1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/kernel.h>
10 #include <linux/blkdev.h>
11 #include <linux/blktrace_api.h>
12 #include <linux/percpu.h>
13 #include <linux/init.h>
14 #include <linux/mutex.h>
15 #include <linux/slab.h>
16 #include <linux/debugfs.h>
17 #include <linux/export.h>
18 #include <linux/time.h>
19 #include <linux/uaccess.h>
20 #include <linux/list.h>
21 #include <linux/blk-cgroup.h>
23 #include "../../block/blk.h"
25 #include <trace/events/block.h>
27 #include "trace_output.h"
29 #ifdef CONFIG_BLK_DEV_IO_TRACE
31 static unsigned int blktrace_seq __read_mostly
= 1;
33 static struct trace_array
*blk_tr
;
34 static bool blk_tracer_enabled __read_mostly
;
36 static LIST_HEAD(running_trace_list
);
37 static __cacheline_aligned_in_smp
DEFINE_SPINLOCK(running_trace_lock
);
39 /* Select an alternative, minimalistic output than the original one */
40 #define TRACE_BLK_OPT_CLASSIC 0x1
41 #define TRACE_BLK_OPT_CGROUP 0x2
42 #define TRACE_BLK_OPT_CGNAME 0x4
44 static struct tracer_opt blk_tracer_opts
[] = {
45 /* Default disable the minimalistic output */
46 { TRACER_OPT(blk_classic
, TRACE_BLK_OPT_CLASSIC
) },
47 #ifdef CONFIG_BLK_CGROUP
48 { TRACER_OPT(blk_cgroup
, TRACE_BLK_OPT_CGROUP
) },
49 { TRACER_OPT(blk_cgname
, TRACE_BLK_OPT_CGNAME
) },
54 static struct tracer_flags blk_tracer_flags
= {
56 .opts
= blk_tracer_opts
,
59 /* Global reference count of probes */
60 static DEFINE_MUTEX(blk_probe_mutex
);
61 static int blk_probes_ref
;
63 static void blk_register_tracepoints(void);
64 static void blk_unregister_tracepoints(void);
67 * Send out a notify message.
69 static void trace_note(struct blk_trace
*bt
, pid_t pid
, int action
,
70 const void *data
, size_t len
, u64 cgid
)
72 struct blk_io_trace
*t
;
73 struct ring_buffer_event
*event
= NULL
;
74 struct trace_buffer
*buffer
= NULL
;
76 int cpu
= smp_processor_id();
77 bool blk_tracer
= blk_tracer_enabled
;
78 ssize_t cgid_len
= cgid
? sizeof(cgid
) : 0;
81 buffer
= blk_tr
->array_buffer
.buffer
;
83 event
= trace_buffer_lock_reserve(buffer
, TRACE_BLK
,
84 sizeof(*t
) + len
+ cgid_len
,
88 t
= ring_buffer_event_data(event
);
95 t
= relay_reserve(bt
->rchan
, sizeof(*t
) + len
+ cgid_len
);
97 t
->magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
;
98 t
->time
= ktime_to_ns(ktime_get());
101 t
->action
= action
| (cgid
? __BLK_TN_CGROUP
: 0);
104 t
->pdu_len
= len
+ cgid_len
;
106 memcpy((void *)t
+ sizeof(*t
), &cgid
, cgid_len
);
107 memcpy((void *) t
+ sizeof(*t
) + cgid_len
, data
, len
);
110 trace_buffer_unlock_commit(blk_tr
, buffer
, event
, 0, pc
);
115 * Send out a notify for this process, if we haven't done so since a trace
118 static void trace_note_tsk(struct task_struct
*tsk
)
121 struct blk_trace
*bt
;
123 tsk
->btrace_seq
= blktrace_seq
;
124 spin_lock_irqsave(&running_trace_lock
, flags
);
125 list_for_each_entry(bt
, &running_trace_list
, running_list
) {
126 trace_note(bt
, tsk
->pid
, BLK_TN_PROCESS
, tsk
->comm
,
127 sizeof(tsk
->comm
), 0);
129 spin_unlock_irqrestore(&running_trace_lock
, flags
);
132 static void trace_note_time(struct blk_trace
*bt
)
134 struct timespec64 now
;
138 /* need to check user space to see if this breaks in y2038 or y2106 */
139 ktime_get_real_ts64(&now
);
140 words
[0] = (u32
)now
.tv_sec
;
141 words
[1] = now
.tv_nsec
;
143 local_irq_save(flags
);
144 trace_note(bt
, 0, BLK_TN_TIMESTAMP
, words
, sizeof(words
), 0);
145 local_irq_restore(flags
);
148 void __trace_note_message(struct blk_trace
*bt
, struct blkcg
*blkcg
,
149 const char *fmt
, ...)
156 if (unlikely(bt
->trace_state
!= Blktrace_running
&&
157 !blk_tracer_enabled
))
161 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
162 * message to the trace.
164 if (!(bt
->act_mask
& BLK_TC_NOTIFY
))
167 local_irq_save(flags
);
168 buf
= this_cpu_ptr(bt
->msg_data
);
170 n
= vscnprintf(buf
, BLK_TN_MAX_MSG
, fmt
, args
);
173 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CGROUP
))
175 #ifdef CONFIG_BLK_CGROUP
176 trace_note(bt
, current
->pid
, BLK_TN_MESSAGE
, buf
, n
,
177 blkcg
? cgroup_id(blkcg
->css
.cgroup
) : 1);
179 trace_note(bt
, current
->pid
, BLK_TN_MESSAGE
, buf
, n
, 0);
181 local_irq_restore(flags
);
183 EXPORT_SYMBOL_GPL(__trace_note_message
);
185 static int act_log_check(struct blk_trace
*bt
, u32 what
, sector_t sector
,
188 if (((bt
->act_mask
<< BLK_TC_SHIFT
) & what
) == 0)
190 if (sector
&& (sector
< bt
->start_lba
|| sector
> bt
->end_lba
))
192 if (bt
->pid
&& pid
!= bt
->pid
)
199 * Data direction bit lookup
201 static const u32 ddir_act
[2] = { BLK_TC_ACT(BLK_TC_READ
),
202 BLK_TC_ACT(BLK_TC_WRITE
) };
204 #define BLK_TC_RAHEAD BLK_TC_AHEAD
205 #define BLK_TC_PREFLUSH BLK_TC_FLUSH
207 /* The ilog2() calls fall out because they're constant */
208 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
209 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
212 * The worker for the various blk_add_trace*() types. Fills out a
213 * blk_io_trace structure and places it in a per-cpu subbuffer.
