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
;
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 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
;
225 unsigned int trace_ctx
= 0;
228 bool blk_tracer
= blk_tracer_enabled
;
229 ssize_t cgid_len
= cgid
? sizeof(cgid
) : 0;
231 if (unlikely(bt
->trace_state
!= Blktrace_running
&& !blk_tracer
))
234 what
|= ddir_act
[op_is_write(op
) ? WRITE
: READ
];
235 what
|= MASK_TC_BIT(op_flags
, SYNC
);
236 what
|= MASK_TC_BIT(op_flags
, RAHEAD
);
237 what
|= MASK_TC_BIT(op_flags
, META
);
238 what
|= MASK_TC_BIT(op_flags
, PREFLUSH
);
239 what
|= MASK_TC_BIT(op_flags
, FUA
);
240 if (op
== REQ_OP_DISCARD
|| op
== REQ_OP_SECURE_ERASE
)
241 what
|= BLK_TC_ACT(BLK_TC_DISCARD
);
242 if (op
== REQ_OP_FLUSH
)
243 what
|= BLK_TC_ACT(BLK_TC_FLUSH
);
245 what
|= __BLK_TA_CGROUP
;
248 if (act_log_check(bt
, what
, sector
, pid
))
250 cpu
= raw_smp_processor_id();
253 tracing_record_cmdline(current
);
255 buffer
= blk_tr
->array_buffer
.buffer
;
256 trace_ctx
= tracing_gen_ctx_flags(0);
257 event
= trace_buffer_lock_reserve(buffer
, TRACE_BLK
,
258 sizeof(*t
) + pdu_len
+ cgid_len
,
262 t
= ring_buffer_event_data(event
);
266 if (unlikely(tsk
->btrace_seq
!= blktrace_seq
))
270 * A word about the locking here - we disable interrupts to reserve
271 * some space in the relay per-cpu buffer, to prevent an irq
272 * from coming in and stepping on our toes.
274 local_irq_save(flags
);
275 t
= relay_reserve(bt
->rchan
, sizeof(*t
) + pdu_len
+ cgid_len
);
277 sequence
= per_cpu_ptr(bt
->sequence
, cpu
);
279 t
->magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
;
280 t
->sequence
= ++(*sequence
);
281 t
->time
= ktime_to_ns(ktime_get());
284 * These two are not needed in ftrace as they are in the
285 * generic trace_entry, filled by tracing_generic_entry_update,
286 * but for the trace_event->bin() synthesizer benefit we do it
297 t
->pdu_len
= pdu_len
+ cgid_len
;
300 memcpy((void *)t
+ sizeof(*t
), &cgid
, cgid_len
);
302 memcpy((void *)t
+ sizeof(*t
) + cgid_len
, pdu_data
, pdu_len
);
305 trace_buffer_unlock_commit(blk_tr
, buffer
, event
, trace_ctx
);
310 local_irq_restore(flags
);
313 static void blk_trace_free(struct blk_trace
*bt
)
315 relay_close(bt
->rchan
);
316 debugfs_remove(bt
->dir
);
317 free_percpu(bt
->sequence
);
318 free_percpu(bt
->msg_data
);
322 static void get_probe_ref(void)
324 mutex_lock(&blk_probe_mutex
);
325 if (++blk_probes_ref
== 1)
326 blk_register_tracepoints();
327 mutex_unlock(&blk_probe_mutex
);
330 static void put_probe_ref(void)
332 mutex_lock(&blk_probe_mutex
);
333 if (!--blk_probes_ref
)
334 blk_unregister_tracepoints();
335 mutex_unlock(&blk_probe_mutex
);
338 static void blk_trace_cleanup(struct blk_trace
*bt
)
345 static int __blk_trace_remove(struct request_queue
*q
)
347 struct blk_trace
*bt
;
349 bt
= rcu_replace_pointer(q
->blk_trace
, NULL
,
350 lockdep_is_held(&q
->debugfs_mutex
));
354 if (bt
->trace_state
!= Blktrace_running
)
355 blk_trace_cleanup(bt
);
360 int blk_trace_remove(struct request_queue
*q
)
364 mutex_lock(&q
->debugfs_mutex
);
365 ret
= __blk_trace_remove(q
);
366 mutex_unlock(&q
->debugfs_mutex
);
370 EXPORT_SYMBOL_GPL(blk_trace_remove
);
372 static ssize_t
blk_dropped_read(struct file
*filp
, char __user
*buffer
,
373 size_t count
, loff_t
*ppos
)
375 struct blk_trace
*bt
= filp
->private_data
;
378 snprintf(buf
, sizeof(buf
), "%u\n", atomic_read(&bt
->dropped
));
380 return simple_read_from_buffer(buffer
, count
, ppos
, buf
, strlen(buf
));
383 static const struct file_operations blk_dropped_fops
= {
384 .owner
= THIS_MODULE
,
386 .read
= blk_dropped_read
,
387 .llseek
= default_llseek
,
390 static ssize_t
blk_msg_write(struct file
*filp
, const char __user
*buffer
,
391 size_t count
, loff_t
*ppos
)
394 struct blk_trace
*bt
;
396 if (count
>= BLK_TN_MAX_MSG
)
399 msg
= memdup_user_nul(buffer
, count
);
403 bt
= filp
->private_data
;
404 __trace_note_message(bt
, NULL
, "%s", msg
);
410 static const struct file_operations blk_msg_fops
= {
411 .owner
= THIS_MODULE
,
413 .write
= blk_msg_write
,
414 .llseek
= noop_llseek
,
418 * Keep track of how many times we encountered a full subbuffer, to aid
419 * the user space app in telling how many lost events there were.
