1 /*- 2 * Copyright (c) 1982, 1986, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)sys_socket.c 8.1 (Berkeley) 6/10/93 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/aio.h> 38 #include <sys/domain.h> 39 #include <sys/file.h> 40 #include <sys/filedesc.h> 41 #include <sys/kernel.h> 42 #include <sys/kthread.h> 43 #include <sys/malloc.h> 44 #include <sys/proc.h> 45 #include <sys/protosw.h> 46 #include <sys/sigio.h> 47 #include <sys/signal.h> 48 #include <sys/signalvar.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/filio.h> /* XXX */ 52 #include <sys/sockio.h> 53 #include <sys/stat.h> 54 #include <sys/sysctl.h> 55 #include <sys/sysproto.h> 56 #include <sys/taskqueue.h> 57 #include <sys/uio.h> 58 #include <sys/ucred.h> 59 #include <sys/un.h> 60 #include <sys/unpcb.h> 61 #include <sys/user.h> 62 63 #include <net/if.h> 64 #include <net/if_var.h> 65 #include <net/route.h> 66 #include <net/vnet.h> 67 68 #include <netinet/in.h> 69 #include <netinet/in_pcb.h> 70 71 #include <security/mac/mac_framework.h> 72 73 #include <vm/vm.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_extern.h> 76 #include <vm/vm_map.h> 77 78 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD, NULL, 79 "socket AIO stats"); 80 81 static int empty_results; 82 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results, 83 0, "socket operation returned EAGAIN"); 84 85 static int empty_retries; 86 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries, 87 0, "socket operation retries"); 88 89 static fo_rdwr_t soo_read; 90 static fo_rdwr_t soo_write; 91 static fo_ioctl_t soo_ioctl; 92 static fo_poll_t soo_poll; 93 extern fo_kqfilter_t soo_kqfilter; 94 static fo_stat_t soo_stat; 95 static fo_close_t soo_close; 96 #ifndef FSTACK 97 static fo_fill_kinfo_t soo_fill_kinfo; 98 static fo_aio_queue_t soo_aio_queue; 99 100 static void soo_aio_cancel(struct kaiocb *job); 101 #endif 102 103 struct fileops socketops = { 104 .fo_read = soo_read, 105 .fo_write = soo_write, 106 .fo_truncate = invfo_truncate, 107 .fo_ioctl = soo_ioctl, 108 .fo_poll = soo_poll, 109 .fo_kqfilter = soo_kqfilter, 110 .fo_stat = soo_stat, 111 .fo_close = soo_close, 112 .fo_chmod = invfo_chmod, 113 .fo_chown = invfo_chown, 114 .fo_sendfile = invfo_sendfile, 115 #ifndef FSTACK 116 .fo_fill_kinfo = soo_fill_kinfo, 117 .fo_aio_queue = soo_aio_queue, 118 #endif 119 .fo_flags = DFLAG_PASSABLE 120 }; 121 122 static int 123 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 124 int flags, struct thread *td) 125 { 126 struct socket *so = fp->f_data; 127 int error; 128 129 #ifdef MAC 130 error = mac_socket_check_receive(active_cred, so); 131 if (error) 132 return (error); 133 #endif 134 error = soreceive(so, 0, uio, 0, 0, 0); 135 return (error); 136 } 137 138 static int 139 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 140 int flags, struct thread *td) 141 { 142 struct socket *so = fp->f_data; 143 int error; 144 145 #ifdef MAC 146 error = mac_socket_check_send(active_cred, so); 147 if (error) 148 return (error); 149 #endif 150 error = sosend(so, 0, uio, 0, 0, 0, uio->uio_td); 151 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) { 152 PROC_LOCK(uio->uio_td->td_proc); 153 tdsignal(uio->uio_td, SIGPIPE); 154 PROC_UNLOCK(uio->uio_td->td_proc); 155 } 156 return (error); 157 } 158 159 static int 160 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, 