Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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 */
12enum 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 BLK_TC_DRV_DATA = 1 << 14, /* binary per-driver data */
28
29 BLK_TC_END = 1 << 15, /* only 16-bits, reminder */
30};
31
32#define BLK_TC_SHIFT (16)
33#define BLK_TC_ACT(act) ((act) << BLK_TC_SHIFT)
34
35/*
36 * Basic trace actions
37 */
38enum blktrace_act {
39 __BLK_TA_QUEUE = 1, /* queued */
40 __BLK_TA_BACKMERGE, /* back merged to existing rq */
41 __BLK_TA_FRONTMERGE, /* front merge to existing rq */
42 __BLK_TA_GETRQ, /* allocated new request */
43 __BLK_TA_SLEEPRQ, /* sleeping on rq allocation */
44 __BLK_TA_REQUEUE, /* request requeued */
45 __BLK_TA_ISSUE, /* sent to driver */
46 __BLK_TA_COMPLETE, /* completed by driver */
47 __BLK_TA_PLUG, /* queue was plugged */
48 __BLK_TA_UNPLUG_IO, /* queue was unplugged by io */
49 __BLK_TA_UNPLUG_TIMER, /* queue was unplugged by timer */
50 __BLK_TA_INSERT, /* insert request */
51 __BLK_TA_SPLIT, /* bio was split */
52 __BLK_TA_BOUNCE, /* bio was bounced */
53 __BLK_TA_REMAP, /* bio was remapped */
54 __BLK_TA_ABORT, /* request aborted */
55 __BLK_TA_DRV_DATA, /* driver-specific binary data */
56};
57
58/*
59 * Notify events.
60 */
61enum blktrace_notify {
62 __BLK_TN_PROCESS = 0, /* establish pid/name mapping */
63 __BLK_TN_TIMESTAMP, /* include system clock */
64 __BLK_TN_MESSAGE, /* Character string message */
65};
66
67
68/*
69 * Trace actions in full. Additionally, read or write is masked
70 */
71#define BLK_TA_QUEUE (__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
72#define BLK_TA_BACKMERGE (__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
73#define BLK_TA_FRONTMERGE (__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
74#define BLK_TA_GETRQ (__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
75#define BLK_TA_SLEEPRQ (__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
76#define BLK_TA_REQUEUE (__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
77#define BLK_TA_ISSUE (__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
78#define BLK_TA_COMPLETE (__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
79#define BLK_TA_PLUG (__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
80#define BLK_TA_UNPLUG_IO (__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
81#define BLK_TA_UNPLUG_TIMER (__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
82#define BLK_TA_INSERT (__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
83#define BLK_TA_SPLIT (__BLK_TA_SPLIT)
84#define BLK_TA_BOUNCE (__BLK_TA_BOUNCE)
85#define BLK_TA_REMAP (__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
86#define BLK_TA_ABORT (__BLK_TA_ABORT | BLK_TC_ACT(BLK_TC_QUEUE))
87#define BLK_TA_DRV_DATA (__BLK_TA_DRV_DATA | BLK_TC_ACT(BLK_TC_DRV_DATA))
88
89#define BLK_TN_PROCESS (__BLK_TN_PROCESS | BLK_TC_ACT(BLK_TC_NOTIFY))
90#define BLK_TN_TIMESTAMP (__BLK_TN_TIMESTAMP | BLK_TC_ACT(BLK_TC_NOTIFY))
91#define BLK_TN_MESSAGE (__BLK_TN_MESSAGE | BLK_TC_ACT(BLK_TC_NOTIFY))
92
93#define BLK_IO_TRACE_MAGIC 0x65617400
94#define BLK_IO_TRACE_VERSION 0x07
95
96/*
97 * The trace itself
98 */
99struct blk_io_trace {
100 __u32 magic; /* MAGIC << 8 | version */
101 __u32 sequence; /* event number */
102 __u64 time; /* in microseconds */
103 __u64 sector; /* disk offset */