421 static int blk_subbuf_start_callback(struct rchan_buf
*buf
, void *subbuf
,
422 void *prev_subbuf
, size_t prev_padding
)
424 struct blk_trace
*bt
;
426 if (!relay_buf_full(buf
))
429 bt
= buf
->chan
->private_data
;
430 atomic_inc(&bt
->dropped
);
434 static int blk_remove_buf_file_callback(struct dentry
*dentry
)
436 debugfs_remove(dentry
);
441 static struct dentry
*blk_create_buf_file_callback(const char *filename
,
442 struct dentry
*parent
,
444 struct rchan_buf
*buf
,
447 return debugfs_create_file(filename
, mode
, parent
, buf
,
448 &relay_file_operations
);
451 static const struct rchan_callbacks blk_relay_callbacks
= {
452 .subbuf_start
= blk_subbuf_start_callback
,
453 .create_buf_file
= blk_create_buf_file_callback
,
454 .remove_buf_file
= blk_remove_buf_file_callback
,
457 static void blk_trace_setup_lba(struct blk_trace
*bt
,
458 struct block_device
*bdev
)
461 bt
->start_lba
= bdev
->bd_start_sect
;
462 bt
->end_lba
= bdev
->bd_start_sect
+ bdev_nr_sectors(bdev
);
470 * Setup everything required to start tracing
472 static int do_blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
473 struct block_device
*bdev
,
474 struct blk_user_trace_setup
*buts
)
476 struct blk_trace
*bt
= NULL
;
477 struct dentry
*dir
= NULL
;
480 lockdep_assert_held(&q
->debugfs_mutex
);
482 if (!buts
->buf_size
|| !buts
->buf_nr
)
485 strncpy(buts
->name
, name
, BLKTRACE_BDEV_SIZE
);
486 buts
->name
[BLKTRACE_BDEV_SIZE
- 1] = '\0';
489 * some device names have larger paths - convert the slashes
490 * to underscores for this to work as expected
492 strreplace(buts
->name
, '/', '_');
495 * bdev can be NULL, as with scsi-generic, this is a helpful as
498 if (rcu_dereference_protected(q
->blk_trace
,
499 lockdep_is_held(&q
->debugfs_mutex
))) {
500 pr_warn("Concurrent blktraces are not allowed on %s\n",
505 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
510 bt
->sequence
= alloc_percpu(unsigned long);
514 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
519 * When tracing the whole disk reuse the existing debugfs directory
520 * created by the block layer on init. For partitions block devices,
521 * and scsi-generic block devices we create a temporary new debugfs
522 * directory that will be removed once the trace ends.
524 if (bdev
&& !bdev_is_partition(bdev
))
525 dir
= q
->debugfs_dir
;
527 bt
->dir
= dir
= debugfs_create_dir(buts
->name
, blk_debugfs_root
);
530 * As blktrace relies on debugfs for its interface the debugfs directory
531 * is required, contrary to the usual mantra of not checking for debugfs
532 * files or directories.
534 if (IS_ERR_OR_NULL(dir
)) {
535 pr_warn("debugfs_dir not present for %s so skipping\n",
542 atomic_set(&bt
->dropped
, 0);
543 INIT_LIST_HEAD(&bt
->running_list
);
546 debugfs_create_file("dropped", 0444, dir
, bt
, &blk_dropped_fops
);
547 debugfs_create_file("msg", 0222, dir
, bt
, &blk_msg_fops
);
549 bt
->rchan
= relay_open("trace", dir
, buts
->buf_size
,
550 buts
->buf_nr
, &blk_relay_callbacks
, bt
);
554 bt
->act_mask
= buts
->act_mask
;
556 bt
->act_mask
= (u16
) -1;
558 blk_trace_setup_lba(bt
, bdev
);
560 /* overwrite with user settings */
562 bt
->start_lba
= buts
->start_lba
;
564 bt
->end_lba
= buts
->end_lba
;
567 bt
->trace_state
= Blktrace_setup
;
569 rcu_assign_pointer(q
->blk_trace
, bt
);
579 static int __blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
580 struct block_device
*bdev
, char __user
*arg
)
582 struct blk_user_trace_setup buts
;
585 ret
= copy_from_user(&buts
, arg
, sizeof(buts
));
589 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
593 if (copy_to_user(arg
, &buts
, sizeof(buts
))) {
594 __blk_trace_remove(q
);
600 int blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
601 struct block_device
*bdev
,
606 mutex_lock(&q
->debugfs_mutex
);
607 ret
= __blk_trace_setup(q
, name
, dev
, bdev
, arg
);
608 mutex_unlock(&q
->debugfs_mutex
);
612 EXPORT_SYMBOL_GPL(blk_trace_setup
);
614 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
615 static int compat_blk_trace_setup(struct request_queue
*q
, char *name
,
616 dev_t dev
, struct block_device
*bdev
,
619 struct blk_user_trace_setup buts
;
620 struct compat_blk_user_trace_setup cbuts
;
623 if (copy_from_user(&cbuts
, arg
, sizeof(cbuts
)))
626 buts
= (struct blk_user_trace_setup
) {
627 .act_mask
= cbuts
.act_mask
,
628 .buf_size
= cbuts
.buf_size
,
629 .buf_nr
= cbuts
.buf_nr
,
630 .start_lba
= cbuts
.start_lba
,
631 .end_lba
= cbuts
.end_lba
,
635 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
639 if (copy_to_user(arg
, &buts
.name
, ARRAY_SIZE(buts
.name
))) {
640 __blk_trace_remove(q
);
648 static int __blk_trace_startstop(struct request_queue
*q
, int start
)
651 struct blk_trace
*bt
;
653 bt
= rcu_dereference_protected(q
->blk_trace
,
654 lockdep_is_held(&q
->debugfs_mutex
));
659 * For starting a trace, we can transition from a setup or stopped
660 * trace. For stopping a trace, the state must be running
664 if (bt
->trace_state
== Blktrace_setup
||
665 bt
->trace_state
== Blktrace_stopped
) {
668 bt
->trace_state
= Blktrace_running
;
669 spin_lock_irq(&running_trace_lock
);
670 list_add(&bt
->running_list
, &running_trace_list
);
671 spin_unlock_irq(&running_trace_lock
);
677 if (bt
->trace_state
== Blktrace_running
) {
678 bt
->trace_state
= Blktrace_stopped
;
679 spin_lock_irq(&running_trace_lock
);
680 list_del_init(&bt
->running_list
);
681 spin_unlock_irq(&running_trace_lock
);
682 relay_flush(bt
->rchan
);
690 int blk_trace_startstop(struct request_queue
*q
, int start
)
694 mutex_lock(&q
->debugfs_mutex
);
695 ret
= __blk_trace_startstop(q
, start
);
696 mutex_unlock(&q
->debugfs_mutex
);
700 EXPORT_SYMBOL_GPL(blk_trace_startstop
);
703 * When reading or writing the blktrace sysfs files, the references to the
704 * opened sysfs or device files should prevent the underlying block device
705 * from being removed. So no further delete protection is really needed.
