]>
Commit | Line | Data |
---|---|---|
3c529d93 AL |
1 | /* |
2 | * QEMU posix-aio emulation | |
3 | * | |
4 | * Copyright IBM, Corp. 2008 | |
5 | * | |
6 | * Authors: | |
7 | * Anthony Liguori <aliguori@us.ibm.com> | |
8 | * | |
9 | * This work is licensed under the terms of the GNU GPL, version 2. See | |
10 | * the COPYING file in the top-level directory. | |
11 | * | |
12 | */ | |
13 | ||
221f715d | 14 | #include <sys/ioctl.h> |
9ef91a67 | 15 | #include <sys/types.h> |
3c529d93 AL |
16 | #include <pthread.h> |
17 | #include <unistd.h> | |
18 | #include <errno.h> | |
30525aff | 19 | #include <time.h> |
8653c015 | 20 | #include <string.h> |
21 | #include <stdlib.h> | |
22 | #include <stdio.h> | |
9ef91a67 | 23 | |
72cf2d4f | 24 | #include "qemu-queue.h" |
3c529d93 | 25 | #include "osdep.h" |
dc786bc9 | 26 | #include "sysemu.h" |
f141eafe | 27 | #include "qemu-common.h" |
6d519a5f | 28 | #include "trace.h" |
9ef91a67 CH |
29 | #include "block_int.h" |
30 | ||
31 | #include "block/raw-posix-aio.h" | |
32 | ||
33 | ||
34 | struct qemu_paiocb { | |
35 | BlockDriverAIOCB common; | |
36 | int aio_fildes; | |
37 | union { | |
38 | struct iovec *aio_iov; | |
b587a52c | 39 | void *aio_ioctl_buf; |
9ef91a67 CH |
40 | }; |
41 | int aio_niov; | |
42 | size_t aio_nbytes; | |
43 | #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */ | |
44 | int ev_signo; | |
45 | off_t aio_offset; | |
46 | ||
72cf2d4f | 47 | QTAILQ_ENTRY(qemu_paiocb) node; |
9ef91a67 CH |
48 | int aio_type; |
49 | ssize_t ret; | |
50 | int active; | |
51 | struct qemu_paiocb *next; | |
52 | }; | |
53 | ||
54 | typedef struct PosixAioState { | |
55 | int rfd, wfd; | |
56 | struct qemu_paiocb *first_aio; | |
57 | } PosixAioState; | |
3c529d93 | 58 | |
3c529d93 AL |
59 | |
60 | static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER; | |
61 | static pthread_cond_t cond = PTHREAD_COND_INITIALIZER; | |
62 | static pthread_t thread_id; | |
a8227a5a | 63 | static pthread_attr_t attr; |
3c529d93 AL |
64 | static int max_threads = 64; |
65 | static int cur_threads = 0; | |
66 | static int idle_threads = 0; | |
72cf2d4f | 67 | static QTAILQ_HEAD(, qemu_paiocb) request_list; |
3c529d93 | 68 | |
2341f9a1 | 69 | #ifdef CONFIG_PREADV |
ceb42de8 AL |
70 | static int preadv_present = 1; |
71 | #else | |
72 | static int preadv_present = 0; | |
73 | #endif | |
74 | ||
8653c015 | 75 | static void die2(int err, const char *what) |
76 | { | |
77 | fprintf(stderr, "%s failed: %s\n", what, strerror(err)); | |
78 | abort(); | |
79 | } | |
80 | ||
81 | static void die(const char *what) | |
82 | { | |
83 | die2(errno, what); | |
84 | } | |
85 | ||
86 | static void mutex_lock(pthread_mutex_t *mutex) | |
87 | { | |
88 | int ret = pthread_mutex_lock(mutex); | |
89 | if (ret) die2(ret, "pthread_mutex_lock"); | |
90 | } | |
91 | ||
92 | static void mutex_unlock(pthread_mutex_t *mutex) | |
93 | { | |
94 | int ret = pthread_mutex_unlock(mutex); | |
95 | if (ret) die2(ret, "pthread_mutex_unlock"); | |
96 | } | |
97 | ||
98 | static int cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex, | |
