blob: 7520cc1ff9e2b9d3947f3cb1ace868b329394555 [file] [log] [blame]
Jens Axboe2056a782006-03-23 20:00:26 +01001#ifndef BLKTRACE_H
2#define BLKTRACE_H
3
Jens Axboe2056a782006-03-23 20:00:26 +01004#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 */
Nathan Scottfc0a75c2006-07-06 09:56:30 +020014 BLK_TC_SYNC = 1 << 3, /* sync IO */
Jens Axboe2056a782006-03-23 20:00:26 +010015 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 */
Nathan Scott40359cc2006-07-06 10:03:28 +020022 BLK_TC_AHEAD = 1 << 11, /* readahead */
Jens Axboe2056a782006-03-23 20:00:26 +010023
24 BLK_TC_END = 1 << 15, /* only 16-bits, reminder */
25};
26
27#define BLK_TC_SHIFT (16)
28#define BLK_TC_ACT(act) ((act) << BLK_TC_SHIFT)
29
30/*
31 * Basic trace actions
32 */
33enum blktrace_act {
34 __BLK_TA_QUEUE = 1, /* queued */
35 __BLK_TA_BACKMERGE, /* back merged to existing rq */
36 __BLK_TA_FRONTMERGE, /* front merge to existing rq */
37 __BLK_TA_GETRQ, /* allocated new request */
38 __BLK_TA_SLEEPRQ, /* sleeping on rq allocation */
39 __BLK_TA_REQUEUE, /* request requeued */
40 __BLK_TA_ISSUE, /* sent to driver */
41 __BLK_TA_COMPLETE, /* completed by driver */
42 __BLK_TA_PLUG, /* queue was plugged */
43 __BLK_TA_UNPLUG_IO, /* queue was unplugged by io */
44 __BLK_TA_UNPLUG_TIMER, /* queue was unplugged by timer */
45 __BLK_TA_INSERT, /* insert request */
46 __BLK_TA_SPLIT, /* bio was split */
47 __BLK_TA_BOUNCE, /* bio was bounced */
48 __BLK_TA_REMAP, /* bio was remapped */
49};
50
51/*
52 * Trace actions in full. Additionally, read or write is masked
53 */
54#define BLK_TA_QUEUE (__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
55#define BLK_TA_BACKMERGE (__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
56#define BLK_TA_FRONTMERGE (__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
57#define BLK_TA_GETRQ (__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
58#define BLK_TA_SLEEPRQ (__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
59#define BLK_TA_REQUEUE (__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
60#define BLK_TA_ISSUE (__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
61#define BLK_TA_COMPLETE (__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
62#define BLK_TA_PLUG (__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
63#define BLK_TA_UNPLUG_IO (__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
64#define BLK_TA_UNPLUG_TIMER (__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
65#define BLK_TA_INSERT (__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
66#define BLK_TA_SPLIT (__BLK_TA_SPLIT)
67#define BLK_TA_BOUNCE (__BLK_TA_BOUNCE)
68#define BLK_TA_REMAP (__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
69
70#define BLK_IO_TRACE_MAGIC 0x65617400
71#define BLK_IO_TRACE_VERSION 0x07
72
73/*
74 * The trace itself
75 */
76struct blk_io_trace {
77 u32 magic; /* MAGIC << 8 | version */
78 u32 sequence; /* event number */
79 u64 time; /* in microseconds */
80 u64 sector; /* disk offset */
81 u32 bytes; /* transfer length */
82 u32 action; /* what happened */
83 u32 pid; /* who did it */
84 u32 device; /* device number */
85 u32 cpu; /* on what cpu did it happen */
86 u16 error; /* completion error */
87 u16 pdu_len; /* length of data after this trace */
88};
89
90/*
91 * The remap event
92 */
93struct blk_io_trace_remap {
Alexey Dobriyanfda151d2006-06-05 12:09:50 +020094 __be32 device;
Jens Axboe2056a782006-03-23 20:00:26 +010095 u32 __pad;
Alexey Dobriyanfda151d2006-06-05 12:09:50 +020096 __be64 sector;
Jens Axboe2056a782006-03-23 20:00:26 +010097};
98
99enum {
100 Blktrace_setup = 1,
101 Blktrace_running,
102 Blktrace_stopped,
103};
104
105struct blk_trace {
106 int trace_state;
107 struct rchan *rchan;
108 unsigned long *sequence;
109 u16 act_mask;
110 u64 start_lba;
111 u64 end_lba;
112 u32 pid;
113 u32 dev;
114 struct dentry *dir;
115 struct dentry *dropped_file;
116 atomic_t dropped;
117};
118
119/*
120 * User setup structure passed with BLKTRACESTART
121 */
122struct blk_user_trace_setup {
123 char name[BDEVNAME_SIZE]; /* output */
124 u16 act_mask; /* input */
125 u32 buf_size; /* input */
126 u32 buf_nr; /* input */
127 u64 start_lba;
128 u64 end_lba;
129 u32 pid;
130};
131
132#if defined(CONFIG_BLK_DEV_IO_TRACE)
133extern int blk_trace_ioctl(struct block_device *, unsigned, char __user *);
134extern void blk_trace_shutdown(request_queue_t *);
135extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
136
137/**
138 * blk_add_trace_rq - Add a trace for a request oriented action
139 * @q: queue the io is for
140 * @rq: the source request
141 * @what: the action
142 *
143 * Description:
144 * Records an action against a request. Will log the bio offset + size.
