blktrace_api.h revision 27b29e86bf3d4b3cf6641a0efd78ed11a9b633b2 
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 */
10enum 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	BLK_TC_DISCARD	= 1 << 13,	/* discard requests */
25
26	BLK_TC_END	= 1 << 15,	/* only 16-bits, reminder */
27};
28
29#define BLK_TC_SHIFT		(16)
30#define BLK_TC_ACT(act)		((act) << BLK_TC_SHIFT)
31
32/*
33 * Basic trace actions
34 */
35enum blktrace_act {
36	__BLK_TA_QUEUE = 1,		/* queued */
37	__BLK_TA_BACKMERGE,		/* back merged to existing rq */
38	__BLK_TA_FRONTMERGE,		/* front merge to existing rq */
39	__BLK_TA_GETRQ,			/* allocated new request */
40	__BLK_TA_SLEEPRQ,		/* sleeping on rq allocation */
41	__BLK_TA_REQUEUE,		/* request requeued */
42	__BLK_TA_ISSUE,			/* sent to driver */
43	__BLK_TA_COMPLETE,		/* completed by driver */
44	__BLK_TA_PLUG,			/* queue was plugged */
45	__BLK_TA_UNPLUG_IO,		/* queue was unplugged by io */
46	__BLK_TA_UNPLUG_TIMER,		/* queue was unplugged by timer */
47	__BLK_TA_INSERT,		/* insert request */
48	__BLK_TA_SPLIT,			/* bio was split */
49	__BLK_TA_BOUNCE,		/* bio was bounced */
50	__BLK_TA_REMAP,			/* bio was remapped */
51};
52
53/*
54 * Notify events.
55 */
56enum blktrace_notify {
57	__BLK_TN_PROCESS = 0,		/* establish pid/name mapping */
58	__BLK_TN_TIMESTAMP,		/* include system clock */
59	__BLK_TN_MESSAGE,		/* Character string message */
60};
61
62
63/*
64 * Trace actions in full. Additionally, read or write is masked
65 */
66#define BLK_TA_QUEUE		(__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
67#define BLK_TA_BACKMERGE	(__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
68#define BLK_TA_FRONTMERGE	(__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
69#define	BLK_TA_GETRQ		(__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
70#define	BLK_TA_SLEEPRQ		(__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
71#define	BLK_TA_REQUEUE		(__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
72#define BLK_TA_ISSUE		(__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
73#define BLK_TA_COMPLETE		(__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
74#define BLK_TA_PLUG		(__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
75#define BLK_TA_UNPLUG_IO	(__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
76#define BLK_TA_UNPLUG_TIMER	(__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
77#define BLK_TA_INSERT		(__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
78#define BLK_TA_SPLIT		(__BLK_TA_SPLIT)
79#define BLK_TA_BOUNCE		(__BLK_TA_BOUNCE)
80#define BLK_TA_REMAP		(__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
81
82#define BLK_TN_PROCESS		(__BLK_TN_PROCESS | BLK_TC_ACT(BLK_TC_NOTIFY))
83#define BLK_TN_TIMESTAMP	(__BLK_TN_TIMESTAMP | BLK_TC_ACT(BLK_TC_NOTIFY))
84#define BLK_TN_MESSAGE		(__BLK_TN_MESSAGE | BLK_TC_ACT(BLK_TC_NOTIFY))
85
86#define BLK_IO_TRACE_MAGIC	0x65617400
87#define BLK_IO_TRACE_VERSION	0x07
88
89/*
90 * The trace itself
91 */
92struct blk_io_trace {
93	u32 magic;		/* MAGIC << 8 | version */
94	u32 sequence;		/* event number */
95	u64 time;		/* in microseconds */
96	u64 sector;		/* disk offset */
97	u32 bytes;		/* transfer length */
98	u32 action;		/* what happened */
99	u32 pid;		/* who did it */
100	u32 device;		/* device number */
101	u32 cpu;		/* on what cpu did it happen */
102	u16 error;		/* completion error */
103	u16 pdu_len;		/* length of data after this trace */
104};
105
106/*
107 * The remap event
108 */
109struct blk_io_trace_remap {
110	__be32 device;
111	__be32 device_from;
112	__be64 sector;
113};
114
115enum {
116	Blktrace_setup = 1,
117	Blktrace_running,
118	Blktrace_stopped,
119};
120
121struct blk_trace {
122	int trace_state;
123	struct rchan *rchan;
124	unsigned long *sequence;
125	unsigned char *msg_data;
126	u16 act_mask;
127	u64 start_lba;
128	u64 end_lba;
129	u32 pid;
130	u32 dev;
131	struct dentry *dir;
132	struct dentry *dropped_file;
133	struct dentry *msg_file;
134	atomic_t dropped;
135};
136
137/*
138 * User setup structure passed with BLKTRACESTART
139 */
140struct blk_user_trace_setup {
141	char name[BDEVNAME_SIZE];	/* output */
142	u16 act_mask;			/* input */
143	u32 buf_size;			/* input */
144	u32 buf_nr;			/* input */
145	u64 start_lba;
146	u64 end_lba;
147	u32 pid;
148};
149
150#ifdef __KERNEL__
151#if defined(CONFIG_BLK_DEV_IO_TRACE)
152extern int blk_trace_ioctl(struct block_device *, unsigned, char __user *);
153extern void blk_trace_shutdown(struct request_queue *);
154extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
155extern int do_blk_trace_setup(struct request_queue *q,
156	char *name, dev_t dev, struct blk_user_trace_setup *buts);
157extern void __trace_note_message(struct blk_trace *, const char *fmt, ...);
158
159/**
160 * blk_add_trace_msg - Add a (simple) message to the blktrace stream
161 * @q:		queue the io is for
162 * @fmt:	format to print message in
163 * args...	Variable argument list for format
164 *
165 * Description:
166 *     Records a (simple) message onto the blktrace stream.