215 static void __blk_add_trace(struct blk_trace
*bt
, sector_t sector
, int bytes
,
216 int op
, int op_flags
, u32 what
, int error
, int pdu_len
,
217 void *pdu_data
, u64 cgid
)
219 struct task_struct
*tsk
= current
;
220 struct ring_buffer_event
*event
= NULL
;
221 struct trace_buffer
*buffer
= NULL
;
222 struct blk_io_trace
*t
;
223 unsigned long flags
= 0;
224 unsigned long *sequence
;
227 bool blk_tracer
= blk_tracer_enabled
;
228 ssize_t cgid_len
= cgid
? sizeof(cgid
) : 0;
230 if (unlikely(bt
->trace_state
!= Blktrace_running
&& !blk_tracer
))
233 what
|= ddir_act
[op_is_write(op
) ? WRITE
: READ
];
234 what
|= MASK_TC_BIT(op_flags
, SYNC
);
235 what
|= MASK_TC_BIT(op_flags
, RAHEAD
);
236 what
|= MASK_TC_BIT(op_flags
, META
);
237 what
|= MASK_TC_BIT(op_flags
, PREFLUSH
);
238 what
|= MASK_TC_BIT(op_flags
, FUA
);
239 if (op
== REQ_OP_DISCARD
|| op
== REQ_OP_SECURE_ERASE
)
240 what
|= BLK_TC_ACT(BLK_TC_DISCARD
);
241 if (op
== REQ_OP_FLUSH
)
242 what
|= BLK_TC_ACT(BLK_TC_FLUSH
);
244 what
|= __BLK_TA_CGROUP
;
247 if (act_log_check(bt
, what
, sector
, pid
))
249 cpu
= raw_smp_processor_id();
252 tracing_record_cmdline(current
);
254 buffer
= blk_tr
->array_buffer
.buffer
;
255 pc
= preempt_count();
256 event
= trace_buffer_lock_reserve(buffer
, TRACE_BLK
,
257 sizeof(*t
) + pdu_len
+ cgid_len
,
261 t
= ring_buffer_event_data(event
);
265 if (unlikely(tsk
->btrace_seq
!= blktrace_seq
))
269 * A word about the locking here - we disable interrupts to reserve
270 * some space in the relay per-cpu buffer, to prevent an irq
271 * from coming in and stepping on our toes.
273 local_irq_save(flags
);
274 t
= relay_reserve(bt
->rchan
, sizeof(*t
) + pdu_len
+ cgid_len
);
276 sequence
= per_cpu_ptr(bt
->sequence
, cpu
);
278 t
->magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
;
279 t
->sequence
= ++(*sequence
);
280 t
->time
= ktime_to_ns(ktime_get());
283 * These two are not needed in ftrace as they are in the
284 * generic trace_entry, filled by tracing_generic_entry_update,
285 * but for the trace_event->bin() synthesizer benefit we do it
296 t
->pdu_len
= pdu_len
+ cgid_len
;
299 memcpy((void *)t
+ sizeof(*t
), &cgid
, cgid_len
);
301 memcpy((void *)t
+ sizeof(*t
) + cgid_len
, pdu_data
, pdu_len
);
304 trace_buffer_unlock_commit(blk_tr
, buffer
, event
, 0, pc
);
309 local_irq_restore(flags
);
312 static void blk_trace_free(struct blk_trace
*bt
)
314 debugfs_remove(bt
->msg_file
);
315 debugfs_remove(bt
->dropped_file
);
316 relay_close(bt
->rchan
);
317 debugfs_remove(bt
->dir
);
318 free_percpu(bt
->sequence
);
319 free_percpu(bt
->msg_data
);
323 static void get_probe_ref(void)
325 mutex_lock(&blk_probe_mutex
);
326 if (++blk_probes_ref
== 1)
327 blk_register_tracepoints();
328 mutex_unlock(&blk_probe_mutex
);
331 static void put_probe_ref(void)
333 mutex_lock(&blk_probe_mutex
);
334 if (!--blk_probes_ref
)
335 blk_unregister_tracepoints();
336 mutex_unlock(&blk_probe_mutex
);
339 static void blk_trace_cleanup(struct blk_trace
*bt
)
346 static int __blk_trace_remove(struct request_queue
*q
)
348 struct blk_trace
*bt
;
350 bt
= rcu_replace_pointer(q
->blk_trace
, NULL
,
351 lockdep_is_held(&q
->debugfs_mutex
));
355 if (bt
->trace_state
!= Blktrace_running
)
356 blk_trace_cleanup(bt
);
361 int blk_trace_remove(struct request_queue
*q
)
365 mutex_lock(&q
->debugfs_mutex
);
366 ret
= __blk_trace_remove(q
);
367 mutex_unlock(&q
->debugfs_mutex
);
371 EXPORT_SYMBOL_GPL(blk_trace_remove
);
373 static ssize_t
blk_dropped_read(struct file
*filp
, char __user
*buffer
,
374 size_t count
, loff_t
*ppos
)
376 struct blk_trace
*bt
= filp
->private_data
;
379 snprintf(buf
, sizeof(buf
), "%u\n", atomic_read(&bt
->dropped
));
381 return simple_read_from_buffer(buffer
, count
, ppos
, buf
, strlen(buf
));
384 static const struct file_operations blk_dropped_fops
= {
385 .owner
= THIS_MODULE
,
387 .read
= blk_dropped_read
,
388 .llseek
= default_llseek
,
391 static ssize_t
blk_msg_write(struct file
*filp
, const char __user
*buffer
,
392 size_t count
, loff_t
*ppos
)
395 struct blk_trace
*bt
;
397 if (count
>= BLK_TN_MAX_MSG
)
400 msg
= memdup_user_nul(buffer
, count
);
404 bt
= filp
->private_data
;
405 __trace_note_message(bt
, NULL
, "%s", msg
);
411 static const struct file_operations blk_msg_fops
= {
412 .owner
= THIS_MODULE
,
414 .write
= blk_msg_write
,
415 .llseek
= noop_llseek
,
419 * Keep track of how many times we encountered a full subbuffer, to aid
420 * the user space app in telling how many lost events there were.
422 static int blk_subbuf_start_callback(struct rchan_buf
*buf
, void *subbuf
,
423 void *prev_subbuf
, size_t prev_padding
)
425 struct blk_trace
*bt
;
427 if (!relay_buf_full(buf
))
430 bt
= buf
->chan
->private_data
;
431 atomic_inc(&bt
->dropped
);
435 static int blk_remove_buf_file_callback(struct dentry
*dentry
)
437 debugfs_remove(dentry
);
442 static struct dentry
*blk_create_buf_file_callback(const char *filename
,
443 struct dentry
*parent
,
445 struct rchan_buf
*buf
,
448 return debugfs_create_file(filename
, mode
, parent
, buf
,
449 &relay_file_operations
);
452 static struct rchan_callbacks blk_relay_callbacks
= {
453 .subbuf_start
= blk_subbuf_start_callback
,
454 .create_buf_file
= blk_create_buf_file_callback
,
455 .remove_buf_file
= blk_remove_buf_file_callback
,
458 static void blk_trace_setup_lba(struct blk_trace
*bt
,
459 struct block_device
*bdev
)
462 bt
->start_lba
= bdev
->bd_start_sect
;
463 bt
->end_lba
= bdev
->bd_start_sect
+ bdev_nr_sectors(bdev
);
471 * Setup everything required to start tracing
473 static int do_blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
474 struct block_device
*bdev
,
475 struct blk_user_trace_setup
*buts
)
477 struct blk_trace
*bt
= NULL
;
478 struct dentry
*dir
= NULL
;
481 lockdep_assert_held(&q
->debugfs_mutex
);
483 if (!buts
->buf_size
|| !buts
->buf_nr
)
486 strncpy(buts
->name
, name
, BLKTRACE_BDEV_SIZE
);
487 buts
->name
[BLKTRACE_BDEV_SIZE
- 1] = '\0';
490 * some device names have larger paths - convert the slashes
491 * to underscores for this to work as expected
493 strreplace(buts
->name
, '/', '_');
496 * bdev can be NULL, as with scsi-generic, this is a helpful as
499 if (rcu_dereference_protected(q
->blk_trace
,
500 lockdep_is_held(&q
->debugfs_mutex
))) {
501 pr_warn("Concurrent blktraces are not allowed on %s\n",
506 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
511 bt
->sequence
= alloc_percpu(unsigned long);
515 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
520 * When tracing the whole disk reuse the existing debugfs directory
521 * created by the block layer on init. For partitions block devices,
522 * and scsi-generic block devices we create a temporary new debugfs
523 * directory that will be removed once the trace ends.
