1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993 5 * The Regents of the University of California. 6 * Copyright (c) 2004 The FreeBSD Foundation 7 * Copyright (c) 2004-2008 Robert N. M. Watson 8 * All rights reserved. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 /* 36 * Comments on the socket life cycle: 37 * 38 * soalloc() sets of socket layer state for a socket, called only by 39 * socreate() and sonewconn(). Socket layer private. 40 * 41 * sodealloc() tears down socket layer state for a socket, called only by 42 * sofree() and sonewconn(). Socket layer private. 43 * 44 * pru_attach() associates protocol layer state with an allocated socket; 45 * called only once, may fail, aborting socket allocation. This is called 46 * from socreate() and sonewconn(). Socket layer private. 47 * 48 * pru_detach() disassociates protocol layer state from an attached socket, 49 * and will be called exactly once for sockets in which pru_attach() has 50 * been successfully called. If pru_attach() returned an error, 51 * pru_detach() will not be called. Socket layer private. 52 * 53 * pru_abort() and pru_close() notify the protocol layer that the last 54 * consumer of a socket is starting to tear down the socket, and that the 55 * protocol should terminate the connection. Historically, pru_abort() also 56 * detached protocol state from the socket state, but this is no longer the 57 * case. 58 * 59 * socreate() creates a socket and attaches protocol state. This is a public 60 * interface that may be used by socket layer consumers to create new 61 * sockets. 62 * 63 * sonewconn() creates a socket and attaches protocol state. This is a 64 * public interface that may be used by protocols to create new sockets when 65 * a new connection is received and will be available for accept() on a 66 * listen socket. 67 * 68 * soclose() destroys a socket after possibly waiting for it to disconnect. 69 * This is a public interface that socket consumers should use to close and 70 * release a socket when done with it. 71 * 72 * soabort() destroys a socket without waiting for it to disconnect (used 73 * only for incoming connections that are already partially or fully 74 * connected). This is used internally by the socket layer when clearing 75 * listen socket queues (due to overflow or close on the listen socket), but 76 * is also a public interface protocols may use to abort connections in 77 * their incomplete listen queues should they no longer be required. Sockets 78 * placed in completed connection listen queues should not be aborted for 79 * reasons described in the comment above the soclose() implementation. This 80 * is not a general purpose close routine, and except in the specific 81 * circumstances described here, should not be used. 82 * 83 * sofree() will free a socket and its protocol state if all references on 84 * the socket have been released, and is the public interface to attempt to 85 * free a socket when a reference is removed. This is a socket layer private 86 * interface. 87 * 88 * NOTE: In addition to socreate() and soclose(), which provide a single 89 * socket reference to the consumer to be managed as required, there are two 90 * calls to explicitly manage socket references, soref(), and sorele(). 91 * Currently, these are generally required only when transitioning a socket 92 * from a listen queue to a file descriptor, in order to prevent garbage 93 * collection of the socket at an untimely moment. For a number of reasons, 94 * these interfaces are not preferred, and should be avoided. 95 * 96 * NOTE: With regard to VNETs the general rule is that callers do not set 97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(), 98 * sofree(), sorele(), sonewconn() and sorflush(), which are usually called 99 * from a pre-set VNET context. sopoll_generic() currently does not need a 100 * VNET context to be set. 101 */ 102 103 #include <sys/cdefs.h> 104 #include "opt_inet.h" 105 #include "opt_inet6.h" 106 #include "opt_kern_tls.h" 107 #include "opt_ktrace.h" 108 #include "opt_sctp.h" 109 110 #include <sys/param.h> 111 #include <sys/systm.h> 112 #include <sys/capsicum.h> 113 #include <sys/fcntl.h> 114 #include <sys/limits.h> 115 #include <sys/lock.h> 116 #include <sys/mac.h> 117 #include <sys/malloc.h> 118 #include <sys/mbuf.h> 119 #include <sys/mutex.h> 120 #include <sys/domain.h> 121 #include <sys/file.h> /* for struct knote */ 122 #include <sys/hhook.h> 123 #include <sys/kernel.h> 124 #include <sys/khelp.h> 125 #include <sys/kthread.h> 126 #include <sys/ktls.h> 127 #include <sys/event.h> 128 #include <sys/eventhandler.h> 129 #include <sys/poll.h> 130 #include <sys/proc.h> 131 #include <sys/protosw.h> 132 #include <sys/sbuf.h> 133 #include <sys/socket.h> 134 #include <sys/socketvar.h> 135 #include <sys/resourcevar.h> 136 #include <net/route.h> 137 #include <sys/sched.h> 138 #include <sys/signalvar.h> 139 #include <sys/smp.h> 140 #include <sys/stat.h> 141 #include <sys/sx.h> 142 #include <sys/sysctl.h> 143 #include <sys/taskqueue.h> 144 #include <sys/uio.h> 145 #include <sys/un.h> 146 #include <sys/unpcb.h> 147 #include <sys/jail.h> 148 #include <sys/syslog.h> 149 #include <netinet/in.h> 150 #include <netinet/in_pcb.h> 151 #include <netinet/tcp.h> 152 153 #include <net/vnet.h> 154 155 #include <security/mac/mac_framework.h> 156 #include <security/mac/mac_internal.h> 157 158 #include <vm/uma.h> 159 160 #ifdef COMPAT_FREEBSD32 161 #include <sys/mount.h> 162 #include <sys/sysent.h> 163 #include <compat/freebsd32/freebsd32.h> 164 #endif 165 166 static int soreceive_generic_locked(struct socket *so, 167 struct sockaddr **psa, struct uio *uio, struct mbuf **mp, 168 struct mbuf **controlp, int *flagsp); 169 static int soreceive_rcvoob(struct socket *so, struct uio *uio, 170 int flags); 171 static int soreceive_stream_locked(struct socket *so, struct sockbuf *sb, 172 struct sockaddr **psa, struct uio *uio, struct mbuf **mp, 173 struct mbuf **controlp, int flags); 174 static int sosend_generic_locked(struct socket *so, struct sockaddr *addr, 175 struct uio *uio, struct mbuf *top, struct mbuf *control, 176 int flags, struct thread *td); 177 static void so_rdknl_lock(void *); 178 static void so_rdknl_unlock(void *); 179 static void so_rdknl_assert_lock(void *, int); 180 static void so_wrknl_lock(void *); 181 static void so_wrknl_unlock(void *); 182 static void so_wrknl_assert_lock(void *, int); 183 184 static void filt_sordetach(struct knote *kn); 185 static int filt_soread(struct knote *kn, long hint); 186 static void filt_sowdetach(struct knote *kn); 187 static int filt_sowrite(struct knote *kn, long hint); 188 static int filt_soempty(struct knote *kn, long hint); 189 fo_kqfilter_t soo_kqfilter; 190 191 static const struct filterops soread_filtops = { 192 .f_isfd = 1, 193 .f_detach = filt_sordetach, 194 .f_event = filt_soread, 195 }; 196 static const struct filterops sowrite_filtops = { 197 .f_isfd = 1, 198 .f_detach = filt_sowdetach, 199 .f_event = filt_sowrite, 200 }; 201 static const struct filterops soempty_filtops = { 202 .f_isfd = 1, 203 .f_detach = filt_sowdetach, 204 .f_event = filt_soempty, 205 }; 206 207 so_gen_t so_gencnt; /* generation count for sockets */ 208 209 MALLOC_DEFINE(M_SONAME, "soname", "socket name"); 210 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block"); 211 212 #define VNET_SO_ASSERT(so) \ 213 VNET_ASSERT(curvnet != NULL, \ 214 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so))); 215 216 #ifdef SOCKET_HHOOK 217 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]); 218 #define V_socket_hhh VNET(socket_hhh) 219 static inline int hhook_run_socket(struct socket *, void *, int32_t); 220 #endif 221 222 #ifdef COMPAT_FREEBSD32 223 #ifdef __amd64__ 224 /* off_t has 4-byte alignment on i386 but not on other 32-bit platforms. */ 225 #define __splice32_packed __packed 226 #else 227 #define __splice32_packed 228 #endif 229 struct splice32 { 230 int32_t sp_fd; 231 int64_t sp_max; 232 struct timeval32 sp_idle; 233 } __splice32_packed; 234 #undef __splice32_packed 235 #endif 236 237 /* 238 * Limit on the number of connections in the listen queue waiting 239 * for accept(2). 240 * NB: The original sysctl somaxconn is still available but hidden 241 * to prevent confusion about the actual purpose of this number. 242 */ 243 VNET_DEFINE_STATIC(u_int, somaxconn) = SOMAXCONN; 244 #define V_somaxconn VNET(somaxconn) 245 246 static int 247 sysctl_somaxconn(SYSCTL_HANDLER_ARGS) 248 { 249 int error; 250 u_int val; 251 252 val = V_somaxconn; 253 error = sysctl_handle_int(oidp, &val, 0, req); 254 if (error || !req->newptr ) 255 return (error); 256 257 /* 258 * The purpose of the UINT_MAX / 3 limit, is so that the formula 259 * 3 * sol_qlimit / 2 260 * below, will not overflow. 261 */ 262 263 if (val < 1 || val > UINT_MAX / 3) 264 return (EINVAL); 265 266 V_somaxconn = val; 267 return (0); 268 } 269 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, 270 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, sizeof(u_int), 271 sysctl_somaxconn, "IU", 272 "Maximum listen socket pending connection accept queue size"); 273 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, 274 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, 275 sizeof(u_int), sysctl_somaxconn, "IU", 276 "Maximum listen socket pending connection accept queue size (compat)"); 277 278 static u_int numopensockets; 279 static int 280 sysctl_numopensockets(SYSCTL_HANDLER_ARGS) 281 { 282 u_int val; 283 284 #ifdef VIMAGE 285 if(!IS_DEFAULT_VNET(curvnet)) 286 val = curvnet->vnet_sockcnt; 287 else 288 #endif 289 val = numopensockets; 290 return (sysctl_handle_int(oidp, &val, 0, req)); 291 } 292 SYSCTL_PROC(_kern_ipc, OID_AUTO, numopensockets, 293 CTLTYPE_UINT | CTLFLAG_RD | CTLFLAG_MPSAFE | CTLFLAG_VNET, 0, sizeof(u_int), 294 sysctl_numopensockets, "IU", "Number of open sockets"); 295 296 /* 297 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket 298 * so_gencnt field. 299 */ 300 static struct mtx so_global_mtx; 301 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF); 302 303 /* 304 * General IPC sysctl name space, used by sockets and a variety of other IPC 305 * types. 306 */ 307 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 308 "IPC"); 309 310 /* 311 * Initialize the socket subsystem and set up the socket 312 * memory allocator. 313 */ 314 static uma_zone_t socket_zone; 315 int maxsockets; 316 317 static void 318 socket_zone_change(void *tag) 319 { 320 321 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 322 } 323 324 static int splice_init_state; 325 static struct sx splice_init_lock; 326 SX_SYSINIT(splice_init_lock, &splice_init_lock, "splice_init"); 327 328 static SYSCTL_NODE(_kern_ipc, OID_AUTO, splice, CTLFLAG_RW, 0, 329 "Settings relating to the SO_SPLICE socket option"); 330 331 static bool splice_receive_stream = true; 332 SYSCTL_BOOL(_kern_ipc_splice, OID_AUTO, receive_stream, CTLFLAG_RWTUN, 333 &splice_receive_stream, 0, 334 "Use soreceive_stream() for stream splices"); 335 336 static uma_zone_t splice_zone; 337 static struct proc *splice_proc; 338 struct splice_wq { 339 struct mtx mtx; 340 STAILQ_HEAD(, so_splice) head; 341 bool running; 342 } __aligned(CACHE_LINE_SIZE); 343 static struct splice_wq *splice_wq; 344 static uint32_t splice_index = 0; 345 346 static void so_splice_timeout(void *arg, int pending); 347 static void so_splice_xfer(struct so_splice *s); 348 static int so_unsplice(struct socket *so, bool timeout); 349 350 static void 351 splice_work_thread(void *ctx) 352 { 353 struct splice_wq *wq = ctx; 354 struct so_splice *s, *s_temp; 355 STAILQ_HEAD(, so_splice) local_head; 356 int cpu; 357 358 cpu = wq - splice_wq; 359 if (bootverbose) 360 printf("starting so_splice worker thread for CPU %d\n", cpu); 361 362 for (;;) { 363 mtx_lock(&wq->mtx); 364 while (STAILQ_EMPTY(&wq->head)) { 365 wq->running = false; 366 mtx_sleep(wq, &wq->mtx, 0, "-", 0); 367 wq->running = true; 368 } 369 STAILQ_INIT(&local_head); 370 STAILQ_CONCAT(&local_head, &wq->head); 371 STAILQ_INIT(&wq->head); 372 mtx_unlock(&wq->mtx); 373 STAILQ_FOREACH_SAFE(s, &local_head, next, s_temp) { 374 mtx_lock(&s->mtx); 375 CURVNET_SET(s->src->so_vnet); 376 so_splice_xfer(s); 377 CURVNET_RESTORE(); 378 } 379 } 380 } 381 382 static void 383 so_splice_dispatch_async(struct so_splice *sp) 384 { 385 struct splice_wq *wq; 386 bool running; 387 388 wq = &splice_wq[sp->wq_index]; 389 mtx_lock(&wq->mtx); 390 STAILQ_INSERT_TAIL(&wq->head, sp, next); 391 running = wq->running; 392 mtx_unlock(&wq->mtx); 393 if (!running) 394 wakeup(wq); 395 } 396 397 void 398 so_splice_dispatch(struct so_splice *sp) 399 { 400 mtx_assert(&sp->mtx, MA_OWNED); 401 402 if (sp->state != SPLICE_IDLE) { 403 mtx_unlock(&sp->mtx); 404 } else { 405 sp->state = SPLICE_QUEUED; 406 mtx_unlock(&sp->mtx); 407 so_splice_dispatch_async(sp); 408 } 409 } 410 411 static int 412 splice_zinit(void *mem, int size __unused, int flags __unused) 413 { 414 struct so_splice *s; 415 416 s = (struct so_splice *)mem; 417 mtx_init(&s->mtx, "so_splice", NULL, MTX_DEF); 418 return (0); 419 } 420 421 static void 422 splice_zfini(void *mem, int size) 423 { 424 struct so_splice *s; 425 426 s = (struct so_splice *)mem; 427 mtx_destroy(&s->mtx); 428 } 429 430 static int 431 splice_init(void) 432 { 433 struct thread *td; 434 int error, i, state; 435 436 state = atomic_load_acq_int(&splice_init_state); 437 if (__predict_true(state > 0)) 438 return (0); 439 if (state < 0) 440 return (ENXIO); 441 sx_xlock(&splice_init_lock); 442 if (splice_init_state != 0) { 443 sx_xunlock(&splice_init_lock); 444 return (0); 445 } 446 447 splice_zone = uma_zcreate("splice", sizeof(struct so_splice), NULL, 448 NULL, splice_zinit, splice_zfini, UMA_ALIGN_CACHE, 0); 449 450 splice_wq = mallocarray(mp_maxid + 1, sizeof(*splice_wq), M_TEMP, 451 M_WAITOK | M_ZERO); 452 453 /* 454 * Initialize the workqueues to run the splice work. We create a 455 * work queue for each CPU. 456 */ 457 CPU_FOREACH(i) { 458 STAILQ_INIT(&splice_wq[i].head); 459 mtx_init(&splice_wq[i].mtx, "splice work queue", NULL, MTX_DEF); 460 } 461 462 /* Start kthreads for each workqueue. */ 463 error = 0; 464 CPU_FOREACH(i) { 465 error = kproc_kthread_add(splice_work_thread, &splice_wq[i], 466 &splice_proc, &td, 0, 0, "so_splice", "thr_%d", i); 467 if (error) { 468 printf("Can't add so_splice thread %d error %d\n", 469 i, error); 470 break; 471 } 472 473 /* 474 * It's possible to create loops with SO_SPLICE; ensure that 475 * worker threads aren't able to starve the system too easily. 