1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1990, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/aio.h> 36 #include <sys/domain.h> 37 #include <sys/file.h> 38 #include <sys/filedesc.h> 39 #include <sys/kernel.h> 40 #include <sys/kthread.h> 41 #include <sys/malloc.h> 42 #include <sys/proc.h> 43 #include <sys/protosw.h> 44 #include <sys/sigio.h> 45 #include <sys/signal.h> 46 #include <sys/signalvar.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/filio.h> /* XXX */ 50 #include <sys/sockio.h> 51 #include <sys/stat.h> 52 #include <sys/sysctl.h> 53 #include <sys/sysproto.h> 54 #include <sys/taskqueue.h> 55 #include <sys/uio.h> 56 #include <sys/ucred.h> 57 #include <sys/un.h> 58 #include <sys/unpcb.h> 59 #include <sys/user.h> 60 61 #include <net/if.h> 62 #include <net/if_var.h> 63 #include <net/route.h> 64 #include <net/vnet.h> 65 66 #include <netinet/in.h> 67 #include <netinet/in_pcb.h> 68 69 #include <security/mac/mac_framework.h> 70 71 #include <vm/vm.h> 72 #include <vm/pmap.h> 73 #include <vm/vm_extern.h> 74 #include <vm/vm_map.h> 75 76 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 77 "socket AIO stats"); 78 79 static int empty_results; 80 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results, 81 0, "socket operation returned EAGAIN"); 82 83 static int empty_retries; 84 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries, 85 0, "socket operation retries"); 86 87 static fo_rdwr_t soo_read; 88 static fo_rdwr_t soo_write; 89 static fo_ioctl_t soo_ioctl; 90 static fo_poll_t soo_poll; 91 extern fo_kqfilter_t soo_kqfilter; 92 static fo_stat_t soo_stat; 93 static fo_close_t soo_close; 94 static fo_fill_kinfo_t soo_fill_kinfo; 95 static fo_aio_queue_t soo_aio_queue; 96 97 static void soo_aio_cancel(struct kaiocb *job); 98 99 struct fileops socketops = { 100 .fo_read = soo_read, 101 .fo_write = soo_write, 102 .fo_truncate = invfo_truncate, 103 .fo_ioctl = soo_ioctl, 104 .fo_poll = soo_poll, 105 .fo_kqfilter = soo_kqfilter, 106 .fo_stat = soo_stat, 107 .fo_close = soo_close, 108 .fo_chmod = invfo_chmod, 109 .fo_chown = invfo_chown, 110 .fo_sendfile = invfo_sendfile, 111 .fo_fill_kinfo = soo_fill_kinfo, 112 .fo_aio_queue = soo_aio_queue, 113 .fo_flags = DFLAG_PASSABLE 114 }; 115 116 static int 117 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 118 int flags, struct thread *td) 119 { 120 struct socket *so = fp->f_data; 121 int error; 122 123 #ifdef MAC 124 error = mac_socket_check_receive(active_cred, so); 125 if (error) 126 return (error); 127 #endif 128 error = soreceive(so, 0, uio, 0, 0, 0); 129 return (error); 130 } 131 132 static int 133 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 134 int flags, struct thread *td) 135 { 136 struct socket *so = fp->f_data; 137 int error; 138 139 #ifdef MAC 140 error = mac_socket_check_send(active_cred, so); 141 if (error) 142 return (error); 143 #endif 144 error = sousrsend(so, NULL, uio, NULL, 0, NULL); 145 return (error); 146 } 147 148 static int 149 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, 150 struct thread *td) 151 { 152 struct socket *so = fp->f_data; 153 int error = 0; 154 155 switch (cmd) { 156 case FIONBIO: 157 SOCK_LOCK(so); 158 if (*(int *)data) 159 so->so_state |= SS_NBIO; 160 else 161 so->so_state &= ~SS_NBIO; 162 SOCK_UNLOCK(so); 163 break; 164 165 case FIOASYNC: 166 if (*(int *)data) { 167 SOCK_LOCK(so); 168 so->so_state |= SS_ASYNC; 169 if (SOLISTENING(so)) { 170 so->sol_sbrcv_flags |= SB_ASYNC; 