709 * blk_trace_ioctl: - handle the ioctls associated with tracing
710 * @bdev: the block device
711 * @cmd: the ioctl cmd
712 * @arg: the argument data, if any
715 int blk_trace_ioctl(struct block_device
*bdev
, unsigned cmd
, char __user
*arg
)
717 struct request_queue
*q
;
719 char b
[BDEVNAME_SIZE
];
721 q
= bdev_get_queue(bdev
);
725 mutex_lock(&q
->debugfs_mutex
);
730 ret
= __blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
732 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
733 case BLKTRACESETUP32
:
735 ret
= compat_blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
742 ret
= __blk_trace_startstop(q
, start
);
744 case BLKTRACETEARDOWN
:
745 ret
= __blk_trace_remove(q
);
752 mutex_unlock(&q
->debugfs_mutex
);
757 * blk_trace_shutdown: - stop and cleanup trace structures
758 * @q: the request queue associated with the device
761 void blk_trace_shutdown(struct request_queue
*q
)
763 mutex_lock(&q
->debugfs_mutex
);
764 if (rcu_dereference_protected(q
->blk_trace
,
765 lockdep_is_held(&q
->debugfs_mutex
))) {
766 __blk_trace_startstop(q
, 0);
767 __blk_trace_remove(q
);
770 mutex_unlock(&q
->debugfs_mutex
);
773 #ifdef CONFIG_BLK_CGROUP
774 static u64
blk_trace_bio_get_cgid(struct request_queue
*q
, struct bio
*bio
)
776 struct blk_trace
*bt
;
778 /* We don't use the 'bt' value here except as an optimization... */
779 bt
= rcu_dereference_protected(q
->blk_trace
, 1);
780 if (!bt
|| !(blk_tracer_flags
.val
& TRACE_BLK_OPT_CGROUP
))
785 return cgroup_id(bio_blkcg(bio
)->css
.cgroup
);
788 static u64
blk_trace_bio_get_cgid(struct request_queue
*q
, struct bio
*bio
)
795 blk_trace_request_get_cgid(struct request
*rq
)
799 /* Use the first bio */
800 return blk_trace_bio_get_cgid(rq
->q
, rq
->bio
);
808 * blk_add_trace_rq - Add a trace for a request oriented action
809 * @rq: the source request
810 * @error: return status to log
811 * @nr_bytes: number of completed bytes
813 * @cgid: the cgroup info
816 * Records an action against a request. Will log the bio offset + size.
819 static void blk_add_trace_rq(struct request
*rq
, int error
,
820 unsigned int nr_bytes
, u32 what
, u64 cgid
)
822 struct blk_trace
*bt
;
825 bt
= rcu_dereference(rq
->q
->blk_trace
);
831 if (blk_rq_is_passthrough(rq
))
832 what
|= BLK_TC_ACT(BLK_TC_PC
);
834 what
|= BLK_TC_ACT(BLK_TC_FS
);
836 __blk_add_trace(bt
, blk_rq_trace_sector(rq
), nr_bytes
, req_op(rq
),
837 rq
->cmd_flags
, what
, error
, 0, NULL
, cgid
);
841 static void blk_add_trace_rq_insert(void *ignore
, struct request
*rq
)
843 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_INSERT
,
844 blk_trace_request_get_cgid(rq
));
847 static void blk_add_trace_rq_issue(void *ignore
, struct request
*rq
)
849 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_ISSUE
,
850 blk_trace_request_get_cgid(rq
));
853 static void blk_add_trace_rq_merge(void *ignore
, struct request
*rq
)
855 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_BACKMERGE
,
856 blk_trace_request_get_cgid(rq
));
859 static void blk_add_trace_rq_requeue(void *ignore
, struct request
*rq
)
861 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_REQUEUE
,
862 blk_trace_request_get_cgid(rq
));
865 static void blk_add_trace_rq_complete(void *ignore
, struct request
*rq
,
866 int error
, unsigned int nr_bytes
)
868 blk_add_trace_rq(rq
, error
, nr_bytes
, BLK_TA_COMPLETE
,
869 blk_trace_request_get_cgid(rq
));
873 * blk_add_trace_bio - Add a trace for a bio oriented action
874 * @q: queue the io is for
875 * @bio: the source bio
877 * @error: error, if any
880 * Records an action against a bio. Will log the bio offset + size.