99 | struct timespec *ts) | |
100 | { | |
101 | int ret = pthread_cond_timedwait(cond, mutex, ts); | |
102 | if (ret && ret != ETIMEDOUT) die2(ret, "pthread_cond_timedwait"); | |
103 | return ret; | |
104 | } | |
105 | ||
5d47e372 | 106 | static void cond_signal(pthread_cond_t *cond) |
8653c015 | 107 | { |
5d47e372 | 108 | int ret = pthread_cond_signal(cond); |
109 | if (ret) die2(ret, "pthread_cond_signal"); | |
8653c015 | 110 | } |
111 | ||
112 | static void thread_create(pthread_t *thread, pthread_attr_t *attr, | |
113 | void *(*start_routine)(void*), void *arg) | |
114 | { | |
115 | int ret = pthread_create(thread, attr, start_routine, arg); | |
116 | if (ret) die2(ret, "pthread_create"); | |
117 | } | |
118 | ||
6769da29 | 119 | static ssize_t handle_aiocb_ioctl(struct qemu_paiocb *aiocb) |
f141eafe | 120 | { |
b587a52c SH |
121 | int ret; |
122 | ||
123 | ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); | |
124 | if (ret == -1) | |
125 | return -errno; | |
126 | ||
127 | /* | |
128 | * This looks weird, but the aio code only consideres a request | |
b0cd712c | 129 | * successful if it has written the number full number of bytes. |
b587a52c SH |
130 | * |
131 | * Now we overload aio_nbytes as aio_ioctl_cmd for the ioctl command, | |
132 | * so in fact we return the ioctl command here to make posix_aio_read() | |
133 | * happy.. | |
134 | */ | |
135 | return aiocb->aio_nbytes; | |
f141eafe AL |
136 | } |
137 | ||
6769da29 | 138 | static ssize_t handle_aiocb_flush(struct qemu_paiocb *aiocb) |
b2e12bc6 CH |
139 | { |
140 | int ret; | |
141 | ||
47faadc6 | 142 | ret = qemu_fdatasync(aiocb->aio_fildes); |
b2e12bc6 CH |
143 | if (ret == -1) |
144 | return -errno; | |
145 | return 0; | |
146 | } | |
147 | ||
2341f9a1 | 148 | #ifdef CONFIG_PREADV |
ceb42de8 AL |
149 | |
150 | static ssize_t | |
151 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
152 | { | |
153 | return preadv(fd, iov, nr_iov, offset); | |
154 | } | |
155 | ||
156 | static ssize_t | |
157 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
158 | { | |
159 | return pwritev(fd, iov, nr_iov, offset); | |
160 | } | |
161 | ||
162 | #else | |
163 | ||
164 | static ssize_t | |
165 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
166 | { | |
167 | return -ENOSYS; | |
168 | } | |
169 | ||
170 | static ssize_t | |
171 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
172 | { | |
173 | return -ENOSYS; | |
174 | } | |
175 | ||
176 | #endif | |
177 | ||
6769da29 | 178 | static ssize_t handle_aiocb_rw_vector(struct qemu_paiocb *aiocb) |
ceb42de8 AL |
179 | { |
180 | size_t offset = 0; | |
181 | ssize_t len; | |
182 | ||
183 | do { | |
9ef91a67 | 184 | if (aiocb->aio_type & QEMU_AIO_WRITE) |
ceb42de8 AL |
185 | len = qemu_pwritev(aiocb->aio_fildes, |
186 | aiocb->aio_iov, | |
187 | aiocb->aio_niov, | |
188 | aiocb->aio_offset + offset); | |
189 | else | |
190 | len = qemu_preadv(aiocb->aio_fildes, | |
191 | aiocb->aio_iov, | |
192 | aiocb->aio_niov, | |
193 | aiocb->aio_offset + offset); | |
194 | } while (len == -1 && errno == EINTR); | |
195 | ||
