]>
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> |
9ef91a67 | 20 | #include <signal.h> |
8653c015 | 21 | #include <string.h> |
22 | #include <stdlib.h> | |
23 | #include <stdio.h> | |
9ef91a67 | 24 | |
72cf2d4f | 25 | #include "qemu-queue.h" |
3c529d93 | 26 | #include "osdep.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; | |
e5f37649 KW |
52 | |
53 | int async_context_id; | |
9ef91a67 CH |
54 | }; |
55 | ||
56 | typedef struct PosixAioState { | |
57 | int rfd, wfd; | |
58 | struct qemu_paiocb *first_aio; | |
59 | } PosixAioState; | |
3c529d93 | 60 | |
3c529d93 AL |
61 | |
62 | static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER; | |
63 | static pthread_cond_t cond = PTHREAD_COND_INITIALIZER; | |
64 | static pthread_t thread_id; | |
a8227a5a | 65 | static pthread_attr_t attr; |
3c529d93 AL |
66 | static int max_threads = 64; |
67 | static int cur_threads = 0; | |
68 | static int idle_threads = 0; | |
72cf2d4f | 69 | static QTAILQ_HEAD(, qemu_paiocb) request_list; |
3c529d93 | 70 | |
2341f9a1 | 71 | #ifdef CONFIG_PREADV |
ceb42de8 AL |
72 | static int preadv_present = 1; |
73 | #else | |
74 | static int preadv_present = 0; | |
75 | #endif | |
76 | ||
8653c015 | 77 | static void die2(int err, const char *what) |
78 | { | |
79 | fprintf(stderr, "%s failed: %s\n", what, strerror(err)); | |
80 | abort(); | |
81 | } | |
82 | ||
83 | static void die(const char *what) | |
84 | { | |
85 | die2(errno, what); | |
86 | } | |
87 | ||
88 | static void mutex_lock(pthread_mutex_t *mutex) | |
89 | { | |
90 | int ret = pthread_mutex_lock(mutex); | |
91 | if (ret) die2(ret, "pthread_mutex_lock"); | |
92 | } | |
93 | ||
94 | static void mutex_unlock(pthread_mutex_t *mutex) | |
95 | { | |
96 | int ret = pthread_mutex_unlock(mutex); | |
97 | if (ret) die2(ret, "pthread_mutex_unlock"); | |
98 | } | |
99 | ||
100 | static int cond_timedwait(pthread_cond_t *cond, pthread_mutex_t *mutex, | |
101 | struct timespec *ts) | |
102 | { | |
103 | int ret = pthread_cond_timedwait(cond, mutex, ts); | |
104 | if (ret && ret != ETIMEDOUT) die2(ret, "pthread_cond_timedwait"); | |
105 | return ret; | |
106 | } | |
107 | ||
5d47e372 | 108 | static void cond_signal(pthread_cond_t *cond) |
8653c015 | 109 | { |
5d47e372 | 110 | int ret = pthread_cond_signal(cond); |
111 | if (ret) die2(ret, "pthread_cond_signal"); | |
8653c015 | 112 | } |
113 | ||
114 | static void thread_create(pthread_t *thread, pthread_attr_t *attr, | |
115 | void *(*start_routine)(void*), void *arg) | |
116 | { | |
117 | int ret = pthread_create(thread, attr, start_routine, arg); | |
118 | if (ret) die2(ret, "pthread_create"); | |
119 | } | |
120 | ||
6769da29 | 121 | static ssize_t handle_aiocb_ioctl(struct qemu_paiocb *aiocb) |
f141eafe | 122 | { |
b587a52c SH |
123 | int ret; |
124 | ||
125 | ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf); | |
126 | if (ret == -1) | |
127 | return -errno; | |
128 | ||
129 | /* | |
130 | * This looks weird, but the aio code only consideres a request | |
131 | * successfull if it has written the number full number of bytes. | |
132 | * | |
133 | * Now we overload aio_nbytes as aio_ioctl_cmd for the ioctl command, | |
134 | * so in fact we return the ioctl command here to make posix_aio_read() | |
135 | * happy.. | |
136 | */ | |
137 | return aiocb->aio_nbytes; | |