883 static void blk_add_trace_bio(struct request_queue
*q
, struct bio
*bio
,
886 struct blk_trace
*bt
;
889 bt
= rcu_dereference(q
->blk_trace
);
895 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
896 bio_op(bio
), bio
->bi_opf
, what
, error
, 0, NULL
,
897 blk_trace_bio_get_cgid(q
, bio
));
901 static void blk_add_trace_bio_bounce(void *ignore
, struct bio
*bio
)
903 blk_add_trace_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
, BLK_TA_BOUNCE
, 0);
906 static void blk_add_trace_bio_complete(void *ignore
,
907 struct request_queue
*q
, struct bio
*bio
)
909 blk_add_trace_bio(q
, bio
, BLK_TA_COMPLETE
,
910 blk_status_to_errno(bio
->bi_status
));
913 static void blk_add_trace_bio_backmerge(void *ignore
, struct bio
*bio
)
915 blk_add_trace_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
, BLK_TA_BACKMERGE
,
919 static void blk_add_trace_bio_frontmerge(void *ignore
, struct bio
*bio
)
921 blk_add_trace_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
, BLK_TA_FRONTMERGE
,
925 static void blk_add_trace_bio_queue(void *ignore
, struct bio
*bio
)
927 blk_add_trace_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
, BLK_TA_QUEUE
, 0);
930 static void blk_add_trace_getrq(void *ignore
, struct bio
*bio
)
932 blk_add_trace_bio(bio
->bi_bdev
->bd_disk
->queue
, bio
, BLK_TA_GETRQ
, 0);
935 static void blk_add_trace_plug(void *ignore
, struct request_queue
*q
)
937 struct blk_trace
*bt
;
940 bt
= rcu_dereference(q
->blk_trace
);
942 __blk_add_trace(bt
, 0, 0, 0, 0, BLK_TA_PLUG
, 0, 0, NULL
, 0);
946 static void blk_add_trace_unplug(void *ignore
, struct request_queue
*q
,
947 unsigned int depth
, bool explicit)
949 struct blk_trace
*bt
;
952 bt
= rcu_dereference(q
->blk_trace
);
954 __be64 rpdu
= cpu_to_be64(depth
);
958 what
= BLK_TA_UNPLUG_IO
;
960 what
= BLK_TA_UNPLUG_TIMER
;
962 __blk_add_trace(bt
, 0, 0, 0, 0, what
, 0, sizeof(rpdu
), &rpdu
, 0);
967 static void blk_add_trace_split(void *ignore
, struct bio
*bio
, unsigned int pdu
)
969 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
970 struct blk_trace
*bt
;
973 bt
= rcu_dereference(q
->blk_trace
);
975 __be64 rpdu
= cpu_to_be64(pdu
);
977 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
,
978 bio
->bi_iter
.bi_size
, bio_op(bio
), bio
->bi_opf
,
980 blk_status_to_errno(bio
->bi_status
),
982 blk_trace_bio_get_cgid(q
, bio
));
988 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
989 * @ignore: trace callback data parameter (not used)
990 * @bio: the source bio
991 * @dev: source device
992 * @from: source sector
994 * Called after a bio is remapped to a different device and/or sector.
996 static void blk_add_trace_bio_remap(void *ignore
, struct bio
*bio
, dev_t dev
,
999 struct request_queue
*q
= bio
->bi_bdev
->bd_disk
->queue
;
1000 struct blk_trace
*bt
;
1001 struct blk_io_trace_remap r
;
1004 bt
= rcu_dereference(q
->blk_trace
);
1010 r
.device_from
= cpu_to_be32(dev
);
1011 r
.device_to
= cpu_to_be32(bio_dev(bio
));
1012 r
.sector_from
= cpu_to_be64(from
);
1014 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
1015 bio_op(bio
), bio
->bi_opf
, BLK_TA_REMAP
,
1016 blk_status_to_errno(bio
->bi_status
),
1017 sizeof(r
), &r
, blk_trace_bio_get_cgid(q
, bio
));
1022 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1023 * @ignore: trace callback data parameter (not used)
1024 * @rq: the source request
1025 * @dev: target device
1026 * @from: source sector
1029 * Device mapper remaps request to other devices.
1030 * Add a trace for that action.
1033 static void blk_add_trace_rq_remap(void *ignore
, struct request
*rq
, dev_t dev
,
1036 struct blk_trace
*bt
;
1037 struct blk_io_trace_remap r
;
1040 bt
= rcu_dereference(rq
->q
->blk_trace
);
1046 r
.device_from
= cpu_to_be32(dev
);
1047 r
.device_to
= cpu_to_be32(disk_devt(rq
->rq_disk
));
1048 r
.sector_from
= cpu_to_be64(from
);
1050 __blk_add_trace(bt
, blk_rq_pos(rq
), blk_rq_bytes(rq
),
1051 rq_data_dir(rq
), 0, BLK_TA_REMAP
, 0,
1052 sizeof(r
), &r
, blk_trace_request_get_cgid(rq
));
1057 * blk_add_driver_data - Add binary message with driver-specific data
1059 * @data: driver-specific data
1060 * @len: length of driver-specific data
1063 * Some drivers might want to write driver-specific data per request.
