct trace_entry
*ent
,
1479 trace_seq_putmem(s
, pdu_start(ent
, has_cg
),
1480 pdu_real_len(ent
, has_cg
));
1481 trace_seq_putc(s
, '\n');
1485 * struct tracer operations
1488 static void blk_tracer_print_header(struct seq_file
*m
)
1490 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1492 seq_puts(m
, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1496 static void blk_tracer_start(struct trace_array
*tr
)
1498 blk_tracer_enabled
= true;
1501 static int blk_tracer_init(struct trace_array
*tr
)
1504 blk_tracer_start(tr
);
1508 static void blk_tracer_stop(struct trace_array
*tr
)
1510 blk_tracer_enabled
= false;
1513 static void blk_tracer_reset(struct trace_array
*tr
)
1515 blk_tracer_stop(tr
);
1518 static const struct {
1520 void (*print
)(struct trace_seq
*s
, const struct trace_entry
*ent
,
1523 [__BLK_TA_QUEUE
] = {{ "Q", "queue" }, blk_log_generic
},
1524 [__BLK_TA_BACKMERGE
] = {{ "M", "backmerge" }, blk_log_generic
},
1525 [__BLK_TA_FRONTMERGE
] = {{ "F", "frontmerge" }, blk_log_generic
},
1526 [__BLK_TA_GETRQ
] = {{ "G", "getrq" }, blk_log_generic
},
1527 [__BLK_TA_SLEEPRQ
] = {{ "S", "sleeprq" }, blk_log_generic
},
1528 [__BLK_TA_REQUEUE
] = {{ "R", "requeue" }, blk_log_with_error
},
1529 [__BLK_TA_ISSUE
] = {{ "D", "issue" }, blk_log_generic
},
1530 [__BLK_TA_COMPLETE
] = {{ "C", "complete" }, blk_log_with_error
},
1531 [__BLK_TA_PLUG
] = {{ "P", "plug" }, blk_log_plug
},
1532 [__BLK_TA_UNPLUG_IO
] = {{ "U", "unplug_io" }, blk_log_unplug
},
1533 [__BLK_TA_UNPLUG_TIMER
] = {{ "UT", "unplug_timer" }, blk_log_unplug
},
1534 [__BLK_TA_INSERT
] = {{ "I", "insert" }, blk_log_generic
},
1535 [__BLK_TA_SPLIT
] = {{ "X", "split" }, blk_log_split
},
1536 [__BLK_TA_BOUNCE
] = {{ "B", "bounce" }, blk_log_generic
},
1537 [__BLK_TA_REMAP
] = {{ "A", "remap" }, blk_log_remap
},
1540 static enum print_line_t
print_one_line(struct trace_iterator
*iter
,
1543 struct trace_array
*tr
= iter
->tr
;
1544 struct trace_seq
*s
= &iter
->seq
;
1545 const struct blk_io_trace
*t
;
1548 blk_log_action_t
*log_action
;
1551 t
= te_blk_io_trace(iter
->ent
);
1552 what
= (t
->action
& ((1 << BLK_TC_SHIFT
) - 1)) & ~__BLK_TA_CGROUP
;
1553 long_act
= !!(tr
->trace_flags
& TRACE_ITER_VERBOSE
);
1554 log_action
= classic
? &blk_log_action_classic
: &blk_log_action
;
1555 has_cg
= t
->action
& __BLK_TA_CGROUP
;
1557 if ((t
->action
& ~__BLK_TN_CGROUP
) == BLK_TN_MESSAGE
) {
1558 log_action(iter
, long_act
? "message" : "m", has_cg
);
1559 blk_log_msg(s
, iter
->ent
, has_cg
);
1560 return trace_handle_return(s
);
1563 if (unlikely(what
== 0 || what
>= ARRAY_SIZE(what2act
)))
1564 trace_seq_printf(s
, "Unknown action %x\n", what
);
1566 log_action(iter
, what2act
[what
].act
[long_act
], has_cg
);
1567 what2act
[what
].print(s
, iter
->ent
, has_cg
);
1570 return trace_handle_return(s
);
1573 static enum print_line_t
blk_trace_event_print(struct trace_iterator
*iter
,
1574 int flags
, struct trace_event
*event
)
1576 return print_one_line(iter
, false);
1579 static void blk_trace_synthesize_old_trace(struct trace_iterator
*iter
)
1581 struct trace_seq
*s
= &iter
->seq
;
1582 struct blk_io_trace
*t
= (struct blk_io_trace
*)iter
->ent
;
1583 const int offset
= offsetof(struct blk_io_trace
, sector
);
1584 struct blk_io_trace old
= {
1585 .magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
,
1589 trace_seq_putmem(s
, &old
, offset
);
1590 trace_seq_putmem(s
, &t
->sector
,
1591 sizeof(old
) - offset
+ t
->pdu_len
);
1594 static enum print_line_t
1595 blk_trace_event_print_binary(struct trace_iterator
*iter
, int flags
,
1596 struct trace_event
*event
)
1598 blk_trace_synthesize_old_trace(iter
);
