xref: /linux-6.15/include/linux/blktrace_api.h (revision 384740dc)
1 #ifndef BLKTRACE_H
2 #define BLKTRACE_H
3 
4 #ifdef __KERNEL__
5 #include <linux/blkdev.h>
6 #include <linux/relay.h>
7 #endif
8 
9 /*
10  * Trace categories
11  */
12 enum blktrace_cat {
13 	BLK_TC_READ	= 1 << 0,	/* reads */
14 	BLK_TC_WRITE	= 1 << 1,	/* writes */
15 	BLK_TC_BARRIER	= 1 << 2,	/* barrier */
16 	BLK_TC_SYNC	= 1 << 3,	/* sync IO */
17 	BLK_TC_QUEUE	= 1 << 4,	/* queueing/merging */
18 	BLK_TC_REQUEUE	= 1 << 5,	/* requeueing */
19 	BLK_TC_ISSUE	= 1 << 6,	/* issue */
20 	BLK_TC_COMPLETE	= 1 << 7,	/* completions */
21 	BLK_TC_FS	= 1 << 8,	/* fs requests */
22 	BLK_TC_PC	= 1 << 9,	/* pc requests */
23 	BLK_TC_NOTIFY	= 1 << 10,	/* special message */
24 	BLK_TC_AHEAD	= 1 << 11,	/* readahead */
25 	BLK_TC_META	= 1 << 12,	/* metadata */
26 	BLK_TC_DISCARD	= 1 << 13,	/* discard requests */
27 
28 	BLK_TC_END	= 1 << 15,	/* only 16-bits, reminder */
29 };
30 
31 #define BLK_TC_SHIFT		(16)
32 #define BLK_TC_ACT(act)		((act) << BLK_TC_SHIFT)
33 
34 /*
35  * Basic trace actions
36  */
37 enum blktrace_act {
38 	__BLK_TA_QUEUE = 1,		/* queued */
39 	__BLK_TA_BACKMERGE,		/* back merged to existing rq */
40 	__BLK_TA_FRONTMERGE,		/* front merge to existing rq */
41 	__BLK_TA_GETRQ,			/* allocated new request */
42 	__BLK_TA_SLEEPRQ,		/* sleeping on rq allocation */
43 	__BLK_TA_REQUEUE,		/* request requeued */
44 	__BLK_TA_ISSUE,			/* sent to driver */
45 	__BLK_TA_COMPLETE,		/* completed by driver */
46 	__BLK_TA_PLUG,			/* queue was plugged */
47 	__BLK_TA_UNPLUG_IO,		/* queue was unplugged by io */
48 	__BLK_TA_UNPLUG_TIMER,		/* queue was unplugged by timer */
49 	__BLK_TA_INSERT,		/* insert request */
50 	__BLK_TA_SPLIT,			/* bio was split */
51 	__BLK_TA_BOUNCE,		/* bio was bounced */
52 	__BLK_TA_REMAP,			/* bio was remapped */
53 	__BLK_TA_ABORT,			/* request aborted */
54 };
55 
56 /*
57  * Notify events.
58  */
59 enum blktrace_notify {
60 	__BLK_TN_PROCESS = 0,		/* establish pid/name mapping */
61 	__BLK_TN_TIMESTAMP,		/* include system clock */
62 	__BLK_TN_MESSAGE,		/* Character string message */
63 };
64 
65 
66 /*
67  * Trace actions in full. Additionally, read or write is masked
68  */
69 #define BLK_TA_QUEUE		(__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
70 #define BLK_TA_BACKMERGE	(__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
71 #define BLK_TA_FRONTMERGE	(__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
72 #define	BLK_TA_GETRQ		(__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
73 #define	BLK_TA_SLEEPRQ		(__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
74 #define	BLK_TA_REQUEUE		(__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
75 #define BLK_TA_ISSUE		(__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
76 #define BLK_TA_COMPLETE		(__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
77 #define BLK_TA_PLUG		(__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
78 #define BLK_TA_UNPLUG_IO	(__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
79 #define BLK_TA_UNPLUG_TIMER	(__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
80 #define BLK_TA_INSERT		(__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
81 #define BLK_TA_SPLIT		(__BLK_TA_SPLIT)
82 #define BLK_TA_BOUNCE		(__BLK_TA_BOUNCE)
83 #define BLK_TA_REMAP		(__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