1066 void blk_add_driver_data(struct request
*rq
, void *data
, size_t len
)
1068 struct blk_trace
*bt
;
1071 bt
= rcu_dereference(rq
->q
->blk_trace
);
1077 __blk_add_trace(bt
, blk_rq_trace_sector(rq
), blk_rq_bytes(rq
), 0, 0,
1078 BLK_TA_DRV_DATA
, 0, len
, data
,
1079 blk_trace_request_get_cgid(rq
));
1082 EXPORT_SYMBOL_GPL(blk_add_driver_data
);
1084 static void blk_register_tracepoints(void)
1088 ret
= register_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1090 ret
= register_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1092 ret
= register_trace_block_rq_merge(blk_add_trace_rq_merge
, NULL
);
1094 ret
= register_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1096 ret
= register_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1098 ret
= register_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1100 ret
= register_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1102 ret
= register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1104 ret
= register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1106 ret
= register_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1108 ret
= register_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1110 ret
= register_trace_block_plug(blk_add_trace_plug
, NULL
);
1112 ret
= register_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1114 ret
= register_trace_block_split(blk_add_trace_split
, NULL
);
1116 ret
= register_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1118 ret
= register_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1122 static void blk_unregister_tracepoints(void)
1124 unregister_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1125 unregister_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1126 unregister_trace_block_split(blk_add_trace_split
, NULL
);
1127 unregister_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1128 unregister_trace_block_plug(blk_add_trace_plug
, NULL
);
1129 unregister_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1130 unregister_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1131 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1132 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1133 unregister_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1134 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1135 unregister_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1136 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1137 unregister_trace_block_rq_merge(blk_add_trace_rq_merge
, NULL
);
1138 unregister_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1139 unregister_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1141 tracepoint_synchronize_unregister();
1145 * struct blk_io_tracer formatting routines
1148 static void fill_rwbs(char *rwbs
, const struct blk_io_trace
*t
)
1151 int tc
= t
->action
>> BLK_TC_SHIFT
;
1153 if ((t
->action
& ~__BLK_TN_CGROUP
) == BLK_TN_MESSAGE
) {
1158 if (tc
& BLK_TC_FLUSH
)
1161 if (tc
& BLK_TC_DISCARD
)
1163 else if (tc
& BLK_TC_WRITE
)
1170 if (tc
& BLK_TC_FUA
)
1172 if (tc
& BLK_TC_AHEAD
)
1174 if (tc
& BLK_TC_SYNC
)
1176 if (tc
& BLK_TC_META
)
1183 const struct blk_io_trace
*te_blk_io_trace(const struct trace_entry
*ent
)
1185 return (const struct blk_io_trace
*)ent
;
1188 static inline const void *pdu_start(const struct trace_entry
*ent
, bool has_cg
)
1190 return (void *)(te_blk_io_trace(ent
) + 1) + (has_cg
? sizeof(u64
) : 0);
1193 static inline u64
t_cgid(const struct trace_entry
*ent
)
1195 return *(u64
*)(te_blk_io_trace(ent
) + 1);
1198 static inline int pdu_real_len(const struct trace_entry
*ent
, bool has_cg
)
1200 return te_blk_io_trace(ent
)->pdu_len
- (has_cg
? sizeof(u64
) : 0);
1203 static inline u32
t_action(const struct trace_entry
*ent
)
1205 return te_blk_io_trace(ent
)->action
;
1208 static inline u32
t_bytes(const struct trace_entry
*ent
)
1210 return te_blk_io_trace(ent
)->bytes
;
1213 static inline u32
t_sec(const struct trace_entry
*ent
)
1215 return te_blk_io_trace(ent
)->bytes
>> 9;
1218 static inline unsigned long long t_sector(const struct trace_entry
*ent
)
1220 return te_blk_io_trace(ent
)->sector
;
1223 static inline __u16
t_error(const struct trace_entry
*ent
)
1225 return te_blk_io_trace(ent
)->error
;
1228 static __u64
get_pdu_int(const struct trace_entry
*ent
, bool has_cg
)
1230 const __be64
*val
= pdu_start(ent
, has_cg
);
1231 return be64_to_cpu(*val
);
1234 typedef void (blk_log_action_t
) (struct trace_iterator
*iter
, const char *act
,
1237 static void blk_log_action_classic(struct trace_iterator
*iter
, const char *act
,
1240 char rwbs
[RWBS_LEN
];
1241 unsigned long long ts
= iter
->ts
;
1242 unsigned long nsec_rem
= do_div(ts
, NSEC_PER_SEC
);
1243 unsigned secs
= (unsigned long)ts
;
1244 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1248 trace_seq_printf(&iter
->seq
,
1249 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1250 MAJOR(t
->device
), MINOR(t
->device
), iter
->cpu
,
1251 secs
, nsec_rem
, iter
->ent
->pid
, act
, rwbs
);
1254 static void blk_log_action(struct trace_iterator
*iter
, const char *act
,
1257 char rwbs
[RWBS_LEN
];
1258 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1262 u64 id
= t_cgid(iter
->ent
);
1264 if (blk_tracer_flags
.val
& TRACE_BLK_OPT_CGNAME
) {
1265 char blkcg_name_buf
[NAME_MAX
+ 1] = "<...>";
1267 cgroup_path_from_kernfs_id(id
, blkcg_name_buf
,
1268 sizeof(blkcg_name_buf
));
1269 trace_seq_printf(&iter
->seq
, "%3d,%-3d %s %2s %3s ",
1270 MAJOR(t
->device
), MINOR(t
->device
),
1271 blkcg_name_buf
, act
, rwbs
);
1274 * The cgid portion used to be "INO,GEN". Userland
1275 * builds a FILEID_INO32_GEN fid out of them and
1276 * opens the cgroup using open_by_handle_at(2).
1277 * While 32bit ino setups are still the same, 64bit
1278 * ones now use the 64bit ino as the whole ID and
1279 * no longer use generation.
1281 * Regardless of the content, always output
1282 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1283 * be mapped back to @id on both 64 and 32bit ino
1284 * setups. See __kernfs_fh_to_dentry().