161 struct thread *td) 162 { 163 struct socket *so = fp->f_data; 164 int error = 0; 165 166 switch (cmd) { 167 case FIONBIO: 168 SOCK_LOCK(so); 169 if (*(int *)data) 170 so->so_state |= SS_NBIO; 171 else 172 so->so_state &= ~SS_NBIO; 173 SOCK_UNLOCK(so); 174 break; 175 176 case FIOASYNC: 177 /* 178 * XXXRW: This code separately acquires SOCK_LOCK(so) and 179 * SOCKBUF_LOCK(&so->so_rcv) even though they are the same 180 * mutex to avoid introducing the assumption that they are 181 * the same. 182 */ 183 if (*(int *)data) { 184 SOCK_LOCK(so); 185 so->so_state |= SS_ASYNC; 186 SOCK_UNLOCK(so); 187 SOCKBUF_LOCK(&so->so_rcv); 188 so->so_rcv.sb_flags |= SB_ASYNC; 189 SOCKBUF_UNLOCK(&so->so_rcv); 190 SOCKBUF_LOCK(&so->so_snd); 191 so->so_snd.sb_flags |= SB_ASYNC; 192 SOCKBUF_UNLOCK(&so->so_snd); 193 } else { 194 SOCK_LOCK(so); 195 so->so_state &= ~SS_ASYNC; 196 SOCK_UNLOCK(so); 197 SOCKBUF_LOCK(&so->so_rcv); 198 so->so_rcv.sb_flags &= ~SB_ASYNC; 199 SOCKBUF_UNLOCK(&so->so_rcv); 200 SOCKBUF_LOCK(&so->so_snd); 201 so->so_snd.sb_flags &= ~SB_ASYNC; 202 SOCKBUF_UNLOCK(&so->so_snd); 203 } 204 break; 205 206 case FIONREAD: 207 /* Unlocked read. */ 208 *(int *)data = sbavail(&so->so_rcv); 209 break; 210 211 case FIONWRITE: 212 /* Unlocked read. */ 213 *(int *)data = sbavail(&so->so_snd); 214 break; 215 216 case FIONSPACE: 217 /* Unlocked read. */ 218 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) || 219 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) 220 *(int *)data = 0; 221 else 222 *(int *)data = sbspace(&so->so_snd); 223 break; 224 225 case FIOSETOWN: 226 error = fsetown(*(int *)data, &so->so_sigio); 227 break; 228 229 case FIOGETOWN: 230 *(int *)data = fgetown(&so->so_sigio); 231 break; 232 233 case SIOCSPGRP: 234 error = fsetown(-(*(int *)data), &so->so_sigio); 235 break; 236 237 case SIOCGPGRP: 238 *(int *)data = -fgetown(&so->so_sigio); 239 break; 240 241 case SIOCATMARK: 242 /* Unlocked read. */ 243 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0; 244 break; 245 default: 246 /* 247 * Interface/routing/protocol specific ioctls: interface and 248 * routing ioctls should have a different entry since a 249 * socket is unnecessary. 250 */ 251 if (IOCGROUP(cmd) == 'i') 252 error = ifioctl(so, cmd, data, td); 253 else if (IOCGROUP(cmd) == 'r') { 254 CURVNET_SET(so->so_vnet); 255 error = rtioctl_fib(cmd, data, so->so_fibnum); 256 CURVNET_RESTORE(); 257 } else { 258 CURVNET_SET(so->so_vnet); 259 error = ((*so->so_proto->pr_usrreqs->pru_control) 260 (so, cmd, data, 0, td)); 261 CURVNET_RESTORE(); 262 } 263 break; 264 } 265 return (error); 266 } 267 268 static int 269 soo_poll(struct file *fp, int events, struct ucred *active_cred, 270 struct thread *td) 271 { 272 struct socket *so = fp->f_data; 273 #ifdef MAC 274 int error; 275 276 error = mac_socket_check_poll(active_cred, so); 277 if (error) 278 return (error); 279 #endif 280 return (sopoll(so, events, fp->f_cred, td)); 281 } 282 283 static int 284 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred, 285 struct thread *td) 286 { 287 struct socket *so = fp->f_data; 288 struct sockbuf *sb; 289 #ifdef MAC 290 int error; 291 #endif 292 293 bzero((caddr_t)ub, sizeof (*ub)); 294 ub->st_mode = S_IFSOCK; 295 #ifdef MAC 296 error = mac_socket_check_stat(active_cred, so); 297 if (error) 298 return (error); 299 #endif 300 /* 301 * If SBS_CANTRCVMORE is set, but there's still data left in the 302 * receive buffer, the socket is still readable. 