84 #define BLK_TA_ABORT		(__BLK_TA_ABORT | BLK_TC_ACT(BLK_TC_QUEUE))
85 
86 #define BLK_TN_PROCESS		(__BLK_TN_PROCESS | BLK_TC_ACT(BLK_TC_NOTIFY))
87 #define BLK_TN_TIMESTAMP	(__BLK_TN_TIMESTAMP | BLK_TC_ACT(BLK_TC_NOTIFY))
88 #define BLK_TN_MESSAGE		(__BLK_TN_MESSAGE | BLK_TC_ACT(BLK_TC_NOTIFY))
89 
90 #define BLK_IO_TRACE_MAGIC	0x65617400
91 #define BLK_IO_TRACE_VERSION	0x07
92 
93 /*
94  * The trace itself
95  */
96 struct blk_io_trace {
97 	__u32 magic;		/* MAGIC << 8 | version */
98 	__u32 sequence;		/* event number */
99 	__u64 time;		/* in microseconds */
100 	__u64 sector;		/* disk offset */
101 	__u32 bytes;		/* transfer length */
102 	__u32 action;		/* what happened */
103 	__u32 pid;		/* who did it */
104 	__u32 device;		/* device number */
105 	__u32 cpu;		/* on what cpu did it happen */
106 	__u16 error;		/* completion error */
107 	__u16 pdu_len;		/* length of data after this trace */
108 };
109 
110 /*
111  * The remap event
112  */
113 struct blk_io_trace_remap {
114 	__be32 device;
115 	__be32 device_from;
116 	__be64 sector;
117 };
118 
119 enum {
120 	Blktrace_setup = 1,
121 	Blktrace_running,
122 	Blktrace_stopped,
123 };
124 
125 #define BLKTRACE_BDEV_SIZE	32
126 
127 /*
128  * User setup structure passed with BLKTRACESTART
129  */
130 struct blk_user_trace_setup {
131 	char name[BLKTRACE_BDEV_SIZE];	/* output */
132 	__u16 act_mask;			/* input */
133 	__u32 buf_size;			/* input */
134 	__u32 buf_nr;			/* input */
135 	__u64 start_lba;
136 	__u64 end_lba;
137 	__u32 pid;
138 };
139 
140 #ifdef __KERNEL__
141 #if defined(CONFIG_BLK_DEV_IO_TRACE)
142 struct blk_trace {
143 	int trace_state;
144 	struct rchan *rchan;
145 	unsigned long *sequence;
146 	unsigned char *msg_data;
147 	u16 act_mask;
148 	u64 start_lba;
149 	u64 end_lba;
150 	u32 pid;
151 	u32 dev;
152 	struct dentry *dir;
153 	struct dentry *dropped_file;
154 	struct dentry *msg_file;
155 	atomic_t dropped;
156 };
157 
158 extern int blk_trace_ioctl(struct block_device *, unsigned, char __user *);
159 extern void blk_trace_shutdown(struct request_queue *);
160 extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
161 extern int do_blk_trace_setup(struct request_queue *q,
162 	char *name, dev_t dev, struct blk_user_trace_setup *buts);
163 extern void __trace_note_message(struct blk_trace *, const char *fmt, ...);
164 
165 /**
166  * blk_add_trace_msg - Add a (simple) message to the blktrace stream
167  * @q:		queue the io is for
168  * @fmt:	format to print message in
169  * args...	Variable argument list for format
170  *
171  * Description:
172  *     Records a (simple) message onto the blktrace stream.
173  *
174  *     NOTE: BLK_TN_MAX_MSG characters are output at most.
175  *     NOTE: Can not use 'static inline' due to presence of var args...
176  *
177  **/
178 #define blk_add_trace_msg(q, fmt, ...)					\
179 	do {								\
180 		struct blk_trace *bt = (q)->blk_trace;			\
181 		if (unlikely(bt))					\
182 			__trace_note_message(bt, fmt, ##__VA_ARGS__);	\
183 	} while (0)
184 #define BLK_TN_MAX_MSG		128
185 
186 /**
187  * blk_add_trace_rq - Add a trace for a request oriented action
188  * @q:		queue the io is for
189  * @rq:		the source request
190  * @what:	the action
191  *
192  * Description:
193  *     Records an action against a request. Will log the bio offset + size.