1286 trace_seq_printf(&iter
->seq
,
1287 "%3d,%-3d %llx,%-llx %2s %3s ",
1288 MAJOR(t
->device
), MINOR(t
->device
),
1289 id
& U32_MAX
, id
>> 32, act
, rwbs
);
1292 trace_seq_printf(&iter
->seq
, "%3d,%-3d %2s %3s ",
1293 MAJOR(t
->device
), MINOR(t
->device
), act
, rwbs
);
1296 static void blk_log_dump_pdu(struct trace_seq
*s
,
1297 const struct trace_entry
*ent
, bool has_cg
)
1299 const unsigned char *pdu_buf
;
1303 pdu_buf
= pdu_start(ent
, has_cg
);
1304 pdu_len
= pdu_real_len(ent
, has_cg
);
1309 /* find the last zero that needs to be printed */
1310 for (end
= pdu_len
- 1; end
>= 0; end
--)
1315 trace_seq_putc(s
, '(');
1317 for (i
= 0; i
< pdu_len
; i
++) {
1319 trace_seq_printf(s
, "%s%02x",
1320 i
== 0 ? "" : " ", pdu_buf
[i
]);
1323 * stop when the rest is just zeros and indicate so
1324 * with a ".." appended
1326 if (i
== end
&& end
!= pdu_len
- 1) {
1327 trace_seq_puts(s
, " ..) ");
1332 trace_seq_puts(s
, ") ");
1335 static void blk_log_generic(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1337 char cmd
[TASK_COMM_LEN
];
1339 trace_find_cmdline(ent
->pid
, cmd
);
1341 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1342 trace_seq_printf(s
, "%u ", t_bytes(ent
));
1343 blk_log_dump_pdu(s
, ent
, has_cg
);
1344 trace_seq_printf(s
, "[%s]\n", cmd
);
1347 trace_seq_printf(s
, "%llu + %u [%s]\n",
1348 t_sector(ent
), t_sec(ent
), cmd
);
1350 trace_seq_printf(s
, "[%s]\n", cmd
);
1354 static void blk_log_with_error(struct trace_seq
*s
,
1355 const struct trace_entry
*ent
, bool has_cg
)
1357 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1358 blk_log_dump_pdu(s
, ent
, has_cg
);
1359 trace_seq_printf(s
, "[%d]\n", t_error(ent
));
1362 trace_seq_printf(s
, "%llu + %u [%d]\n",
1364 t_sec(ent
), t_error(ent
));
1366 trace_seq_printf(s
, "%llu [%d]\n",
1367 t_sector(ent
), t_error(ent
));
1371 static void blk_log_remap(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1373 const struct blk_io_trace_remap
*__r
= pdu_start(ent
, has_cg
);
1375 trace_seq_printf(s
, "%llu + %u <- (%d,%d) %llu\n",
1376 t_sector(ent
), t_sec(ent
),
1377 MAJOR(be32_to_cpu(__r
->device_from
)),
1378 MINOR(be32_to_cpu(__r
->device_from
)),
1379 be64_to_cpu(__r
->sector_from
));
1382 static void blk_log_plug(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1384 char cmd
[TASK_COMM_LEN
];
1386 trace_find_cmdline(ent
->pid
, cmd
);
1388 trace_seq_printf(s
, "[%s]\n", cmd
);
1391 static void blk_log_unplug(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1393 char cmd
[TASK_COMM_LEN
];
1395 trace_find_cmdline(ent
->pid
, cmd
);
1397 trace_seq_printf(s
, "[%s] %llu\n", cmd
, get_pdu_int(ent
, has_cg
));
1400 static void blk_log_split(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 trace_seq_printf(s
, "%llu / %llu [%s]\n", t_sector(ent
),
1407 get_pdu_int(ent
, has_cg
), cmd
);
1410 static void blk_log_msg(struct trace_seq
*s
, const struct trace_entry
*ent
,
1414 trace_seq_putmem(s
, pdu_start(ent
, has_cg
),
1415 pdu_real_len(ent
, has_cg
));
1416 trace_seq_putc(s
, '\n');
1420 * struct tracer operations
1423 static void blk_tracer_print_header(struct seq_file
*m
)
1425 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1427 seq_puts(m
, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1431 static void blk_tracer_start(struct trace_array
*tr
)
1433 blk_tracer_enabled
= true;
1436 static int blk_tracer_init(struct trace_array
*tr
)
1439 blk_tracer_start(tr
);
1443 static void blk_tracer_stop(struct trace_array
*tr
)
1445 blk_tracer_enabled
= false;
1448 static void blk_tracer_reset(struct trace_array
*tr
)
1450 blk_tracer_stop(tr
);
1453 static const struct {
1455 void (*print
)(struct trace_seq
*s
, const struct trace_entry
*ent
,
1458 [__BLK_TA_QUEUE
] = {{ "Q", "queue" }, blk_log_generic
},
1459 [__BLK_TA_BACKMERGE
] = {{ "M", "backmerge" }, blk_log_generic
},
1460 [__BLK_TA_FRONTMERGE
] = {{ "F", "frontmerge" }, blk_log_generic
},
1461 [__BLK_TA_GETRQ
] = {{ "G", "getrq" }, blk_log_generic
},
1462 [__BLK_TA_SLEEPRQ
] = {{ "S", "sleeprq" }, blk_log_generic
},
1463 [__BLK_TA_REQUEUE
] = {{ "R", "requeue" }, blk_log_with_error
},
1464 [__BLK_TA_ISSUE
] = {{ "D", "issue" }, blk_log_generic
},
1465 [__BLK_TA_COMPLETE
] = {{ "C", "complete" }, blk_log_with_error
},
1466 [__BLK_TA_PLUG
] = {{ "P", "plug" }, blk_log_plug
},
1467 [__BLK_TA_UNPLUG_IO
] = {{ "U", "unplug_io" }, blk_log_unplug
},
1468 [__BLK_TA_UNPLUG_TIMER
] = {{ "UT", "unplug_timer" }, blk_log_unplug
},
1469 [__BLK_TA_INSERT
] = {{ "I", "insert" }, blk_log_generic
},
1470 [__BLK_TA_SPLIT
] = {{ "X", "split" }, blk_log_split
},
1471 [__BLK_TA_BOUNCE
] = {{ "B", "bounce" }, blk_log_generic