303 */ 304 sb = &so->so_rcv; 305 SOCKBUF_LOCK(sb); 306 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb)) 307 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH; 308 ub->st_size = sbavail(sb) - sb->sb_ctl; 309 SOCKBUF_UNLOCK(sb); 310 311 sb = &so->so_snd; 312 SOCKBUF_LOCK(sb); 313 if ((sb->sb_state & SBS_CANTSENDMORE) == 0) 314 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH; 315 SOCKBUF_UNLOCK(sb); 316 ub->st_uid = so->so_cred->cr_uid; 317 ub->st_gid = so->so_cred->cr_gid; 318 return (*so->so_proto->pr_usrreqs->pru_sense)(so, ub); 319 } 320 321 /* 322 * API socket close on file pointer. We call soclose() to close the socket 323 * (including initiating closing protocols). soclose() will sorele() the 324 * file reference but the actual socket will not go away until the socket's 325 * ref count hits 0. 326 */ 327 static int 328 soo_close(struct file *fp, struct thread *td) 329 { 330 int error = 0; 331 struct socket *so; 332 333 so = fp->f_data; 334 fp->f_ops = &badfileops; 335 fp->f_data = NULL; 336 337 if (so) 338 error = soclose(so); 339 return (error); 340 } 341 342 #ifndef FSTACK 343 static int 344 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 345 { 346 struct sockaddr *sa; 347 struct inpcb *inpcb; 348 struct unpcb *unpcb; 349 struct socket *so; 350 int error; 351 352 kif->kf_type = KF_TYPE_SOCKET; 353 so = fp->f_data; 354 kif->kf_sock_domain = so->so_proto->pr_domain->dom_family; 355 kif->kf_sock_type = so->so_type; 356 kif->kf_sock_protocol = so->so_proto->pr_protocol; 357 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 358 switch (kif->kf_sock_domain) { 359 case AF_INET: 360 case AF_INET6: 361 if (kif->kf_sock_protocol == IPPROTO_TCP) { 362 if (so->so_pcb != NULL) { 363 inpcb = (struct inpcb *)(so->so_pcb); 364 kif->kf_un.kf_sock.kf_sock_inpcb = 365 (uintptr_t)inpcb->inp_ppcb; 366 } 367 } 368 break; 369 case AF_UNIX: 370 if (so->so_pcb != NULL) { 371 unpcb = (struct unpcb *)(so->so_pcb); 372 if (unpcb->unp_conn) { 373 kif->kf_un.kf_sock.kf_sock_unpconn = 374 (uintptr_t)unpcb->unp_conn; 375 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 376 so->so_rcv.sb_state; 377 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 378 so->so_snd.sb_state; 379 } 380 } 381 break; 382 } 383 error = so->so_proto->pr_usrreqs->pru_sockaddr(so, &sa); 384 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_local)) { 385 bcopy(sa, &kif->kf_sa_local, sa->sa_len); 386 free(sa, M_SONAME); 387 } 388 error = so->so_proto->pr_usrreqs->pru_peeraddr(so, &sa); 389 if (error == 0 && sa->sa_len <= sizeof(kif->kf_sa_peer)) { 390 bcopy(sa, &kif->kf_sa_peer, sa->sa_len); 391 free(sa, M_SONAME); 392 } 393 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 394 sizeof(kif->kf_path)); 395 return (0); 396 } 397 398 /* 399 * Use the 'backend3' field in AIO jobs to store the amount of data 400 * completed by the AIO job so far. 401 */ 402 #define aio_done backend3 403 404 static STAILQ_HEAD(, task) soaio_jobs; 405 static struct mtx soaio_jobs_lock; 406 static struct task soaio_kproc_task; 407 static int soaio_starting, soaio_idle, soaio_queued; 408 static struct unrhdr *soaio_kproc_unr; 409 410 static int soaio_max_procs = MAX_AIO_PROCS; 411 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0, 