525 if (bdev
&& !bdev_is_partition(bdev
))
526 dir
= q
->debugfs_dir
;
528 bt
->dir
= dir
= debugfs_create_dir(buts
->name
, blk_debugfs_root
);
531 * As blktrace relies on debugfs for its interface the debugfs directory
532 * is required, contrary to the usual mantra of not checking for debugfs
533 * files or directories.
535 if (IS_ERR_OR_NULL(dir
)) {
536 pr_warn("debugfs_dir not present for %s so skipping\n",
543 atomic_set(&bt
->dropped
, 0);
544 INIT_LIST_HEAD(&bt
->running_list
);
547 bt
->dropped_file
= debugfs_create_file("dropped", 0444, dir
, bt
,
550 bt
->msg_file
= debugfs_create_file("msg", 0222, dir
, bt
, &blk_msg_fops
);
552 bt
->rchan
= relay_open("trace", dir
, buts
->buf_size
,
553 buts
->buf_nr
, &blk_relay_callbacks
, bt
);
557 bt
->act_mask
= buts
->act_mask
;
559 bt
->act_mask
= (u16
) -1;
561 blk_trace_setup_lba(bt
, bdev
);
563 /* overwrite with user settings */
565 bt
->start_lba
= buts
->start_lba
;
567 bt
->end_lba
= buts
->end_lba
;
570 bt
->trace_state
= Blktrace_setup
;
572 rcu_assign_pointer(q
->blk_trace
, bt
);
582 static int __blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
583 struct block_device
*bdev
, char __user
*arg
)
585 struct blk_user_trace_setup buts
;
588 ret
= copy_from_user(&buts
, arg
, sizeof(buts
));
592 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
596 if (copy_to_user(arg
, &buts
, sizeof(buts
))) {
597 __blk_trace_remove(q
);
603 int blk_trace_setup(struct request_queue
*q
, char *name
, dev_t dev
,
604 struct block_device
*bdev
,
609 mutex_lock(&q
->debugfs_mutex
);
610 ret
= __blk_trace_setup(q
, name
, dev
, bdev
, arg
);
611 mutex_unlock(&q
->debugfs_mutex
);
615 EXPORT_SYMBOL_GPL(blk_trace_setup
);
617 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
618 static int compat_blk_trace_setup(struct request_queue
*q
, char *name
,
619 dev_t dev
, struct block_device
*bdev
,
622 struct blk_user_trace_setup buts
;
623 struct compat_blk_user_trace_setup cbuts
;
626 if (copy_from_user(&cbuts
, arg
, sizeof(cbuts
)))
629 buts
= (struct blk_user_trace_setup
) {
630 .act_mask
= cbuts
.act_mask
,
631 .buf_size
= cbuts
.buf_size
,
632 .buf_nr
= cbuts
.buf_nr
,
633 .start_lba
= cbuts
.start_lba
,
634 .end_lba
= cbuts
.end_lba
,
638 ret
= do_blk_trace_setup(q
, name
, dev
, bdev
, &buts
);
642 if (copy_to_user(arg
, &buts
.name
, ARRAY_SIZE(buts
.name
))) {
643 __blk_trace_remove(q
);
651 static int __blk_trace_startstop(struct request_queue
*q
, int start
)
654 struct blk_trace
*bt
;
656 bt
= rcu_dereference_protected(q
->blk_trace
,
657 lockdep_is_held(&q
->debugfs_mutex
));
662 * For starting a trace, we can transition from a setup or stopped
663 * trace. For stopping a trace, the state must be running
667 if (bt
->trace_state
== Blktrace_setup
||
668 bt
->trace_state
== Blktrace_stopped
) {
671 bt
->trace_state
= Blktrace_running
;
672 spin_lock_irq(&running_trace_lock
);
673 list_add(&bt
->running_list
, &running_trace_list
);
674 spin_unlock_irq(&running_trace_lock
);
680 if (bt
->trace_state
== Blktrace_running
) {
681 bt
->trace_state
= Blktrace_stopped
;
682 spin_lock_irq(&running_trace_lock
);
683 list_del_init(&bt
->running_list
);
684 spin_unlock_irq(&running_trace_lock
);
685 relay_flush(bt
->rchan
);
693 int blk_trace_startstop(struct request_queue
*q
, int start
)
697 mutex_lock(&q
->debugfs_mutex
);
698 ret
= __blk_trace_startstop(q
, start
);
699 mutex_unlock(&q
->debugfs_mutex
);
703 EXPORT_SYMBOL_GPL(blk_trace_startstop
);
706 * When reading or writing the blktrace sysfs files, the references to the
707 * opened sysfs or device files should prevent the underlying block device
708 * from being removed. So no further delete protection is really needed.