476 */ 477 thread_lock(td); 478 sched_prio(td, PUSER); 479 thread_unlock(td); 480 } 481 482 splice_init_state = error != 0 ? -1 : 1; 483 sx_xunlock(&splice_init_lock); 484 485 return (error); 486 } 487 488 /* 489 * Lock a pair of socket's I/O locks for splicing. Avoid blocking while holding 490 * one lock in order to avoid potential deadlocks in case there is some other 491 * code path which acquires more than one I/O lock at a time. 492 */ 493 static void 494 splice_lock_pair(struct socket *so_src, struct socket *so_dst) 495 { 496 int error; 497 498 for (;;) { 499 error = SOCK_IO_SEND_LOCK(so_dst, SBL_WAIT | SBL_NOINTR); 500 KASSERT(error == 0, 501 ("%s: failed to lock send I/O lock: %d", __func__, error)); 502 error = SOCK_IO_RECV_LOCK(so_src, 0); 503 KASSERT(error == 0 || error == EWOULDBLOCK, 504 ("%s: failed to lock recv I/O lock: %d", __func__, error)); 505 if (error == 0) 506 break; 507 SOCK_IO_SEND_UNLOCK(so_dst); 508 509 error = SOCK_IO_RECV_LOCK(so_src, SBL_WAIT | SBL_NOINTR); 510 KASSERT(error == 0, 511 ("%s: failed to lock recv I/O lock: %d", __func__, error)); 512 error = SOCK_IO_SEND_LOCK(so_dst, 0); 513 KASSERT(error == 0 || error == EWOULDBLOCK, 514 ("%s: failed to lock send I/O lock: %d", __func__, error)); 515 if (error == 0) 516 break; 517 SOCK_IO_RECV_UNLOCK(so_src); 518 } 519 } 520 521 static void 522 splice_unlock_pair(struct socket *so_src, struct socket *so_dst) 523 { 524 SOCK_IO_RECV_UNLOCK(so_src); 525 SOCK_IO_SEND_UNLOCK(so_dst); 526 } 527 528 /* 529 * Move data from the source to the sink. Assumes that both of the relevant 530 * socket I/O locks are held. 531 */ 532 static int 533 so_splice_xfer_data(struct socket *so_src, struct socket *so_dst, off_t max, 534 ssize_t *lenp) 535 { 536 struct uio uio; 537 struct mbuf *m; 538 struct sockbuf *sb_src, *sb_dst; 539 ssize_t len; 540 long space; 541 int error, flags; 542 543 SOCK_IO_RECV_ASSERT_LOCKED(so_src); 544 SOCK_IO_SEND_ASSERT_LOCKED(so_dst); 545 546 error = 0; 547 m = NULL; 548 memset(&uio, 0, sizeof(uio)); 549 550 sb_src = &so_src->so_rcv; 551 sb_dst = &so_dst->so_snd; 552 553 space = sbspace(sb_dst); 554 if (space < 0) 555 space = 0; 556 len = MIN(max, MIN(space, sbavail(sb_src))); 557 if (len == 0) { 558 SOCK_RECVBUF_LOCK(so_src); 559 if ((sb_src->sb_state & SBS_CANTRCVMORE) != 0) 560 error = EPIPE; 561 SOCK_RECVBUF_UNLOCK(so_src); 562 } else { 563 flags = MSG_DONTWAIT; 564 uio.uio_resid = len; 565 if (splice_receive_stream && sb_src->sb_tls_info == NULL) { 566 error = soreceive_stream_locked(so_src, sb_src, NULL, 567 &uio, &m, NULL, flags); 568 } else { 569 error = soreceive_generic_locked(so_src, NULL, 570 &uio, &m, NULL, &flags); 571 } 572 if (error != 0 && m != NULL) { 573 m_freem(m); 574 m = NULL; 575 } 576 } 577 if (m != NULL) { 578 len -= uio.uio_resid; 579 error = sosend_generic_locked(so_dst, NULL, NULL, m, NULL, 580 MSG_DONTWAIT, curthread); 581 } else if (error == 0) { 582 len = 0; 583 SOCK_SENDBUF_LOCK(so_dst); 584 if ((sb_dst->sb_state & SBS_CANTSENDMORE) != 0) 585 error = EPIPE; 586 SOCK_SENDBUF_UNLOCK(so_dst); 587 } 588 if (error == 0) 589 *lenp = len; 590 return (error); 591 } 592 593 /* 594 * Transfer data from the source to the sink. 595 */ 596 static void 597 so_splice_xfer(struct so_splice *sp) 598 { 599 struct socket *so_src, *so_dst; 600 off_t max; 601 ssize_t len; 602 int error; 603 604 mtx_assert(&sp->mtx, MA_OWNED); 605 KASSERT(sp->state == SPLICE_QUEUED || sp->state == SPLICE_CLOSING, 606 ("so_splice_xfer: invalid state %d", sp->state)); 607 KASSERT(sp->max != 0, ("so_splice_xfer: max == 0")); 608 609 if (sp->state == SPLICE_CLOSING) { 610 /* Userspace asked us to close the splice. */ 611 goto closing; 612 } 613 614 sp->state = SPLICE_RUNNING; 615 so_src = sp->src; 616 so_dst = sp->dst; 617 max = sp->max > 0 ? sp->max - so_src->so_splice_sent : OFF_MAX; 618 if (max < 0) 619 max = 0; 620 621 /* 622 * Lock the sockets in order to block userspace from doing anything 623 * sneaky. If an error occurs or one of the sockets can no longer 624 * transfer data, we will automatically unsplice. 625 */ 626 mtx_unlock(&sp->mtx); 627 splice_lock_pair(so_src, so_dst); 628 629 error = so_splice_xfer_data(so_src, so_dst, max, &len); 630 631 mtx_lock(&sp->mtx); 632 633 /* 634 * Update our stats while still holding the socket locks. This 635 * synchronizes with getsockopt(SO_SPLICE), see the comment there. 636 */ 637 if (error == 0) { 638 KASSERT(len >= 0, ("%s: len %zd < 0", __func__, len)); 639 so_src->so_splice_sent += len; 640 } 641 splice_unlock_pair(so_src, so_dst); 642 643 switch (sp->state) { 644 case SPLICE_CLOSING: 645 closing: 646 sp->state = SPLICE_CLOSED; 647 wakeup(sp); 648 mtx_unlock(&sp->mtx); 649 break; 650 case SPLICE_RUNNING: 651 if (error != 0 || 652 (sp->max > 0 && so_src->so_splice_sent >= sp->max)) { 653 sp->state = SPLICE_EXCEPTION; 654 soref(so_src); 655 mtx_unlock(&sp->mtx); 656 (void)so_unsplice(so_src, false); 657 sorele(so_src); 658 } else { 659 /* 660 * Locklessly check for additional bytes in the source's 661 * receive buffer and queue more work if possible. We 662 * may end up queuing needless work, but that's ok, and 663 * if we race with a thread inserting more data into the 664 * buffer and observe sbavail() == 0, the splice mutex 665 * ensures that splice_push() will queue more work for 666 * us. 667 */ 668 if (sbavail(&so_src->so_rcv) > 0 && 669 sbspace(&so_dst->so_snd) > 0) { 670 sp->state = SPLICE_QUEUED; 671 mtx_unlock(&sp->mtx); 672 so_splice_dispatch_async(sp); 673 } else { 674 sp->state = SPLICE_IDLE; 675 mtx_unlock(&sp->mtx); 676 } 677 } 678 break; 679 default: 680 __assert_unreachable(); 681 } 682 } 683 684 static void 685 socket_init(void *tag) 686 { 687 688 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL, 689 NULL, NULL, UMA_ALIGN_PTR, 0); 690 maxsockets = uma_zone_set_max(socket_zone, maxsockets); 691 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached"); 692 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL, 693 EVENTHANDLER_PRI_FIRST); 694 } 695 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL); 696 697 #ifdef SOCKET_HHOOK 698 static void 699 socket_hhook_register(int subtype) 700 { 701 702 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype, 703 &V_socket_hhh[subtype], 704 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 705 printf("%s: WARNING: unable to register hook\n", __func__); 706 } 707 708 static void 709 socket_hhook_deregister(int subtype) 710 { 711 712 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0) 713 printf("%s: WARNING: unable to deregister hook\n", __func__); 714 } 715 716 static void 717 socket_vnet_init(const void *unused __unused) 718 { 719 int i; 720 721 /* We expect a contiguous range */ 722 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 723 socket_hhook_register(i); 724 } 725 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 726 socket_vnet_init, NULL); 727 728 static void 729 socket_vnet_uninit(const void *unused __unused) 730 { 731 int i; 732 733 for (i = 0; i <= HHOOK_SOCKET_LAST; i++) 734 socket_hhook_deregister(i); 735 } 736 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 737 socket_vnet_uninit, NULL); 738 #endif /* SOCKET_HHOOK */ 739 740 /* 741 * Initialise maxsockets. This SYSINIT must be run after 742 * tunable_mbinit(). 743 */ 744 static void 745 init_maxsockets(void *ignored) 746 { 747 748 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 749 maxsockets = imax(maxsockets, maxfiles); 750 } 751 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL); 752 753 /* 754 * Sysctl to get and set the maximum global sockets limit. Notify protocols 755 * of the change so that they can update their dependent limits as required. 756 */ 757 static int 758 sysctl_maxsockets(SYSCTL_HANDLER_ARGS) 759 { 760 int error, newmaxsockets; 761 762 newmaxsockets = maxsockets; 763 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req); 764 if (error == 0 && req->newptr && newmaxsockets != maxsockets) { 765 if (newmaxsockets > maxsockets && 766 newmaxsockets <= maxfiles) { 767 maxsockets = newmaxsockets; 768 EVENTHANDLER_INVOKE(maxsockets_change); 769 } else 770 error = EINVAL; 771 } 772 return (error); 773 } 774 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, 775 CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE, 776 &maxsockets, 0, sysctl_maxsockets, "IU", 777 "Maximum number of sockets available"); 778 779 /* 780 * Socket operation routines. These routines are called by the routines in 781 * sys_socket.c or from a system process, and implement the semantics of 782 * socket operations by switching out to the protocol specific routines. 783 */ 784 785 /* 786 * Get a socket structure from our zone, and initialize it. Note that it 787 * would probably be better to allocate socket and PCB at the same time, but 788 * I'm not convinced that all the protocols can be easily modified to do 789 * this. 790 * 791 * soalloc() returns a socket with a ref count of 0. 792 */ 793 static struct socket * 794 soalloc(struct vnet *vnet) 795 { 796 struct socket *so; 797 798 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO); 799 if (so == NULL) 800 return (NULL); 801 #ifdef MAC 802 if (mac_socket_init(so, M_NOWAIT) != 0) { 803 uma_zfree(socket_zone, so); 804 return (NULL); 805 } 806 #endif 807 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) { 808 uma_zfree(socket_zone, so); 809 return (NULL); 810 } 811 812 /* 813 * The socket locking protocol allows to lock 2 sockets at a time, 814 * however, the first one must be a listening socket. WITNESS lacks 815 * a feature to change class of an existing lock, so we use DUPOK. 816 */ 817 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK); 818 mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF); 819 mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF); 820 so->so_rcv.sb_sel = &so->so_rdsel; 821 so->so_snd.sb_sel = &so->so_wrsel; 822 sx_init(&so->so_snd_sx, "so_snd_sx"); 823 sx_init(&so->so_rcv_sx, "so_rcv_sx"); 824 TAILQ_INIT(&so->so_snd.sb_aiojobq); 825 TAILQ_INIT(&so->so_rcv.sb_aiojobq); 826 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so); 827 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so); 828 #ifdef VIMAGE 829 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p", 830 __func__, __LINE__, so)); 831 so->so_vnet = vnet; 832 #endif 833 #ifdef SOCKET_HHOOK 834 /* We shouldn't need the so_global_mtx */ 835 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) { 836 /* Do we need more comprehensive error returns? */ 837 uma_zfree(socket_zone, so); 838 return (NULL); 839 } 840 #endif 841 mtx_lock(&so_global_mtx); 842 so->so_gencnt = ++so_gencnt; 843 ++numopensockets; 844 #ifdef VIMAGE 845 vnet->vnet_sockcnt++; 846 #endif 847 mtx_unlock(&so_global_mtx); 848 849 return (so); 850 } 851 852 /* 853 * Free the storage associated with a socket at the socket layer, tear down 854 * locks, labels, etc. All protocol state is assumed already to have been 855 * torn down (and possibly never set up) by the caller. 856 */ 857 void 858 sodealloc(struct socket *so) 859 { 860 861 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count)); 862 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL")); 863 864 mtx_lock(&so_global_mtx); 865 so->so_gencnt = ++so_gencnt; 866 --numopensockets; /* Could be below, but faster here. */ 867 #ifdef VIMAGE 868 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p", 869 __func__, __LINE__, so)); 870 so->so_vnet->vnet_sockcnt--; 871 #endif 872 mtx_unlock(&so_global_mtx); 873 #ifdef MAC 874 mac_socket_destroy(so); 875 #endif 876 #ifdef SOCKET_HHOOK 877 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE); 878 #endif 879 880 khelp_destroy_osd(&so->osd); 881 if (SOLISTENING(so)) { 882 if (so->sol_accept_filter != NULL) 883 accept_filt_setopt(so, NULL); 884 } else { 885 if (so->so_rcv.sb_hiwat) 886 (void)chgsbsize(so->so_cred->cr_uidinfo, 887 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY); 888 if (so->so_snd.sb_hiwat) 889 (void)chgsbsize(so->so_cred->cr_uidinfo, 890 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY); 891 sx_destroy(&so->so_snd_sx); 892 sx_destroy(&so->so_rcv_sx); 893 mtx_destroy(&so->so_snd_mtx); 894 mtx_destroy(&so->so_rcv_mtx); 895 } 896 crfree(so->so_cred); 897 mtx_destroy(&so->so_lock); 898 uma_zfree(socket_zone, so); 899 } 900 901 /* 902 * socreate returns a socket with a ref count of 1 and a file descriptor 903 * reference. The socket should be closed with soclose(). 904 */ 905 int 906 socreate(int dom, struct socket **aso, int type, int proto, 907 struct ucred *cred, struct thread *td) 908 { 909 struct protosw *prp; 910 struct socket *so; 911 int error; 912 913 /* 914 * XXX: divert(4) historically abused PF_INET. Keep this compatibility 915 * shim until all applications have been updated. 916 */ 917 if (__predict_false(dom == PF_INET && type == SOCK_RAW && 918 proto == IPPROTO_DIVERT)) { 919 dom = PF_DIVERT; 920 printf("%s uses obsolete way to create divert(4) socket\n", 921 td->td_proc->p_comm); 922 } 923 924 prp = pffindproto(dom, type, proto); 925 if (prp == NULL) { 926 /* No support for domain. */ 927 if (pffinddomain(dom) == NULL) 928 return (EAFNOSUPPORT); 929 /* No support for socket type. */ 930 if (proto == 0 && type != 0) 931 return (EPROTOTYPE); 932 return (EPROTONOSUPPORT); 933 } 934 935 MPASS(prp->pr_attach); 936 937 if ((prp->pr_flags & PR_CAPATTACH) == 0) { 938 if (CAP_TRACING(td)) 939 ktrcapfail(CAPFAIL_PROTO, &proto); 940 if (IN_CAPABILITY_MODE(td)) 941 return (ECAPMODE); 942 } 943 944 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0) 945 return (EPROTONOSUPPORT); 946 947 so = soalloc(CRED_TO_VNET(cred)); 948 if (so == NULL) 949 return (ENOBUFS); 950 951 so->so_type = type; 952 so->so_cred = crhold(cred); 953 if ((prp->pr_domain->dom_family == PF_INET) || 954 (prp->pr_domain->dom_family == PF_INET6) || 955 (prp->pr_domain->dom_family == PF_ROUTE)) 956 so->so_fibnum = td->td_proc->p_fibnum; 957 else 958 so->so_fibnum = 0; 959 so->so_proto = prp; 960 #ifdef MAC 961 mac_socket_create(cred, so); 962 #endif 963 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 964 so_rdknl_assert_lock); 965 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 966 so_wrknl_assert_lock); 967 if ((prp->pr_flags & PR_SOCKBUF) == 0) { 968 so->so_snd.sb_mtx = &so->so_snd_mtx; 969 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 970 } 971 /* 972 * Auto-sizing of socket buffers is managed by the protocols and 973 * the appropriate flags must be set in the pru_attach function. 974 */ 975 CURVNET_SET(so->so_vnet); 976 error = prp->pr_attach(so, proto, td); 977 CURVNET_RESTORE(); 978 if (error) { 979 sodealloc(so); 980 return (error); 981 } 982 soref(so); 983 *aso = so; 984 return (0); 985 } 986 987 #ifdef REGRESSION 988 static int regression_sonewconn_earlytest = 1; 989 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW, 990 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test"); 991 #endif 992 993 static int sooverprio = LOG_DEBUG; 994 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW, 995 &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable"); 996 997 static struct timeval overinterval = { 60, 0 }; 998 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW, 999 &overinterval, 1000 "Delay in seconds between warnings for listen socket overflows"); 1001 1002 /* 1003 * When an attempt at a new connection is noted on a socket which supports 1004 * accept(2), the protocol has two options: 1005 * 1) Call legacy sonewconn() function, which would call protocol attach 1006 * method, same as used for socket(2). 1007 * 2) Call solisten_clone(), do attach that is specific to a cloned connection, 1008 * and then call solisten_enqueue(). 1009 * 1010 * Note: the ref count on the socket is 0 on return. 1011 */ 1012 struct socket * 1013 solisten_clone(struct socket *head) 1014 { 1015 struct sbuf descrsb; 1016 struct socket *so; 1017 int len, overcount; 1018 u_int qlen; 1019 const char localprefix[] = "local:"; 1020 char descrbuf[SUNPATHLEN + sizeof(localprefix)]; 1021 #if defined(INET6) 1022 char addrbuf[INET6_ADDRSTRLEN]; 1023 #elif defined(INET) 1024 char addrbuf[INET_ADDRSTRLEN]; 1025 #endif 1026 bool dolog, over; 1027 1028 SOLISTEN_LOCK(head); 1029 over = (head->sol_qlen > 3 * head->sol_qlimit / 2); 1030 #ifdef REGRESSION 1031 if (regression_sonewconn_earlytest && over) { 1032 #else 1033 if (over) { 1034 #endif 1035 head->sol_overcount++; 1036 dolog = (sooverprio >= 0) && 1037 !!ratecheck(&head->sol_lastover, &overinterval); 1038 1039 /* 1040 * If we're going to log, copy the overflow count and queue 1041 * length from the listen socket before dropping the lock. 1042 * Also, reset the overflow count. 1043 */ 1044 if (dolog) { 1045 overcount = head->sol_overcount; 1046 head->sol_overcount = 0; 1047 qlen = head->sol_qlen; 1048 } 1049 SOLISTEN_UNLOCK(head); 1050 1051 if (dolog) { 1052 /* 1053 * Try to print something descriptive about the 1054 * socket for the error message. 