412 "Maximum number of kernel processes to use for async socket IO"); 413 414 static int soaio_num_procs; 415 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0, 416 "Number of active kernel processes for async socket IO"); 417 418 static int soaio_target_procs = TARGET_AIO_PROCS; 419 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD, 420 &soaio_target_procs, 0, 421 "Preferred number of ready kernel processes for async socket IO"); 422 423 static int soaio_lifetime; 424 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0, 425 "Maximum lifetime for idle aiod"); 426 427 static void 428 soaio_kproc_loop(void *arg) 429 { 430 struct proc *p; 431 struct vmspace *myvm; 432 struct task *task; 433 int error, id, pending; 434 435 id = (intptr_t)arg; 436 437 /* 438 * Grab an extra reference on the daemon's vmspace so that it 439 * doesn't get freed by jobs that switch to a different 440 * vmspace. 441 */ 442 p = curproc; 443 myvm = vmspace_acquire_ref(p); 444 445 mtx_lock(&soaio_jobs_lock); 446 MPASS(soaio_starting > 0); 447 soaio_starting--; 448 for (;;) { 449 while (!STAILQ_EMPTY(&soaio_jobs)) { 450 task = STAILQ_FIRST(&soaio_jobs); 451 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link); 452 soaio_queued--; 453 pending = task->ta_pending; 454 task->ta_pending = 0; 455 mtx_unlock(&soaio_jobs_lock); 456 457 task->ta_func(task->ta_context, pending); 458 459 mtx_lock(&soaio_jobs_lock); 460 } 461 MPASS(soaio_queued == 0); 462 463 if (p->p_vmspace != myvm) { 464 mtx_unlock(&soaio_jobs_lock); 465 vmspace_switch_aio(myvm); 466 mtx_lock(&soaio_jobs_lock); 467 continue; 468 } 469 470 soaio_idle++; 471 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-", 472 soaio_lifetime); 473 soaio_idle--; 474 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) && 475 soaio_num_procs > soaio_target_procs) 476 break; 477 } 478 soaio_num_procs--; 479 mtx_unlock(&soaio_jobs_lock); 480 free_unr(soaio_kproc_unr, id); 481 kproc_exit(0); 482 } 483 484 static void 485 soaio_kproc_create(void *context, int pending) 486 { 487 struct proc *p; 488 int error, id; 489 490 mtx_lock(&soaio_jobs_lock); 491 for (;;) { 492 if (soaio_num_procs < soaio_target_procs) { 493 /* Must create */ 494 } else if (soaio_num_procs >= soaio_max_procs) { 495 /* 496 * Hit the limit on kernel processes, don't 497 * create another one. 498 */ 499 break; 500 } else if (soaio_queued <= soaio_idle + soaio_starting) { 501 /* 502 * No more AIO jobs waiting for a process to be 503 * created, so stop. 504 */ 505 break; 506 } 507 soaio_starting++; 508 mtx_unlock(&soaio_jobs_lock); 509 510 id = alloc_unr(soaio_kproc_unr); 511 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id, 512 &p, 0, 0, "soaiod%d", id); 513 if (error != 0) { 514 free_unr(soaio_kproc_unr, id); 515 mtx_lock(&soaio_jobs_lock); 516 soaio_starting--; 517 break; 518 } 519 520 mtx_lock(&soaio_jobs_lock); 521 soaio_num_procs++; 522 } 523 mtx_unlock(&soaio_jobs_lock); 524 } 525 526 void 527 soaio_enqueue(struct task *task) 528 { 529 530 mtx_lock(&soaio_jobs_lock); 531 MPASS(task->ta_pending == 0); 532 task->ta_pending++; 533 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link); 534 soaio_queued++; 535 if (soaio_queued <= soaio_idle) 536 wakeup_one(&soaio_idle); 537 else if (soaio_num_procs < soaio_max_procs) 538 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task); 539 