xref: /linux-6.15/include/linux/blktrace_api.h (revision 2d6ffcca)
1 #ifndef BLKTRACE_H
2 #define BLKTRACE_H
3 
4 #include <linux/blkdev.h>
5 #include <linux/relay.h>
6 
7 /*
8  * Trace categories
9  */
10 enum blktrace_cat {
11 	BLK_TC_READ	= 1 << 0,	/* reads */
12 	BLK_TC_WRITE	= 1 << 1,	/* writes */
13 	BLK_TC_BARRIER	= 1 << 2,	/* barrier */
14 	BLK_TC_SYNC	= 1 << 3,	/* sync IO */
15 	BLK_TC_QUEUE	= 1 << 4,	/* queueing/merging */
16 	BLK_TC_REQUEUE	= 1 << 5,	/* requeueing */
17 	BLK_TC_ISSUE	= 1 << 6,	/* issue */
18 	BLK_TC_COMPLETE	= 1 << 7,	/* completions */
19 	BLK_TC_FS	= 1 << 8,	/* fs requests */
20 	BLK_TC_PC	= 1 << 9,	/* pc requests */
21 	BLK_TC_NOTIFY	= 1 << 10,	/* special message */
22 	BLK_TC_AHEAD	= 1 << 11,	/* readahead */
23 	BLK_TC_META	= 1 << 12,	/* metadata */
24 
25 	BLK_TC_END	= 1 << 15,	/* only 16-bits, reminder */
26 };
27 
28 #define BLK_TC_SHIFT		(16)
29 #define BLK_TC_ACT(act)		((act) << BLK_TC_SHIFT)
30 
31 /*
32  * Basic trace actions
33  */
34 enum blktrace_act {
35 	__BLK_TA_QUEUE = 1,		/* queued */
36 	__BLK_TA_BACKMERGE,		/* back merged to existing rq */
37 	__BLK_TA_FRONTMERGE,		/* front merge to existing rq */
38 	__BLK_TA_GETRQ,			/* allocated new request */
39 	__BLK_TA_SLEEPRQ,		/* sleeping on rq allocation */
40 	__BLK_TA_REQUEUE,		/* request requeued */
41 	__BLK_TA_ISSUE,			/* sent to driver */
42 	__BLK_TA_COMPLETE,		/* completed by driver */
43 	__BLK_TA_PLUG,			/* queue was plugged */
44 	__BLK_TA_UNPLUG_IO,		/* queue was unplugged by io */
45 	__BLK_TA_UNPLUG_TIMER,		/* queue was unplugged by timer */
46 	__BLK_TA_INSERT,		/* insert request */
47 	__BLK_TA_SPLIT,			/* bio was split */
48 	__BLK_TA_BOUNCE,		/* bio was bounced */
49 	__BLK_TA_REMAP,			/* bio was remapped */
50 };
51 
52 /*
53  * Notify events.
54  */
55 enum blktrace_notify {
56 	__BLK_TN_PROCESS = 0,		/* establish pid/name mapping */
57 	__BLK_TN_TIMESTAMP,		/* include system clock */
58 	__BLK_TN_MESSAGE,		/* Character string message */
59 };
60 
61 
62 /*
63  * Trace actions in full. Additionally, read or write is masked
64  */
65 #define BLK_TA_QUEUE		(__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
66 #define BLK_TA_BACKMERGE	(__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
67 #define BLK_TA_FRONTMERGE	(__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
68 #define	BLK_TA_GETRQ		(__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
69 #define	BLK_TA_SLEEPRQ		(__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
70 #define	BLK_TA_REQUEUE		(__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
71 #define BLK_TA_ISSUE		(__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
72 #define BLK_TA_COMPLETE		(__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
73 #define BLK_TA_PLUG		(__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
74 #define BLK_TA_UNPLUG_IO	(__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
75 #define BLK_TA_UNPLUG_TIMER	(__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
76 #define BLK_TA_INSERT		(__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
77 #define BLK_TA_SPLIT		(__BLK_TA_SPLIT)
78 #define BLK_TA_BOUNCE		(__BLK_TA_BOUNCE)
79 #define BLK_TA_REMAP		(__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
