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/param.h> 33 #include <sys/systm.h> 34 #include <sys/aio.h> 35 #include <sys/domain.h> 36 #include <sys/file.h> 37 #include <sys/filedesc.h> 38 #include <sys/kernel.h> 39 #include <sys/kthread.h> 40 #include <sys/malloc.h> 41 #include <sys/proc.h> 42 #include <sys/protosw.h> 43 #include <sys/sigio.h> 44 #include <sys/signal.h> 45 #include <sys/signalvar.h> 46 #include <sys/socket.h> 47 #include <sys/socketvar.h> 48 #include <sys/filio.h> /* XXX */ 49 #include <sys/sockio.h> 50 #include <sys/stat.h> 51 #include <sys/sysctl.h> 52 #include <sys/sysproto.h> 53 #include <sys/taskqueue.h> 54 #include <sys/uio.h> 55 #include <sys/ucred.h> 56 #include <sys/un.h> 57 #include <sys/unpcb.h> 58 #include <sys/user.h> 59 60 #include <net/if.h> 61 #include <net/if_var.h> 62 #include <net/route.h> 63 #include <net/vnet.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_pcb.h> 67 68 #include <security/mac/mac_framework.h> 69 70 #include <vm/vm.h> 71 #include <vm/pmap.h> 72 #include <vm/vm_extern.h> 73 #include <vm/vm_map.h> 74 75 static SYSCTL_NODE(_kern_ipc, OID_AUTO, aio, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 76 "socket AIO stats"); 77 78 static int empty_results; 79 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_results, CTLFLAG_RD, &empty_results, 80 0, "socket operation returned EAGAIN"); 81 82 static int empty_retries; 83 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, empty_retries, CTLFLAG_RD, &empty_retries, 84 0, "socket operation retries"); 85 86 static fo_rdwr_t soo_read; 87 static fo_rdwr_t soo_write; 88 static fo_ioctl_t soo_ioctl; 89 static fo_poll_t soo_poll; 90 static fo_kqfilter_t soo_kqfilter; 91 static fo_stat_t soo_stat; 92 static fo_close_t soo_close; 93 static fo_chmod_t soo_chmod; 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 const 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 = soo_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_cmp = file_kcmp_generic, 114 .fo_flags = DFLAG_PASSABLE 115 }; 116 117 static int 118 soo_read(struct file *fp, struct uio *uio, struct ucred *active_cred, 119 int flags, struct thread *td) 120 { 121 struct socket *so = fp->f_data; 122 int error; 123 124 #ifdef MAC 125 error = mac_socket_check_receive(active_cred, so); 126 if (error) 127 return (error); 128 #endif 129 error = soreceive(so, 0, uio, 0, 0, 0); 130 return (error); 131 } 132 133 static int 134 soo_write(struct file *fp, struct uio *uio, struct ucred *active_cred, 135 int flags, struct thread *td) 136 { 137 struct socket *so = fp->f_data; 138 int error; 139 140 #ifdef MAC 141 error = mac_socket_check_send(active_cred, so); 142 if (error) 143 return (error); 144 #endif 145 error = sousrsend(so, NULL, uio, NULL, 0, NULL); 146 return (error); 147 } 148 149 static int 150 soo_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *active_cred, 151 struct thread *td) 152 { 153 struct socket *so = fp->f_data; 154 int error = 0; 155 156 switch (cmd) { 157 case FIONBIO: 158 SOCK_LOCK(so); 