mtx_unlock(&soaio_jobs_lock); 540 } 541 542 static void 543 soaio_init(void) 544 { 545 546 soaio_lifetime = AIOD_LIFETIME_DEFAULT; 547 STAILQ_INIT(&soaio_jobs); 548 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF); 549 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL); 550 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL); 551 if (soaio_target_procs > 0) 552 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task); 553 } 554 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL); 555 556 static __inline int 557 soaio_ready(struct socket *so, struct sockbuf *sb) 558 { 559 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so)); 560 } 561 562 static void 563 soaio_process_job(struct socket *so, struct sockbuf *sb, struct kaiocb *job) 564 { 565 struct ucred *td_savedcred; 566 struct thread *td; 567 struct file *fp; 568 struct uio uio; 569 struct iovec iov; 570 size_t cnt, done; 571 long ru_before; 572 int error, flags; 573 574 SOCKBUF_UNLOCK(sb); 575 aio_switch_vmspace(job); 576 td = curthread; 577 fp = job->fd_file; 578 retry: 579 td_savedcred = td->td_ucred; 580 td->td_ucred = job->cred; 581 582 done = job->aio_done; 583 cnt = job->uaiocb.aio_nbytes - done; 584 iov.iov_base = (void *)((uintptr_t)job->uaiocb.aio_buf + done); 585 iov.iov_len = cnt; 586 uio.uio_iov = &iov; 587 uio.uio_iovcnt = 1; 588 uio.uio_offset = 0; 589 uio.uio_resid = cnt; 590 uio.uio_segflg = UIO_USERSPACE; 591 uio.uio_td = td; 592 flags = MSG_NBIO; 593 594 /* 595 * For resource usage accounting, only count a completed request 596 * as a single message to avoid counting multiple calls to 597 * sosend/soreceive on a blocking socket. 598 */ 599 600 if (sb == &so->so_rcv) { 601 uio.uio_rw = UIO_READ; 602 ru_before = td->td_ru.ru_msgrcv; 603 #ifdef MAC 604 error = mac_socket_check_receive(fp->f_cred, so); 605 if (error == 0) 606 607 #endif 608 error = soreceive(so, NULL, &uio, NULL, NULL, &flags); 609 if (td->td_ru.ru_msgrcv != ru_before) 610 job->msgrcv = 1; 611 } else { 612 uio.uio_rw = UIO_WRITE; 613 ru_before = td->td_ru.ru_msgsnd; 614 #ifdef MAC 615 error = mac_socket_check_send(fp->f_cred, so); 616 if (error == 0) 617 #endif 618 error = sosend(so, NULL, &uio, NULL, NULL, flags, td); 619 if (td->td_ru.ru_msgsnd != ru_before) 620 job->msgsnd = 1; 621 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0) { 622 PROC_LOCK(job->userproc); 623 kern_psignal(job->userproc, SIGPIPE); 624 PROC_UNLOCK(job->userproc); 625 } 626 } 627 628 done += cnt - uio.uio_resid; 629 job->aio_done = done; 630 td->td_ucred = td_savedcred; 631 632 if (error == EWOULDBLOCK) { 633 /* 634 * The request was either partially completed or not 635 * completed at all due to racing with a read() or 636 * write() on the socket. If the socket is 637 * non-blocking, return with any partial completion. 638 * If the socket is blocking or if no progress has 639 * been made, requeue this request at the head of the 640 * queue to try again when the socket is ready. 641 */ 642 MPASS(done != job->uaiocb.aio_nbytes); 643 SOCKBUF_LOCK(sb); 644 if (done == 0 || !