1055 */ 1056 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf), 1057 SBUF_FIXEDLEN); 1058 switch (head->so_proto->pr_domain->dom_family) { 1059 #if defined(INET) || defined(INET6) 1060 #ifdef INET 1061 case AF_INET: 1062 #endif 1063 #ifdef INET6 1064 case AF_INET6: 1065 if (head->so_proto->pr_domain->dom_family == 1066 AF_INET6 || 1067 (sotoinpcb(head)->inp_inc.inc_flags & 1068 INC_ISIPV6)) { 1069 ip6_sprintf(addrbuf, 1070 &sotoinpcb(head)->inp_inc.inc6_laddr); 1071 sbuf_printf(&descrsb, "[%s]", addrbuf); 1072 } else 1073 #endif 1074 { 1075 #ifdef INET 1076 inet_ntoa_r( 1077 sotoinpcb(head)->inp_inc.inc_laddr, 1078 addrbuf); 1079 sbuf_cat(&descrsb, addrbuf); 1080 #endif 1081 } 1082 sbuf_printf(&descrsb, ":%hu (proto %u)", 1083 ntohs(sotoinpcb(head)->inp_inc.inc_lport), 1084 head->so_proto->pr_protocol); 1085 break; 1086 #endif /* INET || INET6 */ 1087 case AF_UNIX: 1088 sbuf_cat(&descrsb, localprefix); 1089 if (sotounpcb(head)->unp_addr != NULL) 1090 len = 1091 sotounpcb(head)->unp_addr->sun_len - 1092 offsetof(struct sockaddr_un, 1093 sun_path); 1094 else 1095 len = 0; 1096 if (len > 0) 1097 sbuf_bcat(&descrsb, 1098 sotounpcb(head)->unp_addr->sun_path, 1099 len); 1100 else 1101 sbuf_cat(&descrsb, "(unknown)"); 1102 break; 1103 } 1104 1105 /* 1106 * If we can't print something more specific, at least 1107 * print the domain name. 1108 */ 1109 if (sbuf_finish(&descrsb) != 0 || 1110 sbuf_len(&descrsb) <= 0) { 1111 sbuf_clear(&descrsb); 1112 sbuf_cat(&descrsb, 1113 head->so_proto->pr_domain->dom_name ?: 1114 "unknown"); 1115 sbuf_finish(&descrsb); 1116 } 1117 KASSERT(sbuf_len(&descrsb) > 0, 1118 ("%s: sbuf creation failed", __func__)); 1119 /* 1120 * Preserve the historic listen queue overflow log 1121 * message, that starts with "sonewconn:". It has 1122 * been known to sysadmins for years and also test 1123 * sys/kern/sonewconn_overflow checks for it. 1124 */ 1125 if (head->so_cred == 0) { 1126 log(LOG_PRI(sooverprio), 1127 "sonewconn: pcb %p (%s): " 1128 "Listen queue overflow: %i already in " 1129 "queue awaiting acceptance (%d " 1130 "occurrences)\n", head->so_pcb, 1131 sbuf_data(&descrsb), 1132 qlen, overcount); 1133 } else { 1134 log(LOG_PRI(sooverprio), 1135 "sonewconn: pcb %p (%s): " 1136 "Listen queue overflow: " 1137 "%i already in queue awaiting acceptance " 1138 "(%d occurrences), euid %d, rgid %d, jail %s\n", 1139 head->so_pcb, sbuf_data(&descrsb), qlen, 1140 overcount, head->so_cred->cr_uid, 1141 head->so_cred->cr_rgid, 1142 head->so_cred->cr_prison ? 1143 head->so_cred->cr_prison->pr_name : 1144 "not_jailed"); 1145 } 1146 sbuf_delete(&descrsb); 1147 1148 overcount = 0; 1149 } 1150 1151 return (NULL); 1152 } 1153 SOLISTEN_UNLOCK(head); 1154 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL", 1155 __func__, head)); 1156 so = soalloc(head->so_vnet); 1157 if (so == NULL) { 1158 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 1159 "limit reached or out of memory\n", 1160 __func__, head->so_pcb); 1161 return (NULL); 1162 } 1163 so->so_listen = head; 1164 so->so_type = head->so_type; 1165 /* 1166 * POSIX is ambiguous on what options an accept(2)ed socket should 1167 * inherit from the listener. Words "create a new socket" may be 1168 * interpreted as not inheriting anything. Best programming practice 1169 * for application developers is to not rely on such inheritance. 1170 * FreeBSD had historically inherited all so_options excluding 1171 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options, 1172 * including those completely irrelevant to a new born socket. For 1173 * compatibility with older versions we will inherit a list of 1174 * meaningful options. 1175 * The crucial bit to inherit is SO_ACCEPTFILTER. We need it present 1176 * in the child socket for soisconnected() promoting socket from the 1177 * incomplete queue to complete. It will be cleared before the child 1178 * gets available to accept(2). 1179 */ 1180 so->so_options = head->so_options & (SO_ACCEPTFILTER | SO_KEEPALIVE | 1181 SO_DONTROUTE | SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE); 1182 so->so_linger = head->so_linger; 1183 so->so_state = head->so_state; 1184 so->so_fibnum = head->so_fibnum; 1185 so->so_proto = head->so_proto; 1186 so->so_cred = crhold(head->so_cred); 1187 #ifdef SOCKET_HHOOK 1188 if (V_socket_hhh[HHOOK_SOCKET_NEWCONN]->hhh_nhooks > 0) { 1189 if (hhook_run_socket(so, head, HHOOK_SOCKET_NEWCONN)) { 1190 sodealloc(so); 1191 log(LOG_DEBUG, "%s: hhook run failed\n", __func__); 1192 return (NULL); 1193 } 1194 } 1195 #endif 1196 #ifdef MAC 1197 mac_socket_newconn(head, so); 1198 #endif 1199 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 1200 so_rdknl_assert_lock); 1201 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 1202 so_wrknl_assert_lock); 1203 VNET_SO_ASSERT(head); 1204 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) { 1205 sodealloc(so); 1206 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 1207 __func__, head->so_pcb); 1208 return (NULL); 1209 } 1210 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat; 1211 so->so_snd.sb_lowat = head->sol_sbsnd_lowat; 1212 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo; 1213 so->so_snd.sb_timeo = head->sol_sbsnd_timeo; 1214 so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE; 1215 so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE; 1216 if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) { 1217 so->so_snd.sb_mtx = &so->so_snd_mtx; 1218 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 1219 } 1220 1221 return (so); 1222 } 1223 1224 /* Connstatus may be 0 or SS_ISCONNECTED. */ 1225 struct socket * 1226 sonewconn(struct socket *head, int connstatus) 1227 { 1228 struct socket *so; 1229 1230 if ((so = solisten_clone(head)) == NULL) 1231 return (NULL); 1232 1233 if (so->so_proto->pr_attach(so, 0, NULL) != 0) { 1234 sodealloc(so); 1235 log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n", 1236 __func__, head->so_pcb); 1237 return (NULL); 1238 } 1239 1240 (void)solisten_enqueue(so, connstatus); 1241 1242 return (so); 1243 } 1244 1245 /* 1246 * Enqueue socket cloned by solisten_clone() to the listen queue of the 1247 * listener it has been cloned from. 1248 * 1249 * Return 'true' if socket landed on complete queue, otherwise 'false'. 1250 */ 1251 bool 1252 solisten_enqueue(struct socket *so, int connstatus) 1253 { 1254 struct socket *head = so->so_listen; 1255 1256 MPASS(refcount_load(&so->so_count) == 0); 1257 refcount_init(&so->so_count, 1); 1258 1259 SOLISTEN_LOCK(head); 1260 if (head->sol_accept_filter != NULL) 1261 connstatus = 0; 1262 so->so_state |= connstatus; 1263 soref(head); /* A socket on (in)complete queue refs head. */ 1264 if (connstatus) { 1265 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 1266 so->so_qstate = SQ_COMP; 1267 head->sol_qlen++; 1268 solisten_wakeup(head); /* unlocks */ 1269 return (true); 1270 } else { 1271 /* 1272 * Keep removing sockets from the head until there's room for 1273 * us to insert on the tail. In pre-locking revisions, this 1274 * was a simple if(), but as we could be racing with other 1275 * threads and soabort() requires dropping locks, we must 1276 * loop waiting for the condition to be true. 1277 */ 1278 while (head->sol_incqlen > head->sol_qlimit) { 1279 struct socket *sp; 1280 1281 sp = TAILQ_FIRST(&head->sol_incomp); 1282 TAILQ_REMOVE(&head->sol_incomp, sp, so_list); 1283 head->sol_incqlen--; 1284 SOCK_LOCK(sp); 1285 sp->so_qstate = SQ_NONE; 1286 sp->so_listen = NULL; 1287 SOCK_UNLOCK(sp); 1288 sorele_locked(head); /* does SOLISTEN_UNLOCK, head stays */ 1289 soabort(sp); 1290 SOLISTEN_LOCK(head); 1291 } 1292 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list); 1293 so->so_qstate = SQ_INCOMP; 1294 head->sol_incqlen++; 1295 SOLISTEN_UNLOCK(head); 1296 return (false); 1297 } 1298 } 1299 1300 #if defined(SCTP) || defined(SCTP_SUPPORT) 1301 /* 1302 * Socket part of sctp_peeloff(). Detach a new socket from an 1303 * association. The new socket is returned with a reference. 1304 * 1305 * XXXGL: reduce copy-paste with solisten_clone(). 1306 */ 1307 struct socket * 1308 sopeeloff(struct socket *head) 1309 { 1310 struct socket *so; 1311 1312 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p", 1313 __func__, __LINE__, head)); 1314 so = soalloc(head->so_vnet); 1315 if (so == NULL) { 1316 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: " 1317 "limit reached or out of memory\n", 1318 __func__, head->so_pcb); 1319 return (NULL); 1320 } 1321 so->so_type = head->so_type; 1322 so->so_options = head->so_options; 1323 so->so_linger = head->so_linger; 1324 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED; 1325 so->so_fibnum = head->so_fibnum; 1326 so->so_proto = head->so_proto; 1327 so->so_cred = crhold(head->so_cred); 1328 #ifdef MAC 1329 mac_socket_newconn(head, so); 1330 #endif 1331 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock, 1332 so_rdknl_assert_lock); 1333 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock, 1334 so_wrknl_assert_lock); 1335 VNET_SO_ASSERT(head); 1336 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) { 1337 sodealloc(so); 1338 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n", 1339 __func__, head->so_pcb); 1340 return (NULL); 1341 } 1342 if ((*so->so_proto->pr_attach)(so, 0, NULL)) { 1343 sodealloc(so); 1344 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n", 1345 __func__, head->so_pcb); 1346 return (NULL); 1347 } 1348 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat; 1349 so->so_snd.sb_lowat = head->so_snd.sb_lowat; 1350 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo; 1351 so->so_snd.sb_timeo = head->so_snd.sb_timeo; 1352 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE; 1353 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE; 1354 if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) { 1355 so->so_snd.sb_mtx = &so->so_snd_mtx; 1356 so->so_rcv.sb_mtx = &so->so_rcv_mtx; 1357 } 1358 1359 soref(so); 1360 1361 return (so); 1362 } 1363 #endif /* SCTP */ 1364 1365 int 1366 sobind(struct socket *so, struct sockaddr *nam, struct thread *td) 1367 { 1368 int error; 1369 1370 CURVNET_SET(so->so_vnet); 1371 error = so->so_proto->pr_bind(so, nam, td); 1372 CURVNET_RESTORE(); 1373 return (error); 1374 } 1375 1376 int 1377 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 1378 { 1379 int error; 1380 1381 CURVNET_SET(so->so_vnet); 1382 error = so->so_proto->pr_bindat(fd, so, nam, td); 1383 CURVNET_RESTORE(); 1384 return (error); 1385 } 1386 1387 /* 1388 * solisten() transitions a socket from a non-listening state to a listening 1389 * state, but can also be used to update the listen queue depth on an 1390 * existing listen socket. The protocol will call back into the sockets 1391 * layer using solisten_proto_check() and solisten_proto() to check and set 1392 * socket-layer listen state. Call backs are used so that the protocol can 1393 * acquire both protocol and socket layer locks in whatever order is required 1394 * by the protocol. 1395 * 1396 * Protocol implementors are advised to hold the socket lock across the 1397 * socket-layer test and set to avoid races at the socket layer. 1398 */ 1399 int 1400 solisten(struct socket *so, int backlog, struct thread *td) 1401 { 1402 int error; 1403 1404 CURVNET_SET(so->so_vnet); 1405 error = so->so_proto->pr_listen(so, backlog, td); 1406 CURVNET_RESTORE(); 1407 return (error); 1408 } 1409 1410 /* 1411 * Prepare for a call to solisten_proto(). Acquire all socket buffer locks in 1412 * order to interlock with socket I/O. 1413 */ 1414 int 1415 solisten_proto_check(struct socket *so) 1416 { 1417 SOCK_LOCK_ASSERT(so); 1418 1419 if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 1420 SS_ISDISCONNECTING)) != 0) 1421 return (EINVAL); 1422 1423 /* 1424 * Sleeping is not permitted here, so simply fail if userspace is 1425 * attempting to transmit or receive on the socket. This kind of 1426 * transient failure is not ideal, but it should occur only if userspace 1427 * is misusing the socket interfaces. 1428 */ 1429 if (!sx_try_xlock(&so->so_snd_sx)) 1430 return (EAGAIN); 1431 if (!sx_try_xlock(&so->so_rcv_sx)) { 1432 sx_xunlock(&so->so_snd_sx); 1433 return (EAGAIN); 1434 } 1435 mtx_lock(&so->so_snd_mtx); 1436 mtx_lock(&so->so_rcv_mtx); 1437 1438 /* Interlock with soo_aio_queue() and KTLS. */ 1439 if (!SOLISTENING(so)) { 1440 bool ktls; 1441 1442 #ifdef KERN_TLS 1443 ktls = so->so_snd.sb_tls_info != NULL || 1444 so->so_rcv.sb_tls_info != NULL; 1445 #else 1446 ktls = false; 1447 #endif 1448 if (ktls || 1449 (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 || 1450 (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) { 1451 solisten_proto_abort(so); 1452 return (EINVAL); 1453 } 1454 } 1455 1456 return (0); 1457 } 1458 1459 /* 1460 * Undo the setup done by solisten_proto_check(). 1461 */ 1462 void 1463 solisten_proto_abort(struct socket *so) 1464 { 1465 mtx_unlock(&so->so_snd_mtx); 1466 mtx_unlock(&so->so_rcv_mtx); 1467 sx_xunlock(&so->so_snd_sx); 1468 sx_xunlock(&so->so_rcv_sx); 1469 } 1470 1471 void 1472 solisten_proto(struct socket *so, int backlog) 1473 { 1474 int sbrcv_lowat, sbsnd_lowat; 1475 u_int sbrcv_hiwat, sbsnd_hiwat; 1476 short sbrcv_flags, sbsnd_flags; 1477 sbintime_t sbrcv_timeo, sbsnd_timeo; 1478 1479 SOCK_LOCK_ASSERT(so); 1480 KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING | 1481 SS_ISDISCONNECTING)) == 0, 1482 ("%s: bad socket state %p", __func__, so)); 1483 1484 if (SOLISTENING(so)) 1485 goto listening; 1486 1487 /* 1488 * Change this socket to listening state. 1489 */ 1490 sbrcv_lowat = so->so_rcv.sb_lowat; 1491 sbsnd_lowat = so->so_snd.sb_lowat; 1492 sbrcv_hiwat = so->so_rcv.sb_hiwat; 1493 sbsnd_hiwat = so->so_snd.sb_hiwat; 1494 sbrcv_flags = so->so_rcv.sb_flags; 1495 sbsnd_flags = so->so_snd.sb_flags; 1496 sbrcv_timeo = so->so_rcv.sb_timeo; 1497 sbsnd_timeo = so->so_snd.sb_timeo; 1498 1499 #ifdef MAC 1500 mac_socketpeer_label_free(so->so_peerlabel); 1501 #endif 1502 1503 if (!(so->so_proto->pr_flags & PR_SOCKBUF)) { 1504 sbdestroy(so, SO_SND); 1505 sbdestroy(so, SO_RCV); 1506 } 1507 1508 #ifdef INVARIANTS 1509 bzero(&so->so_rcv, 1510 sizeof(struct socket) - offsetof(struct socket, so_rcv)); 1511 #endif 1512 1513 so->sol_sbrcv_lowat = sbrcv_lowat; 1514 so->sol_sbsnd_lowat = sbsnd_lowat; 1515 so->sol_sbrcv_hiwat = sbrcv_hiwat; 1516 so->sol_sbsnd_hiwat = sbsnd_hiwat; 1517 so->sol_sbrcv_flags = sbrcv_flags; 1518 so->sol_sbsnd_flags = sbsnd_flags; 1519 so->sol_sbrcv_timeo = sbrcv_timeo; 1520 so->sol_sbsnd_timeo = sbsnd_timeo; 1521 1522 so->sol_qlen = so->sol_incqlen = 0; 1523 TAILQ_INIT(&so->sol_incomp); 1524 TAILQ_INIT(&so->sol_comp); 1525 1526 so->sol_accept_filter = NULL; 1527 so->sol_accept_filter_arg = NULL; 1528 so->sol_accept_filter_str = NULL; 1529 1530 so->sol_upcall = NULL; 1531 so->sol_upcallarg = NULL; 1532 1533 so->so_options |= SO_ACCEPTCONN; 1534 1535 listening: 1536 if (backlog < 0 || backlog > V_somaxconn) 1537 backlog = V_somaxconn; 1538 so->sol_qlimit = backlog; 1539 1540 mtx_unlock(&so->so_snd_mtx); 1541 mtx_unlock(&so->so_rcv_mtx); 1542 sx_xunlock(&so->so_snd_sx); 1543 sx_xunlock(&so->so_rcv_sx); 1544 } 1545 1546 /* 1547 * Wakeup listeners/subsystems once we have a complete connection. 1548 * Enters with lock, returns unlocked. 1549 */ 1550 void 1551 solisten_wakeup(struct socket *sol) 1552 { 1553 1554 if (sol->sol_upcall != NULL) 1555 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT); 1556 else { 1557 selwakeuppri(&sol->so_rdsel, PSOCK); 1558 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0); 1559 } 1560 SOLISTEN_UNLOCK(sol); 1561 wakeup_one(&sol->sol_comp); 1562 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL) 1563 pgsigio(&sol->so_sigio, SIGIO, 0); 1564 } 1565 1566 /* 1567 * Return single connection off a listening socket queue. Main consumer of 1568 * the function is kern_accept4(). Some modules, that do their own accept 1569 * management also use the function. The socket reference held by the 1570 * listen queue is handed to the caller. 