145 *
146 **/
147static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
148 u32 what)
149{
150 struct blk_trace *bt = q->blk_trace;
Jens Axboe1959d212006-07-06 10:18:05 +0200151 int rw = rq->flags & 0x03;
Jens Axboe2056a782006-03-23 20:00:26 +0100152
153 if (likely(!bt))
154 return;
155
156 if (blk_pc_request(rq)) {
157 what |= BLK_TC_ACT(BLK_TC_PC);
158 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
159 } else {
160 what |= BLK_TC_ACT(BLK_TC_FS);
161 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
162 }
163}
164
165/**
166 * blk_add_trace_bio - Add a trace for a bio oriented action
167 * @q: queue the io is for
168 * @bio: the source bio
169 * @what: the action
170 *
171 * Description:
172 * Records an action against a bio. Will log the bio offset + size.
173 *
174 **/
175static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
176 u32 what)
177{
178 struct blk_trace *bt = q->blk_trace;
179
180 if (likely(!bt))
181 return;
182
183 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
184}
185
186/**
187 * blk_add_trace_generic - Add a trace for a generic action
188 * @q: queue the io is for
189 * @bio: the source bio
190 * @rw: the data direction
191 * @what: the action
192 *
193 * Description:
194 * Records a simple trace
195 *
196 **/
197static inline void blk_add_trace_generic(struct request_queue *q,
198 struct bio *bio, int rw, u32 what)
199{
200 struct blk_trace *bt = q->blk_trace;
201
202 if (likely(!bt))
203 return;
204
205 if (bio)
206 blk_add_trace_bio(q, bio, what);
207 else
208 __blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
209}
210
211/**
212 * blk_add_trace_pdu_int - Add a trace for a bio with an integer payload
213 * @q: queue the io is for
214 * @what: the action
215 * @bio: the source bio
216 * @pdu: the integer payload
217 *
218 * Description:
219 * Adds a trace with some integer payload. This might be an unplug
220 * option given as the action, with the depth at unplug time given
221 * as the payload
222 *
223 **/
224static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
225 struct bio *bio, unsigned int pdu)
226{
227 struct blk_trace *bt = q->blk_trace;
Alexey Dobriyanfda151d2006-06-05 12:09:50 +0200228 __be64 rpdu = cpu_to_be64(pdu);
Jens Axboe2056a782006-03-23 20:00:26 +0100229
230 if (likely(!bt))
231 return;
232
233 if (bio)
234 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
235 else
236 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
237}
238
239/**
240 * blk_add_trace_remap - Add a trace for a remap operation
241 * @q: queue the io is for
242 * @bio: the source bio
243 * @dev: target device
244 * @from: source sector
245 * @to: target sector
246 *
247 * Description:
248 * Device mapper or raid target sometimes need to split a bio because
249 * it spans a stripe (or similar). Add a trace for that action.
250 *
251 **/
252static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
253 dev_t dev, sector_t from, sector_t to)
254{
255 struct blk_trace *bt = q->blk_trace;
256 struct blk_io_trace_remap r;
257
258 if (likely(!bt))
259 return;
260
261 r.device = cpu_to_be32(dev);
262 r.sector = cpu_to_be64(to);
263
264 __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
265}
266
267#else /* !CONFIG_BLK_DEV_IO_TRACE */
268#define blk_trace_ioctl(bdev, cmd, arg) (-ENOTTY)
269#define blk_trace_shutdown(q) do { } while (0)
270#define blk_add_trace_rq(q, rq, what) do { } while (0)
271#define blk_add_trace_bio(q, rq, what) do { } while (0)
272#define blk_add_trace_generic(q, rq, rw, what) do { } while (0)
273#define blk_add_trace_pdu_int(q, what, bio, pdu) do { } while (0)
274#define blk_add_trace_remap(q, bio, dev, f, t) do {} while (0)
275#endif /* CONFIG_BLK_DEV_IO_TRACE */
276
277#endif