80 
81 #define BLK_TN_PROCESS		(__BLK_TN_PROCESS | BLK_TC_ACT(BLK_TC_NOTIFY))
82 #define BLK_TN_TIMESTAMP	(__BLK_TN_TIMESTAMP | BLK_TC_ACT(BLK_TC_NOTIFY))
83 #define BLK_TN_MESSAGE		(__BLK_TN_MESSAGE | BLK_TC_ACT(BLK_TC_NOTIFY))
84 
85 #define BLK_IO_TRACE_MAGIC	0x65617400
86 #define BLK_IO_TRACE_VERSION	0x07
87 
88 /*
89  * The trace itself
90  */
91 struct blk_io_trace {
92 	u32 magic;		/* MAGIC << 8 | version */
93 	u32 sequence;		/* event number */
94 	u64 time;		/* in microseconds */
95 	u64 sector;		/* disk offset */
96 	u32 bytes;		/* transfer length */
97 	u32 action;		/* what happened */
98 	u32 pid;		/* who did it */
99 	u32 device;		/* device number */
100 	u32 cpu;		/* on what cpu did it happen */
101 	u16 error;		/* completion error */
102 	u16 pdu_len;		/* length of data after this trace */
103 };
104 
105 /*
106  * The remap event
107  */
108 struct blk_io_trace_remap {
109 	__be32 device;
110 	__be32 device_from;
111 	__be64 sector;
112 };
113 
114 enum {
115 	Blktrace_setup = 1,
116 	Blktrace_running,
117 	Blktrace_stopped,
118 };
119 
120 struct blk_trace {
121 	int trace_state;
122 	struct rchan *rchan;
123 	unsigned long *sequence;
124 	unsigned char *msg_data;
125 	u16 act_mask;
126 	u64 start_lba;
127 	u64 end_lba;
128 	u32 pid;
129 	u32 dev;
130 	struct dentry *dir;
131 	struct dentry *dropped_file;
132 	struct dentry *msg_file;
133 	atomic_t dropped;
134 };
135 
136 /*
137  * User setup structure passed with BLKTRACESTART
138  */
139 struct blk_user_trace_setup {
140 	char name[BDEVNAME_SIZE];	/* output */
141 	u16 act_mask;			/* input */
142 	u32 buf_size;			/* input */
143 	u32 buf_nr;			/* input */
144 	u64 start_lba;
145 	u64 end_lba;
146 	u32 pid;
147 };
148 
149 #ifdef __KERNEL__
150 #if defined(CONFIG_BLK_DEV_IO_TRACE)
151 extern int blk_trace_ioctl(struct block_device *, unsigned, char __user *);
152 extern void blk_trace_shutdown(struct request_queue *);
153 extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
154 extern int do_blk_trace_setup(struct request_queue *q,
155 	char *name, dev_t dev, struct blk_user_trace_setup *buts);
156 extern void __trace_note_message(struct blk_trace *, const char *fmt, ...);
157 
158 /**
159  * blk_add_trace_msg - Add a (simple) message to the blktrace stream
160  * @q:		queue the io is for
161  * @fmt:	format to print message in
162  * args...	Variable argument list for format
163  *
164  * Description:
165  *     Records a (simple) message onto the blktrace stream.
166  *
167  *     NOTE: BLK_TN_MAX_MSG characters are output at most.
168  *     NOTE: Can not use 'static inline' due to presence of var args...
169  *
170  **/
171 #define blk_add_trace_msg(q, fmt, ...)					\
172 	do {								\
173 		struct blk_trace *bt = (q)->blk_trace;			\
174 		if (unlikely(bt))					\
175 			__trace_note_message(bt, fmt, ##__VA_ARGS__);	\
176 	} while (0)
177 #define BLK_TN_MAX_MSG		128
178 
179 /**
180  * blk_add_trace_rq - Add a trace for a request oriented action
181  * @q:		queue the io is for
182  * @rq:		the source request
183  * @what:	the action
184  *
185  * Description:
186  *     Records an action against a request. Will log the bio offset + size.