196 | if (len == -1) | |
197 | return -errno; | |
198 | return len; | |
199 | } | |
200 | ||
ba1d1afd KW |
201 | /* |
202 | * Read/writes the data to/from a given linear buffer. | |
203 | * | |
204 | * Returns the number of bytes handles or -errno in case of an error. Short | |
205 | * reads are only returned if the end of the file is reached. | |
206 | */ | |
6769da29 | 207 | static ssize_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf) |
221f715d | 208 | { |
6769da29 KW |
209 | ssize_t offset = 0; |
210 | ssize_t len; | |
221f715d AL |
211 | |
212 | while (offset < aiocb->aio_nbytes) { | |
9ef91a67 | 213 | if (aiocb->aio_type & QEMU_AIO_WRITE) |
f141eafe AL |
214 | len = pwrite(aiocb->aio_fildes, |
215 | (const char *)buf + offset, | |
216 | aiocb->aio_nbytes - offset, | |
217 | aiocb->aio_offset + offset); | |
218 | else | |
219 | len = pread(aiocb->aio_fildes, | |
220 | buf + offset, | |
221f715d AL |
221 | aiocb->aio_nbytes - offset, |
222 | aiocb->aio_offset + offset); | |
221f715d | 223 | |
f141eafe AL |
224 | if (len == -1 && errno == EINTR) |
225 | continue; | |
226 | else if (len == -1) { | |
227 | offset = -errno; | |
228 | break; | |
229 | } else if (len == 0) | |
230 | break; | |
231 | ||
232 | offset += len; | |
221f715d AL |
233 | } |
234 | ||
235 | return offset; | |
236 | } | |
237 | ||
6769da29 | 238 | static ssize_t handle_aiocb_rw(struct qemu_paiocb *aiocb) |
221f715d | 239 | { |
6769da29 | 240 | ssize_t nbytes; |
f141eafe AL |
241 | char *buf; |
242 | ||
9ef91a67 | 243 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { |
f141eafe AL |
244 | /* |
245 | * If there is just a single buffer, and it is properly aligned | |
246 | * we can just use plain pread/pwrite without any problems. | |
247 | */ | |
ceb42de8 AL |
248 | if (aiocb->aio_niov == 1) |
249 | return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); | |
250 | ||
251 | /* | |
252 | * We have more than one iovec, and all are properly aligned. | |
253 | * | |
254 | * Try preadv/pwritev first and fall back to linearizing the | |
255 | * buffer if it's not supported. | |
256 | */ | |
b587a52c | 257 | if (preadv_present) { |
ceb42de8 AL |
258 | nbytes = handle_aiocb_rw_vector(aiocb); |
259 | if (nbytes == aiocb->aio_nbytes) | |
b587a52c | 260 | return nbytes; |
ceb42de8 AL |
261 | if (nbytes < 0 && nbytes != -ENOSYS) |
262 | return nbytes; | |
263 | preadv_present = 0; | |
264 | } | |
265 | ||
266 | /* | |
267 | * XXX(hch): short read/write. no easy way to handle the reminder | |
268 | * using these interfaces. For now retry using plain | |
269 | * pread/pwrite? | |
270 | */ | |
f141eafe | 271 | } |
221f715d | 272 | |
f141eafe AL |
273 | /* |
274 | * Ok, we have to do it the hard way, copy all segments into | |
275 | * a single aligned buffer. | |
276 | */ | |
72aef731 | 277 | buf = qemu_blockalign(aiocb->common.bs, aiocb->aio_nbytes); |
9ef91a67 | 278 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
f141eafe AL |
279 | char *p = buf; |
280 | int i; | |
281 | ||
282 | for (i = 0; i < aiocb->aio_niov; ++i) { | |