159 if (*(int *)data) 160 so->so_state |= SS_NBIO; 161 else 162 so->so_state &= ~SS_NBIO; 163 SOCK_UNLOCK(so); 164 break; 165 166 case FIOASYNC: 167 if (*(int *)data) { 168 SOCK_LOCK(so); 169 so->so_state |= SS_ASYNC; 170 if (SOLISTENING(so)) { 171 so->sol_sbrcv_flags |= SB_ASYNC; 172 so->sol_sbsnd_flags |= SB_ASYNC; 173 } else { 174 SOCK_RECVBUF_LOCK(so); 175 so->so_rcv.sb_flags |= SB_ASYNC; 176 SOCK_RECVBUF_UNLOCK(so); 177 SOCK_SENDBUF_LOCK(so); 178 so->so_snd.sb_flags |= SB_ASYNC; 179 SOCK_SENDBUF_UNLOCK(so); 180 } 181 SOCK_UNLOCK(so); 182 } else { 183 SOCK_LOCK(so); 184 so->so_state &= ~SS_ASYNC; 185 if (SOLISTENING(so)) { 186 so->sol_sbrcv_flags &= ~SB_ASYNC; 187 so->sol_sbsnd_flags &= ~SB_ASYNC; 188 } else { 189 SOCK_RECVBUF_LOCK(so); 190 so->so_rcv.sb_flags &= ~SB_ASYNC; 191 SOCK_RECVBUF_UNLOCK(so); 192 SOCK_SENDBUF_LOCK(so); 193 so->so_snd.sb_flags &= ~SB_ASYNC; 194 SOCK_SENDBUF_UNLOCK(so); 195 } 196 SOCK_UNLOCK(so); 197 } 198 break; 199 200 case FIONREAD: 201 SOCK_RECVBUF_LOCK(so); 202 if (SOLISTENING(so)) { 203 error = EINVAL; 204 } else { 205 *(int *)data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 206 } 207 SOCK_RECVBUF_UNLOCK(so); 208 break; 209 210 case FIONWRITE: 211 /* Unlocked read. */ 212 if (SOLISTENING(so)) { 213 error = EINVAL; 214 } else { 215 *(int *)data = sbavail(&so->so_snd); 216 } 217 break; 218 219 case FIONSPACE: 220 /* Unlocked read. */ 221 if (SOLISTENING(so)) { 222 error = EINVAL; 223 } else { 224 if ((so->so_snd.sb_hiwat < sbused(&so->so_snd)) || 225 (so->so_snd.sb_mbmax < so->so_snd.sb_mbcnt)) { 226 *(int *)data = 0; 227 } else { 228 *(int *)data = sbspace(&so->so_snd); 229 } 230 } 231 break; 232 233 case FIOSETOWN: 234 error = fsetown(*(int *)data, &so->so_sigio); 235 break; 236 237 case FIOGETOWN: 238 *(int *)data = fgetown(&so->so_sigio); 239 break; 240 241 case SIOCSPGRP: 242 error = fsetown(-(*(int *)data), &so->so_sigio); 243 break; 244 245 case SIOCGPGRP: 246 *(int *)data = -fgetown(&so->so_sigio); 247 break; 248 249 case SIOCATMARK: 250 /* Unlocked read. */ 251 if (SOLISTENING(so)) { 252 error = EINVAL; 253 } else { 254 *(int *)data = (so->so_rcv.sb_state & SBS_RCVATMARK) != 0; 255 } 256 break; 257 default: 258 /* 259 * Interface/routing/protocol specific ioctls: interface and 260 * routing ioctls should have a different entry since a 261 * socket is unnecessary. 262 */ 263 if (IOCGROUP(cmd) == 'i') 264 error = ifioctl(so, cmd, data, td); 265 else if (IOCGROUP(cmd) == 'r') { 266 CURVNET_SET(so->so_vnet); 267 error = rtioctl_fib(cmd, data, so->so_fibnum); 268 CURVNET_RESTORE(); 269 } else { 270 CURVNET_SET(so->so_vnet); 271 error = so->so_proto->pr_control(so, cmd, data, 0, td); 272 CURVNET_RESTORE(); 273 } 274 break; 275 } 276 return (error); 277 } 278 279 static int 280 soo_poll(struct file *fp, int events, struct ucred *active_cred, 281 struct thread *td) 282 { 283 struct socket *so = fp->f_data; 284 #ifdef MAC 285 int error; 286 287 error = mac_socket_check_poll(active_cred, so); 288 if (error) 289 return (error); 