f141eafe AL |
138 | } |
139 | ||
6769da29 | 140 | static ssize_t handle_aiocb_flush(struct qemu_paiocb *aiocb) |
b2e12bc6 CH |
141 | { |
142 | int ret; | |
143 | ||
47faadc6 | 144 | ret = qemu_fdatasync(aiocb->aio_fildes); |
b2e12bc6 CH |
145 | if (ret == -1) |
146 | return -errno; | |
147 | return 0; | |
148 | } | |
149 | ||
2341f9a1 | 150 | #ifdef CONFIG_PREADV |
ceb42de8 AL |
151 | |
152 | static ssize_t | |
153 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
154 | { | |
155 | return preadv(fd, iov, nr_iov, offset); | |
156 | } | |
157 | ||
158 | static ssize_t | |
159 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
160 | { | |
161 | return pwritev(fd, iov, nr_iov, offset); | |
162 | } | |
163 | ||
164 | #else | |
165 | ||
166 | static ssize_t | |
167 | qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
168 | { | |
169 | return -ENOSYS; | |
170 | } | |
171 | ||
172 | static ssize_t | |
173 | qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset) | |
174 | { | |
175 | return -ENOSYS; | |
176 | } | |
177 | ||
178 | #endif | |
179 | ||
6769da29 | 180 | static ssize_t handle_aiocb_rw_vector(struct qemu_paiocb *aiocb) |
ceb42de8 AL |
181 | { |
182 | size_t offset = 0; | |
183 | ssize_t len; | |
184 | ||
185 | do { | |
9ef91a67 | 186 | if (aiocb->aio_type & QEMU_AIO_WRITE) |
ceb42de8 AL |
187 | len = qemu_pwritev(aiocb->aio_fildes, |
188 | aiocb->aio_iov, | |
189 | aiocb->aio_niov, | |
190 | aiocb->aio_offset + offset); | |
191 | else | |
192 | len = qemu_preadv(aiocb->aio_fildes, | |
193 | aiocb->aio_iov, | |
194 | aiocb->aio_niov, | |
195 | aiocb->aio_offset + offset); | |
196 | } while (len == -1 && errno == EINTR); | |
197 | ||
198 | if (len == -1) | |
199 | return -errno; | |
200 | return len; | |
201 | } | |
202 | ||
6769da29 | 203 | static ssize_t handle_aiocb_rw_linear(struct qemu_paiocb *aiocb, char *buf) |
221f715d | 204 | { |
6769da29 KW |
205 | ssize_t offset = 0; |
206 | ssize_t len; | |
221f715d AL |
207 | |
208 | while (offset < aiocb->aio_nbytes) { | |
9ef91a67 | 209 | if (aiocb->aio_type & QEMU_AIO_WRITE) |
f141eafe AL |
210 | len = pwrite(aiocb->aio_fildes, |
211 | (const char *)buf + offset, | |
212 | aiocb->aio_nbytes - offset, | |
213 | aiocb->aio_offset + offset); | |
214 | else | |
215 | len = pread(aiocb->aio_fildes, | |
216 | buf + offset, | |
221f715d AL |
217 | aiocb->aio_nbytes - offset, |
218 | aiocb->aio_offset + offset); | |
221f715d | 219 | |
f141eafe AL |
220 | if (len == -1 && errno == EINTR) |
221 | continue; | |
222 | else if (len == -1) { | |
223 | offset = -errno; | |
224 | break; | |
225 | } else if (len == 0) | |
226 | break; | |
227 | ||
228 | offset += len; | |
221f715d AL |
229 | } |
230 | ||
231 | return offset; | |
232 | } | |
233 | ||
6769da29 | 234 | static ssize_t handle_aiocb_rw(struct qemu_paiocb *aiocb) |
221f715d | 235 | { |
6769da29 | 236 | ssize_t nbytes; |
f141eafe AL |
237 | char *buf; |
238 | ||
9ef91a67 | 239 | if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) { |
f141eafe AL |
240 | /* |
241 | * If there is just a single buffer, and it is properly aligned | |
242 | * we can just use plain pread/pwrite without any problems. | |
243 | */ | |
ceb42de8 AL |
244 | if (aiocb->aio_niov == 1) |
245 | return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base); | |