171 so->sol_sbsnd_flags |= SB_ASYNC; 172 } else { 173 SOCK_RECVBUF_LOCK(so); 174 so->so_rcv.sb_flags |= SB_ASYNC; 175 SOCK_RECVBUF_UNLOCK(so); 176 SOCK_SENDBUF_LOCK(so); 177 so->so_snd.sb_flags |= SB_ASYNC; 178 SOCK_SENDBUF_UNLOCK(so); 179 } 180 SOCK_UNLOCK(so); 181 } else { 182 SOCK_LOCK(so); 183 so->so_state &= ~SS_ASYNC; 184 if (SOLISTENING(so)) { 185 so->sol_sbrcv_flags &= ~SB_ASYNC; 186 so->sol_sbsnd_flags &= ~SB_ASYNC; 187 } else { 188 SOCK_RECVBUF_LOCK(so); 189 so->so_rcv.sb_flags &= ~SB_ASYNC; 190 SOCK_RECVBUF_UNLOCK(so); 191 SOCK_SENDBUF_LOCK(so); 192 so->so_snd.sb_flags &= ~SB_ASYNC; 193 SOCK_SENDBUF_UNLOCK(so); 194 } 195 SOCK_UNLOCK(so); 196 } 197 break; 198 199 case FIONREAD: 200 SOCK_RECVBUF_LOCK(so); 201 if (SOLISTENING(so)) { 202 error = EINVAL; 203 } else { 204 *(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 205 } 206 SOCK_RECVBUF_UNLOCK(so); 207 break; 208 209 case FIONWRITE: 210 /* Unlocked read. */ 211 if (SOLISTENING(so)) { 212 error = EINVAL; 213 } else { 214 *(int *)data = sbavail(&so->so_snd); 215 } 216 break; 217 218 case FIONSPACE: 219 /* Unlocked read. */ 220 if (SOLISTENING(so)) { 221 error = EINVAL; 222 } else { 223 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) || 224 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) { 225 *(int *)data = 0; 226 } else { 227 *(int *)data = sbspace(&so->so_snd); 228 } 229 } 230 break; 231 232 case FIOSETOWN: 233 error = fsetown(*(int *)data, &so->so_sigio); 234 break; 235 236 case FIOGETOWN: 237 *(int *)data = fgetown(&so->so_sigio); 238 break; 239 240 case SIOCSPGRP: 241 error = fsetown(-(*(int *)data), &so->so_sigio); 242 break; 243 244 case SIOCGPGRP: 245 *(int *)data = -fgetown(&so->so_sigio); 246 break; 247 248 case SIOCATMARK: 249 /* Unlocked read. */ 250 if (SOLISTENING(so)) { 251 error = EINVAL; 252 } else { 253 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0; 254 } 255 break; 256 default: 257 /* 258 * Interface/routing/protocol specific ioctls: interface and 259 * routing ioctls should have a different entry since a 260 * socket is unnecessary. 261 */ 262 if (IOCGROUP(cmd) == 'i') 263 error = ifioctl(so, cmd, data, td); 264 else if (IOCGROUP(cmd) == 'r') { 265 CURVNET_SET(so->so_vnet); 266 error = rtioctl_fib(cmd, data, so->so_fibnum); 267 CURVNET_RESTORE(); 268 } else { 269 CURVNET_SET(so->so_vnet); 270 error = so->so_proto->pr_control(so, cmd, data, 0, td); 271 CURVNET_RESTORE(); 272 } 273 break; 274 } 275 return (error); 276 } 277 278 static int 279 soo_poll(struct file *fp, int events, struct ucred *active_cred, 280 struct thread *td) 281 { 282 struct socket *so = fp->f_data; 283 #ifdef MAC 284 int error; 285 286 error = mac_socket_check_poll(active_cred, so); 287 if (error) 288 return (error); 289 #endif 290 return (sopoll(so, events, fp->f_cred, td)); 291 } 292 293 static int 294 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred) 295 { 296 struct socket *so = fp->f_data; 297 int error = 0; 298 299 bzero((caddr_t)ub, sizeof (*ub)); 300 ub->st_mode = S_IFSOCK; 301 #ifdef MAC 302 error = mac_socket_check_stat(active_cred, so); 303 if (error) 304 return (error); 305 #endif 306 SOCK_LOCK(so); 307 if (!SOLISTENING(so)) { 308 struct sockbuf *sb; 309 310 /* 311 * If SBS_CANTRCVMORE is set, but there's still data left 312 * in the receive buffer, the socket is still readable. 