1600 return trace_handle_return(&iter
->seq
);
1603 static enum print_line_t
blk_tracer_print_line(struct trace_iterator
*iter
)
1605 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1606 return TRACE_TYPE_UNHANDLED
;
1608 return print_one_line(iter
, true);
1612 blk_tracer_set_flag(struct trace_array
*tr
, u32 old_flags
, u32 bit
, int set
)
1614 /* don't output context-info for blk_classic output */
1615 if (bit
== TRACE_BLK_OPT_CLASSIC
) {
1617 tr
->trace_flags
&= ~TRACE_ITER_CONTEXT_INFO
;
1619 tr
->trace_flags
|= TRACE_ITER_CONTEXT_INFO
;
1624 static struct tracer blk_tracer __read_mostly
= {
1626 .init
= blk_tracer_init
,
1627 .reset
= blk_tracer_reset
,
1628 .start
= blk_tracer_start
,
1629 .stop
= blk_tracer_stop
,
1630 .print_header
= blk_tracer_print_header
,
1631 .print_line
= blk_tracer_print_line
,
1632 .flags
= &blk_tracer_flags
,
1633 .set_flag
= blk_tracer_set_flag
,
1636 static struct trace_event_functions trace_blk_event_funcs
= {
1637 .trace
= blk_trace_event_print
,
1638 .binary
= blk_trace_event_print_binary
,
1641 static struct trace_event trace_blk_event
= {
1643 .funcs
= &trace_blk_event_funcs
,
1646 static int __init
init_blk_tracer(void)
1648 if (!register_trace_event(&trace_blk_event
)) {
1649 pr_warn("Warning: could not register block events\n");
1653 if (register_tracer(&blk_tracer
) != 0) {
1654 pr_warn("Warning: could not register the block tracer\n");
1655 unregister_trace_event(&trace_blk_event
);
1662 device_initcall(init_blk_tracer
);
1664 static int blk_trace_remove_queue(struct request_queue
*q
)
1666 struct blk_trace
*bt
;
1668 bt
= rcu_replace_pointer(q
->blk_trace
, NULL
,
1669 lockdep_is_held(&q
->debugfs_mutex
));
1680 * Setup everything required to start tracing
1682 static int blk_trace_setup_queue(struct request_queue
*q
,
1683 struct block_device
*bdev
)
1685 struct blk_trace
*bt
= NULL
;
1688 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
1692 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
1696 bt
->dev
= bdev
->bd_dev
;
1697 bt
->act_mask
= (u16
)-1;
1699 blk_trace_setup_lba(bt
, bdev
);
1701 rcu_assign_pointer(q
->blk_trace
, bt
);
1711 * sysfs interface to enable and configure tracing
1714 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1715 struct device_attribute
*attr
,
1717 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1718 struct device_attribute
*attr
,
1719 const char *buf
, size_t count
);
1720 #define BLK_TRACE_DEVICE_ATTR(_name) \
1721 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1722 sysfs_blk_trace_attr_show, \
1723 sysfs_blk_trace_attr_store)
1725 static BLK_TRACE_DEVICE_ATTR(enable
);
1726 static BLK_TRACE_DEVICE_ATTR(act_mask
);
1727 static BLK_TRACE_DEVICE_ATTR(pid
);
1728 static BLK_TRACE_DEVICE_ATTR(start_lba
);
1729 static BLK_TRACE_DEVICE_ATTR(end_lba
);
1731 static struct attribute
*blk_trace_attrs
[] = {
1732 &dev_attr_enable
.attr
,
1733 &dev_attr_act_mask
.attr
,
1735 &dev_attr_start_lba
.attr
,
1736 &dev_attr_end_lba
.attr
,
1740 struct attribute_group blk_trace_attr_group
= {
1742 .attrs
= blk_trace_attrs
,
1745 static const struct {
1749 { BLK_TC_READ
, "read" },
1750 { BLK_TC_WRITE
, "write" },
1751 { BLK_TC_FLUSH
, "flush" },
1752 { BLK_TC_SYNC
, "sync" },
1753 { BLK_TC_QUEUE
, "queue" },
1754 { BLK_TC_REQUEUE
, "requeue" },
1755 { BLK_TC_ISSUE
, "issue" },
1756 { BLK_TC_COMPLETE
, "complete" },
1757 { BLK_TC_FS
, "fs" },
1758 { BLK_TC_PC
, "pc" },
1759 { BLK_TC_NOTIFY
, "notify" },
1760 { BLK_TC_AHEAD
, "ahead" },
1761 { BLK_TC_META
, "meta" },
1762 { BLK_TC_DISCARD
, "discard" },
1763 { BLK_TC_DRV_DATA
, "drv_data" },
1764 { BLK_TC_FUA
, "fua" },
1767 static int blk_trace_str2mask(const char *str