246 | ||
247 | /* | |
248 | * We have more than one iovec, and all are properly aligned. | |
249 | * | |
250 | * Try preadv/pwritev first and fall back to linearizing the | |
251 | * buffer if it's not supported. | |
252 | */ | |
b587a52c | 253 | if (preadv_present) { |
ceb42de8 AL |
254 | nbytes = handle_aiocb_rw_vector(aiocb); |
255 | if (nbytes == aiocb->aio_nbytes) | |
b587a52c | 256 | return nbytes; |
ceb42de8 AL |
257 | if (nbytes < 0 && nbytes != -ENOSYS) |
258 | return nbytes; | |
259 | preadv_present = 0; | |
260 | } | |
261 | ||
262 | /* | |
263 | * XXX(hch): short read/write. no easy way to handle the reminder | |
264 | * using these interfaces. For now retry using plain | |
265 | * pread/pwrite? | |
266 | */ | |
f141eafe | 267 | } |
221f715d | 268 | |
f141eafe AL |
269 | /* |
270 | * Ok, we have to do it the hard way, copy all segments into | |
271 | * a single aligned buffer. | |
272 | */ | |
273 | buf = qemu_memalign(512, aiocb->aio_nbytes); | |
9ef91a67 | 274 | if (aiocb->aio_type & QEMU_AIO_WRITE) { |
f141eafe AL |
275 | char *p = buf; |
276 | int i; | |
277 | ||
278 | for (i = 0; i < aiocb->aio_niov; ++i) { | |
279 | memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len); | |
280 | p += aiocb->aio_iov[i].iov_len; | |
281 | } | |
282 | } | |
283 | ||
284 | nbytes = handle_aiocb_rw_linear(aiocb, buf); | |
9ef91a67 | 285 | if (!(aiocb->aio_type & QEMU_AIO_WRITE)) { |
f141eafe AL |
286 | char *p = buf; |
287 | size_t count = aiocb->aio_nbytes, copy; | |
288 | int i; | |
289 | ||
290 | for (i = 0; i < aiocb->aio_niov && count; ++i) { | |
291 | copy = count; | |
292 | if (copy > aiocb->aio_iov[i].iov_len) | |
293 | copy = aiocb->aio_iov[i].iov_len; | |
294 | memcpy(aiocb->aio_iov[i].iov_base, p, copy); | |
295 | p += copy; | |
296 | count -= copy; | |
297 | } | |
298 | } | |
299 | qemu_vfree(buf); | |
300 | ||
301 | return nbytes; | |
221f715d AL |
302 | } |
303 | ||
3c529d93 AL |
304 | static void *aio_thread(void *unused) |
305 | { | |
a8227a5a | 306 | pid_t pid; |
3c529d93 | 307 | |
a8227a5a | 308 | pid = getpid(); |
309 | ||
3c529d93 AL |
310 | while (1) { |
311 | struct qemu_paiocb *aiocb; | |
6769da29 | 312 | ssize_t ret = 0; |
30525aff | 313 | qemu_timeval tv; |
314 | struct timespec ts; | |
315 | ||
316 | qemu_gettimeofday(&tv); | |
317 | ts.tv_sec = tv.tv_sec + 10; | |
318 | ts.tv_nsec = 0; | |
3c529d93 | 319 | |
8653c015 | 320 | mutex_lock(&lock); |
3c529d93 | 321 | |
72cf2d4f | 322 | while (QTAILQ_EMPTY(&request_list) && |
3c529d93 | 323 | !(ret == ETIMEDOUT)) { |
8653c015 | 324 | ret = cond_timedwait(&cond, &lock, &ts); |
3c529d93 AL |
325 | } |
326 | ||
72cf2d4f | 327 | if (QTAILQ_EMPTY(&request_list)) |
3c529d93 AL |
328 | break; |
329 | ||
72cf2d4f BS |
330 | aiocb = QTAILQ_FIRST(&request_list); |
331 | QTAILQ_REMOVE(&request_list, aiocb, node); | |
3c529d93 | 332 | aiocb->active = 1; |
3c529d93 | 333 | idle_threads--; |
8653c015 | 334 | mutex_unlock(&lock); |
3c529d93 | 335 | |
9ef91a67 CH |
336 | switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) { |
337 | case QEMU_AIO_READ: | |
338 | case QEMU_AIO_WRITE: | |
b587a52c SH |
339 | ret = handle_aiocb_rw(aiocb); |
340 | break; | |
b2e12bc6 | 341 | case QEMU_AIO_FLUSH: |
b587a52c SH |
342 | ret = handle_aiocb_flush(aiocb); |
343 | break; | |