194  *
195  **/
196 static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
197 				    u32 what)
198 {
199 	struct blk_trace *bt = q->blk_trace;
200 	int rw = rq->cmd_flags & 0x03;
201 
202 	if (likely(!bt))
203 		return;
204 
205 	if (blk_discard_rq(rq))
206 		rw |= (1 << BIO_RW_DISCARD);
207 
208 	if (blk_pc_request(rq)) {
209 		what |= BLK_TC_ACT(BLK_TC_PC);
210 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
211 	} else  {
212 		what |= BLK_TC_ACT(BLK_TC_FS);
213 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
214 	}
215 }
216 
217 /**
218  * blk_add_trace_bio - Add a trace for a bio oriented action
219  * @q:		queue the io is for
220  * @bio:	the source bio
221  * @what:	the action
222  *
223  * Description:
224  *     Records an action against a bio. Will log the bio offset + size.
225  *
226  **/
227 static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
228 				     u32 what)
229 {
230 	struct blk_trace *bt = q->blk_trace;
231 
232 	if (likely(!bt))
233 		return;
234 
235 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
236 }
237 
238 /**
239  * blk_add_trace_generic - Add a trace for a generic action
240  * @q:		queue the io is for
241  * @bio:	the source bio
242  * @rw:		the data direction
243  * @what:	the action
244  *
245  * Description:
246  *     Records a simple trace
247  *
248  **/
249 static inline void blk_add_trace_generic(struct request_queue *q,
250 					 struct bio *bio, int rw, u32 what)
251 {
252 	struct blk_trace *bt = q->blk_trace;
253 
254 	if (likely(!bt))
255 		return;
256 
257 	if (bio)
258 		blk_add_trace_bio(q, bio, what);
259 	else
260 		__blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
261 }
262 
263 /**
264  * blk_add_trace_pdu_int - Add a trace for a bio with an integer payload
265  * @q:		queue the io is for
266  * @what:	the action
267  * @bio:	the source bio
268  * @pdu:	the integer payload
269  *
270  * Description:
271  *     Adds a trace with some integer payload. This might be an unplug
272  *     option given as the action, with the depth at unplug time given
273  *     as the payload
274  *
275  **/
276 static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
277 					 struct bio *bio, unsigned int pdu)
278 {
279 	struct blk_trace *bt = q->blk_trace;
280 	__be64 rpdu = cpu_to_be64(pdu);
281 
282 	if (likely(!bt))
283 		return;
284 
285 	if (bio)
286 		__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
287 	else
288 		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
289 }
290 
291 /**
292  * blk_add_trace_remap - Add a trace for a remap operation
293  * @q:		queue the io is for
294  * @bio:	the source bio
295  * @dev:	target device
296  * @from:	source sector
297  * @to:		target sector
298  *
299  * Description:
300  *     Device mapper or raid target sometimes need to split a bio because
301  *     it spans a stripe (or similar). Add a trace for that action.
302  *
303  **/
304 static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
305 				       dev_t dev, sector_t from, sector_t to)
306 {
307 	struct blk_trace *bt = q->blk_trace;
308 	struct blk_io_trace_remap r;
309 
310 	if (likely(!bt))
311 		return;
312 
313 	r.device = cpu_to_be32(dev);
314 	r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
315 	r.sector = cpu_to_be64(to);
316 
317 	__blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
318 }
319 
320 extern int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
321 			   char __user *arg);
322 extern int blk_trace_startstop(struct request_queue *q, int start);
323 extern int blk_trace_remove(struct request_queue *q);
324 
325 #else /* !CONFIG_BLK_DEV_IO_TRACE */
326 #define blk_trace_ioctl(bdev, cmd, arg)		(-ENOTTY)
327 #define blk_trace_shutdown(q)			do { } while (0)
328 #define blk_add_trace_rq(q, rq, what)		do { } while (0)
329 #define blk_add_trace_bio(q, rq, what)		do { } while (0)
330 #define blk_add_trace_generic(q, rq, rw, what)	do { } while (0)
331 #define blk_add_trace_pdu_int(q, what, bio, pdu)	do { } while (0)
332 #define blk_add_trace_remap(q, bio, dev, f, t)	do {} while (0)
333 #define do_blk_trace_setup(q, name, dev, buts)	(-ENOTTY)
334 #define blk_trace_setup(q, name, dev, arg)	(-ENOTTY)
335 #define blk_trace_startstop(q, start)		(-ENOTTY)
336 #define blk_trace_remove(q)			(-ENOTTY)
337 #define blk_add_trace_msg(q, fmt, ...)		do { } while (0)
338 
339 #endif /* CONFIG_BLK_DEV_IO_TRACE */
340 #endif /* __KERNEL__ */
341 #endif
342