)
1771 char *buf
, *s
, *token
;
1773 buf
= kstrdup(str
, GFP_KERNEL
);
1779 token
= strsep(&s
, ",");
1786 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1787 if (strcasecmp(token
, mask_maps
[i
].str
) == 0) {
1788 mask
|= mask_maps
[i
].mask
;
1792 if (i
== ARRAY_SIZE(mask_maps
)) {
1802 static ssize_t
blk_trace_mask2str(char *buf
, int mask
)
1807 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1808 if (mask
& mask_maps
[i
].mask
) {
1809 p
+= sprintf(p
, "%s%s",
1810 (p
== buf
) ? "" : ",", mask_maps
[i
].str
);
1818 static struct request_queue
*blk_trace_get_queue(struct block_device
*bdev
)
1820 if (bdev
->bd_disk
== NULL
)
1823 return bdev_get_queue(bdev
);
1826 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1827 struct device_attribute
*attr
,
1830 struct block_device
*bdev
= bdget_part(dev_to_part(dev
));
1831 struct request_queue
*q
;
1832 struct blk_trace
*bt
;
1833 ssize_t ret
= -ENXIO
;
1838 q
= blk_trace_get_queue(bdev
);
1842 mutex_lock(&q
->debugfs_mutex
);
1844 bt
= rcu_dereference_protected(q
->blk_trace
,
1845 lockdep_is_held(&q
->debugfs_mutex
));
1846 if (attr
== &dev_attr_enable
) {
1847 ret
= sprintf(buf
, "%u\n", !!bt
);
1848 goto out_unlock_bdev
;
1852 ret
= sprintf(buf
, "disabled\n");
1853 else if (attr
== &dev_attr_act_mask
)
1854 ret
= blk_trace_mask2str(buf
, bt
->act_mask
);
1855 else if (attr
== &dev_attr_pid
)
1856 ret
= sprintf(buf
, "%u\n", bt
->pid
);
1857 else if (attr
== &dev_attr_start_lba
)
1858 ret
= sprintf(buf
, "%llu\n", bt
->start_lba
);
1859 else if (attr
== &dev_attr_end_lba
)
1860 ret
= sprintf(buf
, "%llu\n", bt
->end_lba
);
1863 mutex_unlock(&q
->debugfs_mutex
);
1870 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1871 struct device_attribute
*attr
,
1872 const char *buf
, size_t count
)
1874 struct block_device
*bdev
;
1875 struct request_queue
*q
;
1876 struct blk_trace
*bt
;
1878 ssize_t ret
= -EINVAL
;
1883 if (attr
== &dev_attr_act_mask
) {
1884 if (kstrtoull(buf
, 0, &value
)) {
1885 /* Assume it is a list of trace category names */
1886 ret
= blk_trace_str2mask(buf
);
1891 } else if (kstrtoull(buf
, 0, &value
))
1895 bdev
= bdget_part(dev_to_part(dev
));
1899 q
= blk_trace_get_queue(bdev
);
1903 mutex_lock(&q
->debugfs_mutex
);
1905 bt
= rcu_dereference_protected(q
->blk_trace
,
1906 lockdep_is_held(&q
->debugfs_mutex
));
1907 if (attr
== &dev_attr_enable
) {
1908 if (!!value
== !!bt
) {
1910 goto out_unlock_bdev
;
1913 ret
= blk_trace_setup_queue(q
, bdev
);
1915 ret
= blk_trace_remove_queue(q
);
1916 goto out_unlock_bdev
;
1921 ret
= blk_trace_setup_queue(q
, bdev
);
1922 bt
= rcu_dereference_protected(q
->blk_trace
,
1923 lockdep_is_held(&q
->debugfs_mutex
));
1927 if (attr
== &dev_attr_act_mask
)
1928 bt
->act_mask
= value
;
1929 else if (attr
== &dev_attr_pid
)
1931 else if (attr
== &dev_attr_start_lba
)
1932 bt
->start_lba
= value
;
1933 else if (attr
== &dev_attr_end_lba
)
1934 bt
->end_lba
= value
;
1938 mutex_unlock(&q
->debugfs_mutex
);
1942 return ret
? ret
: count
;
1945 int blk_trace_init_sysfs(struct device
*dev
)
1947 return sysfs_create_group(&dev
->kobj
, &blk_trace_attr_group
);
1950 void blk_trace_remove_sysfs(struct device
*dev
)
1952 sysfs_remove_group(&dev
->kobj
, &blk_trace_attr_group
);
1955 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1957 #ifdef CONFIG_EVENT_TRACING
1959 void blk_fill_rwbs(char *rwbs
, unsigned int op
, int bytes
)
1963 if (op
& REQ_PREFLUSH
)
1966 switch (op
& REQ_OP_MASK
) {
1968 case REQ_OP_WRITE_SAME
:
1971 case REQ_OP_DISCARD
:
1974 case REQ_OP_SECURE_ERASE
:
1990 if (op
& REQ_RAHEAD
)
1999 EXPORT_SYMBOL_GPL(blk_fill_rwbs
);
2001 #endif /* CONFIG_EVENT_TRACING */