Commit | Line | Data |
---|---|---|
2866c82d | 1 | /* |
da751ca9 JA |
2 | * posixaio engine |
3 | * | |
4 | * IO engine that uses the posix defined aio interface. | |
2866c82d JA |
5 | * |
6 | */ | |
7 | #include <stdio.h> | |
8 | #include <stdlib.h> | |
9 | #include <unistd.h> | |
10 | #include <errno.h> | |
bc733f3b | 11 | #include <fcntl.h> |
5f350952 JA |
12 | |
13 | #include "../fio.h" | |
2866c82d | 14 | |
34cfcdaf JA |
15 | #ifdef FIO_HAVE_POSIXAIO |
16 | ||
2866c82d JA |
17 | struct posixaio_data { |
18 | struct io_u **aio_events; | |
207cb0f0 | 19 | unsigned int queued; |
2866c82d JA |
20 | }; |
21 | ||
22 | static int fill_timespec(struct timespec *ts) | |
23 | { | |
24 | #ifdef _POSIX_TIMERS | |
25 | if (!clock_gettime(CLOCK_MONOTONIC, ts)) | |
26 | return 0; | |
27 | ||
28 | perror("clock_gettime"); | |
29 | #endif | |
30 | return 1; | |
31 | } | |
32 | ||
33 | static unsigned long long ts_utime_since_now(struct timespec *t) | |
34 | { | |
35 | long long sec, nsec; | |
36 | struct timespec now; | |
37 | ||
38 | if (fill_timespec(&now)) | |
39 | return 0; | |
40 | ||
41 | sec = now.tv_sec - t->tv_sec; | |
42 | nsec = now.tv_nsec - t->tv_nsec; | |
43 | if (sec > 0 && nsec < 0) { | |
44 | sec--; | |
45 | nsec += 1000000000; | |
46 | } | |
47 | ||
48 | sec *= 1000000; | |
49 | nsec /= 1000; | |
50 | return sec + nsec; | |
51 | } | |
52 | ||
7a16dd02 JA |
53 | static int fio_posixaio_cancel(struct thread_data fio_unused *td, |
54 | struct io_u *io_u) | |
2866c82d | 55 | { |
53cdc686 JA |
56 | struct fio_file *f = io_u->file; |
57 | int r = aio_cancel(f->fd, &io_u->aiocb); | |
2866c82d | 58 | |
2faf9ec8 | 59 | if (r == AIO_ALLDONE || r == AIO_CANCELED) |
2866c82d JA |
60 | return 0; |
61 | ||
62 | return 1; | |
63 | } | |
64 | ||
7a16dd02 JA |
65 | static int fio_posixaio_prep(struct thread_data fio_unused *td, |
66 | struct io_u *io_u) | |
2866c82d JA |
67 | { |
68 | struct aiocb *aiocb = &io_u->aiocb; | |
53cdc686 | 69 | struct fio_file *f = io_u->file; |
2866c82d | 70 | |
53cdc686 | 71 | aiocb->aio_fildes = f->fd; |
cec6b55d JA |
72 | aiocb->aio_buf = io_u->xfer_buf; |
73 | aiocb->aio_nbytes = io_u->xfer_buflen; | |
2866c82d | 74 | aiocb->aio_offset = io_u->offset; |
9918be5a | 75 | aiocb->aio_sigevent.sigev_notify = SIGEV_NONE; |
2866c82d JA |
76 | |
77 | io_u->seen = 0; | |
78 | return 0; | |
79 | } | |
80 | ||
3c77037e JA |
81 | #define SUSPEND_ENTRIES 8 |
82 | ||
e7d2e616 JA |
83 | static int fio_posixaio_getevents(struct thread_data *td, unsigned int min, |
84 | unsigned int max, struct timespec *t) | |
2866c82d JA |
85 | { |
86 | struct posixaio_data *pd = td->io_ops->data; | |
3c77037e | 87 | struct aiocb *suspend_list[SUSPEND_ENTRIES]; |
01743ee1 | 88 | struct flist_head *entry; |
2866c82d | 89 | struct timespec start; |
a3cc770d | 90 | int have_timeout = 0; |
3c77037e | 91 | int suspend_entries = 0; |
a3cc770d | 92 | unsigned int r; |
2866c82d JA |
93 | |
94 | if (t && !fill_timespec(&start)) | |