290 #endif 291 return (so->so_proto->pr_sopoll(so, events, td)); 292 } 293 294 static int 295 soo_kqfilter(struct file *fp, struct knote *kn) 296 { 297 struct socket *so = fp->f_data; 298 299 return (so->so_proto->pr_kqfilter(so, kn)); 300 } 301 302 static int 303 soo_stat(struct file *fp, struct stat *ub, struct ucred *active_cred) 304 { 305 struct socket *so = fp->f_data; 306 int error = 0; 307 308 bzero((caddr_t)ub, sizeof (*ub)); 309 ub->st_mode = S_IFSOCK; 310 #ifdef MAC 311 error = mac_socket_check_stat(active_cred, so); 312 if (error) 313 return (error); 314 #endif 315 SOCK_LOCK(so); 316 if (!SOLISTENING(so)) { 317 struct sockbuf *sb; 318 319 /* 320 * If SBS_CANTRCVMORE is set, but there's still data left 321 * in the receive buffer, the socket is still readable. 322 */ 323 sb = &so->so_rcv; 324 SOCK_RECVBUF_LOCK(so); 325 if ((sb->sb_state & SBS_CANTRCVMORE) == 0 || sbavail(sb)) 326 ub->st_mode |= S_IRUSR | S_IRGRP | S_IROTH; 327 ub->st_size = sbavail(sb) - sb->sb_ctl; 328 SOCK_RECVBUF_UNLOCK(so); 329 330 sb = &so->so_snd; 331 SOCK_SENDBUF_LOCK(so); 332 if ((sb->sb_state & SBS_CANTSENDMORE) == 0) 333 ub->st_mode |= S_IWUSR | S_IWGRP | S_IWOTH; 334 SOCK_SENDBUF_UNLOCK(so); 335 } 336 ub->st_uid = so->so_cred->cr_uid; 337 ub->st_gid = so->so_cred->cr_gid; 338 if (so->so_proto->pr_sense) 339 error = so->so_proto->pr_sense(so, ub); 340 SOCK_UNLOCK(so); 341 return (error); 342 } 343 344 /* 345 * API socket close on file pointer. We call soclose() to close the socket 346 * (including initiating closing protocols). soclose() will sorele() the 347 * file reference but the actual socket will not go away until the socket's 348 * ref count hits 0. 349 */ 350 static int 351 soo_close(struct file *fp, struct thread *td) 352 { 353 int error = 0; 354 struct socket *so; 355 356 so = fp->f_data; 357 fp->f_ops = &badfileops; 358 fp->f_data = NULL; 359 360 if (so) 361 error = soclose(so); 362 return (error); 363 } 364 365 static int 366 soo_chmod(struct file *fp, mode_t mode, struct ucred *cred, struct thread *td) 367 { 368 struct socket *so; 369 int error; 370 371 so = fp->f_data; 372 if (so->so_proto->pr_chmod != NULL) 373 error = so->so_proto->pr_chmod(so, mode, cred, td); 374 else 375 error = EINVAL; 376 return (error); 377 } 378 379 static int 380 soo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) 381 { 382 struct sockaddr_storage ss = { .ss_len = sizeof(ss) }; 383 struct unpcb *unpcb; 384 struct socket *so; 385 int error; 386 387 kif->kf_type = KF_TYPE_SOCKET; 388 so = fp->f_data; 389 CURVNET_SET(so->so_vnet); 390 kif->kf_un.kf_sock.kf_sock_domain0 = 391 so->so_proto->pr_domain->dom_family; 392 kif->kf_un.kf_sock.kf_sock_type0 = so->so_type; 393 kif->kf_un.kf_sock.kf_sock_protocol0 = so->so_proto->pr_protocol; 394 kif->kf_un.kf_sock.kf_sock_pcb = (uintptr_t)so->so_pcb; 395 switch (kif->kf_un.kf_sock.kf_sock_domain0) { 396 case AF_INET: 397 case AF_INET6: 398 /* XXX: kf_sock_inpcb is obsolete. It may be removed. */ 399 kif->kf_un.kf_sock.kf_sock_inpcb = (uintptr_t)so->so_pcb; 