283 | memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); | |
284 | p += aiocb->aio_iov[i].iov_len; | |
285 | } | |
286 | } | |
287 | ||
288 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
9ef91a67 | 289 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { |
f141eafe AL |
290 | char *p = buf; |
291 | size_t count = aiocb->aio_nbytes, copy; | |
292 | int i; | |
293 | ||
294 | for (i = 0; i < aiocb->aio_niov && count; ++i) { | |
295 | copy = count; | |
296 | if (copy > aiocb->aio_iov[i].iov_len) | |
297 | copy = aiocb->aio_iov[i].iov_len; | |
298 | memcpy(aiocb->aio_iov[i].iov_base, p, copy); | |
299 | p += copy; | |
300 | count -= copy; | |
301 | } | |
302 | } | |
303 | qemu_vfree(buf); | |
304 | ||
305 | return nbytes; | |
221f715d AL |
306 | } |
307 | ||
3c529d93 AL |
308 | static void *aio_thread(void *unused) |
309 | { | |
a8227a5a | 310 | pid_t pid; |
3c529d93 | 311 | |
a8227a5a | 312 | pid = getpid(); |
313 | ||
3c529d93 AL |
314 | while (1) { |
315 | struct qemu_paiocb *aiocb; | |
6769da29 | 316 | ssize_t ret = 0; |
30525aff | 317 | qemu_timeval tv; |
318 | struct timespec ts; | |
319 | ||
320 | qemu_gettimeofday(&tv); | |
321 | ts.tv_sec = tv.tv_sec + 10; | |
322 | ts.tv_nsec = 0; | |
3c529d93 | 323 | |
8653c015 | 324 | mutex_lock(&lock); |
3c529d93 | 325 | |
72cf2d4f | 326 | while (QTAILQ_EMPTY(&request_list) && |
3c529d93 | 327 | !(ret == ETIMEDOUT)) { |
5be4aab7 | 328 | idle_threads++; |
8653c015 | 329 | ret = cond_timedwait(&cond, &lock, &ts); |
5be4aab7 | 330 | idle_threads--; |
3c529d93 AL |
331 | } |
332 | ||
72cf2d4f | 333 | if (QTAILQ_EMPTY(&request_list)) |
3c529d93 AL |
334 | break; |
335 | ||
72cf2d4f BS |
336 | aiocb = QTAILQ_FIRST(&request_list); |
337 | QTAILQ_REMOVE(&request_list, aiocb, node); | |
3c529d93 | 338 | aiocb->active = 1; |
8653c015 | 339 | mutex_unlock(&lock); |
3c529d93 | 340 | |
9ef91a67 CH |
341 | switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { |
342 | case QEMU_AIO_READ: | |
ba1d1afd KW |
343 | ret = handle_aiocb_rw(aiocb); |
344 | if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->common.bs->growable) { | |
345 | /* A short read means that we have reached EOF. Pad the buffer | |
346 | * with zeros for bytes after EOF. */ | |
347 | QEMUIOVector qiov; | |
348 | ||
349 | qemu_iovec_init_external(&qiov, aiocb->aio_iov, | |
350 | aiocb->aio_niov); | |
351 | qemu_iovec_memset_skip(&qiov, 0, aiocb->aio_nbytes - ret, ret); | |
352 | ||
353 | ret = aiocb->aio_nbytes; | |
354 | } | |
355 | break; | |
9ef91a67 | 356 | case QEMU_AIO_WRITE: |
b587a52c SH |
357 | ret = handle_aiocb_rw(aiocb); |
358 | break; | |
b2e12bc6 | 359 | case QEMU_AIO_FLUSH: |
b587a52c SH |
360 | ret = handle_aiocb_flush(aiocb); |
361 | break; | |
9ef91a67 | 362 | case QEMU_AIO_IOCTL: |
b587a52c SH |
363 | ret = handle_aiocb_ioctl(aiocb); |
364 | break; | |
365 | default: | |
366 | fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); | |
367 | ret = -EINVAL; | |
368 | break; | |
369 | } | |
3c529d93 | 370 | |
8653c015 | 371 | mutex_lock(&lock); |
221f715d | 372 | aiocb->ret = ret; |
8653c015 | 373 | mutex_unlock(&lock); |