95 | have_timeout = 1; | |
96 | ||
97 | r = 0; | |
565cc35f | 98 | memset(suspend_list, 0, sizeof(*suspend_list)); |
2866c82d | 99 | restart: |
01743ee1 JA |
100 | flist_for_each(entry, &td->io_u_busylist) { |
101 | struct io_u *io_u = flist_entry(entry, struct io_u, list); | |
2866c82d JA |
102 | int err; |
103 | ||
104 | if (io_u->seen) | |
105 | continue; | |
106 | ||
107 | err = aio_error(&io_u->aiocb); | |
3c77037e JA |
108 | if (err == EINPROGRESS) { |
109 | if (suspend_entries < SUSPEND_ENTRIES) { | |
110 | suspend_list[suspend_entries] = &io_u->aiocb; | |
111 | suspend_entries++; | |
112 | } | |
3f344316 | 113 | continue; |
3c77037e | 114 | } |
3f344316 JA |
115 | |
116 | io_u->seen = 1; | |
207cb0f0 | 117 | pd->queued--; |
3f344316 JA |
118 | pd->aio_events[r++] = io_u; |
119 | ||
120 | if (err == ECANCELED) | |
121 | io_u->resid = io_u->xfer_buflen; | |
122 | else if (!err) { | |
123 | ssize_t retval = aio_return(&io_u->aiocb); | |
124 | ||
125 | io_u->resid = io_u->xfer_buflen - retval; | |
126 | } else | |
127 | io_u->error = err; | |
2866c82d JA |
128 | } |
129 | ||
130 | if (r >= min) | |
131 | return r; | |
132 | ||
133 | if (have_timeout) { | |
134 | unsigned long long usec; | |
135 | ||
136 | usec = (t->tv_sec * 1000000) + (t->tv_nsec / 1000); | |
137 | if (ts_utime_since_now(&start) > usec) | |
138 | return r; | |
139 | } | |
140 | ||
141 | /* | |
3c77037e | 142 | * must have some in-flight, wait for at least one |
2866c82d | 143 | */ |
3c77037e JA |
144 | aio_suspend((const struct aiocb * const *)suspend_list, |
145 | suspend_entries, t); | |
2866c82d JA |
146 | goto restart; |
147 | } | |
148 | ||
149 | static struct io_u *fio_posixaio_event(struct thread_data *td, int event) | |
150 | { | |
151 | struct posixaio_data *pd = td->io_ops->data; | |
152 | ||
153 | return pd->aio_events[event]; | |
154 | } | |
155 | ||
03e20d68 | 156 | static int fio_posixaio_queue(struct thread_data *td, |
2866c82d JA |
157 | struct io_u *io_u) |
158 | { | |
207cb0f0 | 159 | struct posixaio_data *pd = td->io_ops->data; |
2866c82d JA |
160 | struct aiocb *aiocb = &io_u->aiocb; |
161 | int ret; | |
162 | ||
7101d9c2 JA |
163 | fio_ro_check(td, io_u); |
164 | ||
2866c82d JA |
165 | if (io_u->ddir == DDIR_READ) |
166 | ret = aio_read(aiocb); | |
87dc1ab1 | 167 | else if (io_u->ddir == DDIR_WRITE) |
2866c82d | 168 | ret = aio_write(aiocb); |
a5f3027c JA |
169 | else if (io_u->ddir == DDIR_TRIM) { |
170 | if (pd->queued) | |
171 | return FIO_Q_BUSY; | |
172 | ||
173 | do_io_u_trim(td, io_u); | |
174 | return FIO_Q_COMPLETED; | |
175 | } else { | |
207cb0f0 | 176 | #ifdef FIO_HAVE_POSIXAIO_FSYNC |
87dc1ab1 | 177 | ret = aio_fsync(O_SYNC, aiocb); |
207cb0f0 JA |
178 | #else |
179 | if (pd->queued) | |
180 | return FIO_Q_BUSY; | |
181 | ||
f011531e | 182 | do_io_u_sync(td, io_u); |
207cb0f0 JA |
183 | return FIO_Q_COMPLETED; |
184 | #endif | |
185 | } | |
186 | ||
95bcd815 | 187 | if (ret) { |