400 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 401 so->so_rcv.sb_state; 402 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 403 so->so_snd.sb_state; 404 kif->kf_un.kf_sock.kf_sock_sendq = 405 sbused(&so->so_snd); 406 kif->kf_un.kf_sock.kf_sock_recvq = 407 sbused(&so->so_rcv); 408 break; 409 case AF_UNIX: 410 if (so->so_pcb != NULL) { 411 unpcb = (struct unpcb *)(so->so_pcb); 412 if (unpcb->unp_conn) { 413 kif->kf_un.kf_sock.kf_sock_unpconn = 414 (uintptr_t)unpcb->unp_conn; 415 kif->kf_un.kf_sock.kf_sock_rcv_sb_state = 416 so->so_rcv.sb_state; 417 kif->kf_un.kf_sock.kf_sock_snd_sb_state = 418 so->so_snd.sb_state; 419 kif->kf_un.kf_sock.kf_sock_sendq = 420 sbused(&so->so_snd); 421 kif->kf_un.kf_sock.kf_sock_recvq = 422 sbused(&so->so_rcv); 423 } 424 } 425 break; 426 } 427 error = sosockaddr(so, (struct sockaddr *)&ss); 428 if (error == 0 && 429 ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_local)) { 430 bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_local, ss.ss_len); 431 } 432 ss.ss_len = sizeof(ss); 433 error = sopeeraddr(so, (struct sockaddr *)&ss); 434 if (error == 0 && 435 ss.ss_len <= sizeof(kif->kf_un.kf_sock.kf_sa_peer)) { 436 bcopy(&ss, &kif->kf_un.kf_sock.kf_sa_peer, ss.ss_len); 437 } 438 strncpy(kif->kf_path, so->so_proto->pr_domain->dom_name, 439 sizeof(kif->kf_path)); 440 CURVNET_RESTORE(); 441 return (0); 442 } 443 444 /* 445 * Use the 'backend3' field in AIO jobs to store the amount of data 446 * completed by the AIO job so far. 447 */ 448 #define aio_done backend3 449 450 static STAILQ_HEAD(, task) soaio_jobs; 451 static struct mtx soaio_jobs_lock; 452 static struct task soaio_kproc_task; 453 static int soaio_starting, soaio_idle, soaio_queued; 454 static struct unrhdr *soaio_kproc_unr; 455 456 static int soaio_max_procs = MAX_AIO_PROCS; 457 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, max_procs, CTLFLAG_RW, &soaio_max_procs, 0, 458 "Maximum number of kernel processes to use for async socket IO"); 459 460 static int soaio_num_procs; 461 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, num_procs, CTLFLAG_RD, &soaio_num_procs, 0, 462 "Number of active kernel processes for async socket IO"); 463 464 static int soaio_target_procs = TARGET_AIO_PROCS; 465 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, target_procs, CTLFLAG_RD, 466 &soaio_target_procs, 0, 467 "Preferred number of ready kernel processes for async socket IO"); 468 469 static int soaio_lifetime; 470 SYSCTL_INT(_kern_ipc_aio, OID_AUTO, lifetime, CTLFLAG_RW, &soaio_lifetime, 0, 471 "Maximum lifetime for idle aiod"); 472 473 static void 474 soaio_kproc_loop(void *arg) 475 { 476 struct proc *p; 477 struct vmspace *myvm; 478 struct task *task; 479 int error, id, pending; 480 481 id = (intptr_t)arg; 482 483 /* 484 * Grab an extra reference on the daemon's vmspace so that it 485 * doesn't get freed by jobs that switch to a different 486 * vmspace. 