167 *
168 *     NOTE: BLK_TN_MAX_MSG characters are output at most.
169 *     NOTE: Can not use 'static inline' due to presence of var args...
170 *
171 **/
172#define blk_add_trace_msg(q, fmt, ...)					\
173	do {								\
174		struct blk_trace *bt = (q)->blk_trace;			\
175		if (unlikely(bt))					\
176			__trace_note_message(bt, fmt, ##__VA_ARGS__);	\
177	} while (0)
178#define BLK_TN_MAX_MSG		128
179
180/**
181 * blk_add_trace_rq - Add a trace for a request oriented action
182 * @q:		queue the io is for
183 * @rq:		the source request
184 * @what:	the action
185 *
186 * Description:
187 *     Records an action against a request. Will log the bio offset + size.
188 *
189 **/
190static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
191				    u32 what)
192{
193	struct blk_trace *bt = q->blk_trace;
194	int rw = rq->cmd_flags & 0x03;
195
196	if (likely(!bt))
197		return;
198
199	if (blk_discard_rq(rq))
200		rw |= (1 << BIO_RW_DISCARD);
201
202	if (blk_pc_request(rq)) {
203		what |= BLK_TC_ACT(BLK_TC_PC);
204		__blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
205	} else  {
206		what |= BLK_TC_ACT(BLK_TC_FS);
207		__blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
208	}
209}
210
211/**
212 * blk_add_trace_bio - Add a trace for a bio oriented action
213 * @q:		queue the io is for
214 * @bio:	the source bio
215 * @what:	the action
216 *
217 * Description:
218 *     Records an action against a bio. Will log the bio offset + size.
219 *
220 **/
221static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
222				     u32 what)
223{
224	struct blk_trace *bt = q->blk_trace;
225
226	if (likely(!bt))
227		return;
228
229	__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
230}
231
232/**
233 * blk_add_trace_generic - Add a trace for a generic action
234 * @q:		queue the io is for
235 * @bio:	the source bio
236 * @rw:		the data direction
237 * @what:	the action
238 *
239 * Description:
240 *     Records a simple trace
241 *
242 **/
243static inline void blk_add_trace_generic(struct request_queue *q,
244					 struct bio *bio, int rw, u32 what)
245{
246	struct blk_trace *bt = q->blk_trace;
247
248	if (likely(!bt))
249		return;
250
251	if (bio)
252		blk_add_trace_bio(q, bio, what);
253	else
254		__blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
255}
256
257/**
258 * blk_add_trace_pdu_int - Add a trace for a bio with an integer payload
259 * @q:		queue the io is for
260 * @what:	the action
261 * @bio:	the source bio
262 * @pdu:	the integer payload
263 *
264 * Description:
265 *     Adds a trace with some integer payload. This might be an unplug
266 *     option given as the action, with the depth at unplug time given
267 *     as the payload
268 *
269 **/
270static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
271					 struct bio *bio, unsigned int pdu)
272{
273	struct blk_trace *bt = q->blk_trace;
274	__be64 rpdu = cpu_to_be64(pdu);
275
276	if (likely(!bt))
277		return;
278
279	if (bio)
280		__blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
281	else
282		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
283}
284
285/**
286 * blk_add_trace_remap - Add a trace for a remap operation
287 * @q:		queue the io is for
288 * @bio:	the source bio
289 * @dev:	target device
290 * @from:	source sector
291 * @to:		target sector
292 *
293 * Description:
294 *     Device mapper or raid target sometimes need to split a bio because
295 *     it spans a stripe (or similar). Add a trace for that action.
296 *
297 **/
298static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
299				       dev_t dev, sector_t from, sector_t to)
300{
301	struct blk_trace *bt = q->blk_trace;
302	struct blk_io_trace_remap r;
303
304	if (likely(!bt))
305		return;
306
307	r.device = cpu_to_be32(dev);
308	r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
309	r.sector = cpu_to_be64(to);
310
311	__blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
312}
313
314extern int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
315			   char __user *arg);
316extern int blk_trace_startstop(struct request_queue *q, int start);
317extern int blk_trace_remove(struct request_queue *q);
318
319#else /* !CONFIG_BLK_DEV_IO_TRACE */
320#define blk_trace_ioctl(bdev, cmd, arg)		(-ENOTTY)
321#define blk_trace_shutdown(q)			do { } while (0)
322#define blk_add_trace_rq(q, rq, what)		do { } while (0)
323#define blk_add_trace_bio(q, rq, what)		do { } while (0)
324#define blk_add_trace_generic(q, rq, rw, what)	do { } while (0)
325#define blk_add_trace_pdu_int(q, what, bio, pdu)	do { } while (0)
326#define blk_add_trace_remap(q, bio, dev, f, t)	do {} while (0)
327#define do_blk_trace_setup(q, name, dev, buts)	(-ENOTTY)
328#define blk_trace_setup(q, name, dev, arg)	(-ENOTTY)
329#define blk_trace_startstop(q, start)		(-ENOTTY)
330#define blk_trace_remove(q)			(-ENOTTY)
331#define blk_add_trace_msg(q, fmt, ...)		do { } while (0)
332
333#endif /* CONFIG_BLK_DEV_IO_TRACE */
334#endif /* __KERNEL__ */
335#endif
336