3c529d93 | 374 | |
a8227a5a | 375 | if (kill(pid, aiocb->ev_signo)) die("kill failed"); |
3c529d93 AL |
376 | } |
377 | ||
3c529d93 | 378 | cur_threads--; |
8653c015 | 379 | mutex_unlock(&lock); |
3c529d93 AL |
380 | |
381 | return NULL; | |
382 | } | |
383 | ||
8653c015 | 384 | static void spawn_thread(void) |
3c529d93 | 385 | { |
ee399306 | 386 | sigset_t set, oldset; |
387 | ||
3c529d93 | 388 | cur_threads++; |
ee399306 | 389 | |
390 | /* block all signals */ | |
391 | if (sigfillset(&set)) die("sigfillset"); | |
392 | if (sigprocmask(SIG_SETMASK, &set, &oldset)) die("sigprocmask"); | |
393 | ||
8653c015 | 394 | thread_create(&thread_id, &attr, aio_thread, NULL); |
ee399306 | 395 | |
396 | if (sigprocmask(SIG_SETMASK, &oldset, NULL)) die("sigprocmask restore"); | |
3c529d93 AL |
397 | } |
398 | ||
9ef91a67 | 399 | static void qemu_paio_submit(struct qemu_paiocb *aiocb) |
3c529d93 | 400 | { |
3c529d93 AL |
401 | aiocb->ret = -EINPROGRESS; |
402 | aiocb->active = 0; | |
8653c015 | 403 | mutex_lock(&lock); |
3c529d93 AL |
404 | if (idle_threads == 0 && cur_threads < max_threads) |
405 | spawn_thread(); | |
72cf2d4f | 406 | QTAILQ_INSERT_TAIL(&request_list, aiocb, node); |
8653c015 | 407 | mutex_unlock(&lock); |
5d47e372 | 408 | cond_signal(&cond); |
3c529d93 AL |
409 | } |
410 | ||
9ef91a67 | 411 | static ssize_t qemu_paio_return(struct qemu_paiocb *aiocb) |
3c529d93 AL |
412 | { |
413 | ssize_t ret; | |
414 | ||
8653c015 | 415 | mutex_lock(&lock); |
3c529d93 | 416 | ret = aiocb->ret; |
8653c015 | 417 | mutex_unlock(&lock); |
3c529d93 AL |
418 | |
419 | return ret; | |
420 | } | |
421 | ||
9ef91a67 | 422 | static int qemu_paio_error(struct qemu_paiocb *aiocb) |
3c529d93 AL |
423 | { |
424 | ssize_t ret = qemu_paio_return(aiocb); | |
425 | ||
426 | if (ret < 0) | |
427 | ret = -ret; | |
428 | else | |
429 | ret = 0; | |
430 | ||
431 | return ret; | |
432 | } | |
433 | ||
59c7b155 | 434 | static int posix_aio_process_queue(void *opaque) |
3c529d93 | 435 | { |
9ef91a67 CH |
436 | PosixAioState *s = opaque; |
437 | struct qemu_paiocb *acb, **pacb; | |
3c529d93 | 438 | int ret; |
59c7b155 | 439 | int result = 0; |
9ef91a67 CH |
440 | |
441 | for(;;) { | |
442 | pacb = &s->first_aio; | |
443 | for(;;) { | |
444 | acb = *pacb; | |
445 | if (!acb) | |
59c7b155 | 446 | return result; |
e5f37649 | 447 | |
9ef91a67 CH |
448 | ret = qemu_paio_error(acb); |
449 | if (ret == ECANCELED) { | |
450 | /* remove the request */ | |
451 | *pacb = acb->next; | |
452 | qemu_aio_release(acb); | |
59c7b155 | 453 | result = 1; |
9ef91a67 CH |
454 | } else if (ret != EINPROGRESS) { |
455 | /* end of aio */ | |
456 | if (ret == 0) { | |
457 | ret = qemu_paio_return(acb); | |
458 | if (ret == acb->aio_nbytes) | |
459 | ret = 0; | |
460 | else | |
461 | ret = -EINVAL; | |
462 | } else { | |
463 | ret = -ret; | |
464 | } | |
ddca9fb2 SH |
465 | |
466 | trace_paio_complete(acb, acb->common.opaque, ret); | |
467 | ||
9ef91a67 CH |
468 | /* remove the request */ |
469 | *pacb = acb->next; | |
470 | /* call the callback */ | |
471 | acb->common.cb(acb->common.opaque, ret); | |