1571 * 1572 * Listening socket must be locked on entry and is returned unlocked on 1573 * return. 1574 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT. 1575 */ 1576 int 1577 solisten_dequeue(struct socket *head, struct socket **ret, int flags) 1578 { 1579 struct socket *so; 1580 int error; 1581 1582 SOLISTEN_LOCK_ASSERT(head); 1583 1584 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) && 1585 head->so_error == 0) { 1586 error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH, 1587 "accept", 0); 1588 if (error != 0) { 1589 SOLISTEN_UNLOCK(head); 1590 return (error); 1591 } 1592 } 1593 if (head->so_error) { 1594 error = head->so_error; 1595 head->so_error = 0; 1596 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp)) 1597 error = EWOULDBLOCK; 1598 else 1599 error = 0; 1600 if (error) { 1601 SOLISTEN_UNLOCK(head); 1602 return (error); 1603 } 1604 so = TAILQ_FIRST(&head->sol_comp); 1605 SOCK_LOCK(so); 1606 KASSERT(so->so_qstate == SQ_COMP, 1607 ("%s: so %p not SQ_COMP", __func__, so)); 1608 head->sol_qlen--; 1609 so->so_qstate = SQ_NONE; 1610 so->so_listen = NULL; 1611 TAILQ_REMOVE(&head->sol_comp, so, so_list); 1612 if (flags & ACCEPT4_INHERIT) 1613 so->so_state |= (head->so_state & SS_NBIO); 1614 else 1615 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0; 1616 SOCK_UNLOCK(so); 1617 sorele_locked(head); 1618 1619 *ret = so; 1620 return (0); 1621 } 1622 1623 static struct so_splice * 1624 so_splice_alloc(off_t max) 1625 { 1626 struct so_splice *sp; 1627 1628 sp = uma_zalloc(splice_zone, M_WAITOK); 1629 sp->src = NULL; 1630 sp->dst = NULL; 1631 sp->max = max > 0 ? max : -1; 1632 do { 1633 sp->wq_index = atomic_fetchadd_32(&splice_index, 1) % 1634 (mp_maxid + 1); 1635 } while (CPU_ABSENT(sp->wq_index)); 1636 sp->state = SPLICE_INIT; 1637 TIMEOUT_TASK_INIT(taskqueue_thread, &sp->timeout, 0, so_splice_timeout, 1638 sp); 1639 return (sp); 1640 } 1641 1642 static void 1643 so_splice_free(struct so_splice *sp) 1644 { 1645 KASSERT(sp->state == SPLICE_CLOSED, 1646 ("so_splice_free: sp %p not closed", sp)); 1647 uma_zfree(splice_zone, sp); 1648 } 1649 1650 static void 1651 so_splice_timeout(void *arg, int pending __unused) 1652 { 1653 struct so_splice *sp; 1654 1655 sp = arg; 1656 (void)so_unsplice(sp->src, true); 1657 } 1658 1659 /* 1660 * Splice the output from so to the input of so2. 1661 */ 1662 static int 1663 so_splice(struct socket *so, struct socket *so2, struct splice *splice) 1664 { 1665 struct so_splice *sp; 1666 int error; 1667 1668 if (splice->sp_max < 0) 1669 return (EINVAL); 1670 /* Handle only TCP for now; TODO: other streaming protos */ 1671 if (so->so_proto->pr_protocol != IPPROTO_TCP || 1672 so2->so_proto->pr_protocol != IPPROTO_TCP) 1673 return (EPROTONOSUPPORT); 1674 if (so->so_vnet != so2->so_vnet) 1675 return (EINVAL); 1676 1677 /* so_splice_xfer() assumes that we're using these implementations. */ 1678 KASSERT(so->so_proto->pr_sosend == sosend_generic, 1679 ("so_splice: sosend not sosend_generic")); 1680 KASSERT(so2->so_proto->pr_soreceive == soreceive_generic || 1681 so2->so_proto->pr_soreceive == soreceive_stream, 1682 ("so_splice: soreceive not soreceive_generic/stream")); 1683 1684 sp = so_splice_alloc(splice->sp_max); 1685 so->so_splice_sent = 0; 1686 sp->src = so; 1687 sp->dst = so2; 1688 1689 error = 0; 1690 SOCK_LOCK(so); 1691 if (SOLISTENING(so)) 1692 error = EINVAL; 1693 else if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0) 1694 error = ENOTCONN; 1695 else if (so->so_splice != NULL) 1696 error = EBUSY; 1697 if (error != 0) { 1698 SOCK_UNLOCK(so); 1699 uma_zfree(splice_zone, sp); 1700 return (error); 1701 } 1702 soref(so); 1703 so->so_splice = sp; 1704 SOCK_RECVBUF_LOCK(so); 1705 so->so_rcv.sb_flags |= SB_SPLICED; 1706 SOCK_RECVBUF_UNLOCK(so); 1707 SOCK_UNLOCK(so); 1708 1709 error = 0; 1710 SOCK_LOCK(so2); 1711 if (SOLISTENING(so2)) 1712 error = EINVAL; 1713 else if ((so2->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) == 0) 1714 error = ENOTCONN; 1715 else if (so2->so_splice_back != NULL) 1716 error = EBUSY; 1717 if (error != 0) { 1718 SOCK_UNLOCK(so2); 1719 SOCK_LOCK(so); 1720 so->so_splice = NULL; 1721 SOCK_RECVBUF_LOCK(so); 1722 so->so_rcv.sb_flags &= ~SB_SPLICED; 1723 SOCK_RECVBUF_UNLOCK(so); 1724 SOCK_UNLOCK(so); 1725 sorele(so); 1726 uma_zfree(splice_zone, sp); 1727 return (error); 1728 } 1729 soref(so2); 1730 so2->so_splice_back = sp; 1731 SOCK_SENDBUF_LOCK(so2); 1732 so2->so_snd.sb_flags |= SB_SPLICED; 1733 mtx_lock(&sp->mtx); 1734 SOCK_SENDBUF_UNLOCK(so2); 1735 SOCK_UNLOCK(so2); 1736 1737 if (splice->sp_idle.tv_sec != 0 || splice->sp_idle.tv_usec != 0) { 1738 taskqueue_enqueue_timeout_sbt(taskqueue_thread, &sp->timeout, 1739 tvtosbt(splice->sp_idle), 0, C_PREL(4)); 1740 } 1741 1742 /* 1743 * Transfer any data already present in the socket buffer. 1744 */ 1745 sp->state = SPLICE_QUEUED; 1746 so_splice_xfer(sp); 1747 return (0); 1748 } 1749 1750 static int 1751 so_unsplice(struct socket *so, bool timeout) 1752 { 1753 struct socket *so2; 1754 struct so_splice *sp; 1755 bool drain; 1756 1757 /* 1758 * First unset SB_SPLICED and hide the splice structure so that 1759 * wakeup routines will stop enqueuing work. This also ensures that 1760 * a only a single thread will proceed with the unsplice. 1761 */ 1762 SOCK_LOCK(so); 1763 if (SOLISTENING(so)) { 1764 SOCK_UNLOCK(so); 1765 return (EINVAL); 1766 } 1767 SOCK_RECVBUF_LOCK(so); 1768 if ((so->so_rcv.sb_flags & SB_SPLICED) == 0) { 1769 SOCK_RECVBUF_UNLOCK(so); 1770 SOCK_UNLOCK(so); 1771 return (ENOTCONN); 1772 } 1773 so->so_rcv.sb_flags &= ~SB_SPLICED; 1774 sp = so->so_splice; 1775 so->so_splice = NULL; 1776 SOCK_RECVBUF_UNLOCK(so); 1777 SOCK_UNLOCK(so); 1778 1779 so2 = sp->dst; 1780 SOCK_LOCK(so2); 1781 KASSERT(!SOLISTENING(so2), ("%s: so2 is listening", __func__)); 1782 SOCK_SENDBUF_LOCK(so2); 1783 KASSERT((so2->so_snd.sb_flags & SB_SPLICED) != 0, 1784 ("%s: so2 is not spliced", __func__)); 1785 KASSERT(so2->so_splice_back == sp, 1786 ("%s: so_splice_back != sp", __func__)); 1787 so2->so_snd.sb_flags &= ~SB_SPLICED; 1788 so2->so_splice_back = NULL; 1789 SOCK_SENDBUF_UNLOCK(so2); 1790 SOCK_UNLOCK(so2); 1791 1792 /* 1793 * No new work is being enqueued. The worker thread might be 1794 * splicing data right now, in which case we want to wait for it to 1795 * finish before proceeding. 1796 */ 1797 mtx_lock(&sp->mtx); 1798 switch (sp->state) { 1799 case SPLICE_QUEUED: 1800 case SPLICE_RUNNING: 1801 sp->state = SPLICE_CLOSING; 1802 while (sp->state == SPLICE_CLOSING) 1803 msleep(sp, &sp->mtx, PSOCK, "unsplice", 0); 1804 break; 1805 case SPLICE_IDLE: 1806 case SPLICE_EXCEPTION: 1807 sp->state = SPLICE_CLOSED; 1808 break; 1809 default: 1810 __assert_unreachable(); 1811 } 1812 if (!timeout) { 1813 drain = taskqueue_cancel_timeout(taskqueue_thread, &sp->timeout, 1814 NULL) != 0; 1815 } else { 1816 drain = false; 1817 } 1818 mtx_unlock(&sp->mtx); 1819 if (drain) 1820 taskqueue_drain_timeout(taskqueue_thread, &sp->timeout); 1821 1822 /* 1823 * Now we hold the sole reference to the splice structure. 1824 * Clean up: signal userspace and release socket references. 1825 */ 1826 sorwakeup(so); 1827 CURVNET_SET(so->so_vnet); 1828 sorele(so); 1829 sowwakeup(so2); 1830 sorele(so2); 1831 CURVNET_RESTORE(); 1832 so_splice_free(sp); 1833 return (0); 1834 } 1835 1836 /* 1837 * Free socket upon release of the very last reference. 1838 */ 1839 static void 1840 sofree(struct socket *so) 1841 { 1842 struct protosw *pr = so->so_proto; 1843 1844 SOCK_LOCK_ASSERT(so); 1845 KASSERT(refcount_load(&so->so_count) == 0, 1846 ("%s: so %p has references", __func__, so)); 1847 KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE, 1848 ("%s: so %p is on listen queue", __func__, so)); 1849 KASSERT(SOLISTENING(so) || (so->so_rcv.sb_flags & SB_SPLICED) == 0, 1850 ("%s: so %p rcvbuf is spliced", __func__, so)); 1851 KASSERT(SOLISTENING(so) || (so->so_snd.sb_flags & SB_SPLICED) == 0, 1852 ("%s: so %p sndbuf is spliced", __func__, so)); 1853 KASSERT(so->so_splice == NULL && so->so_splice_back == NULL, 1854 ("%s: so %p has spliced data", __func__, so)); 1855 1856 SOCK_UNLOCK(so); 1857 1858 if (so->so_dtor != NULL) 1859 so->so_dtor(so); 1860 1861 VNET_SO_ASSERT(so); 1862 if (pr->pr_detach != NULL) 1863 pr->pr_detach(so); 1864 1865 /* 1866 * From this point on, we assume that no other references to this 1867 * socket exist anywhere else in the stack. Therefore, no locks need 1868 * to be acquired or held. 1869 */ 1870 if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) { 1871 sbdestroy(so, SO_SND); 1872 sbdestroy(so, SO_RCV); 1873 } 1874 seldrain(&so->so_rdsel); 1875 seldrain(&so->so_wrsel); 1876 knlist_destroy(&so->so_rdsel.si_note); 1877 knlist_destroy(&so->so_wrsel.si_note); 1878 sodealloc(so); 1879 } 1880 1881 /* 1882 * Release a reference on a socket while holding the socket lock. 1883 * Unlocks the socket lock before returning. 1884 */ 1885 void 1886 sorele_locked(struct socket *so) 1887 { 1888 SOCK_LOCK_ASSERT(so); 1889 if (refcount_release(&so->so_count)) 1890 sofree(so); 1891 else 1892 SOCK_UNLOCK(so); 1893 } 1894 1895 /* 1896 * Close a socket on last file table reference removal. Initiate disconnect 1897 * if connected. Free socket when disconnect complete. 1898 * 1899 * This function will sorele() the socket. Note that soclose() may be called 1900 * prior to the ref count reaching zero. The actual socket structure will 1901 * not be freed until the ref count reaches zero. 1902 */ 1903 int 1904 soclose(struct socket *so) 1905 { 1906 struct accept_queue lqueue; 1907 int error = 0; 1908 bool listening, last __diagused; 1909 1910 CURVNET_SET(so->so_vnet); 1911 funsetown(&so->so_sigio); 1912 if (so->so_state & SS_ISCONNECTED) { 1913 if ((so->so_state & SS_ISDISCONNECTING) == 0) { 1914 error = sodisconnect(so); 1915 if (error) { 1916 if (error == ENOTCONN) 1917 error = 0; 1918 goto drop; 1919 } 1920 } 1921 1922 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) { 1923 if ((so->so_state & SS_ISDISCONNECTING) && 1924 (so->so_state & SS_NBIO)) 1925 goto drop; 1926 while (so->so_state & SS_ISCONNECTED) { 1927 error = tsleep(&so->so_timeo, 1928 PSOCK | PCATCH, "soclos", 1929 so->so_linger * hz); 1930 if (error) 1931 break; 1932 } 1933 } 1934 } 1935 1936 drop: 1937 if (so->so_proto->pr_close != NULL) 1938 so->so_proto->pr_close(so); 1939 1940 SOCK_LOCK(so); 1941 if ((listening = SOLISTENING(so))) { 1942 struct socket *sp; 1943 1944 TAILQ_INIT(&lqueue); 1945 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list); 1946 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list); 1947 1948 so->sol_qlen = so->sol_incqlen = 0; 1949 1950 TAILQ_FOREACH(sp, &lqueue, so_list) { 1951 SOCK_LOCK(sp); 1952 sp->so_qstate = SQ_NONE; 1953 sp->so_listen = NULL; 1954 SOCK_UNLOCK(sp); 1955 last = refcount_release(&so->so_count); 1956 KASSERT(!last, ("%s: released last reference for %p", 1957 __func__, so)); 1958 } 1959 } 1960 sorele_locked(so); 1961 if (listening) { 1962 struct socket *sp, *tsp; 1963 1964 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) 1965 soabort(sp); 1966 } 1967 CURVNET_RESTORE(); 1968 return (error); 1969 } 1970 1971 /* 1972 * soabort() is used to abruptly tear down a connection, such as when a 1973 * resource limit is reached (listen queue depth exceeded), or if a listen 1974 * socket is closed while there are sockets waiting to be accepted. 1975 * 1976 * This interface is tricky, because it is called on an unreferenced socket, 1977 * and must be called only by a thread that has actually removed the socket 1978 * from the listen queue it was on. Likely this thread holds the last 1979 * reference on the socket and soabort() will proceed with sofree(). But 1980 * it might be not the last, as the sockets on the listen queues are seen 1981 * from the protocol side. 1982 * 1983 * This interface will call into the protocol code, so must not be called 1984 * with any socket locks held. Protocols do call it while holding their own 1985 * recursible protocol mutexes, but this is something that should be subject 1986 * to review in the future. 1987 * 1988 * Usually socket should have a single reference left, but this is not a 1989 * requirement. In the past, when we have had named references for file 1990 * descriptor and protocol, we asserted that none of them are being held. 1991 */ 1992 void 1993 soabort(struct socket *so) 1994 { 1995 1996 VNET_SO_ASSERT(so); 1997 1998 if (so->so_proto->pr_abort != NULL) 1999 so->so_proto->pr_abort(so); 2000 SOCK_LOCK(so); 2001 sorele_locked(so); 2002 } 2003 2004 int 2005 soaccept(struct socket *so, struct sockaddr *sa) 2006 { 2007 #ifdef INVARIANTS 2008 u_char len = sa->sa_len; 2009 #endif 2010 int error; 2011 2012 CURVNET_SET(so->so_vnet); 2013 error = so->so_proto->pr_accept(so, sa); 2014 KASSERT(sa->sa_len <= len, 2015 ("%s: protocol %p sockaddr overflow", __func__, so->so_proto)); 2016 CURVNET_RESTORE(); 2017 return (error); 2018 } 2019 2020 int 2021 sopeeraddr(struct socket *so, struct sockaddr *sa) 2022 { 2023 #ifdef INVARIANTS 2024 u_char len = sa->sa_len; 2025 #endif 2026 int error; 2027 2028 CURVNET_ASSERT_SET(); 2029 2030 error = so->so_proto->pr_peeraddr(so, sa); 2031 KASSERT(sa->sa_len <= len, 2032 ("%s: protocol %p sockaddr overflow", __func__, so->so_proto)); 2033 2034 return (error); 2035 } 2036 2037 int 2038 sosockaddr(struct socket *so, struct sockaddr *sa) 2039 { 2040 #ifdef INVARIANTS 2041 u_char len = sa->sa_len; 2042 #endif 2043 int error; 2044 2045 CURVNET_SET(so->so_vnet); 2046 error = so->so_proto->pr_sockaddr(so, sa); 2047 KASSERT(sa->sa_len <= len, 2048 ("%s: protocol %p sockaddr overflow", __func__, so->so_proto)); 2049 CURVNET_RESTORE(); 2050 2051 return (error); 2052 } 2053 2054 int 2055 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td) 2056 { 2057 2058 return (soconnectat(AT_FDCWD, so, nam, td)); 2059 } 2060 2061 int 2062 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) 2063 { 2064 int error; 2065 2066 CURVNET_SET(so->so_vnet); 2067 2068 /* 2069 * If protocol is connection-based, can only connect once. 2070 * Otherwise, if connected, try to disconnect first. This allows 2071 * user to disconnect by connecting to, e.g., a null address. 2072 * 2073 * Note, this check is racy and may need to be re-evaluated at the 2074 * protocol layer. 2075 */ 2076 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) && 2077 ((so->so_proto->pr_flags & PR_CONNREQUIRED) || 2078 (error = sodisconnect(so)))) { 2079 error = EISCONN; 2080 } else { 2081 /* 2082 * Prevent accumulated error from previous connection from 2083 * biting us. 2084 */ 2085 so->so_error = 0; 2086 if (fd == AT_FDCWD) { 2087 error = so->so_proto->pr_connect(so, nam, td); 2088 } else { 2089 error = so->so_proto->pr_connectat(fd, so, nam, td); 2090 } 2091 } 2092 CURVNET_RESTORE(); 2093 2094 return (error); 2095 } 2096 2097 int 2098 soconnect2(struct socket *so1, struct socket *so2) 2099 { 2100 int error; 2101 2102 CURVNET_SET(so1->so_vnet); 2103 error = so1->so_proto->pr_connect2(so1, so2); 2104 CURVNET_RESTORE(); 2105 return (error); 2106 } 2107 2108 int 2109 sodisconnect(struct socket *so) 2110 { 2111 int error; 2112 2113 if ((so->so_state & SS_ISCONNECTED) == 0) 2114 return (ENOTCONN); 2115 if (so->so_state & SS_ISDISCONNECTING) 2116 return (EALREADY); 2117 VNET_SO_ASSERT(so); 2118 error = so->so_proto->pr_disconnect(so); 2119 return (error); 2120 } 2121 2122 int 2123 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio, 2124 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 2125 { 2126 long space; 2127 ssize_t resid; 2128 int clen = 0, error, dontroute; 2129 2130 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM")); 2131 KASSERT(so->so_proto->pr_flags & PR_ATOMIC, 2132 ("sosend_dgram: !PR_ATOMIC")); 2133 2134 if (uio != NULL) 2135 resid = uio->uio_resid; 2136 else 2137 resid = top->m_pkthdr.len; 2138 /* 2139 * In theory resid should be unsigned. However, space must be 2140 * signed, as it might be less than 0 if we over-committed, and we 2141 * must use a signed comparison of space and resid. On the other 2142 * hand, a negative resid causes us to loop sending 0-length 2143 * segments to the protocol. 