313 */ 314 sb = &so->so_rcv; 315 SOCK_RECVBUF_LOCK(so); 316 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb)) 317 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH; 318 ub->st_size = sbavail(sb) - sb->sb_ctl; 319 SOCK_RECVBUF_UNLOCK(so); 320 321 sb = &so->so_snd; 322 SOCK_SENDBUF_LOCK(so); 323 if ((sb->sb_state & SBS_CANTSENDMORE) == 0) 324 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH; 325 SOCK_SENDBUF_UNLOCK(so); 326 } 327 ub->st_uid = so->so_cred->cr_uid; 328 ub->st_gid = so->so_cred->cr_gid; 329 if (so->so_proto->pr_sense) 330 error = so->so_proto->pr_sense(so, ub); 331 SOCK_UNLOCK(so); 332 return (error); 333 } 334 335 /* 336 * API socket close on file pointer. We call soclose() to close the socket 337 * (including initiating closing protocols). soclose() will sorele() the 338 * file reference but the actual socket will not go away until the socket's 339 * ref count hits 0. 340 */ 341 static int 342 soo_close(struct file *fp, struct thread *td) 343 { 344 int error = 0; 345 struct socket *so; 346 347 so = fp->f_data; 348 fp->f_ops = &badfileops; 349 fp->f_data = NULL; 350 351 if (so) 352 error = soclose(so); 353 return (error); 354 } 355 356 static int 357 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 358 { 359 struct sockaddr *sa; 360 struct inpcb *inpcb; 361 struct unpcb *unpcb; 362 struct socket *so; 363 int error; 364 365 kif->kf_type = KF_TYPE_SOCKET; 366 so = fp->f_data; 367 CURVNET_SET(so->so_vnet); 368 kif->kf_un.kf_sock.kf_sock_domain0 = 369 so->so_proto->pr_domain->dom_family; 370 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type; 371 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol; 372 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 373 switch (kif->kf_un.kf_sock.kf_sock_domain0) { 374 case AF_INET: 375 case AF_INET6: 376 if (so->so_pcb != NULL) { 377 inpcb = (struct inpcb *)(so->so_pcb); 378 kif->kf_un.kf_sock.kf_sock_inpcb = 379 (uintptr_t)inpcb->inp_ppcb; 380 } 381 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 382 so->so_rcv.sb_state; 383 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 384 so->so_snd.sb_state; 385 kif->kf_un.kf_sock.kf_sock_sendq = 386 sbused(&so->so_snd); 387 kif->kf_un.kf_sock.kf_sock_recvq = 388 sbused(&so->so_rcv); 389 break; 390 case AF_UNIX: 391 if (so->so_pcb != NULL) { 392 unpcb = (struct unpcb *)(so->so_pcb); 393 if (unpcb->unp_conn) { 394 kif->kf_un.kf_sock.kf_sock_unpconn = 395 (uintptr_t)unpcb->unp_conn; 396 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 397 so->so_rcv.sb_state; 398 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 399 so->so_snd.sb_state; 400 kif->kf_un.kf_sock.kf_sock_sendq = 401 sbused(&so->so_snd); 402 kif->kf_un.kf_sock.kf_sock_recvq = 403 sbused(&so->so_rcv); 404 } 405 } 406 break; 407 } 408 error = so->so_proto->pr_sockaddr(so, &sa); 409 if (error == 0 && 410 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) { 411 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_local, sa->sa_len); 412 free(sa, M_SONAME); 413 } 414 error = so->so_proto->pr_peeraddr(so, &sa); 415 if (error == 0 && 416 sa->sa_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) { 417 bcopy(sa, &kif->kf_un.kf_sock.kf_sa_peer, sa->sa_len); 418 free(sa, M_SONAME); 419 } 420 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 421 sizeof(kif->kf_path)); 422 CURVNET_RESTORE(); 423 return (0); 424 } 425 426 /* 427 * Use the 'backend3' field in AIO jobs to store the amount of data 428 * completed by the AIO job so far. 429 */ 430 #define aio_done backend3 431 432 static STAILQ_HEAD(, task) soaio_jobs; 433 