472 | qemu_aio_release(acb); | |
59c7b155 | 473 | result = 1; |
9ef91a67 CH |
474 | break; |
475 | } else { | |
476 | pacb = &acb->next; | |
477 | } | |
478 | } | |
479 | } | |
59c7b155 KW |
480 | |
481 | return result; | |
482 | } | |
483 | ||
484 | static void posix_aio_read(void *opaque) | |
485 | { | |
486 | PosixAioState *s = opaque; | |
487 | ssize_t len; | |
488 | ||
489 | /* read all bytes from signal pipe */ | |
490 | for (;;) { | |
491 | char bytes[16]; | |
492 | ||
493 | len = read(s->rfd, bytes, sizeof(bytes)); | |
494 | if (len == -1 && errno == EINTR) | |
495 | continue; /* try again */ | |
496 | if (len == sizeof(bytes)) | |
497 | continue; /* more to read */ | |
498 | break; | |
499 | } | |
500 | ||
501 | posix_aio_process_queue(s); | |
9ef91a67 CH |
502 | } |
503 | ||
504 | static int posix_aio_flush(void *opaque) | |
505 | { | |
506 | PosixAioState *s = opaque; | |
507 | return !!s->first_aio; | |
508 | } | |
509 | ||
510 | static PosixAioState *posix_aio_state; | |
511 | ||
512 | static void aio_signal_handler(int signum) | |
513 | { | |
514 | if (posix_aio_state) { | |
515 | char byte = 0; | |
4817d327 | 516 | ssize_t ret; |
9ef91a67 | 517 | |
4817d327 KS |
518 | ret = write(posix_aio_state->wfd, &byte, sizeof(byte)); |
519 | if (ret < 0 && errno != EAGAIN) | |
520 | die("write()"); | |
9ef91a67 CH |
521 | } |
522 | ||
523 | qemu_service_io(); | |
524 | } | |
525 | ||
526 | static void paio_remove(struct qemu_paiocb *acb) | |
527 | { | |
528 | struct qemu_paiocb **pacb; | |
529 | ||
530 | /* remove the callback from the queue */ | |
531 | pacb = &posix_aio_state->first_aio; | |
532 | for(;;) { | |
533 | if (*pacb == NULL) { | |
534 | fprintf(stderr, "paio_remove: aio request not found!\n"); | |
535 | break; | |
536 | } else if (*pacb == acb) { | |
537 | *pacb = acb->next; | |
538 | qemu_aio_release(acb); | |
539 | break; | |
540 | } | |
541 | pacb = &(*pacb)->next; | |
542 | } | |
543 | } | |
544 | ||
545 | static void paio_cancel(BlockDriverAIOCB *blockacb) | |
546 | { | |
547 | struct qemu_paiocb *acb = (struct qemu_paiocb *)blockacb; | |
548 | int active = 0; | |
3c529d93 | 549 | |
ddca9fb2 SH |
550 | trace_paio_cancel(acb, acb->common.opaque); |
551 | ||
8653c015 | 552 | mutex_lock(&lock); |
9ef91a67 | 553 | if (!acb->active) { |
72cf2d4f | 554 | QTAILQ_REMOVE(&request_list, acb, node); |
9ef91a67 CH |
555 | acb->ret = -ECANCELED; |
556 | } else if (acb->ret == -EINPROGRESS) { | |
557 | active = 1; | |
558 | } | |
8653c015 | 559 | mutex_unlock(&lock); |
3c529d93 | 560 | |
9ef91a67 CH |
561 | if (active) { |
562 | /* fail safe: if the aio could not be canceled, we wait for | |
563 | it */ | |
564 | while (qemu_paio_error(acb) == EINPROGRESS) | |
565 | ; | |
566 | } | |
567 | ||
568 | paio_remove(acb); | |
569 | } | |
570 | ||
571 | static AIOPool raw_aio_pool = { | |
572 | .aiocb_size = sizeof(struct qemu_paiocb), | |
573 | .cancel = paio_cancel, | |
574 | }; | |
575 | ||
1e5b9d2f | 576 | BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd, |
9ef91a67 CH |
577 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