104 __u32 bytes; /* transfer length */
105 __u32 action; /* what happened */
106 __u32 pid; /* who did it */
107 __u32 device; /* device number */
108 __u32 cpu; /* on what cpu did it happen */
109 __u16 error; /* completion error */
110 __u16 pdu_len; /* length of data after this trace */
111};
112
113/*
114 * The remap event
115 */
116struct blk_io_trace_remap {
117 __be32 device;
118 __be32 device_from;
119 __be64 sector;
120};
121
122enum {
123 Blktrace_setup = 1,
124 Blktrace_running,
125 Blktrace_stopped,
126};
127
128#define BLKTRACE_BDEV_SIZE 32
129
130/*
131 * User setup structure passed with BLKTRACESTART
132 */
133struct blk_user_trace_setup {
134 char name[BLKTRACE_BDEV_SIZE]; /* output */
135 __u16 act_mask; /* input */
136 __u32 buf_size; /* input */
137 __u32 buf_nr; /* input */
138 __u64 start_lba;
139 __u64 end_lba;
140 __u32 pid;
141};
142
143#ifdef __KERNEL__
144#if defined(CONFIG_BLK_DEV_IO_TRACE)
145struct blk_trace {
146 int trace_state;
147 struct rchan *rchan;
148 unsigned long *sequence;
149 unsigned char *msg_data;
150 u16 act_mask;
151 u64 start_lba;
152 u64 end_lba;
153 u32 pid;
154 u32 dev;
155 struct dentry *dir;
156 struct dentry *dropped_file;
157 struct dentry *msg_file;
158 atomic_t dropped;
159};
160
161extern int blk_trace_ioctl(struct block_device *, unsigned, char __user *);
162extern void blk_trace_shutdown(struct request_queue *);
163extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
164extern int do_blk_trace_setup(struct request_queue *q,
165 char *name, dev_t dev, struct blk_user_trace_setup *buts);
166extern void __trace_note_message(struct blk_trace *, const char *fmt, ...);
167
168/**
169 * blk_add_trace_msg - Add a (simple) message to the blktrace stream
170 * @q: queue the io is for
171 * @fmt: format to print message in
172 * args... Variable argument list for format
173 *
174 * Description:
175 * Records a (simple) message onto the blktrace stream.
176 *
177 * NOTE: BLK_TN_MAX_MSG characters are output at most.
178 * NOTE: Can not use 'static inline' due to presence of var args...
179 *
180 **/
181#define blk_add_trace_msg(q, fmt, ...) \
182 do { \
183 struct blk_trace *bt = (q)->blk_trace; \
184 if (unlikely(bt)) \
185 __trace_note_message(bt, fmt, ##__VA_ARGS__); \
186 } while (0)
187#define BLK_TN_MAX_MSG 128
188
189/**
190 * blk_add_trace_rq - Add a trace for a request oriented action
191 * @q: queue the io is for
192 * @rq: the source request
193 * @what: the action
194 *
195 * Description:
196 * Records an action against a request. Will log the bio offset + size.
197 *
198 **/
199static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
200 u32 what)
201{
202 struct blk_trace *bt = q->blk_trace;
203 int rw = rq->cmd_flags & 0x03;
204
205 if (likely(!bt))
206 return;
207
208 if (blk_discard_rq(rq))
209 rw |= (1 << BIO_RW_DISCARD);
210
211 if (blk_pc_request(rq)) {
212 what |= BLK_TC_ACT(BLK_TC_PC);
213 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
214 } else {
215 what |= BLK_TC_ACT(BLK_TC_FS);
216 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
217 }
218}
219
220/**
221 * blk_add_trace_bio - Add a trace for a bio oriented action
222 * @q: queue the io is for
223 * @bio: the source bio
224 * @what: the action
225 *
226 * Description:
227 * Records an action against a bio. Will log the bio offset + size.