712 * blk_trace_ioctl: - handle the ioctls associated with tracing
713 * @bdev: the block device
714 * @cmd: the ioctl cmd
715 * @arg: the argument data, if any
718 int blk_trace_ioctl(struct block_device
*bdev
, unsigned cmd
, char __user
*arg
)
720 struct request_queue
*q
;
722 char b
[BDEVNAME_SIZE
];
724 q
= bdev_get_queue(bdev
);
728 mutex_lock(&q
->debugfs_mutex
);
733 ret
= __blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
735 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
736 case BLKTRACESETUP32
:
738 ret
= compat_blk_trace_setup(q
, b
, bdev
->bd_dev
, bdev
, arg
);
745 ret
= __blk_trace_startstop(q
, start
);
747 case BLKTRACETEARDOWN
:
748 ret
= __blk_trace_remove(q
);
755 mutex_unlock(&q
->debugfs_mutex
);
760 * blk_trace_shutdown: - stop and cleanup trace structures
761 * @q: the request queue associated with the device
764 void blk_trace_shutdown(struct request_queue
*q
)
766 mutex_lock(&q
->debugfs_mutex
);
767 if (rcu_dereference_protected(q
->blk_trace
,
768 lockdep_is_held(&q
->debugfs_mutex
))) {
769 __blk_trace_startstop(q
, 0);
770 __blk_trace_remove(q
);
773 mutex_unlock(&q
->debugfs_mutex
);
776 #ifdef CONFIG_BLK_CGROUP
777 static u64
blk_trace_bio_get_cgid(struct request_queue
*q
, struct bio
*bio
)
779 struct blk_trace
*bt
;
781 /* We don't use the 'bt' value here except as an optimization... */
782 bt
= rcu_dereference_protected(q
->blk_trace
, 1);
783 if (!bt
|| !(blk_tracer_flags
.val
& TRACE_BLK_OPT_CGROUP
))
788 return cgroup_id(bio_blkcg(bio
)->css
.cgroup
);
791 static u64
blk_trace_bio_get_cgid(struct request_queue
*q
, struct bio
*bio
)
798 blk_trace_request_get_cgid(struct request_queue
*q
, struct request
*rq
)
802 /* Use the first bio */
803 return blk_trace_bio_get_cgid(q
, rq
->bio
);
811 * blk_add_trace_rq - Add a trace for a request oriented action
812 * @rq: the source request
813 * @error: return status to log
814 * @nr_bytes: number of completed bytes
816 * @cgid: the cgroup info
819 * Records an action against a request. Will log the bio offset + size.
822 static void blk_add_trace_rq(struct request
*rq
, int error
,
823 unsigned int nr_bytes
, u32 what
, u64 cgid
)
825 struct blk_trace
*bt
;
828 bt
= rcu_dereference(rq
->q
->blk_trace
);
834 if (blk_rq_is_passthrough(rq
))
835 what
|= BLK_TC_ACT(BLK_TC_PC
);
837 what
|= BLK_TC_ACT(BLK_TC_FS
);
839 __blk_add_trace(bt
, blk_rq_trace_sector(rq
), nr_bytes
, req_op(rq
),
840 rq
->cmd_flags
, what
, error
, 0, NULL
, cgid
);
844 static void blk_add_trace_rq_insert(void *ignore
,
845 struct request_queue
*q
, struct request
*rq
)
847 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_INSERT
,
848 blk_trace_request_get_cgid(q
, rq
));
851 static void blk_add_trace_rq_issue(void *ignore
,
852 struct request_queue
*q
, struct request
*rq
)
854 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_ISSUE
,
855 blk_trace_request_get_cgid(q
, rq
));
858 static void blk_add_trace_rq_merge(void *ignore
,
859 struct request_queue
*q
, struct request
*rq
)
861 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_BACKMERGE
,
862 blk_trace_request_get_cgid(q
, rq
));
865 static void blk_add_trace_rq_requeue(void *ignore
,
866 struct request_queue
*q
,
869 blk_add_trace_rq(rq
, 0, blk_rq_bytes(rq
), BLK_TA_REQUEUE
,
870 blk_trace_request_get_cgid(q
, rq
));
873 static void blk_add_trace_rq_complete(void *ignore
, struct request
*rq
,
874 int error
, unsigned int nr_bytes
)
876 blk_add_trace_rq(rq
, error
, nr_bytes
, BLK_TA_COMPLETE
,
877 blk_trace_request_get_cgid(rq
->q
, rq
));
881 * blk_add_trace_bio - Add a trace for a bio oriented action
882 * @q: queue the io is for
883 * @bio: the source bio
885 * @error: error, if any
888 * Records an action against a bio. Will log the bio offset + size.
891 static void blk_add_trace_bio(struct request_queue
*q
, struct bio
*bio
,
894 struct blk_trace
*bt
;
897 bt
= rcu_dereference(q
->blk_trace
);
903 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
904 bio_op(bio
), bio
->bi_opf
, what
, error
, 0, NULL
,
905 blk_trace_bio_get_cgid(q
, bio
));
909 static void blk_add_trace_bio_bounce(void *ignore
, struct bio
*bio
)
911 blk_add_trace_bio(bio
->bi_disk
->queue
, bio
, BLK_TA_BOUNCE
, 0);
914 static void blk_add_trace_bio_complete(void *ignore
,
915 struct request_queue
*q
, struct bio
*bio
)
917 blk_add_trace_bio(q
, bio
, BLK_TA_COMPLETE
,
918 blk_status_to_errno(bio
->bi_status
));
921 static void blk_add_trace_bio_backmerge(void *ignore
, struct bio
*bio
)
923 blk_add_trace_bio(bio
->bi_disk
->queue
, bio
, BLK_TA_BACKMERGE
, 0);
926 static void blk_add_trace_bio_frontmerge(void *ignore
, struct bio
*bio
)
928 blk_add_trace_bio(bio
->bi_disk
->queue
, bio
, BLK_TA_FRONTMERGE
, 0);
931 static void blk_add_trace_bio_queue(void *ignore
, struct bio
*bio
)
933 blk_add_trace_bio(bio
->bi_disk
->queue
, bio
, BLK_TA_QUEUE
, 0);
936 static void blk_add_trace_getrq(void *ignore
, struct bio
*bio
)
938 blk_add_trace_bio(bio
->bi_disk
->queue
, bio
, BLK_TA_GETRQ
, 0);
941 static void blk_add_trace_plug(void *ignore
, struct request_queue
*q
)
943 struct blk_trace
*bt
;
946 bt
= rcu_dereference(q
->blk_trace
);
948 __blk_add_trace(bt
, 0, 0, 0, 0, BLK_TA_PLUG
, 0, 0, NULL
, 0);
952 static void blk_add_trace_unplug(void *ignore
, struct request_queue
*q
,
953 unsigned int depth
, bool explicit)
955 struct blk_trace
*bt
;
958 bt
= rcu_dereference(q
->blk_trace
);
960 __be64 rpdu
= cpu_to_be64(depth
);
964 what
= BLK_TA_UNPLUG_IO
;
966 what
= BLK_TA_UNPLUG_TIMER
;
968 __blk_add_trace(bt
, 0, 0, 0, 0, what
, 0, sizeof(rpdu
), &rpdu
, 0);
973 static void blk_add_trace_split(void *ignore
,
974 struct request_queue
*q
, struct bio
*bio
,
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_op(bio
), bio
->bi_opf
,
987 blk_status_to_errno(bio
->bi_status
),
989 blk_trace_bio_get_cgid(q
, bio
));
995 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
996 * @ignore: trace callback data parameter (not used)
997 * @q: queue the io is for
998 * @bio: the source bio
999 * @dev: target device
1000 * @from: source sector
1003 * Device mapper or raid target sometimes need to split a bio because
1004 * it spans a stripe (or similar). Add a trace for that action.