487 */ 488 p = curproc; 489 myvm = vmspace_acquire_ref(p); 490 491 mtx_lock(&soaio_jobs_lock); 492 MPASS(soaio_starting > 0); 493 soaio_starting--; 494 for (;;) { 495 while (!STAILQ_EMPTY(&soaio_jobs)) { 496 task = STAILQ_FIRST(&soaio_jobs); 497 STAILQ_REMOVE_HEAD(&soaio_jobs, ta_link); 498 soaio_queued--; 499 pending = task->ta_pending; 500 task->ta_pending = 0; 501 mtx_unlock(&soaio_jobs_lock); 502 503 task->ta_func(task->ta_context, pending); 504 505 mtx_lock(&soaio_jobs_lock); 506 } 507 MPASS(soaio_queued == 0); 508 509 if (p->p_vmspace != myvm) { 510 mtx_unlock(&soaio_jobs_lock); 511 vmspace_switch_aio(myvm); 512 mtx_lock(&soaio_jobs_lock); 513 continue; 514 } 515 516 soaio_idle++; 517 error = mtx_sleep(&soaio_idle, &soaio_jobs_lock, 0, "-", 518 soaio_lifetime); 519 soaio_idle--; 520 if (error == EWOULDBLOCK && STAILQ_EMPTY(&soaio_jobs) && 521 soaio_num_procs > soaio_target_procs) 522 break; 523 } 524 soaio_num_procs--; 525 mtx_unlock(&soaio_jobs_lock); 526 free_unr(soaio_kproc_unr, id); 527 kproc_exit(0); 528 } 529 530 static void 531 soaio_kproc_create(void *context, int pending) 532 { 533 struct proc *p; 534 int error, id; 535 536 mtx_lock(&soaio_jobs_lock); 537 for (;;) { 538 if (soaio_num_procs < soaio_target_procs) { 539 /* Must create */ 540 } else if (soaio_num_procs >= soaio_max_procs) { 541 /* 542 * Hit the limit on kernel processes, don't 543 * create another one. 544 */ 545 break; 546 } else if (soaio_queued <= soaio_idle + soaio_starting) { 547 /* 548 * No more AIO jobs waiting for a process to be 549 * created, so stop. 550 */ 551 break; 552 } 553 soaio_starting++; 554 mtx_unlock(&soaio_jobs_lock); 555 556 id = alloc_unr(soaio_kproc_unr); 557 error = kproc_create(soaio_kproc_loop, (void *)(intptr_t)id, 558 &p, 0, 0, "soaiod%d", id); 559 if (error != 0) { 560 free_unr(soaio_kproc_unr, id); 561 mtx_lock(&soaio_jobs_lock); 562 soaio_starting--; 563 break; 564 } 565 566 mtx_lock(&soaio_jobs_lock); 567 soaio_num_procs++; 568 } 569 mtx_unlock(&soaio_jobs_lock); 570 } 571 572 void 573 soaio_enqueue(struct task *task) 574 { 575 576 mtx_lock(&soaio_jobs_lock); 577 MPASS(task->ta_pending == 0); 578 task->ta_pending++; 579 STAILQ_INSERT_TAIL(&soaio_jobs, task, ta_link); 580 soaio_queued++; 581 if (soaio_queued <= soaio_idle) 582 wakeup_one(&soaio_idle); 583 else if (soaio_num_procs < soaio_max_procs) 584 taskqueue_enqueue(taskqueue_thread, &soaio_kproc_task); 585 mtx_unlock(&soaio_jobs_lock); 586 } 587 588 static void 589 soaio_init(void) 590 { 591 592 soaio_lifetime = AIOD_LIFETIME_DEFAULT; 593 STAILQ_INIT(&soaio_jobs); 594 mtx_init(&soaio_jobs_lock, "soaio jobs", NULL, MTX_DEF); 595 soaio_kproc_unr = new_unrhdr(1, INT_MAX, NULL); 596 TASK_INIT(&soaio_kproc_task, 0, soaio_kproc_create, NULL); 597 } 598 SYSINIT(soaio, SI_SUB_VFS, SI_ORDER_ANY, soaio_init, NULL); 599 600 static __inline int 601 soaio_ready(struct socket *so, struct sockbuf *sb) 602 { 603 return (sb == &so->so_rcv ? soreadable(so) : sowriteable(so)); 604 } 605 606 static void 607 soaio_process_job(struct socket *so, sb_which which, struct kaiocb *job) 608 { 609 struct ucred *td_savedcred; 610 struct thread *td; 611 struct sockbuf *sb = sobuf(so, which); 612 #ifdef MAC 613 struct file *fp = job->fd_file; 614 #endif 615 size_t cnt, done, job_total_nbytes __diagused; 616 long ru_before; 617 int error, flags; 618 619 SOCK_BUF_UNLOCK(so, which); 620 aio_switch_vmspace(job); 621 td = curthread; 622 retry: 623 td_savedcred = td->td_ucred; 624 td->td_ucred = job->cred; 625 626 job_total_nbytes = job->uiop->uio_resid + job->aio_done; 627 done = job->aio_done; 628 cnt = job->uiop->uio_resid; 629 job->uiop->uio_offset = 0; 630 job->uiop->uio_td = td; 631 flags = MSG_NBIO; 632 633 /* 634 * For resource usage accounting, only count a completed request 635 * as a single message to avoid counting multiple calls to 636 * sosend/soreceive on a blocking socket. 637 */ 638 639 if (sb == &so->so_rcv) { 640 ru_before = td->td_ru.ru_msgrcv; 641 #ifdef MAC 642 error = mac_socket_check_receive(fp->f_cred, so); 643 if (error == 0) 644 645 #endif 646 error = soreceive(so, NULL, job->uiop, NULL, NULL, 647 &flags); 648 if (td->td_ru.ru_msgrcv != ru_before) 649 job->msgrcv = 1; 650 } else { 651 if (!TAILQ_EMPTY(&sb->sb_aiojobq)) 652 flags |= MSG_MORETOCOME; 653 ru_before = td->td_ru.ru_msgsnd; 654 #ifdef MAC 655 error = mac_socket_check_send(fp->f_cred, so); 656 if (error == 0) 657 #endif 658 error = sousrsend(so, NULL, job->uiop, NULL, flags, 659 job->userproc); 660 if (td->td_ru.ru_msgsnd != ru_before) 661 job->msgsnd = 1; 662 } 663 664 done += cnt - job->uiop->uio_resid; 665 job->aio_done = done; 666 td->td_ucred = td_savedcred; 667 668 if (error == EWOULDBLOCK) { 669 /* 670 * The request was either partially completed or not 671 * completed at all due to racing with a read() or 672 * write() on the socket. If the socket is 673 * non-blocking, return with any partial completion. 674 * If the socket is blocking or if no progress has 675 * been made, requeue this request at the head of the 676 * queue to try again when the socket is ready. 677 */ 678 MPASS(done != job_total_nbytes); 679 SOCK_BUF_LOCK(so, which); 680 if (done == 0 || !(so->so_state & SS_NBIO)) { 681 empty_results++; 682 if (soaio_ready(so, sb)) { 683 empty_retries++; 684 SOCK_BUF_UNLOCK(so, which); 685 goto retry; 686 } 687 688 if (!aio_set_cancel_function(job, soo_aio_cancel)) { 689 SOCK_BUF_UNLOCK(so, which); 690 if (done != 0) 691 aio_complete(job, done, 0); 692 else 693 aio_cancel(job); 694 SOCK_BUF_LOCK(so, which); 695 } else { 696 TAILQ_INSERT_HEAD(&sb->sb_aiojobq, job, list); 697 } 698 return; 699 } 700 SOCK_BUF_UNLOCK(so, which); 701 } 702 if (done != 0 && (error == ERESTART || error == EINTR || 703 error == EWOULDBLOCK)) 704 error = 0; 705 if (error) 706 aio_complete(job, -1, error); 707 else 708 aio_complete(job, done, 0); 709 SOCK_BUF_LOCK(so, which); 710 } 711 712 static void 713 soaio_process_sb(struct socket *so, sb_which which) 714 { 715 struct kaiocb *job; 716 struct sockbuf *sb = sobuf(so, which); 717 718 CURVNET_SET(so->so_vnet); 719 SOCK_BUF_LOCK(so, which); 720 while (!TAILQ_EMPTY(&sb->sb_aiojobq) && soaio_ready(so, sb)) { 721 job = TAILQ_FIRST(&sb->sb_aiojobq); 722 TAILQ_REMOVE(&sb->sb_aiojobq, job, list); 723 if (!aio_clear_cancel_function(job)) 724 continue; 725 726 soaio_process_job(so, which, job); 727 } 728 729 /* 730 * If there are still pending requests, the socket must not be 731 * ready so set SB_AIO to request a wakeup when the socket 732 * becomes ready. 