blob: b99a714fcac67adf74e85650abcfa7ecaf6ef7ab [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 Axboe7457e6e2006-07-23 02:12:01 +020023 BLK_TC_META = 1 << 12, /* metadata */
Jens Axboe2056a782006-03-23 20:00:26 +010024
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 */
34enum 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 * Trace actions in full. Additionally, read or write is masked
54 */
55#define BLK_TA_QUEUE (__BLK_TA_QUEUE | BLK_TC_ACT(BLK_TC_QUEUE))
56#define BLK_TA_BACKMERGE (__BLK_TA_BACKMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
57#define BLK_TA_FRONTMERGE (__BLK_TA_FRONTMERGE | BLK_TC_ACT(BLK_TC_QUEUE))
58#define BLK_TA_GETRQ (__BLK_TA_GETRQ | BLK_TC_ACT(BLK_TC_QUEUE))
59#define BLK_TA_SLEEPRQ (__BLK_TA_SLEEPRQ | BLK_TC_ACT(BLK_TC_QUEUE))
60#define BLK_TA_REQUEUE (__BLK_TA_REQUEUE | BLK_TC_ACT(BLK_TC_REQUEUE))
61#define BLK_TA_ISSUE (__BLK_TA_ISSUE | BLK_TC_ACT(BLK_TC_ISSUE))
62#define BLK_TA_COMPLETE (__BLK_TA_COMPLETE| BLK_TC_ACT(BLK_TC_COMPLETE))
63#define BLK_TA_PLUG (__BLK_TA_PLUG | BLK_TC_ACT(BLK_TC_QUEUE))
64#define BLK_TA_UNPLUG_IO (__BLK_TA_UNPLUG_IO | BLK_TC_ACT(BLK_TC_QUEUE))
65#define BLK_TA_UNPLUG_TIMER (__BLK_TA_UNPLUG_TIMER | BLK_TC_ACT(BLK_TC_QUEUE))
66#define BLK_TA_INSERT (__BLK_TA_INSERT | BLK_TC_ACT(BLK_TC_QUEUE))
67#define BLK_TA_SPLIT (__BLK_TA_SPLIT)
68#define BLK_TA_BOUNCE (__BLK_TA_BOUNCE)
69#define BLK_TA_REMAP (__BLK_TA_REMAP | BLK_TC_ACT(BLK_TC_QUEUE))
70
71#define BLK_IO_TRACE_MAGIC 0x65617400
72#define BLK_IO_TRACE_VERSION 0x07
73
74/*
75 * The trace itself
76 */
77struct blk_io_trace {
78 u32 magic; /* MAGIC << 8 | version */
79 u32 sequence; /* event number */
80 u64 time; /* in microseconds */
81 u64 sector; /* disk offset */
82 u32 bytes; /* transfer length */
83 u32 action; /* what happened */
84 u32 pid; /* who did it */
85 u32 device; /* device number */
86 u32 cpu; /* on what cpu did it happen */
87 u16 error; /* completion error */
88 u16 pdu_len; /* length of data after this trace */
89};
90
91/*
92 * The remap event
93 */
94struct blk_io_trace_remap {
Alexey Dobriyanfda151d2006-06-05 12:09:50 +020095 __be32 device;
Jens Axboe2056a782006-03-23 20:00:26 +010096 u32 __pad;
Alexey Dobriyanfda151d2006-06-05 12:09:50 +020097 __be64 sector;
Jens Axboe2056a782006-03-23 20:00:26 +010098};
99
100enum {
101 Blktrace_setup = 1,
102 Blktrace_running,
103 Blktrace_stopped,
104};
105
106struct blk_trace {
107 int trace_state;
108 struct rchan *rchan;
109 unsigned long *sequence;
110 u16 act_mask;
111 u64 start_lba;
112 u64 end_lba;
113 u32 pid;
114 u32 dev;
115 struct dentry *dir;
116 struct dentry *dropped_file;
117 atomic_t dropped;
118};
119
120/*
121 * User setup structure passed with BLKTRACESTART
122 */
123struct blk_user_trace_setup {
124 char name[BDEVNAME_SIZE]; /* output */
125 u16 act_mask; /* input */
126 u32 buf_size; /* input */
127 u32 buf_nr; /* input */
128 u64 start_lba;
129 u64 end_lba;
130 u32 pid;
131};
132
133#if defined(CONFIG_BLK_DEV_IO_TRACE)
134extern int blk_trace_ioctl(struct block_device *, unsigned, char __user *);
135extern void blk_trace_shutdown(request_queue_t *);
136extern void __blk_add_trace(struct blk_trace *, sector_t, int, int, u32, int, int, void *);
137
138/**
139 * blk_add_trace_rq - Add a trace for a request oriented action
140 * @q: queue the io is for
141 * @rq: the source request
142 * @what: the action
143 *
144 * Description:
145 * Records an action against a request. Will log the bio offset + size.