1007 static void blk_add_trace_bio_remap(void *ignore
,
1008 struct request_queue
*q
, struct bio
*bio
,
1009 dev_t dev
, sector_t from
)
1011 struct blk_trace
*bt
;
1012 struct blk_io_trace_remap r
;
1015 bt
= rcu_dereference(q
->blk_trace
);
1021 r
.device_from
= cpu_to_be32(dev
);
1022 r
.device_to
= cpu_to_be32(bio_dev(bio
));
1023 r
.sector_from
= cpu_to_be64(from
);
1025 __blk_add_trace(bt
, bio
->bi_iter
.bi_sector
, bio
->bi_iter
.bi_size
,
1026 bio_op(bio
), bio
->bi_opf
, BLK_TA_REMAP
,
1027 blk_status_to_errno(bio
->bi_status
),
1028 sizeof(r
), &r
, blk_trace_bio_get_cgid(q
, bio
));
1033 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1034 * @ignore: trace callback data parameter (not used)
1035 * @q: queue the io is for
1036 * @rq: the source request
1037 * @dev: target device
1038 * @from: source sector
1041 * Device mapper remaps request to other devices.
1042 * Add a trace for that action.
1045 static void blk_add_trace_rq_remap(void *ignore
,
1046 struct request_queue
*q
,
1047 struct request
*rq
, dev_t dev
,
1050 struct blk_trace
*bt
;
1051 struct blk_io_trace_remap r
;
1054 bt
= rcu_dereference(q
->blk_trace
);
1060 r
.device_from
= cpu_to_be32(dev
);
1061 r
.device_to
= cpu_to_be32(disk_devt(rq
->rq_disk
));
1062 r
.sector_from
= cpu_to_be64(from
);
1064 __blk_add_trace(bt
, blk_rq_pos(rq
), blk_rq_bytes(rq
),
1065 rq_data_dir(rq
), 0, BLK_TA_REMAP
, 0,
1066 sizeof(r
), &r
, blk_trace_request_get_cgid(q
, rq
));
1071 * blk_add_driver_data - Add binary message with driver-specific data
1072 * @q: queue the io is for
1074 * @data: driver-specific data
1075 * @len: length of driver-specific data
1078 * Some drivers might want to write driver-specific data per request.
1081 void blk_add_driver_data(struct request_queue
*q
,
1083 void *data
, size_t len
)
1085 struct blk_trace
*bt
;
1088 bt
= rcu_dereference(q
->blk_trace
);
1094 __blk_add_trace(bt
, blk_rq_trace_sector(rq
), blk_rq_bytes(rq
), 0, 0,
1095 BLK_TA_DRV_DATA
, 0, len
, data
,
1096 blk_trace_request_get_cgid(q
, rq
));
1099 EXPORT_SYMBOL_GPL(blk_add_driver_data
);
1101 static void blk_register_tracepoints(void)
1105 ret
= register_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1107 ret
= register_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1109 ret
= register_trace_block_rq_merge(blk_add_trace_rq_merge
, NULL
);
1111 ret
= register_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1113 ret
= register_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1115 ret
= register_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1117 ret
= register_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1119 ret
= register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1121 ret
= register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1123 ret
= register_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1125 ret
= register_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1127 ret
= register_trace_block_plug(blk_add_trace_plug
, NULL
);
1129 ret
= register_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1131 ret
= register_trace_block_split(blk_add_trace_split
, NULL
);
1133 ret
= register_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1135 ret
= register_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1139 static void blk_unregister_tracepoints(void)
1141 unregister_trace_block_rq_remap(blk_add_trace_rq_remap
, NULL
);
1142 unregister_trace_block_bio_remap(blk_add_trace_bio_remap
, NULL
);
1143 unregister_trace_block_split(blk_add_trace_split
, NULL
);
1144 unregister_trace_block_unplug(blk_add_trace_unplug
, NULL
);
1145 unregister_trace_block_plug(blk_add_trace_plug
, NULL
);
1146 unregister_trace_block_getrq(blk_add_trace_getrq
, NULL
);
1147 unregister_trace_block_bio_queue(blk_add_trace_bio_queue
, NULL
);
1148 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge
, NULL
);
1149 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge
, NULL
);
1150 unregister_trace_block_bio_complete(blk_add_trace_bio_complete
, NULL
);
1151 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce
, NULL
);
1152 unregister_trace_block_rq_complete(blk_add_trace_rq_complete
, NULL
);
1153 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue
, NULL
);
1154 unregister_trace_block_rq_merge(blk_add_trace_rq_merge
, NULL
);
1155 unregister_trace_block_rq_issue(blk_add_trace_rq_issue
, NULL
);
1156 unregister_trace_block_rq_insert(blk_add_trace_rq_insert
, NULL
);
1158 tracepoint_synchronize_unregister();
1162 * struct blk_io_tracer formatting routines
1165 static void fill_rwbs(char *rwbs
, const struct blk_io_trace
*t
)
1168 int tc
= t
->action
>> BLK_TC_SHIFT
;
1170 if ((t
->action
& ~__BLK_TN_CGROUP
) == BLK_TN_MESSAGE
) {
1175 if (tc
& BLK_TC_FLUSH
)
1178 if (tc
& BLK_TC_DISCARD
)
1180 else if (tc
& BLK_TC_WRITE
)
1187 if (tc
& BLK_TC_FUA
)
1189 if (tc
& BLK_TC_AHEAD
)
1191 if (tc
& BLK_TC_SYNC
)
1193 if (tc
& BLK_TC_META
)
1200 const struct blk_io_trace
*te_blk_io_trace(const struct trace_entry
*ent
)
1202 return (const struct blk_io_trace
*)ent
;
1205 static inline const void *pdu_start(const struct trace_entry
*ent
, bool has_cg
)
1207 return (void *)(te_blk_io_trace(ent
) + 1) + (has_cg
? sizeof(u64
) : 0);
1210 static inline u64
t_cgid(const struct trace_entry
*ent
)
1212 return *(u64
*)(te_blk_io_trace(ent
) + 1);
1215 static inline int pdu_real_len(const struct trace_entry
*ent
, bool has_cg
)
1217 return te_blk_io_trace(ent
)->pdu_len
- (has_cg
? sizeof(u64
) : 0);
1220 static inline u32
t_action(const struct trace_entry
*ent
)
1222 return te_blk_io_trace(ent
)->action
;
1225 static inline u32
t_bytes(const struct trace_entry
*ent
)
1227 return te_blk_io_trace(ent
)->bytes
;
1230 static inline u32
t_sec(const struct trace_entry
*ent
)
1232 return te_blk_io_trace(ent
)->bytes
>> 9;
1235 static inline unsigned long long t_sector(const struct trace_entry
*ent
)
1237 return te_blk_io_trace(ent
)->sector
;
1240 static inline __u16
t_error(const struct trace_entry
*ent
)
1242 return te_blk_io_trace(ent
)->error
;
1245 static __u64
get_pdu_int(const struct trace_entry
*ent
, bool has_cg
)
1247 const __be64
*val
= pdu_start(ent
, has_cg
);
1248 return be64_to_cpu(*val
);
1251 typedef void (blk_log_action_t
) (struct trace_iterator
*iter
, const char *act
,
1254 static void blk_log_action_classic(struct trace_iterator
*iter
, const char *act
,
1257 char rwbs
[RWBS_LEN
];
1258 unsigned long long ts
= iter
->ts
;
1259 unsigned long nsec_rem
= do_div(ts
, NSEC_PER_SEC
);
1260 unsigned secs
= (unsigned long)ts
;
1261 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1265 trace_seq_printf(&iter
->seq
,
1266 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1267 MAJOR(t
->device
), MINOR(t
->device
), iter
->cpu
,
1268 secs
, nsec_rem
, iter
->ent
->pid
, act
, rwbs
);
1271 static void blk_log_action(struct trace_iterator
*iter
, const char *act
,
1274 char rwbs
[RWBS_LEN
];
1275 const struct blk_io_trace
*t
= te_blk_io_trace(iter
->ent
);
1279 u64 id
= t_cgid(iter
->ent
);
1281 if (blk_tracer_flags
.val
& TRACE_BLK_OPT_CGNAME
) {
1282 char blkcg_name_buf
[NAME_MAX
+ 1] = "<...>";
1284 cgroup_path_from_kernfs_id(id
, blkcg_name_buf
,
1285 sizeof(blkcg_name_buf
));
1286 trace_seq_printf(&iter
->seq
, "%3d,%-3d %s %2s %3s ",
1287 MAJOR(t
->device
), MINOR(t
->device
),
1288 blkcg_name_buf
, act
, rwbs
);
1291 * The cgid portion used to be "INO,GEN". Userland
1292 * builds a FILEID_INO32_GEN fid out of them and
1293 * opens the cgroup using open_by_handle_at(2).