2144 */ 2145 if (resid < 0) { 2146 error = EINVAL; 2147 goto out; 2148 } 2149 2150 dontroute = 2151 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0; 2152 if (td != NULL) 2153 td->td_ru.ru_msgsnd++; 2154 if (control != NULL) 2155 clen = control->m_len; 2156 2157 SOCKBUF_LOCK(&so->so_snd); 2158 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 2159 SOCKBUF_UNLOCK(&so->so_snd); 2160 error = EPIPE; 2161 goto out; 2162 } 2163 if (so->so_error) { 2164 error = so->so_error; 2165 so->so_error = 0; 2166 SOCKBUF_UNLOCK(&so->so_snd); 2167 goto out; 2168 } 2169 if ((so->so_state & SS_ISCONNECTED) == 0) { 2170 /* 2171 * `sendto' and `sendmsg' is allowed on a connection-based 2172 * socket if it supports implied connect. Return ENOTCONN if 2173 * not connected and no address is supplied. 2174 */ 2175 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 2176 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 2177 if (!(resid == 0 && clen != 0)) { 2178 SOCKBUF_UNLOCK(&so->so_snd); 2179 error = ENOTCONN; 2180 goto out; 2181 } 2182 } else if (addr == NULL) { 2183 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 2184 error = ENOTCONN; 2185 else 2186 error = EDESTADDRREQ; 2187 SOCKBUF_UNLOCK(&so->so_snd); 2188 goto out; 2189 } 2190 } 2191 2192 /* 2193 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a 2194 * problem and need fixing. 2195 */ 2196 space = sbspace(&so->so_snd); 2197 if (flags & MSG_OOB) 2198 space += 1024; 2199 space -= clen; 2200 SOCKBUF_UNLOCK(&so->so_snd); 2201 if (resid > space) { 2202 error = EMSGSIZE; 2203 goto out; 2204 } 2205 if (uio == NULL) { 2206 resid = 0; 2207 if (flags & MSG_EOR) 2208 top->m_flags |= M_EOR; 2209 } else { 2210 /* 2211 * Copy the data from userland into a mbuf chain. 2212 * If no data is to be copied in, a single empty mbuf 2213 * is returned. 2214 */ 2215 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr, 2216 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0))); 2217 if (top == NULL) { 2218 error = EFAULT; /* only possible error */ 2219 goto out; 2220 } 2221 space -= resid - uio->uio_resid; 2222 resid = uio->uio_resid; 2223 } 2224 KASSERT(resid == 0, ("sosend_dgram: resid != 0")); 2225 /* 2226 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock 2227 * than with. 2228 */ 2229 if (dontroute) { 2230 SOCK_LOCK(so); 2231 so->so_options |= SO_DONTROUTE; 2232 SOCK_UNLOCK(so); 2233 } 2234 /* 2235 * XXX all the SBS_CANTSENDMORE checks previously done could be out 2236 * of date. We could have received a reset packet in an interrupt or 2237 * maybe we slept while doing page faults in uiomove() etc. We could 2238 * probably recheck again inside the locking protection here, but 2239 * there are probably other places that this also happens. We must 2240 * rethink this. 2241 */ 2242 VNET_SO_ASSERT(so); 2243 error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB : 2244 /* 2245 * If the user set MSG_EOF, the protocol understands this flag and 2246 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND. 2247 */ 2248 ((flags & MSG_EOF) && 2249 (so->so_proto->pr_flags & PR_IMPLOPCL) && 2250 (resid <= 0)) ? 2251 PRUS_EOF : 2252 /* If there is more to send set PRUS_MORETOCOME */ 2253 (flags & MSG_MORETOCOME) || 2254 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0, 2255 top, addr, control, td); 2256 if (dontroute) { 2257 SOCK_LOCK(so); 2258 so->so_options &= ~SO_DONTROUTE; 2259 SOCK_UNLOCK(so); 2260 } 2261 clen = 0; 2262 control = NULL; 2263 top = NULL; 2264 out: 2265 if (top != NULL) 2266 m_freem(top); 2267 if (control != NULL) 2268 m_freem(control); 2269 return (error); 2270 } 2271 2272 /* 2273 * Send on a socket. If send must go all at once and message is larger than 2274 * send buffering, then hard error. Lock against other senders. If must go 2275 * all at once and not enough room now, then inform user that this would 2276 * block and do nothing. Otherwise, if nonblocking, send as much as 2277 * possible. The data to be sent is described by "uio" if nonzero, otherwise 2278 * by the mbuf chain "top" (which must be null if uio is not). Data provided 2279 * in mbuf chain must be small enough to send all at once. 2280 * 2281 * Returns nonzero on error, timeout or signal; callers must check for short 2282 * counts if EINTR/ERESTART are returned. Data and control buffers are freed 2283 * on return. 2284 */ 2285 static int 2286 sosend_generic_locked(struct socket *so, struct sockaddr *addr, struct uio *uio, 2287 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 2288 { 2289 long space; 2290 ssize_t resid; 2291 int clen = 0, error, dontroute; 2292 int atomic = sosendallatonce(so) || top; 2293 int pr_send_flag; 2294 #ifdef KERN_TLS 2295 struct ktls_session *tls; 2296 int tls_enq_cnt, tls_send_flag; 2297 uint8_t tls_rtype; 2298 2299 tls = NULL; 2300 tls_rtype = TLS_RLTYPE_APP; 2301 #endif 2302 2303 SOCK_IO_SEND_ASSERT_LOCKED(so); 2304 2305 if (uio != NULL) 2306 resid = uio->uio_resid; 2307 else if ((top->m_flags & M_PKTHDR) != 0) 2308 resid = top->m_pkthdr.len; 2309 else 2310 resid = m_length(top, NULL); 2311 /* 2312 * In theory resid should be unsigned. However, space must be 2313 * signed, as it might be less than 0 if we over-committed, and we 2314 * must use a signed comparison of space and resid. On the other 2315 * hand, a negative resid causes us to loop sending 0-length 2316 * segments to the protocol. 2317 * 2318 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM 2319 * type sockets since that's an error. 2320 */ 2321 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) { 2322 error = EINVAL; 2323 goto out; 2324 } 2325 2326 dontroute = 2327 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 && 2328 (so->so_proto->pr_flags & PR_ATOMIC); 2329 if (td != NULL) 2330 td->td_ru.ru_msgsnd++; 2331 if (control != NULL) 2332 clen = control->m_len; 2333 2334 #ifdef KERN_TLS 2335 tls_send_flag = 0; 2336 tls = ktls_hold(so->so_snd.sb_tls_info); 2337 if (tls != NULL) { 2338 if (tls->mode == TCP_TLS_MODE_SW) 2339 tls_send_flag = PRUS_NOTREADY; 2340 2341 if (control != NULL) { 2342 struct cmsghdr *cm = mtod(control, struct cmsghdr *); 2343 2344 if (clen >= sizeof(*cm) && 2345 cm->cmsg_type == TLS_SET_RECORD_TYPE) { 2346 tls_rtype = *((uint8_t *)CMSG_DATA(cm)); 2347 clen = 0; 2348 m_freem(control); 2349 control = NULL; 2350 atomic = 1; 2351 } 2352 } 2353 2354 if (resid == 0 && !ktls_permit_empty_frames(tls)) { 2355 error = EINVAL; 2356 goto out; 2357 } 2358 } 2359 #endif 2360 2361 restart: 2362 do { 2363 SOCKBUF_LOCK(&so->so_snd); 2364 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 2365 SOCKBUF_UNLOCK(&so->so_snd); 2366 error = EPIPE; 2367 goto out; 2368 } 2369 if (so->so_error) { 2370 error = so->so_error; 2371 so->so_error = 0; 2372 SOCKBUF_UNLOCK(&so->so_snd); 2373 goto out; 2374 } 2375 if ((so->so_state & SS_ISCONNECTED) == 0) { 2376 /* 2377 * `sendto' and `sendmsg' is allowed on a connection- 2378 * based socket if it supports implied connect. 2379 * Return ENOTCONN if not connected and no address is 2380 * supplied. 2381 */ 2382 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) && 2383 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) { 2384 if (!(resid == 0 && clen != 0)) { 2385 SOCKBUF_UNLOCK(&so->so_snd); 2386 error = ENOTCONN; 2387 goto out; 2388 } 2389 } else if (addr == NULL) { 2390 SOCKBUF_UNLOCK(&so->so_snd); 2391 if (so->so_proto->pr_flags & PR_CONNREQUIRED) 2392 error = ENOTCONN; 2393 else 2394 error = EDESTADDRREQ; 2395 goto out; 2396 } 2397 } 2398 space = sbspace(&so->so_snd); 2399 if (flags & MSG_OOB) 2400 space += 1024; 2401 if ((atomic && resid > so->so_snd.sb_hiwat) || 2402 clen > so->so_snd.sb_hiwat) { 2403 SOCKBUF_UNLOCK(&so->so_snd); 2404 error = EMSGSIZE; 2405 goto out; 2406 } 2407 if (space < resid + clen && 2408 (atomic || space < so->so_snd.sb_lowat || space < clen)) { 2409 if ((so->so_state & SS_NBIO) || 2410 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 2411 SOCKBUF_UNLOCK(&so->so_snd); 2412 error = EWOULDBLOCK; 2413 goto out; 2414 } 2415 error = sbwait(so, SO_SND); 2416 SOCKBUF_UNLOCK(&so->so_snd); 2417 if (error) 2418 goto out; 2419 goto restart; 2420 } 2421 SOCKBUF_UNLOCK(&so->so_snd); 2422 space -= clen; 2423 do { 2424 if (uio == NULL) { 2425 resid = 0; 2426 if (flags & MSG_EOR) 2427 top->m_flags |= M_EOR; 2428 #ifdef KERN_TLS 2429 if (tls != NULL) { 2430 ktls_frame(top, tls, &tls_enq_cnt, 2431 tls_rtype); 2432 tls_rtype = TLS_RLTYPE_APP; 2433 } 2434 #endif 2435 } else { 2436 /* 2437 * Copy the data from userland into a mbuf 2438 * chain. If resid is 0, which can happen 2439 * only if we have control to send, then 2440 * a single empty mbuf is returned. This 2441 * is a workaround to prevent protocol send 2442 * methods to panic. 2443 */ 2444 #ifdef KERN_TLS 2445 if (tls != NULL) { 2446 top = m_uiotombuf(uio, M_WAITOK, space, 2447 tls->params.max_frame_len, 2448 M_EXTPG | 2449 ((flags & MSG_EOR) ? M_EOR : 0)); 2450 if (top != NULL) { 2451 ktls_frame(top, tls, 2452 &tls_enq_cnt, tls_rtype); 2453 } 2454 tls_rtype = TLS_RLTYPE_APP; 2455 } else 2456 #endif 2457 top = m_uiotombuf(uio, M_WAITOK, space, 2458 (atomic ? max_hdr : 0), 2459 (atomic ? M_PKTHDR : 0) | 2460 ((flags & MSG_EOR) ? M_EOR : 0)); 2461 if (top == NULL) { 2462 error = EFAULT; /* only possible error */ 2463 goto out; 2464 } 2465 space -= resid - uio->uio_resid; 2466 resid = uio->uio_resid; 2467 } 2468 if (dontroute) { 2469 SOCK_LOCK(so); 2470 so->so_options |= SO_DONTROUTE; 2471 SOCK_UNLOCK(so); 2472 } 2473 /* 2474 * XXX all the SBS_CANTSENDMORE checks previously 2475 * done could be out of date. We could have received 2476 * a reset packet in an interrupt or maybe we slept 2477 * while doing page faults in uiomove() etc. We 2478 * could probably recheck again inside the locking 2479 * protection here, but there are probably other 2480 * places that this also happens. We must rethink 2481 * this. 2482 */ 2483 VNET_SO_ASSERT(so); 2484 2485 pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB : 2486 /* 2487 * If the user set MSG_EOF, the protocol understands 2488 * this flag and nothing left to send then use 2489 * PRU_SEND_EOF instead of PRU_SEND. 2490 */ 2491 ((flags & MSG_EOF) && 2492 (so->so_proto->pr_flags & PR_IMPLOPCL) && 2493 (resid <= 0)) ? 2494 PRUS_EOF : 2495 /* If there is more to send set PRUS_MORETOCOME. */ 2496 (flags & MSG_MORETOCOME) || 2497 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0; 2498 2499 #ifdef KERN_TLS 2500 pr_send_flag |= tls_send_flag; 2501 #endif 2502 2503 error = so->so_proto->pr_send(so, pr_send_flag, top, 2504 addr, control, td); 2505 2506 if (dontroute) { 2507 SOCK_LOCK(so); 2508 so->so_options &= ~SO_DONTROUTE; 2509 SOCK_UNLOCK(so); 2510 } 2511 2512 #ifdef KERN_TLS 2513 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) { 2514 if (error != 0) { 2515 m_freem(top); 2516 top = NULL; 2517 } else { 2518 soref(so); 2519 ktls_enqueue(top, so, tls_enq_cnt); 2520 } 2521 } 2522 #endif 2523 clen = 0; 2524 control = NULL; 2525 top = NULL; 2526 if (error) 2527 goto out; 2528 } while (resid && space > 0); 2529 } while (resid); 2530 2531 out: 2532 #ifdef KERN_TLS 2533 if (tls != NULL) 2534 ktls_free(tls); 2535 #endif 2536 if (top != NULL) 2537 m_freem(top); 2538 if (control != NULL) 2539 m_freem(control); 2540 return (error); 2541 } 2542 2543 int 2544 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio, 2545 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 2546 { 2547 int error; 2548 2549 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags)); 2550 if (error) 2551 return (error); 2552 error = sosend_generic_locked(so, addr, uio, top, control, flags, td); 2553 SOCK_IO_SEND_UNLOCK(so); 2554 return (error); 2555 } 2556 2557 /* 2558 * Send to a socket from a kernel thread. 2559 * 2560 * XXXGL: in almost all cases uio is NULL and the mbuf is supplied. 2561 * Exception is nfs/bootp_subr.c. It is arguable that the VNET context needs 2562 * to be set at all. This function should just boil down to a static inline 2563 * calling the protocol method. 2564 */ 2565 int 2566 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 2567 struct mbuf *top, struct mbuf *control, int flags, struct thread *td) 2568 { 2569 int error; 2570 2571 CURVNET_SET(so->so_vnet); 2572 error = so->so_proto->pr_sosend(so, addr, uio, 2573 top, control, flags, td); 2574 CURVNET_RESTORE(); 2575 return (error); 2576 } 2577 2578 /* 2579 * send(2), write(2) or aio_write(2) on a socket. 2580 */ 2581 int 2582 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio, 2583 struct mbuf *control, int flags, struct proc *userproc) 2584 { 2585 struct thread *td; 2586 ssize_t len; 2587 int error; 2588 2589 td = uio->uio_td; 2590 len = uio->uio_resid; 2591 CURVNET_SET(so->so_vnet); 2592 error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags, 2593 td); 2594 CURVNET_RESTORE(); 2595 if (error != 0) { 2596 /* 2597 * Clear transient errors for stream protocols if they made 2598 * some progress. Make exclusion for aio(4) that would 2599 * schedule a new write in case of EWOULDBLOCK and clear 2600 * error itself. See soaio_process_job(). 2601 */ 2602 if (uio->uio_resid != len && 2603 (so->so_proto->pr_flags & PR_ATOMIC) == 0 && 2604 userproc == NULL && 2605 (error == ERESTART || error == EINTR || 2606 error == EWOULDBLOCK)) 2607 error = 0; 2608 /* Generation of SIGPIPE can be controlled per socket. */ 2609 if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 && 2610 (flags & MSG_NOSIGNAL) == 0) { 2611 if (userproc != NULL) { 2612 /* aio(4) job */ 2613 PROC_LOCK(userproc); 2614 kern_psignal(userproc, SIGPIPE); 2615 PROC_UNLOCK(userproc); 2616 } else { 2617 PROC_LOCK(td->td_proc); 2618 tdsignal(td, SIGPIPE); 2619 PROC_UNLOCK(td->td_proc); 2620 } 2621 } 2622 } 2623 return (error); 2624 } 2625 2626 /* 2627 * The part of soreceive() that implements reading non-inline out-of-band 2628 * data from a socket. For more complete comments, see soreceive(), from 2629 * which this code originated. 2630 * 2631 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is 2632 * unable to return an mbuf chain to the caller. 2633 */ 2634 static int 2635 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags) 2636 { 2637 struct protosw *pr = so->so_proto; 2638 struct mbuf *m; 2639 int error; 2640 2641 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0")); 2642 VNET_SO_ASSERT(so); 2643 2644 m = m_get(M_WAITOK, MT_DATA); 2645 error = pr->pr_rcvoob(so, m, flags & MSG_PEEK); 2646 if (error) 2647 goto bad; 2648 do { 2649 error = uiomove(mtod(m, void *), 2650 (int) min(uio->uio_resid, m->m_len), uio); 2651 m = m_free(m); 2652 } while (uio->uio_resid && error == 0 && m); 2653 bad: 2654 if (m != NULL) 2655 m_freem(m); 2656 return (error); 2657 } 2658 2659 /* 2660 * Following replacement or removal of the first mbuf on the first mbuf chain 2661 * of a socket buffer, push necessary state changes back into the socket 2662 * buffer so that other consumers see the values consistently. 