afa16407 JA |
188 | /* |
189 | * At least OSX has a very low limit on the number of pending | |
def1d8e3 JA |
190 | * IOs, so if it returns EAGAIN, we are out of resources |
191 | * to queue more. Just return FIO_Q_BUSY to naturally | |
192 | * drop off at this depth. | |
afa16407 JA |
193 | */ |
194 | if (errno == EAGAIN) | |
195 | return FIO_Q_BUSY; | |
196 | ||
2866c82d | 197 | io_u->error = errno; |
e1161c32 | 198 | td_verror(td, io_u->error, "xfer"); |
36167d82 | 199 | return FIO_Q_COMPLETED; |
95bcd815 | 200 | } |
36167d82 | 201 | |
207cb0f0 | 202 | pd->queued++; |
36167d82 | 203 | return FIO_Q_QUEUED; |
2866c82d JA |
204 | } |
205 | ||
206 | static void fio_posixaio_cleanup(struct thread_data *td) | |
207 | { | |
208 | struct posixaio_data *pd = td->io_ops->data; | |
209 | ||
210 | if (pd) { | |
211 | free(pd->aio_events); | |
212 | free(pd); | |
2866c82d JA |
213 | } |
214 | } | |
215 | ||
216 | static int fio_posixaio_init(struct thread_data *td) | |
217 | { | |
218 | struct posixaio_data *pd = malloc(sizeof(*pd)); | |
219 | ||
cb781c75 | 220 | memset(pd, 0, sizeof(*pd)); |
2dc1bbeb JA |
221 | pd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u *)); |
222 | memset(pd->aio_events, 0, td->o.iodepth * sizeof(struct io_u *)); | |
2866c82d JA |
223 | |
224 | td->io_ops->data = pd; | |
225 | return 0; | |
226 | } | |
227 | ||
5f350952 | 228 | static struct ioengine_ops ioengine = { |
2866c82d JA |
229 | .name = "posixaio", |
230 | .version = FIO_IOOPS_VERSION, | |
231 | .init = fio_posixaio_init, | |
232 | .prep = fio_posixaio_prep, | |
233 | .queue = fio_posixaio_queue, | |
234 | .cancel = fio_posixaio_cancel, | |
235 | .getevents = fio_posixaio_getevents, | |
236 | .event = fio_posixaio_event, | |
237 | .cleanup = fio_posixaio_cleanup, | |
b5af8293 JA |
238 | .open_file = generic_open_file, |
239 | .close_file = generic_close_file, | |
df9c26b1 | 240 | .get_file_size = generic_get_file_size, |
2866c82d | 241 | }; |
34cfcdaf JA |
242 | |
243 | #else /* FIO_HAVE_POSIXAIO */ | |
244 | ||
245 | /* | |
246 | * When we have a proper configure system in place, we simply wont build | |
247 | * and install this io engine. For now install a crippled version that | |
248 | * just complains and fails to load. | |
249 | */ | |
250 | static int fio_posixaio_init(struct thread_data fio_unused *td) | |
251 | { | |
a3edaf76 | 252 | log_err("fio: posixaio not available\n"); |
34cfcdaf JA |
253 | return 1; |
254 | } | |
255 | ||
5f350952 | 256 | static struct ioengine_ops ioengine = { |
34cfcdaf JA |
257 | .name = "posixaio", |
258 | .version = FIO_IOOPS_VERSION, | |
259 | .init = fio_posixaio_init, | |
260 | }; | |
261 | ||
262 | #endif | |
5f350952 JA |
263 | |
264 | static void fio_init fio_posixaio_register(void) | |
265 | { | |
266 | register_ioengine(&ioengine); | |
267 | } | |
268 | ||
269 | static void fio_exit fio_posixaio_unregister(void) | |
270 | { | |
271 | unregister_ioengine(&ioengine); | |
272 | } |