1294 * While 32bit ino setups are still the same, 64bit
1295 * ones now use the 64bit ino as the whole ID and
1296 * no longer use generation.
1298 * Regarldess of the content, always output
1299 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1300 * be mapped back to @id on both 64 and 32bit ino
1301 * setups. See __kernfs_fh_to_dentry().
1303 trace_seq_printf(&iter
->seq
,
1304 "%3d,%-3d %llx,%-llx %2s %3s ",
1305 MAJOR(t
->device
), MINOR(t
->device
),
1306 id
& U32_MAX
, id
>> 32, act
, rwbs
);
1309 trace_seq_printf(&iter
->seq
, "%3d,%-3d %2s %3s ",
1310 MAJOR(t
->device
), MINOR(t
->device
), act
, rwbs
);
1313 static void blk_log_dump_pdu(struct trace_seq
*s
,
1314 const struct trace_entry
*ent
, bool has_cg
)
1316 const unsigned char *pdu_buf
;
1320 pdu_buf
= pdu_start(ent
, has_cg
);
1321 pdu_len
= pdu_real_len(ent
, has_cg
);
1326 /* find the last zero that needs to be printed */
1327 for (end
= pdu_len
- 1; end
>= 0; end
--)
1332 trace_seq_putc(s
, '(');
1334 for (i
= 0; i
< pdu_len
; i
++) {
1336 trace_seq_printf(s
, "%s%02x",
1337 i
== 0 ? "" : " ", pdu_buf
[i
]);
1340 * stop when the rest is just zeroes and indicate so
1341 * with a ".." appended
1343 if (i
== end
&& end
!= pdu_len
- 1) {
1344 trace_seq_puts(s
, " ..) ");
1349 trace_seq_puts(s
, ") ");
1352 static void blk_log_generic(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1354 char cmd
[TASK_COMM_LEN
];
1356 trace_find_cmdline(ent
->pid
, cmd
);
1358 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1359 trace_seq_printf(s
, "%u ", t_bytes(ent
));
1360 blk_log_dump_pdu(s
, ent
, has_cg
);
1361 trace_seq_printf(s
, "[%s]\n", cmd
);
1364 trace_seq_printf(s
, "%llu + %u [%s]\n",
1365 t_sector(ent
), t_sec(ent
), cmd
);
1367 trace_seq_printf(s
, "[%s]\n", cmd
);
1371 static void blk_log_with_error(struct trace_seq
*s
,
1372 const struct trace_entry
*ent
, bool has_cg
)
1374 if (t_action(ent
) & BLK_TC_ACT(BLK_TC_PC
)) {
1375 blk_log_dump_pdu(s
, ent
, has_cg
);
1376 trace_seq_printf(s
, "[%d]\n", t_error(ent
));
1379 trace_seq_printf(s
, "%llu + %u [%d]\n",
1381 t_sec(ent
), t_error(ent
));
1383 trace_seq_printf(s
, "%llu [%d]\n",
1384 t_sector(ent
), t_error(ent
));
1388 static void blk_log_remap(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1390 const struct blk_io_trace_remap
*__r
= pdu_start(ent
, has_cg
);
1392 trace_seq_printf(s
, "%llu + %u <- (%d,%d) %llu\n",
1393 t_sector(ent
), t_sec(ent
),
1394 MAJOR(be32_to_cpu(__r
->device_from
)),
1395 MINOR(be32_to_cpu(__r
->device_from
)),
1396 be64_to_cpu(__r
->sector_from
));
1399 static void blk_log_plug(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1401 char cmd
[TASK_COMM_LEN
];
1403 trace_find_cmdline(ent
->pid
, cmd
);
1405 trace_seq_printf(s
, "[%s]\n", cmd
);
1408 static void blk_log_unplug(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1410 char cmd
[TASK_COMM_LEN
];
1412 trace_find_cmdline(ent
->pid
, cmd
);
1414 trace_seq_printf(s
, "[%s] %llu\n", cmd
, get_pdu_int(ent
, has_cg
));
1417 static void blk_log_split(struct trace_seq
*s
, const struct trace_entry
*ent
, bool has_cg
)
1419 char cmd
[TASK_COMM_LEN
];
1421 trace_find_cmdline(ent
->pid
, cmd
);
1423 trace_seq_printf(s
, "%llu / %llu [%s]\n", t_sector(ent
),
1424 get_pdu_int(ent
, has_cg
), cmd
);
1427 static void blk_log_msg(struct trace_seq
*s
, const struct trace_entry
*ent
,
1431 trace_seq_putmem(s
, pdu_start(ent
, has_cg
),
1432 pdu_real_len(ent
, has_cg
));
1433 trace_seq_putc(s
, '\n');
1437 * struct tracer operations
1440 static void blk_tracer_print_header(struct seq_file
*m
)
1442 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1444 seq_puts(m
, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1448 static void blk_tracer_start(struct trace_array
*tr
)
1450 blk_tracer_enabled
= true;
1453 static int blk_tracer_init(struct trace_array
*tr
)
1456 blk_tracer_start(tr
);
1460 static void blk_tracer_stop(struct trace_array
*tr
)
1462 blk_tracer_enabled
= false;
1465 static void blk_tracer_reset(struct trace_array
*tr
)
1467 blk_tracer_stop(tr
);
1470 static const struct {
1472 void (*print
)(struct trace_seq
*s
, const struct trace_entry
*ent
,
1475 [__BLK_TA_QUEUE
] = {{ "Q", "queue" }, blk_log_generic
},
1476 [__BLK_TA_BACKMERGE
] = {{ "M", "backmerge" }, blk_log_generic
},
1477 [__BLK_TA_FRONTMERGE
] = {{ "F", "frontmerge" }, blk_log_generic
},
1478 [__BLK_TA_GETRQ
] = {{ "G", "getrq" }, blk_log_generic
},
1479 [__BLK_TA_SLEEPRQ
] = {{ "S", "sleeprq" }, blk_log_generic
},
1480 [__BLK_TA_REQUEUE
] = {{ "R", "requeue" }, blk_log_with_error
},
1481 [__BLK_TA_ISSUE
] = {{ "D", "issue" }, blk_log_generic
},
1482 [__BLK_TA_COMPLETE
] = {{ "C", "complete" }, blk_log_with_error
},
1483 [__BLK_TA_PLUG