'nextrecord' 2663 * is the callers locally stored value of the original value of 2664 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes. 2665 * NOTE: 'nextrecord' may be NULL. 2666 */ 2667 static __inline void 2668 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord) 2669 { 2670 2671 SOCKBUF_LOCK_ASSERT(sb); 2672 /* 2673 * First, update for the new value of nextrecord. If necessary, make 2674 * it the first record. 2675 */ 2676 if (sb->sb_mb != NULL) 2677 sb->sb_mb->m_nextpkt = nextrecord; 2678 else 2679 sb->sb_mb = nextrecord; 2680 2681 /* 2682 * Now update any dependent socket buffer fields to reflect the new 2683 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the 2684 * addition of a second clause that takes care of the case where 2685 * sb_mb has been updated, but remains the last record. 2686 */ 2687 if (sb->sb_mb == NULL) { 2688 sb->sb_mbtail = NULL; 2689 sb->sb_lastrecord = NULL; 2690 } else if (sb->sb_mb->m_nextpkt == NULL) 2691 sb->sb_lastrecord = sb->sb_mb; 2692 } 2693 2694 /* 2695 * Implement receive operations on a socket. We depend on the way that 2696 * records are added to the sockbuf by sbappend. In particular, each record 2697 * (mbufs linked through m_next) must begin with an address if the protocol 2698 * so specifies, followed by an optional mbuf or mbufs containing ancillary 2699 * data, and then zero or more mbufs of data. In order to allow parallelism 2700 * between network receive and copying to user space, as well as avoid 2701 * sleeping with a mutex held, we release the socket buffer mutex during the 2702 * user space copy. Although the sockbuf is locked, new data may still be 2703 * appended, and thus we must maintain consistency of the sockbuf during that 2704 * time. 2705 * 2706 * The caller may receive the data as a single mbuf chain by supplying an 2707 * mbuf **mp0 for use in returning the chain. The uio is then used only for 2708 * the count in uio_resid. 2709 */ 2710 static int 2711 soreceive_generic_locked(struct socket *so, struct sockaddr **psa, 2712 struct uio *uio, struct mbuf **mp, struct mbuf **controlp, int *flagsp) 2713 { 2714 struct mbuf *m; 2715 int flags, error, offset; 2716 ssize_t len; 2717 struct protosw *pr = so->so_proto; 2718 struct mbuf *nextrecord; 2719 int moff, type = 0; 2720 ssize_t orig_resid = uio->uio_resid; 2721 bool report_real_len = false; 2722 2723 SOCK_IO_RECV_ASSERT_LOCKED(so); 2724 2725 error = 0; 2726 if (flagsp != NULL) { 2727 report_real_len = *flagsp & MSG_TRUNC; 2728 *flagsp &= ~MSG_TRUNC; 2729 flags = *flagsp &~ MSG_EOR; 2730 } else 2731 flags = 0; 2732 2733 restart: 2734 SOCKBUF_LOCK(&so->so_rcv); 2735 m = so->so_rcv.sb_mb; 2736 /* 2737 * If we have less data than requested, block awaiting more (subject 2738 * to any timeout) if: 2739 * 1. the current count is less than the low water mark, or 2740 * 2. MSG_DONTWAIT is not set 2741 */ 2742 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 && 2743 sbavail(&so->so_rcv) < uio->uio_resid) && 2744 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat && 2745 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) { 2746 KASSERT(m != NULL || !sbavail(&so->so_rcv), 2747 ("receive: m == %p sbavail == %u", 2748 m, sbavail(&so->so_rcv))); 2749 if (so->so_error || so->so_rerror) { 2750 if (m != NULL) 2751 goto dontblock; 2752 if (so->so_error) 2753 error = so->so_error; 2754 else 2755 error = so->so_rerror; 2756 if ((flags & MSG_PEEK) == 0) { 2757 if (so->so_error) 2758 so->so_error = 0; 2759 else 2760 so->so_rerror = 0; 2761 } 2762 SOCKBUF_UNLOCK(&so->so_rcv); 2763 goto release; 2764 } 2765 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2766 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2767 if (m != NULL) 2768 goto dontblock; 2769 #ifdef KERN_TLS 2770 else if (so->so_rcv.sb_tlsdcc == 0 && 2771 so->so_rcv.sb_tlscc == 0) { 2772 #else 2773 else { 2774 #endif 2775 SOCKBUF_UNLOCK(&so->so_rcv); 2776 goto release; 2777 } 2778 } 2779 for (; m != NULL; m = m->m_next) 2780 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) { 2781 m = so->so_rcv.sb_mb; 2782 goto dontblock; 2783 } 2784 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED | 2785 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 && 2786 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 2787 SOCKBUF_UNLOCK(&so->so_rcv); 2788 error = ENOTCONN; 2789 goto release; 2790 } 2791 if (uio->uio_resid == 0 && !report_real_len) { 2792 SOCKBUF_UNLOCK(&so->so_rcv); 2793 goto release; 2794 } 2795 if ((so->so_state & SS_NBIO) || 2796 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 2797 SOCKBUF_UNLOCK(&so->so_rcv); 2798 error = EWOULDBLOCK; 2799 goto release; 2800 } 2801 SBLASTRECORDCHK(&so->so_rcv); 2802 SBLASTMBUFCHK(&so->so_rcv); 2803 error = sbwait(so, SO_RCV); 2804 SOCKBUF_UNLOCK(&so->so_rcv); 2805 if (error) 2806 goto release; 2807 goto restart; 2808 } 2809 dontblock: 2810 /* 2811 * From this point onward, we maintain 'nextrecord' as a cache of the 2812 * pointer to the next record in the socket buffer. We must keep the 2813 * various socket buffer pointers and local stack versions of the 2814 * pointers in sync, pushing out modifications before dropping the 2815 * socket buffer mutex, and re-reading them when picking it up. 2816 * 2817 * Otherwise, we will race with the network stack appending new data 2818 * or records onto the socket buffer by using inconsistent/stale 2819 * versions of the field, possibly resulting in socket buffer 2820 * corruption. 2821 * 2822 * By holding the high-level sblock(), we prevent simultaneous 2823 * readers from pulling off the front of the socket buffer. 2824 */ 2825 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2826 if (uio->uio_td) 2827 uio->uio_td->td_ru.ru_msgrcv++; 2828 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb")); 2829 SBLASTRECORDCHK(&so->so_rcv); 2830 SBLASTMBUFCHK(&so->so_rcv); 2831 nextrecord = m->m_nextpkt; 2832 if (pr->pr_flags & PR_ADDR) { 2833 KASSERT(m->m_type == MT_SONAME, 2834 ("m->m_type == %d", m->m_type)); 2835 orig_resid = 0; 2836 if (psa != NULL) 2837 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 2838 M_NOWAIT); 2839 if (flags & MSG_PEEK) { 2840 m = m->m_next; 2841 } else { 2842 sbfree(&so->so_rcv, m); 2843 so->so_rcv.sb_mb = m_free(m); 2844 m = so->so_rcv.sb_mb; 2845 sockbuf_pushsync(&so->so_rcv, nextrecord); 2846 } 2847 } 2848 2849 /* 2850 * Process one or more MT_CONTROL mbufs present before any data mbufs 2851 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we 2852 * just copy the data; if !MSG_PEEK, we call into the protocol to 2853 * perform externalization (or freeing if controlp == NULL). 2854 */ 2855 if (m != NULL && m->m_type == MT_CONTROL) { 2856 struct mbuf *cm = NULL, *cmn; 2857 struct mbuf **cme = &cm; 2858 #ifdef KERN_TLS 2859 struct cmsghdr *cmsg; 2860 struct tls_get_record tgr; 2861 2862 /* 2863 * For MSG_TLSAPPDATA, check for an alert record. 2864 * If found, return ENXIO without removing 2865 * it from the receive queue. This allows a subsequent 2866 * call without MSG_TLSAPPDATA to receive it. 2867 * Note that, for TLS, there should only be a single 2868 * control mbuf with the TLS_GET_RECORD message in it. 2869 */ 2870 if (flags & MSG_TLSAPPDATA) { 2871 cmsg = mtod(m, struct cmsghdr *); 2872 if (cmsg->cmsg_type == TLS_GET_RECORD && 2873 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) { 2874 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr)); 2875 if (__predict_false(tgr.tls_type == 2876 TLS_RLTYPE_ALERT)) { 2877 SOCKBUF_UNLOCK(&so->so_rcv); 2878 error = ENXIO; 2879 goto release; 2880 } 2881 } 2882 } 2883 #endif 2884 2885 do { 2886 if (flags & MSG_PEEK) { 2887 if (controlp != NULL) { 2888 *controlp = m_copym(m, 0, m->m_len, 2889 M_NOWAIT); 2890 controlp = &(*controlp)->m_next; 2891 } 2892 m = m->m_next; 2893 } else { 2894 sbfree(&so->so_rcv, m); 2895 so->so_rcv.sb_mb = m->m_next; 2896 m->m_next = NULL; 2897 *cme = m; 2898 cme = &(*cme)->m_next; 2899 m = so->so_rcv.sb_mb; 2900 } 2901 } while (m != NULL && m->m_type == MT_CONTROL); 2902 if ((flags & MSG_PEEK) == 0) 2903 sockbuf_pushsync(&so->so_rcv, nextrecord); 2904 while (cm != NULL) { 2905 cmn = cm->m_next; 2906 cm->m_next = NULL; 2907 if (pr->pr_domain->dom_externalize != NULL) { 2908 SOCKBUF_UNLOCK(&so->so_rcv); 2909 VNET_SO_ASSERT(so); 2910 error = (*pr->pr_domain->dom_externalize) 2911 (cm, controlp, flags); 2912 SOCKBUF_LOCK(&so->so_rcv); 2913 } else if (controlp != NULL) 2914 *controlp = cm; 2915 else 2916 m_freem(cm); 2917 if (controlp != NULL) { 2918 while (*controlp != NULL) 2919 controlp = &(*controlp)->m_next; 2920 } 2921 cm = cmn; 2922 } 2923 if (m != NULL) 2924 nextrecord = so->so_rcv.sb_mb->m_nextpkt; 2925 else 2926 nextrecord = so->so_rcv.sb_mb; 2927 orig_resid = 0; 2928 } 2929 if (m != NULL) { 2930 if ((flags & MSG_PEEK) == 0) { 2931 KASSERT(m->m_nextpkt == nextrecord, 2932 ("soreceive: post-control, nextrecord !sync")); 2933 if (nextrecord == NULL) { 2934 KASSERT(so->so_rcv.sb_mb == m, 2935 ("soreceive: post-control, sb_mb!=m")); 2936 KASSERT(so->so_rcv.sb_lastrecord == m, 2937 ("soreceive: post-control, lastrecord!=m")); 2938 } 2939 } 2940 type = m->m_type; 2941 if (type == MT_OOBDATA) 2942 flags |= MSG_OOB; 2943 } else { 2944 if ((flags & MSG_PEEK) == 0) { 2945 KASSERT(so->so_rcv.sb_mb == nextrecord, 2946 ("soreceive: sb_mb != nextrecord")); 2947 if (so->so_rcv.sb_mb == NULL) { 2948 KASSERT(so->so_rcv.sb_lastrecord == NULL, 2949 ("soreceive: sb_lastercord != NULL")); 2950 } 2951 } 2952 } 2953 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2954 SBLASTRECORDCHK(&so->so_rcv); 2955 SBLASTMBUFCHK(&so->so_rcv); 2956 2957 /* 2958 * Now continue to read any data mbufs off of the head of the socket 2959 * buffer until the read request is satisfied. Note that 'type' is 2960 * used to store the type of any mbuf reads that have happened so far 2961 * such that soreceive() can stop reading if the type changes, which 2962 * causes soreceive() to return only one of regular data and inline 2963 * out-of-band data in a single socket receive operation. 2964 */ 2965 moff = 0; 2966 offset = 0; 2967 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0 2968 && error == 0) { 2969 /* 2970 * If the type of mbuf has changed since the last mbuf 2971 * examined ('type'), end the receive operation. 2972 */ 2973 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2974 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) { 2975 if (type != m->m_type) 2976 break; 2977 } else if (type == MT_OOBDATA) 2978 break; 2979 else 2980 KASSERT(m->m_type == MT_DATA, 2981 ("m->m_type == %d", m->m_type)); 2982 so->so_rcv.sb_state &= ~SBS_RCVATMARK; 2983 len = uio->uio_resid; 2984 if (so->so_oobmark && len > so->so_oobmark - offset) 2985 len = so->so_oobmark - offset; 2986 if (len > m->m_len - moff) 2987 len = m->m_len - moff; 2988 /* 2989 * If mp is set, just pass back the mbufs. Otherwise copy 2990 * them out via the uio, then free. Sockbuf must be 2991 * consistent here (points to current mbuf, it points to next 2992 * record) when we drop priority; we must note any additions 2993 * to the sockbuf when we block interrupts again. 2994 */ 2995 if (mp == NULL) { 2996 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 2997 SBLASTRECORDCHK(&so->so_rcv); 2998 SBLASTMBUFCHK(&so->so_rcv); 2999 SOCKBUF_UNLOCK(&so->so_rcv); 3000 if ((m->m_flags & M_EXTPG) != 0) 3001 error = m_unmapped_uiomove(m, moff, uio, 3002 (int)len); 3003 else 3004 error = uiomove(mtod(m, char *) + moff, 3005 (int)len, uio); 3006 SOCKBUF_LOCK(&so->so_rcv); 3007 if (error) { 3008 /* 3009 * The MT_SONAME mbuf has already been removed 3010 * from the record, so it is necessary to 3011 * remove the data mbufs, if any, to preserve 3012 * the invariant in the case of PR_ADDR that 3013 * requires MT_SONAME mbufs at the head of 3014 * each record. 3015 */ 3016 if (pr->pr_flags & PR_ATOMIC && 3017 ((flags & MSG_PEEK) == 0)) 3018 (void)sbdroprecord_locked(&so->so_rcv); 3019 SOCKBUF_UNLOCK(&so->so_rcv); 3020 goto release; 3021 } 3022 } else 3023 uio->uio_resid -= len; 3024 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3025 if (len == m->m_len - moff) { 3026 if (m->m_flags & M_EOR) 3027 flags |= MSG_EOR; 3028 if (flags & MSG_PEEK) { 3029 m = m->m_next; 3030 moff = 0; 3031 } else { 3032 nextrecord = m->m_nextpkt; 3033 sbfree(&so->so_rcv, m); 3034 if (mp != NULL) { 3035 m->m_nextpkt = NULL; 3036 *mp = m; 3037 mp = &m->m_next; 3038 so->so_rcv.sb_mb = m = m->m_next; 3039 *mp = NULL; 3040 } else { 3041 so->so_rcv.sb_mb = m_free(m); 3042 m = so->so_rcv.sb_mb; 3043 } 3044 sockbuf_pushsync(&so->so_rcv, nextrecord); 3045 SBLASTRECORDCHK(&so->so_rcv); 3046 SBLASTMBUFCHK(&so->so_rcv); 3047 } 3048 } else { 3049 if (flags & MSG_PEEK) 3050 moff += len; 3051 else { 3052 if (mp != NULL) { 3053 if (flags & MSG_DONTWAIT) { 3054 *mp = m_copym(m, 0, len, 3055 M_NOWAIT); 3056 if (*mp == NULL) { 3057 /* 3058 * m_copym() couldn't 3059 * allocate an mbuf. 3060 * Adjust uio_resid back 3061 * (it was adjusted 3062 * down by len bytes, 3063 * which we didn't end 3064 * up "copying" over). 3065 */ 3066 uio->uio_resid += len; 3067 break; 3068 } 3069 } else { 3070 SOCKBUF_UNLOCK(&so->so_rcv); 3071 *mp = m_copym(m, 0, len, 3072 M_WAITOK); 3073 SOCKBUF_LOCK(&so->so_rcv); 3074 } 3075 } 3076 sbcut_locked(&so->so_rcv, len); 3077 } 3078 } 3079 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3080 if (so->so_oobmark) { 3081 if ((flags & MSG_PEEK) == 0) { 3082 so->so_oobmark -= len; 3083 if (so->so_oobmark == 0) { 3084 so->so_rcv.sb_state |= SBS_RCVATMARK; 3085 break; 3086 } 3087 } else { 3088 offset += len; 3089 if (offset == so->so_oobmark) 3090 break; 3091 } 3092 } 3093 if (flags & MSG_EOR) 3094 break; 3095 /* 3096 * If the MSG_WAITALL flag is set (for non-atomic socket), we 3097 * must not quit until "uio->uio_resid == 0" or an error 3098 * termination. If a signal/timeout occurs, return with a 3099 * short count but without error. Keep sockbuf locked 3100 * against other readers. 3101 */ 3102 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 && 3103 !sosendallatonce(so) && nextrecord == NULL) { 3104 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3105 if (so->so_error || so->so_rerror || 3106 so->so_rcv.sb_state & SBS_CANTRCVMORE) 3107 break; 3108 /* 3109 * Notify the protocol that some data has been 3110 * drained before blocking. 3111 */ 3112 if (pr->pr_flags & PR_WANTRCVD) { 3113 SOCKBUF_UNLOCK(&so->so_rcv); 3114 VNET_SO_ASSERT(so); 3115 pr->pr_rcvd(so, flags); 3116 SOCKBUF_LOCK(&so->so_rcv); 3117 if (__predict_false(so->so_rcv.sb_mb == NULL && 3118 (so->so_error || so->so_rerror || 3119 so->so_rcv.sb_state & SBS_CANTRCVMORE))) 3120 break; 3121 } 3122 SBLASTRECORDCHK(&so->so_rcv); 3123 SBLASTMBUFCHK(&so->so_rcv); 3124 /* 3125 * We could receive some data while was notifying 3126 * the protocol. Skip blocking in this case. 3127 */ 3128 if (so->so_rcv.sb_mb == NULL) { 3129 error = sbwait(so, SO_RCV); 3130 if (error) { 3131 SOCKBUF_UNLOCK(&so->so_rcv); 3132 goto release; 3133 } 3134 } 3135 m = so->so_rcv.sb_mb; 3136 if (m != NULL) 3137 nextrecord = m->m_nextpkt; 3138 } 3139 } 3140 3141 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3142 if (m != NULL && pr->pr_flags & PR_ATOMIC) { 3143 if (report_real_len) 3144 uio->uio_resid -= m_length(m, NULL) - moff; 3145 flags |= MSG_TRUNC; 3146 if ((flags & MSG_PEEK) == 0) 3147 (void) sbdroprecord_locked(&so->so_rcv); 3148 } 3149 if ((flags & MSG_PEEK) == 0) { 3150 if (m == NULL) { 3151 /* 3152 * First part is an inline SB_EMPTY_FIXUP(). Second 3153 * part makes sure sb_lastrecord is up-to-date if 3154 * there is still data in the socket buffer. 