static struct mtx soaio_jobs_lock; 434 static struct task soaio_kproc_task; 435 static int soaio_starting, soaio_idle, soaio_queued; 436 static struct unrhdr *soaio_kproc_unr; 437 438 static int soaio_max_procs = MAX_AIO_PROCS; 439 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0, 440 "Maximum number of kernel processes to use for async socket IO"); 441 442 static int soaio_num_procs; 443 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0, 444 "Number of active kernel processes for async socket IO"); 445 446 static int soaio_target_procs = TARGET_AIO_PROCS; 447 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD, 448 &soaio_target_procs, 0, 449 "Preferred number of ready kernel processes for async socket IO"); 450 451 static int soaio_lifetime; 452 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0, 453 "Maximum lifetime for idle aiod"); 454 455 static void 456 soaio_kproc_loop(void *arg) 457 { 458 struct proc *p; 459 struct vmspace *myvm; 460 struct task *task; 461 int error, id, pending; 462 463 id = (intptr_t)arg; 464 465 /* 466 * Grab an extra reference on the daemon's vmspace so that it 467 * doesn't get freed by jobs that switch to a different 468 * vmspace. 469 */ 470 p = curproc; 471 myvm = vmspace_acquire_ref(p); 472 473 mtx_lock(&soaio_jobs_lock); 474 MPASS(soaio_starting > 0); 475 soaio_starting--; 476 for (;;) { 477 while (!STAILQ_EMPTY(&soaio_jobs)) { 478 task = STAILQ_FIRST(&soaio_jobs); 479 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link); 480 soaio_queued--; 481 pending = task->ta_pending; 482 task->ta_pending = 0; 483 mtx_unlock(&soaio_jobs_lock); 484 485 task->ta_func(task->ta_context, pending); 486 487 mtx_lock(&soaio_jobs_lock); 488 } 489 MPASS(soaio_queued == 0); 490 491 if (p->p_vmspace != myvm) { 492 mtx_unlock(&soaio_jobs_lock); 493 vmspace_switch_aio(myvm); 494 mtx_lock(&soaio_jobs_lock); 495 continue; 496 } 497 498 soaio_idle++; 499 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-", 500 soaio_lifetime); 501 soaio_idle--; 502 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) && 503 soaio_num_procs > soaio_target_procs) 504 break; 505 } 506 soaio_num_procs--; 507 mtx_unlock(&soaio_jobs_lock); 508 free_unr(soaio_kproc_unr, id); 509 kproc_exit(0); 510 } 511 512 static void 513 soaio_kproc_create(void *context, int pending) 514 { 515 struct proc *p; 516 int error, id; 517 518 mtx_lock(&soaio_jobs_lock); 519 for (;;) { 520 if (soaio_num_procs < soaio_target_procs) { 521 /* Must create */ 522 } else if (soaio_num_procs >= soaio_max_procs) { 523 /* 524 * Hit the limit on kernel processes, don't 525 * create another one. 526 */ 527 break; 528 } else if (soaio_queued <= soaio_idle + soaio_starting) { 529 /* 530 * No more AIO jobs waiting for a process to be 531 * created, so stop. 532 */ 533 break; 534 } 535 soaio_starting++; 536 mtx_unlock(&soaio_jobs_lock); 537 538 id = alloc_unr(soaio_kproc_unr); 539 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id, 540 &p, 0, 0, "soaiod%d", id); 541 if (error != 0) { 542 free_unr(soaio_kproc_unr, id); 543 mtx_lock(&soaio_jobs_lock); 544 soaio_starting--; 545 break; 546 } 547 548 mtx_lock(&soaio_jobs_lock); 549 soaio_num_procs++; 550 } 551 mtx_unlock(&soaio_jobs_lock); 552 } 553 554 void 555 soaio_enqueue(struct task *task) 556 { 557 558 mtx_lock(&soaio_jobs_lock); 559 MPASS(task->ta_pending == 0); 560 task->ta_pending++; 561 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link); 562 soaio_queued++; 