9ef91a67 | 344 | case QEMU_AIO_IOCTL: |
b587a52c SH |
345 | ret = handle_aiocb_ioctl(aiocb); |
346 | break; | |
347 | default: | |
348 | fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type); | |
349 | ret = -EINVAL; | |
350 | break; | |
351 | } | |
3c529d93 | 352 | |
8653c015 | 353 | mutex_lock(&lock); |
221f715d | 354 | aiocb->ret = ret; |
3c529d93 | 355 | idle_threads++; |
8653c015 | 356 | mutex_unlock(&lock); |
3c529d93 | 357 | |
a8227a5a | 358 | if (kill(pid, aiocb->ev_signo)) die("kill failed"); |
3c529d93 AL |
359 | } |
360 | ||
361 | idle_threads--; | |
362 | cur_threads--; | |
8653c015 | 363 | mutex_unlock(&lock); |
3c529d93 AL |
364 | |
365 | return NULL; | |
366 | } | |
367 | ||
8653c015 | 368 | static void spawn_thread(void) |
3c529d93 | 369 | { |
ee399306 | 370 | sigset_t set, oldset; |
371 | ||
3c529d93 AL |
372 | cur_threads++; |
373 | idle_threads++; | |
ee399306 | 374 | |
375 | /* block all signals */ | |
376 | if (sigfillset(&set)) die("sigfillset"); | |
377 | if (sigprocmask(SIG_SETMASK, &set, &oldset)) die("sigprocmask"); | |
378 | ||
8653c015 | 379 | thread_create(&thread_id, &attr, aio_thread, NULL); |
ee399306 | 380 | |
381 | if (sigprocmask(SIG_SETMASK, &oldset, NULL)) die("sigprocmask restore"); | |
3c529d93 AL |
382 | } |
383 | ||
9ef91a67 | 384 | static void qemu_paio_submit(struct qemu_paiocb *aiocb) |
3c529d93 | 385 | { |
3c529d93 AL |
386 | aiocb->ret = -EINPROGRESS; |
387 | aiocb->active = 0; | |
8653c015 | 388 | mutex_lock(&lock); |
3c529d93 AL |
389 | if (idle_threads == 0 && cur_threads < max_threads) |
390 | spawn_thread(); | |
72cf2d4f | 391 | QTAILQ_INSERT_TAIL(&request_list, aiocb, node); |
8653c015 | 392 | mutex_unlock(&lock); |
5d47e372 | 393 | cond_signal(&cond); |
3c529d93 AL |
394 | } |
395 | ||
9ef91a67 | 396 | static ssize_t qemu_paio_return(struct qemu_paiocb *aiocb) |
3c529d93 AL |
397 | { |
398 | ssize_t ret; | |
399 | ||
8653c015 | 400 | mutex_lock(&lock); |
3c529d93 | 401 | ret = aiocb->ret; |
8653c015 | 402 | mutex_unlock(&lock); |
3c529d93 AL |
403 | |
404 | return ret; | |
405 | } | |
406 | ||
9ef91a67 | 407 | static int qemu_paio_error(struct qemu_paiocb *aiocb) |
3c529d93 AL |
408 | { |
409 | ssize_t ret = qemu_paio_return(aiocb); | |
410 | ||
411 | if (ret < 0) | |
412 | ret = -ret; | |
413 | else | |
414 | ret = 0; | |
415 | ||
416 | return ret; | |
417 | } | |
418 | ||
59c7b155 | 419 | static int posix_aio_process_queue(void *opaque) |
3c529d93 | 420 | { |
9ef91a67 CH |
421 | PosixAioState *s = opaque; |
422 | struct qemu_paiocb *acb, **pacb; | |
3c529d93 | 423 | int ret; |
59c7b155 | 424 | int result = 0; |
e5f37649 | 425 | int async_context_id = get_async_context_id(); |
9ef91a67 CH |
426 | |
427 | for(;;) { | |
428 | pacb = &s->first_aio; | |
429 | for(;;) { | |
430 | acb = *pacb; | |
431 | if (!acb) | |
59c7b155 | 432 | return result; |
e5f37649 KW |
433 | |
434 | /* we're only interested in requests in the right context */ | |
435 | if (acb->async_context_id != async_context_id) { | |
436 | pacb = &acb->next; | |
437 | continue; | |
438 | } | |
439 | ||
9ef91a67 CH |
440 | ret = qemu_paio_error(acb); |
441 | if (ret == ECANCELED) { | |
442 | /* remove the request */ | |
443 | *pacb = acb->next; | |
444 | qemu_aio_release(acb); | |
59c7b155 | 445 | result = 1; |
9ef91a67 CH |
446 | } else if (ret != EINPROGRESS) { |