187  *
188  **/
189 static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
190 				    u32 what)
191 {
192 	struct blk_trace *bt = q->blk_trace;
193 	int rw = rq->cmd_flags & 0x03;
194 
195 	if (likely(!bt))
196 		return;
197 
198 	if (blk_pc_request(rq)) {
199 		what |= BLK_TC_ACT(BLK_TC_PC);
200 		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
201 	} else  {
202 		what |= BLK_TC_ACT(BLK_TC_FS);
203 		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
204 	}
205 }
206 
207 /**
208  * blk_add_trace_bio - Add a trace for a bio oriented action
209  * @q:		queue the io is for
210  * @bio:	the source bio
211  * @what:	the action
212  *
213  * Description:
214  *     Records an action against a bio. Will log the bio offset + size.
215  *
216  **/
217 static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
218 				     u32 what)
219 {
220 	struct blk_trace *bt = q->blk_trace;
221 
222 	if (likely(!bt))
223 		return;
224 
225 	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
226 }
227 
228 /**
229  * blk_add_trace_generic - Add a trace for a generic action
230  * @q:		queue the io is for
231  * @bio:	the source bio
232  * @rw:		the data direction
233  * @what:	the action
234  *
235  * Description:
236  *     Records a simple trace
237  *
238  **/
239 static inline void blk_add_trace_generic(struct request_queue *q,
240 					 struct bio *bio, int rw, u32 what)
241 {
242 	struct blk_trace *bt = q->blk_trace;
243 
244 	if (likely(!bt))
245 		return;
246 
247 	if (bio)
248 		blk_add_trace_bio(q, bio, what);
249 	else
250 		__blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
251 }
252 
253 /**
254  * blk_add_trace_pdu_int - Add a trace for a bio with an integer payload
255  * @q:		queue the io is for
256  * @what:	the action
257  * @bio:	the source bio
258  * @pdu:	the integer payload
259  *
260  * Description:
261  *     Adds a trace with some integer payload. This might be an unplug
262  *     option given as the action, with the depth at unplug time given
263  *     as the payload
264  *
265  **/
266 static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
267 					 struct bio *bio, unsigned int pdu)
268 {
269 	struct blk_trace *bt = q->blk_trace;
270 	__be64 rpdu = cpu_to_be64(pdu);
271 
272 	if (likely(!bt))
273 		return;
274 
275 	if (bio)
276 		__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
277 	else
278 		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
279 }
280 
281 /**
282  * blk_add_trace_remap - Add a trace for a remap operation
283  * @q:		queue the io is for
284  * @bio:	the source bio
285  * @dev:	target device
286  * @from:	source sector
287  * @to:		target sector
288  *
289  * Description:
290  *     Device mapper or raid target sometimes need to split a bio because
291  *     it spans a stripe (or similar). Add a trace for that action.
292  *
293  **/
294 static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
295 				       dev_t dev, sector_t from, sector_t to)
296 {
297 	struct blk_trace *bt = q->blk_trace;
298 	struct blk_io_trace_remap r;
299 
300 	if (likely(!bt))
301 		return;
302 
303 	r.device = cpu_to_be32(dev);
304 	r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
305 	r.sector = cpu_to_be64(to);
306 
307 	__blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
308 }
309 
310 extern int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
311 			   char __user *arg);
312 extern int blk_trace_startstop(struct request_queue *q, int start);
313 extern int blk_trace_remove(struct request_queue *q);
314 
315 #else /* !CONFIG_BLK_DEV_IO_TRACE */
316 #define blk_trace_ioctl(bdev, cmd, arg)		(-ENOTTY)
317 #define blk_trace_shutdown(q)			do { } while (0)
318 #define blk_add_trace_rq(q, rq, what)		do { } while (0)
319 #define blk_add_trace_bio(q, rq, what)		do { } while (0)
320 #define blk_add_trace_generic(q, rq, rw, what)	do { } while (0)
321 #define blk_add_trace_pdu_int(q, what, bio, pdu)	do { } while (0)
322 #define blk_add_trace_remap(q, bio, dev, f, t)	do {} while (0)
323 #define do_blk_trace_setup(q, name, dev, buts)	(-ENOTTY)
324 #define blk_trace_setup(q, name, dev, arg)	(-ENOTTY)
325 #define blk_trace_startstop(q, start)		(-ENOTTY)
326 #define blk_trace_remove(q)			(-ENOTTY)
327 #define blk_add_trace_msg(q, fmt, ...)		do { } while (0)
328 
329 #endif /* CONFIG_BLK_DEV_IO_TRACE */
330 #endif /* __KERNEL__ */
331 #endif
332