563 if (soaio_queued <= soaio_idle) 564 wakeup_one(&soaio_idle); 565 else if (soaio_num_procs < soaio_max_procs) 566 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task); 567 mtx_unlock(&soaio_jobs_lock); 568 } 569 570 static void 571 soaio_init(void) 572 { 573 574 soaio_lifetime = AIOD_LIFETIME_DEFAULT; 575 STAILQ_INIT(&soaio_jobs); 576 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF); 577 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL); 578 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL); 579 } 580 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL); 581 582 static __inline int 583 soaio_ready(struct socket *so, struct sockbuf *sb) 584 { 585 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so)); 586 } 587 588 static void 589 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job) 590 { 591 struct ucred *td_savedcred; 592 struct thread *td; 593 struct sockbuf *sb = sobuf(so, which); 594 #ifdef MAC 595 struct file *fp = job->fd_file; 596 #endif 597 size_t cnt, done, job_total_nbytes __diagused; 598 long ru_before; 599 int error, flags; 600 601 SOCK_BUF_UNLOCK(so, which); 602 aio_switch_vmspace(job); 603 td = curthread; 604 retry: 605 td_savedcred = td->td_ucred; 606 td->td_ucred = job->cred; 607 608 job_total_nbytes = job->uiop->uio_resid + job->aio_done; 609 done = job->aio_done; 610 cnt = job->uiop->uio_resid; 611 job->uiop->uio_offset = 0; 612 job->uiop->uio_td = td; 613 flags = MSG_NBIO; 614 615 /* 616 * For resource usage accounting, only count a completed request 617 * as a single message to avoid counting multiple calls to 618 * sosend/soreceive on a blocking socket. 619 */ 620 621 if (sb == &so->so_rcv) { 622 ru_before = td->td_ru.ru_msgrcv; 623 #ifdef MAC 624 error = mac_socket_check_receive(fp->f_cred, so); 625 if (error == 0) 626 627 #endif 628 error = soreceive(so, NULL, job->uiop, NULL, NULL, 629 &flags); 630 if (td->td_ru.ru_msgrcv != ru_before) 631 job->msgrcv = 1; 632 } else { 633 if (!TAILQ_EMPTY(&sb->sb_aiojobq)) 634 flags |= MSG_MORETOCOME; 635 ru_before = td->td_ru.ru_msgsnd; 636 #ifdef MAC 637 error = mac_socket_check_send(fp->f_cred, so); 638 if (error == 0) 639 #endif 640 error = sousrsend(so, NULL, job->uiop, NULL, flags, 641 job->userproc); 642 if (td->td_ru.ru_msgsnd != ru_before) 643 job->msgsnd = 1; 644 } 645 646 done += cnt - job->uiop->uio_resid; 647 job->aio_done = done; 648 td->td_ucred = td_savedcred; 649 650 if (error == EWOULDBLOCK) { 651 /* 652 * The request was either partially completed or not 653 * completed at all due to racing with a read() or 654 * write() on the socket. If the socket is 655 * non-blocking, return with any partial completion. 656 * If the socket is blocking or if no progress has 657 * been made, requeue this request at the head of the 658 * queue to try again when the socket is ready. 659 */ 660 MPASS(done != job_total_nbytes); 661 SOCK_BUF_LOCK(so, which); 662 if (done == 0 || !(so->so_state & SS_NBIO)) { 663 empty_results++; 664 if (soaio_ready(so, sb)) { 665 empty_retries++; 666 SOCK_BUF_UNLOCK(so, which); 667 goto retry; 668 } 669 670 if (!aio_set_cancel_function(job, soo_aio_cancel)) { 671 SOCK_BUF_UNLOCK(so, which); 672 if (done != 0) 673 aio_complete(job, done, 0); 674 else 675 aio_cancel(job); 676 SOCK_BUF_LOCK(so, which); 677 } else { 678 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list); 679 } 680 return; 681 } 682 SOCK_BUF_UNLOCK(so, which); 683 } 684 if (done != 0 && (error == ERESTART || error == EINTR || 685 error == EWOULDBLOCK)) 686 error = 0; 687 if (error) 688 aio_complete(job, -1, error); 689 else 690 aio_complete(job, done, 0); 691 SOCK_BUF_LOCK(so, which); 692 } 693 694 static void 695 soaio_process_sb(struct socket *so, sb_which which) 696 { 697 struct kaiocb *job; 698 struct sockbuf *sb = sobuf(so, which); 699 700 CURVNET_SET(so->so_vnet); 701 SOCK_BUF_LOCK(so, which); 702 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) { 703 job = TAILQ_FIRST(&sb->sb_aiojobq); 704 TAILQ_REMOVE(&sb->sb_aiojobq, job, list); 705 if (!aio_clear_cancel_function(job)) 706 continue; 707 708 soaio_process_job(so, which, job); 709 } 710 711 /* 712 * If there are still pending requests, the socket must not be 713 * ready so set SB_AIO to request a wakeup when the socket 714 * becomes ready. 715 */ 716 if (!TAILQ_EMPTY(&sb->sb_aiojobq)) 717 sb->sb_flags |= SB_AIO; 718 sb->sb_flags &= ~SB_AIO_RUNNING; 719 SOCK_BUF_UNLOCK(so, which); 720 721 sorele(so); 722 CURVNET_RESTORE(); 723 } 724 725 void 726 soaio_rcv(void *context, int pending) 727 { 728 struct socket *so; 729 730 so = context; 731 soaio_process_sb(so, SO_RCV); 732 } 733 734 void 735 soaio_snd(void *context, int pending) 736 { 737 struct socket *so; 738 739 so = context; 740 soaio_process_sb(so, SO_SND); 741 } 742 743 void 744 sowakeup_aio(struct socket *so, sb_which which) 745 { 746 struct sockbuf *sb = sobuf(so, which); 747 748 SOCK_BUF_LOCK_ASSERT(so, which); 749 750 sb->sb_flags &= ~SB_AIO; 751 if (sb->sb_flags & SB_AIO_RUNNING) 752 return; 753 sb->sb_flags |= SB_AIO_RUNNING; 754 soref(so); 755 soaio_enqueue(&sb->sb_aiotask); 756 } 757 758 static void 759 soo_aio_cancel(struct kaiocb *job) 760 { 761 struct socket *so; 762 struct sockbuf *sb; 763 long done; 764 int opcode; 765 sb_which which; 766 767 so = job->fd_file->f_data; 768 opcode = job->uaiocb.aio_lio_opcode; 769 if (opcode & LIO_READ) { 770 sb = &so->so_rcv; 771 which = SO_RCV; 772 } else { 773 MPASS(opcode & LIO_WRITE); 774 sb = &so->so_snd; 775 which = SO_SND; 776 } 777 778 SOCK_BUF_LOCK(so, which); 779 if (!aio_cancel_cleared(job)) 780 TAILQ_REMOVE(&sb->sb_aiojobq, job, list); 781 if (TAILQ_EMPTY(&sb->sb_aiojobq)) 782 sb->sb_flags &= ~SB_AIO; 783 SOCK_BUF_UNLOCK(so, which); 784 785 done = job->aio_done; 786 if (done != 0) 787 aio_complete(job, done, 0); 788 else 789 aio_cancel(job); 790 } 791 792 static int 793 soo_aio_queue(struct file *fp, struct kaiocb *job) 794 { 795 struct socket *so; 796 struct sockbuf *sb; 797 sb_which which; 798 int error; 799 800 so = fp->f_data; 801 error = so->so_proto->pr_aio_queue(so, job); 802 if (error == 0) 803 return (0); 804 805 /* Lock through the socket, since this may be a listening socket. */ 806 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) { 807 case LIO_READ: 808 SOCK_RECVBUF_LOCK(so); 809 sb = &so->so_rcv; 810 which = SO_RCV; 811 break; 812 case LIO_WRITE: 813 SOCK_SENDBUF_LOCK(so); 814 sb = &so->so_snd; 815 which = SO_SND; 816 break; 817 default: 818 return (EINVAL); 819 } 820 821 if (SOLISTENING(so)) { 822 SOCK_BUF_UNLOCK(so, which); 823 return (EINVAL); 824 } 825 826 if (!aio_set_cancel_function(job, soo_aio_cancel)) 827 panic("new job was cancelled"); 828 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list); 829 if (!(sb->sb_flags & SB_AIO_RUNNING)) { 830 if (soaio_ready(so, sb)) 831 sowakeup_aio(so, which); 832 else 833 sb->sb_flags |= SB_AIO; 834 } 835 SOCK_BUF_UNLOCK(so, which); 836 return (0); 837 } 838