447 | /* end of aio */ | |
448 | if (ret == 0) { | |
449 | ret = qemu_paio_return(acb); | |
450 | if (ret == acb->aio_nbytes) | |
451 | ret = 0; | |
452 | else | |
453 | ret = -EINVAL; | |
454 | } else { | |
455 | ret = -ret; | |
456 | } | |
457 | /* remove the request */ | |
458 | *pacb = acb->next; | |
459 | /* call the callback */ | |
460 | acb->common.cb(acb->common.opaque, ret); | |
461 | qemu_aio_release(acb); | |
59c7b155 | 462 | result = 1; |
9ef91a67 CH |
463 | break; |
464 | } else { | |
465 | pacb = &acb->next; | |
466 | } | |
467 | } | |
468 | } | |
59c7b155 KW |
469 | |
470 | return result; | |
471 | } | |
472 | ||
473 | static void posix_aio_read(void *opaque) | |
474 | { | |
475 | PosixAioState *s = opaque; | |
476 | ssize_t len; | |
477 | ||
478 | /* read all bytes from signal pipe */ | |
479 | for (;;) { | |
480 | char bytes[16]; | |
481 | ||
482 | len = read(s->rfd, bytes, sizeof(bytes)); | |
483 | if (len == -1 && errno == EINTR) | |
484 | continue; /* try again */ | |
485 | if (len == sizeof(bytes)) | |
486 | continue; /* more to read */ | |
487 | break; | |
488 | } | |
489 | ||
490 | posix_aio_process_queue(s); | |
9ef91a67 CH |
491 | } |
492 | ||
493 | static int posix_aio_flush(void *opaque) | |
494 | { | |
495 | PosixAioState *s = opaque; | |
496 | return !!s->first_aio; | |
497 | } | |
498 | ||
499 | static PosixAioState *posix_aio_state; | |
500 | ||
501 | static void aio_signal_handler(int signum) | |
502 | { | |
503 | if (posix_aio_state) { | |
504 | char byte = 0; | |
4817d327 | 505 | ssize_t ret; |
9ef91a67 | 506 | |
4817d327 KS |
507 | ret = write(posix_aio_state->wfd, &byte, sizeof(byte)); |
508 | if (ret < 0 && errno != EAGAIN) | |
509 | die("write()"); | |
9ef91a67 CH |
510 | } |
511 | ||
512 | qemu_service_io(); | |
513 | } | |
514 | ||
515 | static void paio_remove(struct qemu_paiocb *acb) | |
516 | { | |
517 | struct qemu_paiocb **pacb; | |
518 | ||
519 | /* remove the callback from the queue */ | |
520 | pacb = &posix_aio_state->first_aio; | |
521 | for(;;) { | |
522 | if (*pacb == NULL) { | |
523 | fprintf(stderr, "paio_remove: aio request not found!\n"); | |
524 | break; | |
525 | } else if (*pacb == acb) { | |
526 | *pacb = acb->next; | |
527 | qemu_aio_release(acb); | |
528 | break; | |
529 | } | |
530 | pacb = &(*pacb)->next; | |
531 | } | |
532 | } | |
533 | ||
534 | static void paio_cancel(BlockDriverAIOCB *blockacb) | |
535 | { | |
536 | struct qemu_paiocb *acb = (struct qemu_paiocb *)blockacb; | |
537 | int active = 0; | |
3c529d93 | 538 | |
8653c015 | 539 | mutex_lock(&lock); |
9ef91a67 | 540 | if (!acb->active) { |
72cf2d4f | 541 | QTAILQ_REMOVE(&request_list, acb, node); |
9ef91a67 CH |
542 | acb->ret = -ECANCELED; |
543 | } else if (acb->ret == -EINPROGRESS) { | |
544 | active = 1; | |
545 | } | |
8653c015 | 546 | mutex_unlock(&lock); |
3c529d93 | 547 | |
9ef91a67 CH |
548 | if (active) { |
549 | /* fail safe: if the aio could not be canceled, we wait for | |
550 | it */ | |
551 | while (qemu_paio_error(acb) == EINPROGRESS) | |
552 | ; | |
553 | } | |
554 | ||
555 | paio_remove(acb); | |
556 | } | |
557 | ||
558 | static AIOPool raw_aio_pool = { | |
559 | .aiocb_size = sizeof(struct qemu_paiocb), | |
560 | .cancel = paio_cancel, | |
561 | }; | |
562 | ||