146 *
147 **/
148static inline void blk_add_trace_rq(struct request_queue *q, struct request *rq,
149 u32 what)
150{
151 struct blk_trace *bt = q->blk_trace;
Jens Axboe4aff5e22006-08-10 08:44:47 +0200152 int rw = rq->cmd_flags & 0x03;
Jens Axboe2056a782006-03-23 20:00:26 +0100153
154 if (likely(!bt))
155 return;
156
157 if (blk_pc_request(rq)) {
158 what |= BLK_TC_ACT(BLK_TC_PC);
159 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors, sizeof(rq->cmd), rq->cmd);
160 } else {
161 what |= BLK_TC_ACT(BLK_TC_FS);
162 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9, rw, what, rq->errors, 0, NULL);
163 }
164}
165
166/**
167 * blk_add_trace_bio - Add a trace for a bio oriented action
168 * @q: queue the io is for
169 * @bio: the source bio
170 * @what: the action
171 *
172 * Description:
173 * Records an action against a bio. Will log the bio offset + size.
174 *
175 **/
176static inline void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
177 u32 what)
178{
179 struct blk_trace *bt = q->blk_trace;
180
181 if (likely(!bt))
182 return;
183
184 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
185}
186
187/**
188 * blk_add_trace_generic - Add a trace for a generic action
189 * @q: queue the io is for
190 * @bio: the source bio
191 * @rw: the data direction
192 * @what: the action
193 *
194 * Description:
195 * Records a simple trace
196 *
197 **/
198static inline void blk_add_trace_generic(struct request_queue *q,
199 struct bio *bio, int rw, u32 what)
200{
201 struct blk_trace *bt = q->blk_trace;
202
203 if (likely(!bt))
204 return;
205
206 if (bio)
207 blk_add_trace_bio(q, bio, what);
208 else
209 __blk_add_trace(bt, 0, 0, rw, what, 0, 0, NULL);
210}
211
212/**
213 * blk_add_trace_pdu_int - Add a trace for a bio with an integer payload
214 * @q: queue the io is for
215 * @what: the action
216 * @bio: the source bio
217 * @pdu: the integer payload
218 *
219 * Description:
220 * Adds a trace with some integer payload. This might be an unplug
221 * option given as the action, with the depth at unplug time given
222 * as the payload
223 *
224 **/
225static inline void blk_add_trace_pdu_int(struct request_queue *q, u32 what,
226 struct bio *bio, unsigned int pdu)
227{
228 struct blk_trace *bt = q->blk_trace;
Alexey Dobriyanfda151d2006-06-05 12:09:50 +0200229 __be64 rpdu = cpu_to_be64(pdu);
Jens Axboe2056a782006-03-23 20:00:26 +0100230
231 if (likely(!bt))
232 return;
233
234 if (bio)
235 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, !bio_flagged(bio, BIO_UPTODATE), sizeof(rpdu), &rpdu);
236 else
237 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
238}
239
240/**
241 * blk_add_trace_remap - Add a trace for a remap operation
242 * @q: queue the io is for
243 * @bio: the source bio
244 * @dev: target device
245 * @from: source sector
246 * @to: target sector
247 *
248 * Description:
249 * Device mapper or raid target sometimes need to split a bio because
250 * it spans a stripe (or similar). Add a trace for that action.
251 *
252 **/
253static inline void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
254 dev_t dev, sector_t from, sector_t to)
255{
256 struct blk_trace *bt = q->blk_trace;
257 struct blk_io_trace_remap r;
258
259 if (likely(!bt))
260 return;
261
262 r.device = cpu_to_be32(dev);
263 r.sector = cpu_to_be64(to);
264
265 __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
266}
267
268#else /* !CONFIG_BLK_DEV_IO_TRACE */
269#define blk_trace_ioctl(bdev, cmd, arg) (-ENOTTY)
270#define blk_trace_shutdown(q) do { } while (0)
271#define blk_add_trace_rq(q, rq, what) do { } while (0)
272#define blk_add_trace_bio(q, rq, what) do { } while (0)
273#define blk_add_trace_generic(q, rq, rw, what) do { } while (0)
274#define blk_add_trace_pdu_int(q, what, bio, pdu) do { } while (0)
275#define blk_add_trace_remap(q, bio, dev, f, t) do {} while (0)
276#endif /* CONFIG_BLK_DEV_IO_TRACE */
277
278#endif