3155 */ 3156 so->so_rcv.sb_mb = nextrecord; 3157 if (so->so_rcv.sb_mb == NULL) { 3158 so->so_rcv.sb_mbtail = NULL; 3159 so->so_rcv.sb_lastrecord = NULL; 3160 } else if (nextrecord->m_nextpkt == NULL) 3161 so->so_rcv.sb_lastrecord = nextrecord; 3162 } 3163 SBLASTRECORDCHK(&so->so_rcv); 3164 SBLASTMBUFCHK(&so->so_rcv); 3165 /* 3166 * If soreceive() is being done from the socket callback, 3167 * then don't need to generate ACK to peer to update window, 3168 * since ACK will be generated on return to TCP. 3169 */ 3170 if (!(flags & MSG_SOCALLBCK) && 3171 (pr->pr_flags & PR_WANTRCVD)) { 3172 SOCKBUF_UNLOCK(&so->so_rcv); 3173 VNET_SO_ASSERT(so); 3174 pr->pr_rcvd(so, flags); 3175 SOCKBUF_LOCK(&so->so_rcv); 3176 } 3177 } 3178 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3179 if (orig_resid == uio->uio_resid && orig_resid && 3180 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 3181 SOCKBUF_UNLOCK(&so->so_rcv); 3182 goto restart; 3183 } 3184 SOCKBUF_UNLOCK(&so->so_rcv); 3185 3186 if (flagsp != NULL) 3187 *flagsp |= flags; 3188 release: 3189 return (error); 3190 } 3191 3192 int 3193 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio, 3194 struct mbuf **mp, struct mbuf **controlp, int *flagsp) 3195 { 3196 int error, flags; 3197 3198 if (psa != NULL) 3199 *psa = NULL; 3200 if (controlp != NULL) 3201 *controlp = NULL; 3202 if (flagsp != NULL) { 3203 flags = *flagsp; 3204 if ((flags & MSG_OOB) != 0) 3205 return (soreceive_rcvoob(so, uio, flags)); 3206 } else { 3207 flags = 0; 3208 } 3209 if (mp != NULL) 3210 *mp = NULL; 3211 3212 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 3213 if (error) 3214 return (error); 3215 error = soreceive_generic_locked(so, psa, uio, mp, controlp, flagsp); 3216 SOCK_IO_RECV_UNLOCK(so); 3217 return (error); 3218 } 3219 3220 /* 3221 * Optimized version of soreceive() for stream (TCP) sockets. 3222 */ 3223 static int 3224 soreceive_stream_locked(struct socket *so, struct sockbuf *sb, 3225 struct sockaddr **psa, struct uio *uio, struct mbuf **mp0, 3226 struct mbuf **controlp, int flags) 3227 { 3228 int len = 0, error = 0, oresid; 3229 struct mbuf *m, *n = NULL; 3230 3231 SOCK_IO_RECV_ASSERT_LOCKED(so); 3232 3233 /* Easy one, no space to copyout anything. */ 3234 if (uio->uio_resid == 0) 3235 return (EINVAL); 3236 oresid = uio->uio_resid; 3237 3238 SOCKBUF_LOCK(sb); 3239 /* We will never ever get anything unless we are or were connected. */ 3240 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) { 3241 error = ENOTCONN; 3242 goto out; 3243 } 3244 3245 restart: 3246 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3247 3248 /* Abort if socket has reported problems. */ 3249 if (so->so_error) { 3250 if (sbavail(sb) > 0) 3251 goto deliver; 3252 if (oresid > uio->uio_resid) 3253 goto out; 3254 error = so->so_error; 3255 if (!(flags & MSG_PEEK)) 3256 so->so_error = 0; 3257 goto out; 3258 } 3259 3260 /* Door is closed. Deliver what is left, if any. */ 3261 if (sb->sb_state & SBS_CANTRCVMORE) { 3262 if (sbavail(sb) > 0) 3263 goto deliver; 3264 else 3265 goto out; 3266 } 3267 3268 /* Socket buffer is empty and we shall not block. */ 3269 if (sbavail(sb) == 0 && 3270 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) { 3271 error = EAGAIN; 3272 goto out; 3273 } 3274 3275 /* Socket buffer got some data that we shall deliver now. */ 3276 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) && 3277 ((so->so_state & SS_NBIO) || 3278 (flags & (MSG_DONTWAIT|MSG_NBIO)) || 3279 sbavail(sb) >= sb->sb_lowat || 3280 sbavail(sb) >= uio->uio_resid || 3281 sbavail(sb) >= sb->sb_hiwat) ) { 3282 goto deliver; 3283 } 3284 3285 /* On MSG_WAITALL we must wait until all data or error arrives. */ 3286 if ((flags & MSG_WAITALL) && 3287 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat)) 3288 goto deliver; 3289 3290 /* 3291 * Wait and block until (more) data comes in. 3292 * NB: Drops the sockbuf lock during wait. 3293 */ 3294 error = sbwait(so, SO_RCV); 3295 if (error) 3296 goto out; 3297 goto restart; 3298 3299 deliver: 3300 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3301 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__)); 3302 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__)); 3303 3304 /* Statistics. */ 3305 if (uio->uio_td) 3306 uio->uio_td->td_ru.ru_msgrcv++; 3307 3308 /* Fill uio until full or current end of socket buffer is reached. */ 3309 len = min(uio->uio_resid, sbavail(sb)); 3310 if (mp0 != NULL) { 3311 /* Dequeue as many mbufs as possible. */ 3312 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) { 3313 if (*mp0 == NULL) 3314 *mp0 = sb->sb_mb; 3315 else 3316 m_cat(*mp0, sb->sb_mb); 3317 for (m = sb->sb_mb; 3318 m != NULL && m->m_len <= len; 3319 m = m->m_next) { 3320 KASSERT(!(m->m_flags & M_NOTAVAIL), 3321 ("%s: m %p not available", __func__, m)); 3322 len -= m->m_len; 3323 uio->uio_resid -= m->m_len; 3324 sbfree(sb, m); 3325 n = m; 3326 } 3327 n->m_next = NULL; 3328 sb->sb_mb = m; 3329 sb->sb_lastrecord = sb->sb_mb; 3330 if (sb->sb_mb == NULL) 3331 SB_EMPTY_FIXUP(sb); 3332 } 3333 /* Copy the remainder. */ 3334 if (len > 0) { 3335 KASSERT(sb->sb_mb != NULL, 3336 ("%s: len > 0 && sb->sb_mb empty", __func__)); 3337 3338 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT); 3339 if (m == NULL) 3340 len = 0; /* Don't flush data from sockbuf. */ 3341 else 3342 uio->uio_resid -= len; 3343 if (*mp0 != NULL) 3344 m_cat(*mp0, m); 3345 else 3346 *mp0 = m; 3347 if (*mp0 == NULL) { 3348 error = ENOBUFS; 3349 goto out; 3350 } 3351 } 3352 } else { 3353 /* NB: Must unlock socket buffer as uiomove may sleep. */ 3354 SOCKBUF_UNLOCK(sb); 3355 error = m_mbuftouio(uio, sb->sb_mb, len); 3356 SOCKBUF_LOCK(sb); 3357 if (error) 3358 goto out; 3359 } 3360 SBLASTRECORDCHK(sb); 3361 SBLASTMBUFCHK(sb); 3362 3363 /* 3364 * Remove the delivered data from the socket buffer unless we 3365 * were only peeking. 3366 */ 3367 if (!(flags & MSG_PEEK)) { 3368 if (len > 0) 3369 sbdrop_locked(sb, len); 3370 3371 /* Notify protocol that we drained some data. */ 3372 if ((so->so_proto->pr_flags & PR_WANTRCVD) && 3373 (((flags & MSG_WAITALL) && uio->uio_resid > 0) || 3374 !(flags & MSG_SOCALLBCK))) { 3375 SOCKBUF_UNLOCK(sb); 3376 VNET_SO_ASSERT(so); 3377 so->so_proto->pr_rcvd(so, flags); 3378 SOCKBUF_LOCK(sb); 3379 } 3380 } 3381 3382 /* 3383 * For MSG_WAITALL we may have to loop again and wait for 3384 * more data to come in. 3385 */ 3386 if ((flags & MSG_WAITALL) && uio->uio_resid > 0) 3387 goto restart; 3388 out: 3389 SBLASTRECORDCHK(sb); 3390 SBLASTMBUFCHK(sb); 3391 SOCKBUF_UNLOCK(sb); 3392 return (error); 3393 } 3394 3395 int 3396 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio, 3397 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3398 { 3399 struct sockbuf *sb; 3400 int error, flags; 3401 3402 sb = &so->so_rcv; 3403 3404 /* We only do stream sockets. */ 3405 if (so->so_type != SOCK_STREAM) 3406 return (EINVAL); 3407 if (psa != NULL) 3408 *psa = NULL; 3409 if (flagsp != NULL) 3410 flags = *flagsp & ~MSG_EOR; 3411 else 3412 flags = 0; 3413 if (controlp != NULL) 3414 *controlp = NULL; 3415 if (flags & MSG_OOB) 3416 return (soreceive_rcvoob(so, uio, flags)); 3417 if (mp0 != NULL) 3418 *mp0 = NULL; 3419 3420 #ifdef KERN_TLS 3421 /* 3422 * KTLS store TLS records as records with a control message to 3423 * describe the framing. 3424 * 3425 * We check once here before acquiring locks to optimize the 3426 * common case. 3427 */ 3428 if (sb->sb_tls_info != NULL) 3429 return (soreceive_generic(so, psa, uio, mp0, controlp, 3430 flagsp)); 3431 #endif 3432 3433 /* 3434 * Prevent other threads from reading from the socket. This lock may be 3435 * dropped in order to sleep waiting for data to arrive. 3436 */ 3437 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags)); 3438 if (error) 3439 return (error); 3440 #ifdef KERN_TLS 3441 if (__predict_false(sb->sb_tls_info != NULL)) { 3442 SOCK_IO_RECV_UNLOCK(so); 3443 return (soreceive_generic(so, psa, uio, mp0, controlp, 3444 flagsp)); 3445 } 3446 #endif 3447 error = soreceive_stream_locked(so, sb, psa, uio, mp0, controlp, flags); 3448 SOCK_IO_RECV_UNLOCK(so); 3449 return (error); 3450 } 3451 3452 /* 3453 * Optimized version of soreceive() for simple datagram cases from userspace. 3454 * Unlike in the stream case, we're able to drop a datagram if copyout() 3455 * fails, and because we handle datagrams atomically, we don't need to use a 3456 * sleep lock to prevent I/O interlacing. 3457 */ 3458 int 3459 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio, 3460 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3461 { 3462 struct mbuf *m, *m2; 3463 int flags, error; 3464 ssize_t len; 3465 struct protosw *pr = so->so_proto; 3466 struct mbuf *nextrecord; 3467 3468 if (psa != NULL) 3469 *psa = NULL; 3470 if (controlp != NULL) 3471 *controlp = NULL; 3472 if (flagsp != NULL) 3473 flags = *flagsp &~ MSG_EOR; 3474 else 3475 flags = 0; 3476 3477 /* 3478 * For any complicated cases, fall back to the full 3479 * soreceive_generic(). 3480 */ 3481 if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC))) 3482 return (soreceive_generic(so, psa, uio, mp0, controlp, 3483 flagsp)); 3484 3485 /* 3486 * Enforce restrictions on use. 3487 */ 3488 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0, 3489 ("soreceive_dgram: wantrcvd")); 3490 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic")); 3491 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0, 3492 ("soreceive_dgram: SBS_RCVATMARK")); 3493 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0, 3494 ("soreceive_dgram: P_CONNREQUIRED")); 3495 3496 /* 3497 * Loop blocking while waiting for a datagram. 3498 */ 3499 SOCKBUF_LOCK(&so->so_rcv); 3500 while ((m = so->so_rcv.sb_mb) == NULL) { 3501 KASSERT(sbavail(&so->so_rcv) == 0, 3502 ("soreceive_dgram: sb_mb NULL but sbavail %u", 3503 sbavail(&so->so_rcv))); 3504 if (so->so_error) { 3505 error = so->so_error; 3506 so->so_error = 0; 3507 SOCKBUF_UNLOCK(&so->so_rcv); 3508 return (error); 3509 } 3510 if (so->so_rcv.sb_state & SBS_CANTRCVMORE || 3511 uio->uio_resid == 0) { 3512 SOCKBUF_UNLOCK(&so->so_rcv); 3513 return (0); 3514 } 3515 if ((so->so_state & SS_NBIO) || 3516 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 3517 SOCKBUF_UNLOCK(&so->so_rcv); 3518 return (EWOULDBLOCK); 3519 } 3520 SBLASTRECORDCHK(&so->so_rcv); 3521 SBLASTMBUFCHK(&so->so_rcv); 3522 error = sbwait(so, SO_RCV); 3523 if (error) { 3524 SOCKBUF_UNLOCK(&so->so_rcv); 3525 return (error); 3526 } 3527 } 3528 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 3529 3530 if (uio->uio_td) 3531 uio->uio_td->td_ru.ru_msgrcv++; 3532 SBLASTRECORDCHK(&so->so_rcv); 3533 SBLASTMBUFCHK(&so->so_rcv); 3534 nextrecord = m->m_nextpkt; 3535 if (nextrecord == NULL) { 3536 KASSERT(so->so_rcv.sb_lastrecord == m, 3537 ("soreceive_dgram: lastrecord != m")); 3538 } 3539 3540 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord, 3541 ("soreceive_dgram: m_nextpkt != nextrecord")); 3542 3543 /* 3544 * Pull 'm' and its chain off the front of the packet queue. 3545 */ 3546 so->so_rcv.sb_mb = NULL; 3547 sockbuf_pushsync(&so->so_rcv, nextrecord); 3548 3549 /* 3550 * Walk 'm's chain and free that many bytes from the socket buffer. 3551 */ 3552 for (m2 = m; m2 != NULL; m2 = m2->m_next) 3553 sbfree(&so->so_rcv, m2); 3554 3555 /* 3556 * Do a few last checks before we let go of the lock. 3557 */ 3558 SBLASTRECORDCHK(&so->so_rcv); 3559 SBLASTMBUFCHK(&so->so_rcv); 3560 SOCKBUF_UNLOCK(&so->so_rcv); 3561 3562 if (pr->pr_flags & PR_ADDR) { 3563 KASSERT(m->m_type == MT_SONAME, 3564 ("m->m_type == %d", m->m_type)); 3565 if (psa != NULL) 3566 *psa = sodupsockaddr(mtod(m, struct sockaddr *), 3567 M_WAITOK); 3568 m = m_free(m); 3569 } 3570 KASSERT(m, ("%s: no data or control after soname", __func__)); 3571 3572 /* 3573 * Packet to copyout() is now in 'm' and it is disconnected from the 3574 * queue. 3575 * 3576 * Process one or more MT_CONTROL mbufs present before any data mbufs 3577 * in the first mbuf chain on the socket buffer. We call into the 3578 * protocol to perform externalization (or freeing if controlp == 3579 * NULL). In some cases there can be only MT_CONTROL mbufs without 3580 * MT_DATA mbufs. 3581 */ 3582 if (m->m_type == MT_CONTROL) { 3583 struct mbuf *cm = NULL, *cmn; 3584 struct mbuf **cme = &cm; 3585 3586 do { 3587 m2 = m->m_next; 3588 m->m_next = NULL; 3589 *cme = m; 3590 cme = &(*cme)->m_next; 3591 m = m2; 3592 } while (m != NULL && m->m_type == MT_CONTROL); 3593 while (cm != NULL) { 3594 cmn = cm->m_next; 3595 cm->m_next = NULL; 3596 if (pr->pr_domain->dom_externalize != NULL) { 3597 error = (*pr->pr_domain->dom_externalize) 3598 (cm, controlp, flags); 3599 } else if (controlp != NULL) 3600 *controlp = cm; 3601 else 3602 m_freem(cm); 3603 if (controlp != NULL) { 3604 while (*controlp != NULL) 3605 controlp = &(*controlp)->m_next; 3606 } 3607 cm = cmn; 3608 } 3609 } 3610 KASSERT(m == NULL || m->m_type == MT_DATA, 3611 ("soreceive_dgram: !data")); 3612 while (m != NULL && uio->uio_resid > 0) { 3613 len = uio->uio_resid; 3614 if (len > m->m_len) 3615 len = m->m_len; 3616 error = uiomove(mtod(m, char *), (int)len, uio); 3617 if (error) { 3618 m_freem(m); 3619 return (error); 3620 } 3621 if (len == m->m_len) 3622 m = m_free(m); 3623 else { 3624 m->m_data += len; 3625 m->m_len -= len; 3626 } 3627 } 3628 if (m != NULL) { 3629 flags |= MSG_TRUNC; 3630 m_freem(m); 3631 } 3632 if (flagsp != NULL) 3633 *flagsp |= flags; 3634 return (0); 3635 } 3636 3637 int 3638 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio, 3639 struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 3640 { 3641 int error; 3642 3643 CURVNET_SET(so->so_vnet); 3644 error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp); 3645 CURVNET_RESTORE(); 3646 return (error); 3647 } 3648 3649 int 3650 soshutdown(struct socket *so, enum shutdown_how how) 3651 { 3652 int error; 3653 3654 CURVNET_SET(so->so_vnet); 3655 error = so->so_proto->pr_shutdown(so, how); 3656 CURVNET_RESTORE(); 3657 3658 return (error); 3659 } 3660 3661 /* 3662 * Used by several pr_shutdown implementations that use generic socket buffers. 3663 */ 3664 void 3665 sorflush(struct socket *so) 3666 { 3667 int error; 3668 3669 VNET_SO_ASSERT(so); 3670 3671 /* 3672 * Dislodge threads currently blocked in receive and wait to acquire 3673 * a lock against other simultaneous readers before clearing the 3674 * socket buffer. Don't let our acquire be interrupted by a signal 3675 * despite any existing socket disposition on interruptable waiting. 3676 * 3677 * The SOCK_IO_RECV_LOCK() is important here as there some pr_soreceive 3678 * methods that read the top of the socket buffer without acquisition 3679 * of the socket buffer mutex, assuming that top of the buffer 3680 * exclusively belongs to the read(2) syscall. This is handy when 3681 * performing MSG_PEEK. 3682 */ 3683 socantrcvmore(so); 3684 3685 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR); 3686 if (error != 0) { 3687 KASSERT(SOLISTENING(so), 3688 ("%s: soiolock(%p) failed", __func__, so)); 3689 return; 3690 } 3691 3692 sbrelease(so, SO_RCV); 3693 SOCK_IO_RECV_UNLOCK(so); 3694 3695 } 3696 3697 int 3698 sosetfib(struct socket *so, int fibnum) 3699 { 3700 if (fibnum < 0 || fibnum >= rt_numfibs) 3701 return (EINVAL); 3702 3703 SOCK_LOCK(so); 3704 so->so_fibnum = fibnum; 3705 SOCK_UNLOCK(so); 3706 3707 return (0); 3708 } 3709 3710 #ifdef SOCKET_HHOOK 3711 /* 3712 * Wrapper for Socket established helper hook. 3713 * Parameters: socket, context of the hook point, hook id. 3714 */ 3715 static inline int 3716 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id) 3717 { 3718 struct socket_hhook_data hhook_data = { 3719 .so = so, 3720 .hctx = hctx, 3721 .m = NULL, 3722 .status = 0 3723 }; 3724 3725 CURVNET_SET(so->so_vnet); 3726 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd); 3727 CURVNET_RESTORE(); 3728 3729 /* Ugly but needed, since hhooks return void for now */ 3730 return (hhook_data.status); 3731 } 3732 #endif 3733 3734 /* 3735 * Perhaps this routine, and sooptcopyout(), below, ought to come in an 3736 * additional variant to handle the case where the option value needs to be 3737 * some kind of integer, but not a specific size. In addition to their use 3738 * here, these functions are also called by the protocol-level pr_ctloutput() 3739 * routines. 