228 *
229 **/
230static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
231 u32 what)
232{
233 struct blk_trace *bt = q->blk_trace;
234
235 if (likely(!bt))
236 return;
237
238 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
239}
240
241/**
242 * blk_add_trace_generic - Add a trace for a generic action
243 * @q: queue the io is for
244 * @bio: the source bio
245 * @rw: the data direction
246 * @what: the action
247 *
248 * Description:
249 * Records a simple trace
250 *
251 **/
252static inline void blk_add_trace_generic(struct request_queue *q,
253 struct bio *bio, int rw, u32 what)
254{
255 struct blk_trace *bt = q->blk_trace;
256
257 if (likely(!bt))
258 return;
259
260 if (bio)
261 blk_add_trace_bio(q, bio, what);
262 else
263 __blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
264}
265
266/**
267 * blk_add_trace_pdu_int - Add a trace for a bio with an integer payload
268 * @q: queue the io is for
269 * @what: the action
270 * @bio: the source bio
271 * @pdu: the integer payload
272 *
273 * Description:
274 * Adds a trace with some integer payload. This might be an unplug
275 * option given as the action, with the depth at unplug time given
276 * as the payload
277 *
278 **/
279static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
280 struct bio *bio, unsigned int pdu)
281{
282 struct blk_trace *bt = q->blk_trace;
283 __be64 rpdu = cpu_to_be64(pdu);
284
285 if (likely(!bt))
286 return;
287
288 if (bio)
289 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
290 else
291 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
292}
293
294/**
295 * blk_add_trace_remap - Add a trace for a remap operation
296 * @q: queue the io is for
297 * @bio: the source bio
298 * @dev: target device
299 * @from: source sector
300 * @to: target sector
301 *
302 * Description:
303 * Device mapper or raid target sometimes need to split a bio because
304 * it spans a stripe (or similar). Add a trace for that action.
305 *
306 **/
307static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
308 dev_t dev, sector_t from, sector_t to)
309{
310 struct blk_trace *bt = q->blk_trace;
311 struct blk_io_trace_remap r;
312
313 if (likely(!bt))
314 return;
315
316 r.device = cpu_to_be32(dev);
317 r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
318 r.sector = cpu_to_be64(to);
319
320 __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
321}
322
323/**
324 * blk_add_driver_data - Add binary message with driver-specific data
325 * @q: queue the io is for
326 * @rq: io request
327 * @data: driver-specific data
328 * @len: length of driver-specific data
329 *
330 * Description:
331 * Some drivers might want to write driver-specific data per request.
332 *
333 **/
334static inline void blk_add_driver_data(struct request_queue *q,
335 struct request *rq,
336 void *data, size_t len)
337{
338 struct blk_trace *bt = q->blk_trace;
339
340 if (likely(!bt))
341 return;
342
343 if (blk_pc_request(rq))
344 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
345 rq->errors, len, data);
346 else
347 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
348 0, BLK_TA_DRV_DATA, rq->errors, len, data);
349}
350
351extern int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
352 char __user *arg);
353extern int blk_trace_startstop(struct request_queue *q, int start);
354extern int blk_trace_remove(struct request_queue *q);
355
356#else /* !CONFIG_BLK_DEV_IO_TRACE */
357#define blk_trace_ioctl(bdev, cmd, arg) (-ENOTTY)
358#define blk_trace_shutdown(q) do { } while (0)
359#define blk_add_trace_rq(q, rq, what) do { } while (0)
360#define blk_add_trace_bio(q, rq, what) do { } while (0)
361#define blk_add_trace_generic(q, rq, rw, what) do { } while (0)
362#define blk_add_trace_pdu_int(q, what, bio, pdu) do { } while (0)
363#define blk_add_trace_remap(q, bio, dev, f, t) do {} while (0)
364#define blk_add_driver_data(q, rq, data, len) do {} while (0)
365#define do_blk_trace_setup(q, name, dev, buts) (-ENOTTY)
366#define blk_trace_setup(q, name, dev, arg) (-ENOTTY)
367#define blk_trace_startstop(q, start) (-ENOTTY)
368#define blk_trace_remove(q) (-ENOTTY)
369#define blk_add_trace_msg(q, fmt, ...) do { } while (0)
370
371#endif /* CONFIG_BLK_DEV_IO_TRACE */
372#endif /* __KERNEL__ */
373#endif