},
1472 [__BLK_TA_REMAP
] = {{ "A", "remap" }, blk_log_remap
},
1475 static enum print_line_t
print_one_line(struct trace_iterator
*iter
,
1478 struct trace_array
*tr
= iter
->tr
;
1479 struct trace_seq
*s
= &iter
->seq
;
1480 const struct blk_io_trace
*t
;
1483 blk_log_action_t
*log_action
;
1486 t
= te_blk_io_trace(iter
->ent
);
1487 what
= (t
->action
& ((1 << BLK_TC_SHIFT
) - 1)) & ~__BLK_TA_CGROUP
;
1488 long_act
= !!(tr
->trace_flags
& TRACE_ITER_VERBOSE
);
1489 log_action
= classic
? &blk_log_action_classic
: &blk_log_action
;
1490 has_cg
= t
->action
& __BLK_TA_CGROUP
;
1492 if ((t
->action
& ~__BLK_TN_CGROUP
) == BLK_TN_MESSAGE
) {
1493 log_action(iter
, long_act
? "message" : "m", has_cg
);
1494 blk_log_msg(s
, iter
->ent
, has_cg
);
1495 return trace_handle_return(s
);
1498 if (unlikely(what
== 0 || what
>= ARRAY_SIZE(what2act
)))
1499 trace_seq_printf(s
, "Unknown action %x\n", what
);
1501 log_action(iter
, what2act
[what
].act
[long_act
], has_cg
);
1502 what2act
[what
].print(s
, iter
->ent
, has_cg
);
1505 return trace_handle_return(s
);
1508 static enum print_line_t
blk_trace_event_print(struct trace_iterator
*iter
,
1509 int flags
, struct trace_event
*event
)
1511 return print_one_line(iter
, false);
1514 static void blk_trace_synthesize_old_trace(struct trace_iterator
*iter
)
1516 struct trace_seq
*s
= &iter
->seq
;
1517 struct blk_io_trace
*t
= (struct blk_io_trace
*)iter
->ent
;
1518 const int offset
= offsetof(struct blk_io_trace
, sector
);
1519 struct blk_io_trace old
= {
1520 .magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
,
1524 trace_seq_putmem(s
, &old
, offset
);
1525 trace_seq_putmem(s
, &t
->sector
,
1526 sizeof(old
) - offset
+ t
->pdu_len
);
1529 static enum print_line_t
1530 blk_trace_event_print_binary(struct trace_iterator
*iter
, int flags
,
1531 struct trace_event
*event
)
1533 blk_trace_synthesize_old_trace(iter
);
1535 return trace_handle_return(&iter
->seq
);
1538 static enum print_line_t
blk_tracer_print_line(struct trace_iterator
*iter
)
1540 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1541 return TRACE_TYPE_UNHANDLED
;
1543 return print_one_line(iter
, true);
1547 blk_tracer_set_flag(struct trace_array
*tr
, u32 old_flags
, u32 bit
, int set
)
1549 /* don't output context-info for blk_classic output */
1550 if (bit
== TRACE_BLK_OPT_CLASSIC
) {
1552 tr
->trace_flags
&= ~TRACE_ITER_CONTEXT_INFO
;
1554 tr
->trace_flags
|= TRACE_ITER_CONTEXT_INFO
;
1559 static struct tracer blk_tracer __read_mostly
= {
1561 .init
= blk_tracer_init
,
1562 .reset
= blk_tracer_reset
,
1563 .start
= blk_tracer_start
,
1564 .stop
= blk_tracer_stop
,
1565 .print_header
= blk_tracer_print_header
,
1566 .print_line
= blk_tracer_print_line
,
1567 .flags
= &blk_tracer_flags
,
1568 .set_flag
= blk_tracer_set_flag
,
1571 static struct trace_event_functions trace_blk_event_funcs
= {
1572 .trace
= blk_trace_event_print
,
1573 .binary
= blk_trace_event_print_binary
,
1576 static struct trace_event trace_blk_event
= {
1578 .funcs
= &trace_blk_event_funcs
,
1581 static int __init
init_blk_tracer(void)
1583 if (!register_trace_event(&trace_blk_event
)) {
1584 pr_warn("Warning: could not register block events\n");
1588 if (register_tracer(&blk_tracer
) != 0) {
1589 pr_warn("Warning: could not register the block tracer\n");
1590 unregister_trace_event(&trace_blk_event
);
1597 device_initcall(init_blk_tracer
);
1599 static int blk_trace_remove_queue(struct request_queue
*q
)
1601 struct blk_trace
*bt
;
1603 bt
= rcu_replace_pointer(q
->blk_trace
, NULL
,
1604 lockdep_is_held(&q
->debugfs_mutex
));
1615 * Setup everything required to start tracing
1617 static int blk_trace_setup_queue(struct request_queue
*q
,
1618 struct block_device
*bdev
)
1620 struct blk_trace
*bt
= NULL
;
1623 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
1627 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
1631 bt
->dev
= bdev
->bd_dev
;
1632 bt
->act_mask
= (u16
)-1;
1634 blk_trace_setup_lba(bt
, bdev
);
1636 rcu_assign_pointer(q
->blk_trace
, bt
);
1646 * sysfs interface to enable and configure tracing
1649 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1650 struct device_attribute
*attr
,
1652 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1653 struct device_attribute
*attr
,
1654 const char *buf
, size_t count
);
1655 #define BLK_TRACE_DEVICE_ATTR(_name) \
1656 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1657 sysfs_blk_trace_attr_show, \
1658 sysfs_blk_trace_attr_store)
1660 static BLK_TRACE_DEVICE_ATTR(enable
);
1661 static BLK_TRACE_DEVICE_ATTR(act_mask
);
1662 static BLK_TRACE_DEVICE_ATTR(pid
);