] = {{ "P", "plug" }, blk_log_plug
},
1484 [__BLK_TA_UNPLUG_IO
] = {{ "U", "unplug_io" }, blk_log_unplug
},
1485 [__BLK_TA_UNPLUG_TIMER
] = {{ "UT", "unplug_timer" }, blk_log_unplug
},
1486 [__BLK_TA_INSERT
] = {{ "I", "insert" }, blk_log_generic
},
1487 [__BLK_TA_SPLIT
] = {{ "X", "split" }, blk_log_split
},
1488 [__BLK_TA_BOUNCE
] = {{ "B", "bounce" }, blk_log_generic
},
1489 [__BLK_TA_REMAP
] = {{ "A", "remap" }, blk_log_remap
},
1492 static enum print_line_t
print_one_line(struct trace_iterator
*iter
,
1495 struct trace_array
*tr
= iter
->tr
;
1496 struct trace_seq
*s
= &iter
->seq
;
1497 const struct blk_io_trace
*t
;
1500 blk_log_action_t
*log_action
;
1503 t
= te_blk_io_trace(iter
->ent
);
1504 what
= (t
->action
& ((1 << BLK_TC_SHIFT
) - 1)) & ~__BLK_TA_CGROUP
;
1505 long_act
= !!(tr
->trace_flags
& TRACE_ITER_VERBOSE
);
1506 log_action
= classic
? &blk_log_action_classic
: &blk_log_action
;
1507 has_cg
= t
->action
& __BLK_TA_CGROUP
;
1509 if ((t
->action
& ~__BLK_TN_CGROUP
) == BLK_TN_MESSAGE
) {
1510 log_action(iter
, long_act
? "message" : "m", has_cg
);
1511 blk_log_msg(s
, iter
->ent
, has_cg
);
1512 return trace_handle_return(s
);
1515 if (unlikely(what
== 0 || what
>= ARRAY_SIZE(what2act
)))
1516 trace_seq_printf(s
, "Unknown action %x\n", what
);
1518 log_action(iter
, what2act
[what
].act
[long_act
], has_cg
);
1519 what2act
[what
].print(s
, iter
->ent
, has_cg
);
1522 return trace_handle_return(s
);
1525 static enum print_line_t
blk_trace_event_print(struct trace_iterator
*iter
,
1526 int flags
, struct trace_event
*event
)
1528 return print_one_line(iter
, false);
1531 static void blk_trace_synthesize_old_trace(struct trace_iterator
*iter
)
1533 struct trace_seq
*s
= &iter
->seq
;
1534 struct blk_io_trace
*t
= (struct blk_io_trace
*)iter
->ent
;
1535 const int offset
= offsetof(struct blk_io_trace
, sector
);
1536 struct blk_io_trace old
= {
1537 .magic
= BLK_IO_TRACE_MAGIC
| BLK_IO_TRACE_VERSION
,
1541 trace_seq_putmem(s
, &old
, offset
);
1542 trace_seq_putmem(s
, &t
->sector
,
1543 sizeof(old
) - offset
+ t
->pdu_len
);
1546 static enum print_line_t
1547 blk_trace_event_print_binary(struct trace_iterator
*iter
, int flags
,
1548 struct trace_event
*event
)
1550 blk_trace_synthesize_old_trace(iter
);
1552 return trace_handle_return(&iter
->seq
);
1555 static enum print_line_t
blk_tracer_print_line(struct trace_iterator
*iter
)
1557 if (!(blk_tracer_flags
.val
& TRACE_BLK_OPT_CLASSIC
))
1558 return TRACE_TYPE_UNHANDLED
;
1560 return print_one_line(iter
, true);
1564 blk_tracer_set_flag(struct trace_array
*tr
, u32 old_flags
, u32 bit
, int set
)
1566 /* don't output context-info for blk_classic output */
1567 if (bit
== TRACE_BLK_OPT_CLASSIC
) {
1569 tr
->trace_flags
&= ~TRACE_ITER_CONTEXT_INFO
;
1571 tr
->trace_flags
|= TRACE_ITER_CONTEXT_INFO
;
1576 static struct tracer blk_tracer __read_mostly
= {
1578 .init
= blk_tracer_init
,
1579 .reset
= blk_tracer_reset
,
1580 .start
= blk_tracer_start
,
1581 .stop
= blk_tracer_stop
,
1582 .print_header
= blk_tracer_print_header
,
1583 .print_line
= blk_tracer_print_line
,
1584 .flags
= &blk_tracer_flags
,
1585 .set_flag
= blk_tracer_set_flag
,
1588 static struct trace_event_functions trace_blk_event_funcs
= {
1589 .trace
= blk_trace_event_print
,
1590 .binary
= blk_trace_event_print_binary
,
1593 static struct trace_event trace_blk_event
= {
1595 .funcs
= &trace_blk_event_funcs
,
1598 static int __init
init_blk_tracer(void)
1600 if (!register_trace_event(&trace_blk_event
)) {
1601 pr_warn("Warning: could not register block events\n");
1605 if (register_tracer(&blk_tracer
) != 0) {
1606 pr_warn("Warning: could not register the block tracer\n");
1607 unregister_trace_event(&trace_blk_event
);
1614 device_initcall(init_blk_tracer
);
1616 static int blk_trace_remove_queue(struct request_queue
*q
)
1618 struct blk_trace
*bt
;
1620 bt
= rcu_replace_pointer(q
->blk_trace
, NULL
,
1621 lockdep_is_held(&q
->debugfs_mutex
));
1632 * Setup everything required to start tracing
1634 static int blk_trace_setup_queue(struct request_queue
*q
,
1635 struct block_device
*bdev
)
1637 struct blk_trace
*bt
= NULL
;
1640 bt
= kzalloc(sizeof(*bt
), GFP_KERNEL
);
1644 bt
->msg_data
= __alloc_percpu(BLK_TN_MAX_MSG
, __alignof__(char));
1648 bt
->dev
= bdev
->bd_dev
;
1649 bt
->act_mask
= (u16
)-1;
1651 blk_trace_setup_lba(bt
, bdev
);
1653 rcu_assign_pointer(q
->blk_trace
, bt
);
1663 * sysfs interface to enable and configure tracing
1666 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1667 struct device_attribute
*attr
,
1669 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1670 struct device_attribute
*attr
,
1671 const char *buf
, size_t count
);