(so->so_state & SS_NBIO)) { 645 empty_results++; 646 if (soaio_ready(so, sb)) { 647 empty_retries++; 648 SOCKBUF_UNLOCK(sb); 649 goto retry; 650 } 651 652 if (!aio_set_cancel_function(job, soo_aio_cancel)) { 653 SOCKBUF_UNLOCK(sb); 654 if (done != 0) 655 aio_complete(job, done, 0); 656 else 657 aio_cancel(job); 658 SOCKBUF_LOCK(sb); 659 } else { 660 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list); 661 } 662 return; 663 } 664 SOCKBUF_UNLOCK(sb); 665 } 666 if (done != 0 && (error == ERESTART || error == EINTR || 667 error == EWOULDBLOCK)) 668 error = 0; 669 if (error) 670 aio_complete(job, -1, error); 671 else 672 aio_complete(job, done, 0); 673 SOCKBUF_LOCK(sb); 674 } 675 676 static void 677 soaio_process_sb(struct socket *so, struct sockbuf *sb) 678 { 679 struct kaiocb *job; 680 681 CURVNET_SET(so->so_vnet); 682 SOCKBUF_LOCK(sb); 683 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) { 684 job = TAILQ_FIRST(&sb->sb_aiojobq); 685 TAILQ_REMOVE(&sb->sb_aiojobq, job, list); 686 if (!aio_clear_cancel_function(job)) 687 continue; 688 689 soaio_process_job(so, sb, job); 690 } 691 692 /* 693 * If there are still pending requests, the socket must not be 694 * ready so set SB_AIO to request a wakeup when the socket 695 * becomes ready. 696 */ 697 if (!TAILQ_EMPTY(&sb->sb_aiojobq)) 698 sb->sb_flags |= SB_AIO; 699 sb->sb_flags &= ~SB_AIO_RUNNING; 700 SOCKBUF_UNLOCK(sb); 701 702 ACCEPT_LOCK(); 703 SOCK_LOCK(so); 704 sorele(so); 705 CURVNET_RESTORE(); 706 } 707 708 void 709 soaio_rcv(void *context, int pending) 710 { 711 struct socket *so; 712 713 so = context; 714 soaio_process_sb(so, &so->so_rcv); 715 } 716 717 void 718 soaio_snd(void *context, int pending) 719 { 720 struct socket *so; 721 722 so = context; 723 soaio_process_sb(so, &so->so_snd); 724 } 725 726 void 727 sowakeup_aio(struct socket *so, struct sockbuf *sb) 728 { 729 730 SOCKBUF_LOCK_ASSERT(sb); 731 sb->sb_flags &= ~SB_AIO; 732 if (sb->sb_flags & SB_AIO_RUNNING) 733 return; 734 sb->sb_flags |= SB_AIO_RUNNING; 735 if (sb == &so->so_snd) 736 SOCK_LOCK(so); 737 soref(so); 738 if (sb == &so->so_snd) 739 SOCK_UNLOCK(so); 740 soaio_enqueue(&sb->sb_aiotask); 741 } 742 743 static void 744 soo_aio_cancel(struct kaiocb *job) 745 { 746 struct socket *so; 747 struct sockbuf *sb; 748 long done; 749 int opcode; 750 751 so = job->fd_file->f_data; 752 opcode = job->uaiocb.aio_lio_opcode; 753 if (opcode == LIO_READ) 754 sb = &so->so_rcv; 755 else { 756 MPASS(opcode == LIO_WRITE); 757 sb = &so->so_snd; 758 } 759 760 SOCKBUF_LOCK(sb); 761 if (!aio_cancel_cleared(job)) 762 TAILQ_REMOVE(&sb->sb_aiojobq, job, list); 763 if (TAILQ_EMPTY(&sb->sb_aiojobq)) 764 sb->sb_flags &= ~SB_AIO; 765 SOCKBUF_UNLOCK(sb); 766 767 done = job->aio_done; 768 if (done != 0) 769 aio_complete(job, done, 0); 770 else 771 aio_cancel(job); 772 } 773 774 static int 775 soo_aio_queue(struct file *fp, struct kaiocb *job) 776 { 777 struct socket *so; 778 struct sockbuf *sb; 779 int error; 780 781 so = fp->f_data; 782 error = (*so->so_proto->pr_usrreqs->pru_aio_queue)(so, job); 783 if (error == 0) 784 return (0); 785 786 switch (job->uaiocb.aio_lio_opcode) { 787 case LIO_READ: 788 sb = &so->so_rcv; 789 break; 790 case LIO_WRITE: 791 sb = &so->so_snd; 792 break; 793 default: 794 return (EINVAL); 795 } 796 797 SOCKBUF_LOCK(sb); 798 if (!aio_set_cancel_function(job, soo_aio_cancel)) 799 panic("new job was cancelled"); 800 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list); 801 if (!(sb->sb_flags & SB_AIO_RUNNING)) { 802 if (soaio_ready(so, sb)) 803 sowakeup_aio(so, sb); 804 else 805 sb->sb_flags |= SB_AIO; 806 } 807 SOCKBUF_UNLOCK(sb); 808 return (0); 809 } 810 #endif 811