1663 static BLK_TRACE_DEVICE_ATTR(start_lba
);
1664 static BLK_TRACE_DEVICE_ATTR(end_lba
);
1666 static struct attribute
*blk_trace_attrs
[] = {
1667 &dev_attr_enable
.attr
,
1668 &dev_attr_act_mask
.attr
,
1670 &dev_attr_start_lba
.attr
,
1671 &dev_attr_end_lba
.attr
,
1675 struct attribute_group blk_trace_attr_group
= {
1677 .attrs
= blk_trace_attrs
,
1680 static const struct {
1684 { BLK_TC_READ
, "read" },
1685 { BLK_TC_WRITE
, "write" },
1686 { BLK_TC_FLUSH
, "flush" },
1687 { BLK_TC_SYNC
, "sync" },
1688 { BLK_TC_QUEUE
, "queue" },
1689 { BLK_TC_REQUEUE
, "requeue" },
1690 { BLK_TC_ISSUE
, "issue" },
1691 { BLK_TC_COMPLETE
, "complete" },
1692 { BLK_TC_FS
, "fs" },
1693 { BLK_TC_PC
, "pc" },
1694 { BLK_TC_NOTIFY
, "notify" },
1695 { BLK_TC_AHEAD
, "ahead" },
1696 { BLK_TC_META
, "meta" },
1697 { BLK_TC_DISCARD
, "discard" },
1698 { BLK_TC_DRV_DATA
, "drv_data" },
1699 { BLK_TC_FUA
, "fua" },
1702 static int blk_trace_str2mask(const char *str
)
1706 char *buf
, *s
, *token
;
1708 buf
= kstrdup(str
, GFP_KERNEL
);
1714 token
= strsep(&s
, ",");
1721 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1722 if (strcasecmp(token
, mask_maps
[i
].str
) == 0) {
1723 mask
|= mask_maps
[i
].mask
;
1727 if (i
== ARRAY_SIZE(mask_maps
)) {
1737 static ssize_t
blk_trace_mask2str(char *buf
, int mask
)
1742 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1743 if (mask
& mask_maps
[i
].mask
) {
1744 p
+= sprintf(p
, "%s%s",
1745 (p
== buf
) ? "" : ",", mask_maps
[i
].str
);
1753 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1754 struct device_attribute
*attr
,
1757 struct block_device
*bdev
= dev_to_bdev(dev
);
1758 struct request_queue
*q
= bdev_get_queue(bdev
);
1759 struct blk_trace
*bt
;
1760 ssize_t ret
= -ENXIO
;
1762 mutex_lock(&q
->debugfs_mutex
);
1764 bt
= rcu_dereference_protected(q
->blk_trace
,
1765 lockdep_is_held(&q
->debugfs_mutex
));
1766 if (attr
== &dev_attr_enable
) {
1767 ret
= sprintf(buf
, "%u\n", !!bt
);
1768 goto out_unlock_bdev
;
1772 ret
= sprintf(buf
, "disabled\n");
1773 else if (attr
== &dev_attr_act_mask
)
1774 ret
= blk_trace_mask2str(buf
, bt
->act_mask
);
1775 else if (attr
== &dev_attr_pid
)
1776 ret
= sprintf(buf
, "%u\n", bt
->pid
);
1777 else if (attr
== &dev_attr_start_lba
)
1778 ret
= sprintf(buf
, "%llu\n", bt
->start_lba
);
1779 else if (attr
== &dev_attr_end_lba
)
1780 ret
= sprintf(buf
, "%llu\n", bt
->end_lba
);
1783 mutex_unlock(&q
->debugfs_mutex
);
1787 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1788 struct device_attribute
*attr
,
1789 const char *buf
, size_t count
)
1791 struct block_device
*bdev
= dev_to_bdev(dev
);
1792 struct request_queue
*q
= bdev_get_queue(bdev
);
1793 struct blk_trace
*bt
;
1795 ssize_t ret
= -EINVAL
;
1800 if (attr
== &dev_attr_act_mask
) {
1801 if (kstrtoull(buf
, 0, &value
)) {
1802 /* Assume it is a list of trace category names */
1803 ret
= blk_trace_str2mask(buf
);
1809 if (kstrtoull(buf
, 0, &value
))
1813 mutex_lock(&q
->debugfs_mutex
);
1815 bt
= rcu_dereference_protected(q
->blk_trace
,
1816 lockdep_is_held(&q
->debugfs_mutex
));
1817 if (attr
== &dev_attr_enable
) {
1818 if (!!value
== !!bt
) {
1820 goto out_unlock_bdev
;
1823 ret
= blk_trace_setup_queue(q
, bdev
);
1825 ret
= blk_trace_remove_queue(q
);
1826 goto out_unlock_bdev
;
1831 ret
= blk_trace_setup_queue(q
, bdev
);
1832 bt
= rcu_dereference_protected(q
->blk_trace
,
1833 lockdep_is_held(&q
->debugfs_mutex
));
1837 if (attr
== &dev_attr_act_mask
)
1838 bt
->act_mask
= value
;
1839 else if (attr
== &dev_attr_pid
)
1841 else if (attr
== &dev_attr_start_lba
)
1842 bt
->start_lba
= value
;
1843 else if (attr
== &dev_attr_end_lba
)
1844 bt
->end_lba
= value
;
1848 mutex_unlock(&q
->debugfs_mutex
);
1850 return ret
? ret
: count
;
1853 int blk_trace_init_sysfs(struct device
*dev
)
1855 return sysfs_create_group(&dev
->kobj
, &blk_trace_attr_group
);
1858 void blk_trace_remove_sysfs(struct device
*dev
)
1860 sysfs_remove_group(&dev
->kobj
, &blk_trace_attr_group
);
1863 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1865 #ifdef CONFIG_EVENT_TRACING
1868 * blk_fill_rwbs - Fill the buffer rwbs by mapping op to character string.
1869 * @rwbs: buffer to be filled
1870 * @op: REQ_OP_XXX for the tracepoint
1873 * Maps the REQ_OP_XXX to character and fills the buffer provided by the
1874 * caller with resulting string.
1877 void blk_fill_rwbs(char *rwbs
, unsigned int op
)
1881 if (op
& REQ_PREFLUSH
)
1884 switch (op
& REQ_OP_MASK
) {
1886 case REQ_OP_WRITE_SAME
:
1889 case REQ_OP_DISCARD
:
1892 case REQ_OP_SECURE_ERASE
:
1908 if (op
& REQ_RAHEAD
)
1917 EXPORT_SYMBOL_GPL(blk_fill_rwbs
);
1919 #endif /* CONFIG_EVENT_TRACING */