578 | BlockDriverCompletionFunc *cb, void *opaque, int type) | |
579 | { | |
580 | struct qemu_paiocb *acb; | |
581 | ||
582 | acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque); | |
583 | if (!acb) | |
584 | return NULL; | |
585 | acb->aio_type = type; | |
586 | acb->aio_fildes = fd; | |
587 | acb->ev_signo = SIGUSR2; | |
e5f37649 | 588 | |
b2e12bc6 CH |
589 | if (qiov) { |
590 | acb->aio_iov = qiov->iov; | |
591 | acb->aio_niov = qiov->niov; | |
592 | } | |
9ef91a67 CH |
593 | acb->aio_nbytes = nb_sectors * 512; |
594 | acb->aio_offset = sector_num * 512; | |
595 | ||
596 | acb->next = posix_aio_state->first_aio; | |
597 | posix_aio_state->first_aio = acb; | |
598 | ||
6d519a5f | 599 | trace_paio_submit(acb, opaque, sector_num, nb_sectors, type); |
9ef91a67 CH |
600 | qemu_paio_submit(acb); |
601 | return &acb->common; | |
602 | } | |
603 | ||
604 | BlockDriverAIOCB *paio_ioctl(BlockDriverState *bs, int fd, | |
605 | unsigned long int req, void *buf, | |
606 | BlockDriverCompletionFunc *cb, void *opaque) | |
607 | { | |
608 | struct qemu_paiocb *acb; | |
609 | ||
610 | acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque); | |
611 | if (!acb) | |
612 | return NULL; | |
613 | acb->aio_type = QEMU_AIO_IOCTL; | |
614 | acb->aio_fildes = fd; | |
615 | acb->ev_signo = SIGUSR2; | |
616 | acb->aio_offset = 0; | |
617 | acb->aio_ioctl_buf = buf; | |
618 | acb->aio_ioctl_cmd = req; | |
619 | ||
620 | acb->next = posix_aio_state->first_aio; | |
621 | posix_aio_state->first_aio = acb; | |
622 | ||
623 | qemu_paio_submit(acb); | |
624 | return &acb->common; | |
625 | } | |
626 | ||
1e5b9d2f | 627 | int paio_init(void) |
9ef91a67 CH |
628 | { |
629 | struct sigaction act; | |
630 | PosixAioState *s; | |
631 | int fds[2]; | |
632 | int ret; | |
633 | ||
634 | if (posix_aio_state) | |
1e5b9d2f | 635 | return 0; |
9ef91a67 CH |
636 | |
637 | s = qemu_malloc(sizeof(PosixAioState)); | |
638 | ||
639 | sigfillset(&act.sa_mask); | |
640 | act.sa_flags = 0; /* do not restart syscalls to interrupt select() */ | |
641 | act.sa_handler = aio_signal_handler; | |
642 | sigaction(SIGUSR2, &act, NULL); | |
643 | ||
644 | s->first_aio = NULL; | |
40ff6d7e | 645 | if (qemu_pipe(fds) == -1) { |
9ef91a67 | 646 | fprintf(stderr, "failed to create pipe\n"); |
1e5b9d2f | 647 | return -1; |
9ef91a67 CH |
648 | } |
649 | ||
650 | s->rfd = fds[0]; | |
651 | s->wfd = fds[1]; | |
652 | ||
653 | fcntl(s->rfd, F_SETFL, O_NONBLOCK); | |
654 | fcntl(s->wfd, F_SETFL, O_NONBLOCK); | |
655 | ||
8febfa26 KW |
656 | qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, |
657 | posix_aio_process_queue, s); | |
9ef91a67 CH |
658 | |
659 | ret = pthread_attr_init(&attr); | |
660 | if (ret) | |
661 | die2(ret, "pthread_attr_init"); | |
662 | ||
663 | ret = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); | |
664 | if (ret) | |
665 | die2(ret, "pthread_attr_setdetachstate"); | |
666 | ||
72cf2d4f | 667 | QTAILQ_INIT(&request_list); |
9ef91a67 CH |
668 | |
669 | posix_aio_state = s; | |
1e5b9d2f | 670 | return 0; |
3c529d93 | 671 | } |