1e5b9d2f | 563 | BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd, |
9ef91a67 CH |
564 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, |
565 | BlockDriverCompletionFunc *cb, void *opaque, int type) | |
566 | { | |
567 | struct qemu_paiocb *acb; | |
568 | ||
569 | acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque); | |
570 | if (!acb) | |
571 | return NULL; | |
572 | acb->aio_type = type; | |
573 | acb->aio_fildes = fd; | |
574 | acb->ev_signo = SIGUSR2; | |
e5f37649 KW |
575 | acb->async_context_id = get_async_context_id(); |
576 | ||
b2e12bc6 CH |
577 | if (qiov) { |
578 | acb->aio_iov = qiov->iov; | |
579 | acb->aio_niov = qiov->niov; | |
580 | } | |
9ef91a67 CH |
581 | acb->aio_nbytes = nb_sectors * 512; |
582 | acb->aio_offset = sector_num * 512; | |
583 | ||
584 | acb->next = posix_aio_state->first_aio; | |
585 | posix_aio_state->first_aio = acb; | |
586 | ||
6d519a5f | 587 | trace_paio_submit(acb, opaque, sector_num, nb_sectors, type); |
9ef91a67 CH |
588 | qemu_paio_submit(acb); |
589 | return &acb->common; | |
590 | } | |
591 | ||
592 | BlockDriverAIOCB *paio_ioctl(BlockDriverState *bs, int fd, | |
593 | unsigned long int req, void *buf, | |
594 | BlockDriverCompletionFunc *cb, void *opaque) | |
595 | { | |
596 | struct qemu_paiocb *acb; | |
597 | ||
598 | acb = qemu_aio_get(&raw_aio_pool, bs, cb, opaque); | |
599 | if (!acb) | |
600 | return NULL; | |
601 | acb->aio_type = QEMU_AIO_IOCTL; | |
602 | acb->aio_fildes = fd; | |
603 | acb->ev_signo = SIGUSR2; | |
34cf0081 | 604 | acb->async_context_id = get_async_context_id(); |
9ef91a67 CH |
605 | acb->aio_offset = 0; |
606 | acb->aio_ioctl_buf = buf; | |
607 | acb->aio_ioctl_cmd = req; | |
608 | ||
609 | acb->next = posix_aio_state->first_aio; | |
610 | posix_aio_state->first_aio = acb; | |
611 | ||
612 | qemu_paio_submit(acb); | |
613 | return &acb->common; | |
614 | } | |
615 | ||
1e5b9d2f | 616 | int paio_init(void) |
9ef91a67 CH |
617 | { |
618 | struct sigaction act; | |
619 | PosixAioState *s; | |
620 | int fds[2]; | |
621 | int ret; | |
622 | ||
623 | if (posix_aio_state) | |
1e5b9d2f | 624 | return 0; |
9ef91a67 CH |
625 | |
626 | s = qemu_malloc(sizeof(PosixAioState)); | |
627 | ||
628 | sigfillset(&act.sa_mask); | |
629 | act.sa_flags = 0; /* do not restart syscalls to interrupt select() */ | |
630 | act.sa_handler = aio_signal_handler; | |
631 | sigaction(SIGUSR2, &act, NULL); | |
632 | ||
633 | s->first_aio = NULL; | |
40ff6d7e | 634 | if (qemu_pipe(fds) == -1) { |
9ef91a67 | 635 | fprintf(stderr, "failed to create pipe\n"); |
1e5b9d2f | 636 | return -1; |
9ef91a67 CH |
637 | } |
638 | ||
639 | s->rfd = fds[0]; | |
640 | s->wfd = fds[1]; | |
641 | ||
642 | fcntl(s->rfd, F_SETFL, O_NONBLOCK); | |
643 | fcntl(s->wfd, F_SETFL, O_NONBLOCK); | |
644 | ||
8febfa26 KW |
645 | qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, |
646 | posix_aio_process_queue, s); | |
9ef91a67 CH |
647 | |
648 | ret = pthread_attr_init(&attr); | |
649 | if (ret) | |
650 | die2(ret, "pthread_attr_init"); | |
651 | ||
652 | ret = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); | |
653 | if (ret) | |
654 | die2(ret, "pthread_attr_setdetachstate"); | |
655 | ||
72cf2d4f | 656 | QTAILQ_INIT(&request_list); |
9ef91a67 CH |
657 | |
658 | posix_aio_state = s; | |
1e5b9d2f | 659 | return 0; |
3c529d93 | 660 | } |