1672 #define BLK_TRACE_DEVICE_ATTR(_name) \
1673 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1674 sysfs_blk_trace_attr_show, \
1675 sysfs_blk_trace_attr_store)
1677 static BLK_TRACE_DEVICE_ATTR(enable
);
1678 static BLK_TRACE_DEVICE_ATTR(act_mask
);
1679 static BLK_TRACE_DEVICE_ATTR(pid
);
1680 static BLK_TRACE_DEVICE_ATTR(start_lba
);
1681 static BLK_TRACE_DEVICE_ATTR(end_lba
);
1683 static struct attribute
*blk_trace_attrs
[] = {
1684 &dev_attr_enable
.attr
,
1685 &dev_attr_act_mask
.attr
,
1687 &dev_attr_start_lba
.attr
,
1688 &dev_attr_end_lba
.attr
,
1692 struct attribute_group blk_trace_attr_group
= {
1694 .attrs
= blk_trace_attrs
,
1697 static const struct {
1701 { BLK_TC_READ
, "read" },
1702 { BLK_TC_WRITE
, "write" },
1703 { BLK_TC_FLUSH
, "flush" },
1704 { BLK_TC_SYNC
, "sync" },
1705 { BLK_TC_QUEUE
, "queue" },
1706 { BLK_TC_REQUEUE
, "requeue" },
1707 { BLK_TC_ISSUE
, "issue" },
1708 { BLK_TC_COMPLETE
, "complete" },
1709 { BLK_TC_FS
, "fs" },
1710 { BLK_TC_PC
, "pc" },
1711 { BLK_TC_NOTIFY
, "notify" },
1712 { BLK_TC_AHEAD
, "ahead" },
1713 { BLK_TC_META
, "meta" },
1714 { BLK_TC_DISCARD
, "discard" },
1715 { BLK_TC_DRV_DATA
, "drv_data" },
1716 { BLK_TC_FUA
, "fua" },
1719 static int blk_trace_str2mask(const char *str
)
1723 char *buf
, *s
, *token
;
1725 buf
= kstrdup(str
, GFP_KERNEL
);
1731 token
= strsep(&s
, ",");
1738 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1739 if (strcasecmp(token
, mask_maps
[i
].str
) == 0) {
1740 mask
|= mask_maps
[i
].mask
;
1744 if (i
== ARRAY_SIZE(mask_maps
)) {
1754 static ssize_t
blk_trace_mask2str(char *buf
, int mask
)
1759 for (i
= 0; i
< ARRAY_SIZE(mask_maps
); i
++) {
1760 if (mask
& mask_maps
[i
].mask
) {
1761 p
+= sprintf(p
, "%s%s",
1762 (p
== buf
) ? "" : ",", mask_maps
[i
].str
);
1770 static ssize_t
sysfs_blk_trace_attr_show(struct device
*dev
,
1771 struct device_attribute
*attr
,
1774 struct block_device
*bdev
= dev_to_bdev(dev
);
1775 struct request_queue
*q
= bdev_get_queue(bdev
);
1776 struct blk_trace
*bt
;
1777 ssize_t ret
= -ENXIO
;
1779 mutex_lock(&q
->debugfs_mutex
);
1781 bt
= rcu_dereference_protected(q
->blk_trace
,
1782 lockdep_is_held(&q
->debugfs_mutex
));
1783 if (attr
== &dev_attr_enable
) {
1784 ret
= sprintf(buf
, "%u\n", !!bt
);
1785 goto out_unlock_bdev
;
1789 ret
= sprintf(buf
, "disabled\n");
1790 else if (attr
== &dev_attr_act_mask
)
1791 ret
= blk_trace_mask2str(buf
, bt
->act_mask
);
1792 else if (attr
== &dev_attr_pid
)
1793 ret
= sprintf(buf
, "%u\n", bt
->pid
);
1794 else if (attr
== &dev_attr_start_lba
)
1795 ret
= sprintf(buf
, "%llu\n", bt
->start_lba
);
1796 else if (attr
== &dev_attr_end_lba
)
1797 ret
= sprintf(buf
, "%llu\n", bt
->end_lba
);
1800 mutex_unlock(&q
->debugfs_mutex
);
1804 static ssize_t
sysfs_blk_trace_attr_store(struct device
*dev
,
1805 struct device_attribute
*attr
,
1806 const char *buf
, size_t count
)
1808 struct block_device
*bdev
= dev_to_bdev(dev
);
1809 struct request_queue
*q
= bdev_get_queue(bdev
);
1810 struct blk_trace
*bt
;
1812 ssize_t ret
= -EINVAL
;
1817 if (attr
== &dev_attr_act_mask
) {
1818 if (kstrtoull(buf
, 0, &value
)) {
1819 /* Assume it is a list of trace category names */
1820 ret
= blk_trace_str2mask(buf
);
1826 if (kstrtoull(buf
, 0, &value
))
1830 mutex_lock(&q
->debugfs_mutex
);
1832 bt
= rcu_dereference_protected(q
->blk_trace
,
1833 lockdep_is_held(&q
->debugfs_mutex
));
1834 if (attr
== &dev_attr_enable
) {
1835 if (!!value
== !!bt
) {
1837 goto out_unlock_bdev
;
1840 ret
= blk_trace_setup_queue(q
, bdev
);
1842 ret
= blk_trace_remove_queue(q
);
1843 goto out_unlock_bdev
;
1848 ret
= blk_trace_setup_queue(q
, bdev
);
1849 bt
= rcu_dereference_protected(q
->blk_trace
,
1850 lockdep_is_held(&q
->debugfs_mutex
));
1854 if (attr
== &dev_attr_act_mask
)
1855 bt
->act_mask
= value
;
1856 else if (attr
== &dev_attr_pid
)
1858 else if (attr
== &dev_attr_start_lba
)
1859 bt
->start_lba
= value
;
1860 else if (attr
== &dev_attr_end_lba
)
1861 bt
->end_lba
= value
;
1865 mutex_unlock(&q
->debugfs_mutex
);
1867 return ret
? ret
: count
;
1870 int blk_trace_init_sysfs(struct device
*dev
)
1872 return sysfs_create_group(&dev
->kobj
, &blk_trace_attr_group
);
1875 void blk_trace_remove_sysfs(struct device
*dev
)
1877 sysfs_remove_group(&dev
->kobj
, &blk_trace_attr_group
);
1880 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1882 #ifdef CONFIG_EVENT_TRACING
1884 void blk_fill_rwbs(char *rwbs
, unsigned int op
, int bytes
)
1888 if (op
& REQ_PREFLUSH
)
1891 switch (op
& REQ_OP_MASK
) {
1893 case REQ_OP_WRITE_SAME
:
1896 case REQ_OP_DISCARD
:
1899 case REQ_OP_SECURE_ERASE
:
1915 if (op
& REQ_RAHEAD
)
1924 EXPORT_SYMBOL_GPL(blk_fill_rwbs
);
1926 #endif /* CONFIG_EVENT_TRACING */