733 */ 734 if (!TAILQ_EMPTY(&sb->sb_aiojobq)) 735 sb->sb_flags |= SB_AIO; 736 sb->sb_flags &= ~SB_AIO_RUNNING; 737 SOCK_BUF_UNLOCK(so, which); 738 739 sorele(so); 740 CURVNET_RESTORE(); 741 } 742 743 void 744 soaio_rcv(void *context, int pending) 745 { 746 struct socket *so; 747 748 so = context; 749 soaio_process_sb(so, SO_RCV); 750 } 751 752 void 753 soaio_snd(void *context, int pending) 754 { 755 struct socket *so; 756 757 so = context; 758 soaio_process_sb(so, SO_SND); 759 } 760 761 void 762 sowakeup_aio(struct socket *so, sb_which which) 763 { 764 struct sockbuf *sb = sobuf(so, which); 765 766 SOCK_BUF_LOCK_ASSERT(so, which); 767 768 sb->sb_flags &= ~SB_AIO; 769 if (sb->sb_flags & SB_AIO_RUNNING) 770 return; 771 sb->sb_flags |= SB_AIO_RUNNING; 772 soref(so); 773 soaio_enqueue(&sb->sb_aiotask); 774 } 775 776 static void 777 soo_aio_cancel(struct kaiocb *job) 778 { 779 struct socket *so; 780 struct sockbuf *sb; 781 long done; 782 int opcode; 783 sb_which which; 784 785 so = job->fd_file->f_data; 786 opcode = job->uaiocb.aio_lio_opcode; 787 if (opcode & LIO_READ) { 788 sb = &so->so_rcv; 789 which = SO_RCV; 790 } else { 791 MPASS(opcode & LIO_WRITE); 792 sb = &so->so_snd; 793 which = SO_SND; 794 } 795 796 SOCK_BUF_LOCK(so, which); 797 if (!aio_cancel_cleared(job)) 798 TAILQ_REMOVE(&sb->sb_aiojobq, job, list); 799 if (TAILQ_EMPTY(&sb->sb_aiojobq)) 800 sb->sb_flags &= ~SB_AIO; 801 SOCK_BUF_UNLOCK(so, which); 802 803 done = job->aio_done; 804 if (done != 0) 805 aio_complete(job, done, 0); 806 else 807 aio_cancel(job); 808 } 809 810 static int 811 soo_aio_queue(struct file *fp, struct kaiocb *job) 812 { 813 struct socket *so = fp->f_data; 814 815 return (so->so_proto->pr_aio_queue(so, job)); 816 } 817 818 int 819 soaio_queue_generic(struct socket *so, struct kaiocb *job) 820 { 821 struct sockbuf *sb; 822 sb_which which; 823 824 /* Lock through the socket, since this may be a listening socket. */ 825 switch (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) { 826 case LIO_READ: 827 SOCK_RECVBUF_LOCK(so); 828 sb = &so->so_rcv; 829 which = SO_RCV; 830 break; 831 case LIO_WRITE: 832 SOCK_SENDBUF_LOCK(so); 833 sb = &so->so_snd; 834 which = SO_SND; 835 break; 836 default: 837 return (EINVAL); 838 } 839 840 if (SOLISTENING(so)) { 841 SOCK_BUF_UNLOCK(so, which); 842 return (EINVAL); 843 } 844 845 if (!aio_set_cancel_function(job, soo_aio_cancel)) 846 panic("new job was cancelled"); 847 TAILQ_INSERT_TAIL(&sb->sb_aiojobq, job, list); 848 if (!(sb->sb_flags & SB_AIO_RUNNING)) { 849 if (soaio_ready(so, sb)) 850 sowakeup_aio(so, which); 851 else 852 sb->sb_flags |= SB_AIO; 853 } 854 SOCK_BUF_UNLOCK(so, which); 855 return (0); 856 } 857