3740 */ 3741 int 3742 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen) 3743 { 3744 size_t valsize; 3745 3746 /* 3747 * If the user gives us more than we wanted, we ignore it, but if we 3748 * don't get the minimum length the caller wants, we return EINVAL. 3749 * On success, sopt->sopt_valsize is set to however much we actually 3750 * retrieved. 3751 */ 3752 if ((valsize = sopt->sopt_valsize) < minlen) 3753 return EINVAL; 3754 if (valsize > len) 3755 sopt->sopt_valsize = valsize = len; 3756 3757 if (sopt->sopt_td != NULL) 3758 return (copyin(sopt->sopt_val, buf, valsize)); 3759 3760 bcopy(sopt->sopt_val, buf, valsize); 3761 return (0); 3762 } 3763 3764 /* 3765 * Kernel version of setsockopt(2). 3766 * 3767 * XXX: optlen is size_t, not socklen_t 3768 */ 3769 int 3770 so_setsockopt(struct socket *so, int level, int optname, void *optval, 3771 size_t optlen) 3772 { 3773 struct sockopt sopt; 3774 3775 sopt.sopt_level = level; 3776 sopt.sopt_name = optname; 3777 sopt.sopt_dir = SOPT_SET; 3778 sopt.sopt_val = optval; 3779 sopt.sopt_valsize = optlen; 3780 sopt.sopt_td = NULL; 3781 return (sosetopt(so, &sopt)); 3782 } 3783 3784 int 3785 sosetopt(struct socket *so, struct sockopt *sopt) 3786 { 3787 int error, optval; 3788 struct linger l; 3789 struct timeval tv; 3790 sbintime_t val, *valp; 3791 uint32_t val32; 3792 #ifdef MAC 3793 struct mac extmac; 3794 #endif 3795 3796 CURVNET_SET(so->so_vnet); 3797 error = 0; 3798 if (sopt->sopt_level != SOL_SOCKET) { 3799 if (so->so_proto->pr_ctloutput != NULL) 3800 error = (*so->so_proto->pr_ctloutput)(so, sopt); 3801 else 3802 error = ENOPROTOOPT; 3803 } else { 3804 switch (sopt->sopt_name) { 3805 case SO_ACCEPTFILTER: 3806 error = accept_filt_setopt(so, sopt); 3807 if (error) 3808 goto bad; 3809 break; 3810 3811 case SO_LINGER: 3812 error = sooptcopyin(sopt, &l, sizeof l, sizeof l); 3813 if (error) 3814 goto bad; 3815 if (l.l_linger < 0 || 3816 l.l_linger > USHRT_MAX || 3817 l.l_linger > (INT_MAX / hz)) { 3818 error = EDOM; 3819 goto bad; 3820 } 3821 SOCK_LOCK(so); 3822 so->so_linger = l.l_linger; 3823 if (l.l_onoff) 3824 so->so_options |= SO_LINGER; 3825 else 3826 so->so_options &= ~SO_LINGER; 3827 SOCK_UNLOCK(so); 3828 break; 3829 3830 case SO_DEBUG: 3831 case SO_KEEPALIVE: 3832 case SO_DONTROUTE: 3833 case SO_USELOOPBACK: 3834 case SO_BROADCAST: 3835 case SO_REUSEADDR: 3836 case SO_REUSEPORT: 3837 case SO_REUSEPORT_LB: 3838 case SO_OOBINLINE: 3839 case SO_TIMESTAMP: 3840 case SO_BINTIME: 3841 case SO_NOSIGPIPE: 3842 case SO_NO_DDP: 3843 case SO_NO_OFFLOAD: 3844 case SO_RERROR: 3845 error = sooptcopyin(sopt, &optval, sizeof optval, 3846 sizeof optval); 3847 if (error) 3848 goto bad; 3849 SOCK_LOCK(so); 3850 if (optval) 3851 so->so_options |= sopt->sopt_name; 3852 else 3853 so->so_options &= ~sopt->sopt_name; 3854 SOCK_UNLOCK(so); 3855 break; 3856 3857 case SO_SETFIB: 3858 error = so->so_proto->pr_ctloutput(so, sopt); 3859 break; 3860 3861 case SO_USER_COOKIE: 3862 error = sooptcopyin(sopt, &val32, sizeof val32, 3863 sizeof val32); 3864 if (error) 3865 goto bad; 3866 so->so_user_cookie = val32; 3867 break; 3868 3869 case SO_SNDBUF: 3870 case SO_RCVBUF: 3871 case SO_SNDLOWAT: 3872 case SO_RCVLOWAT: 3873 error = so->so_proto->pr_setsbopt(so, sopt); 3874 if (error) 3875 goto bad; 3876 break; 3877 3878 case SO_SNDTIMEO: 3879 case SO_RCVTIMEO: 3880 #ifdef COMPAT_FREEBSD32 3881 if (SV_CURPROC_FLAG(SV_ILP32)) { 3882 struct timeval32 tv32; 3883 3884 error = sooptcopyin(sopt, &tv32, sizeof tv32, 3885 sizeof tv32); 3886 CP(tv32, tv, tv_sec); 3887 CP(tv32, tv, tv_usec); 3888 } else 3889 #endif 3890 error = sooptcopyin(sopt, &tv, sizeof tv, 3891 sizeof tv); 3892 if (error) 3893 goto bad; 3894 if (tv.tv_sec < 0 || tv.tv_usec < 0 || 3895 tv.tv_usec >= 1000000) { 3896 error = EDOM; 3897 goto bad; 3898 } 3899 if (tv.tv_sec > INT32_MAX) 3900 val = SBT_MAX; 3901 else 3902 val = tvtosbt(tv); 3903 SOCK_LOCK(so); 3904 valp = sopt->sopt_name == SO_SNDTIMEO ? 3905 (SOLISTENING(so) ? &so->sol_sbsnd_timeo : 3906 &so->so_snd.sb_timeo) : 3907 (SOLISTENING(so) ? &so->sol_sbrcv_timeo : 3908 &so->so_rcv.sb_timeo); 3909 *valp = val; 3910 SOCK_UNLOCK(so); 3911 break; 3912 3913 case SO_LABEL: 3914 #ifdef MAC 3915 error = sooptcopyin(sopt, &extmac, sizeof extmac, 3916 sizeof extmac); 3917 if (error) 3918 goto bad; 3919 error = mac_setsockopt_label(sopt->sopt_td->td_ucred, 3920 so, &extmac); 3921 #else 3922 error = EOPNOTSUPP; 3923 #endif 3924 break; 3925 3926 case SO_TS_CLOCK: 3927 error = sooptcopyin(sopt, &optval, sizeof optval, 3928 sizeof optval); 3929 if (error) 3930 goto bad; 3931 if (optval < 0 || optval > SO_TS_CLOCK_MAX) { 3932 error = EINVAL; 3933 goto bad; 3934 } 3935 so->so_ts_clock = optval; 3936 break; 3937 3938 case SO_MAX_PACING_RATE: 3939 error = sooptcopyin(sopt, &val32, sizeof(val32), 3940 sizeof(val32)); 3941 if (error) 3942 goto bad; 3943 so->so_max_pacing_rate = val32; 3944 break; 3945 3946 case SO_SPLICE: { 3947 struct splice splice; 3948 3949 #ifdef COMPAT_FREEBSD32 3950 if (SV_CURPROC_FLAG(SV_ILP32)) { 3951 struct splice32 splice32; 3952 3953 error = sooptcopyin(sopt, &splice32, 3954 sizeof(splice32), sizeof(splice32)); 3955 if (error == 0) { 3956 splice.sp_fd = splice32.sp_fd; 3957 splice.sp_max = splice32.sp_max; 3958 CP(splice32.sp_idle, splice.sp_idle, 3959 tv_sec); 3960 CP(splice32.sp_idle, splice.sp_idle, 3961 tv_usec); 3962 } 3963 } else 3964 #endif 3965 { 3966 error = sooptcopyin(sopt, &splice, 3967 sizeof(splice), sizeof(splice)); 3968 } 3969 if (error) 3970 goto bad; 3971 #ifdef KTRACE 3972 if (KTRPOINT(curthread, KTR_STRUCT)) 3973 ktrsplice(&splice); 3974 #endif 3975 3976 error = splice_init(); 3977 if (error != 0) 3978 goto bad; 3979 3980 if (splice.sp_fd >= 0) { 3981 struct file *fp; 3982 struct socket *so2; 3983 3984 if (!cap_rights_contains(sopt->sopt_rights, 3985 &cap_recv_rights)) { 3986 error = ENOTCAPABLE; 3987 goto bad; 3988 } 3989 error = getsock(sopt->sopt_td, splice.sp_fd, 3990 &cap_send_rights, &fp); 3991 if (error != 0) 3992 goto bad; 3993 so2 = fp->f_data; 3994 3995 error = so_splice(so, so2, &splice); 3996 fdrop(fp, sopt->sopt_td); 3997 } else { 3998 error = so_unsplice(so, false); 3999 } 4000 break; 4001 } 4002 default: 4003 #ifdef SOCKET_HHOOK 4004 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 4005 error = hhook_run_socket(so, sopt, 4006 HHOOK_SOCKET_OPT); 4007 else 4008 #endif 4009 error = ENOPROTOOPT; 4010 break; 4011 } 4012 if (error == 0 && so->so_proto->pr_ctloutput != NULL) 4013 (void)(*so->so_proto->pr_ctloutput)(so, sopt); 4014 } 4015 bad: 4016 CURVNET_RESTORE(); 4017 return (error); 4018 } 4019 4020 /* 4021 * Helper routine for getsockopt. 4022 */ 4023 int 4024 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len) 4025 { 4026 int error; 4027 size_t valsize; 4028 4029 error = 0; 4030 4031 /* 4032 * Documented get behavior is that we always return a value, possibly 4033 * truncated to fit in the user's buffer. Traditional behavior is 4034 * that we always tell the user precisely how much we copied, rather 4035 * than something useful like the total amount we had available for 4036 * her. Note that this interface is not idempotent; the entire 4037 * answer must be generated ahead of time. 4038 */ 4039 valsize = min(len, sopt->sopt_valsize); 4040 sopt->sopt_valsize = valsize; 4041 if (sopt->sopt_val != NULL) { 4042 if (sopt->sopt_td != NULL) 4043 error = copyout(buf, sopt->sopt_val, valsize); 4044 else 4045 bcopy(buf, sopt->sopt_val, valsize); 4046 } 4047 return (error); 4048 } 4049 4050 int 4051 sogetopt(struct socket *so, struct sockopt *sopt) 4052 { 4053 int error, optval; 4054 struct linger l; 4055 struct timeval tv; 4056 #ifdef MAC 4057 struct mac extmac; 4058 #endif 4059 4060 CURVNET_SET(so->so_vnet); 4061 error = 0; 4062 if (sopt->sopt_level != SOL_SOCKET) { 4063 if (so->so_proto->pr_ctloutput != NULL) 4064 error = (*so->so_proto->pr_ctloutput)(so, sopt); 4065 else 4066 error = ENOPROTOOPT; 4067 CURVNET_RESTORE(); 4068 return (error); 4069 } else { 4070 switch (sopt->sopt_name) { 4071 case SO_ACCEPTFILTER: 4072 error = accept_filt_getopt(so, sopt); 4073 break; 4074 4075 case SO_LINGER: 4076 SOCK_LOCK(so); 4077 l.l_onoff = so->so_options & SO_LINGER; 4078 l.l_linger = so->so_linger; 4079 SOCK_UNLOCK(so); 4080 error = sooptcopyout(sopt, &l, sizeof l); 4081 break; 4082 4083 case SO_USELOOPBACK: 4084 case SO_DONTROUTE: 4085 case SO_DEBUG: 4086 case SO_KEEPALIVE: 4087 case SO_REUSEADDR: 4088 case SO_REUSEPORT: 4089 case SO_REUSEPORT_LB: 4090 case SO_BROADCAST: 4091 case SO_OOBINLINE: 4092 case SO_ACCEPTCONN: 4093 case SO_TIMESTAMP: 4094 case SO_BINTIME: 4095 case SO_NOSIGPIPE: 4096 case SO_NO_DDP: 4097 case SO_NO_OFFLOAD: 4098 case SO_RERROR: 4099 optval = so->so_options & sopt->sopt_name; 4100 integer: 4101 error = sooptcopyout(sopt, &optval, sizeof optval); 4102 break; 4103 4104 case SO_FIB: 4105 SOCK_LOCK(so); 4106 optval = so->so_fibnum; 4107 SOCK_UNLOCK(so); 4108 goto integer; 4109 4110 case SO_DOMAIN: 4111 optval = so->so_proto->pr_domain->dom_family; 4112 goto integer; 4113 4114 case SO_TYPE: 4115 optval = so->so_type; 4116 goto integer; 4117 4118 case SO_PROTOCOL: 4119 optval = so->so_proto->pr_protocol; 4120 goto integer; 4121 4122 case SO_ERROR: 4123 SOCK_LOCK(so); 4124 if (so->so_error) { 4125 optval = so->so_error; 4126 so->so_error = 0; 4127 } else { 4128 optval = so->so_rerror; 4129 so->so_rerror = 0; 4130 } 4131 SOCK_UNLOCK(so); 4132 goto integer; 4133 4134 case SO_SNDBUF: 4135 SOCK_LOCK(so); 4136 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat : 4137 so->so_snd.sb_hiwat; 4138 SOCK_UNLOCK(so); 4139 goto integer; 4140 4141 case SO_RCVBUF: 4142 SOCK_LOCK(so); 4143 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat : 4144 so->so_rcv.sb_hiwat; 4145 SOCK_UNLOCK(so); 4146 goto integer; 4147 4148 case SO_SNDLOWAT: 4149 SOCK_LOCK(so); 4150 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat : 4151 so->so_snd.sb_lowat; 4152 SOCK_UNLOCK(so); 4153 goto integer; 4154 4155 case SO_RCVLOWAT: 4156 SOCK_LOCK(so); 4157 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat : 4158 so->so_rcv.sb_lowat; 4159 SOCK_UNLOCK(so); 4160 goto integer; 4161 4162 case SO_SNDTIMEO: 4163 case SO_RCVTIMEO: 4164 SOCK_LOCK(so); 4165 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ? 4166 (SOLISTENING(so) ? so->sol_sbsnd_timeo : 4167 so->so_snd.sb_timeo) : 4168 (SOLISTENING(so) ? so->sol_sbrcv_timeo : 4169 so->so_rcv.sb_timeo)); 4170 SOCK_UNLOCK(so); 4171 #ifdef COMPAT_FREEBSD32 4172 if (SV_CURPROC_FLAG(SV_ILP32)) { 4173 struct timeval32 tv32; 4174 4175 CP(tv, tv32, tv_sec); 4176 CP(tv, tv32, tv_usec); 4177 error = sooptcopyout(sopt, &tv32, sizeof tv32); 4178 } else 4179 #endif 4180 error = sooptcopyout(sopt, &tv, sizeof tv); 4181 break; 4182 4183 case SO_LABEL: 4184 #ifdef MAC 4185 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 4186 sizeof(extmac)); 4187 if (error) 4188 goto bad; 4189 error = mac_getsockopt_label(sopt->sopt_td->td_ucred, 4190 so, &extmac); 4191 if (error) 4192 goto bad; 4193 /* Don't copy out extmac, it is unchanged. */ 4194 #else 4195 error = EOPNOTSUPP; 4196 #endif 4197 break; 4198 4199 case SO_PEERLABEL: 4200 #ifdef MAC 4201 error = sooptcopyin(sopt, &extmac, sizeof(extmac), 4202 sizeof(extmac)); 4203 if (error) 4204 goto bad; 4205 error = mac_getsockopt_peerlabel( 4206 sopt->sopt_td->td_ucred, so, &extmac); 4207 if (error) 4208 goto bad; 4209 /* Don't copy out extmac, it is unchanged. */ 4210 #else 4211 error = EOPNOTSUPP; 4212 #endif 4213 break; 4214 4215 case SO_LISTENQLIMIT: 4216 SOCK_LOCK(so); 4217 optval = SOLISTENING(so) ? so->sol_qlimit : 0; 4218 SOCK_UNLOCK(so); 4219 goto integer; 4220 4221 case SO_LISTENQLEN: 4222 SOCK_LOCK(so); 4223 optval = SOLISTENING(so) ? so->sol_qlen : 0; 4224 SOCK_UNLOCK(so); 4225 goto integer; 4226 4227 case SO_LISTENINCQLEN: 4228 SOCK_LOCK(so); 4229 optval = SOLISTENING(so) ? so->sol_incqlen : 0; 4230 SOCK_UNLOCK(so); 4231 goto integer; 4232 4233 case SO_TS_CLOCK: 4234 optval = so->so_ts_clock; 4235 goto integer; 4236 4237 case SO_MAX_PACING_RATE: 4238 optval = so->so_max_pacing_rate; 4239 goto integer; 4240 4241 case SO_SPLICE: { 4242 off_t n; 4243 4244 /* 4245 * Acquire the I/O lock to serialize with 4246 * so_splice_xfer(). This is not required for 4247 * correctness, but makes testing simpler: once a byte 4248 * has been transmitted to the sink and observed (e.g., 4249 * by reading from the socket to which the sink is 4250 * connected), a subsequent getsockopt(SO_SPLICE) will 4251 * return an up-to-date value. 4252 */ 4253 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT); 4254 if (error != 0) 4255 goto bad; 4256 SOCK_LOCK(so); 4257 if (SOLISTENING(so)) { 4258 n = 0; 4259 } else { 4260 n = so->so_splice_sent; 4261 } 4262 SOCK_UNLOCK(so); 4263 SOCK_IO_RECV_UNLOCK(so); 4264 error = sooptcopyout(sopt, &n, sizeof(n)); 4265 break; 4266 } 4267 4268 default: 4269 #ifdef SOCKET_HHOOK 4270 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0) 4271 error = hhook_run_socket(so, sopt, 4272 HHOOK_SOCKET_OPT); 4273 else 4274 #endif 4275 error = ENOPROTOOPT; 4276 break; 4277 } 4278 } 4279 bad: 4280 CURVNET_RESTORE(); 4281 return (error); 4282 } 4283 4284 int 4285 soopt_getm(struct sockopt *sopt, struct mbuf **mp) 4286 { 4287 struct mbuf *m, *m_prev; 4288 int sopt_size = sopt->sopt_valsize; 4289 4290 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 4291 if (m == NULL) 4292 return ENOBUFS; 4293 if (sopt_size > MLEN) { 4294 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT); 4295 if ((m->m_flags & M_EXT) == 0) { 4296 m_free(m); 4297 return ENOBUFS; 4298 } 4299 m->m_len = min(MCLBYTES, sopt_size); 4300 } else { 4301 m->m_len = min(MLEN, sopt_size); 4302 } 4303 sopt_size -= m->m_len; 4304 *mp = m; 4305 m_prev = m; 4306 4307 while (sopt_size) { 4308 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA); 4309 if (m == NULL) { 4310 m_freem(*mp); 4311 return ENOBUFS; 4312 } 4313 if (sopt_size > MLEN) { 4314 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK : 4315 M_NOWAIT); 4316 if ((m->m_flags & M_EXT) == 0) { 4317 m_freem(m); 4318 m_freem(*mp); 4319 return ENOBUFS; 4320 } 4321 m->m_len = min(MCLBYTES, sopt_size); 4322 } else { 4323 m->m_len = min(MLEN, sopt_size); 4324 } 4325 sopt_size -= m->m_len; 4326 m_prev->m_next = m; 4327 m_prev = m; 4328 } 4329 return (0); 4330 } 4331 4332 int 4333 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m) 4334 { 4335 struct mbuf *m0 = m; 4336 4337 if (sopt->sopt_val == NULL) 4338 return (0); 4339 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 4340 if (sopt->sopt_td != NULL) { 4341 int error; 4342 4343 error = copyin(sopt->sopt_val, mtod(m, char *), 4344 m->m_len); 4345 if (error != 0) { 4346 m_freem(m0); 4347 return(error); 4348 } 4349 } else 4350 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len); 4351 sopt->sopt_valsize -= m->m_len; 4352 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 4353 m = m->m_next; 4354 } 4355 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */ 4356 panic("ip6_sooptmcopyin"); 4357 return (0); 4358 } 4359 4360 int 4361 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m) 4362 { 4363 struct mbuf *m0 = m; 4364 size_t valsize = 0; 4365 4366 if (sopt->sopt_val == NULL) 4367 return (0); 4368 while (m != NULL && sopt->sopt_valsize >= m->m_len) { 4369 if (sopt->sopt_td != NULL) { 4370 int error; 4371 4372 error = copyout(mtod(m, char *), sopt->sopt_val, 4373 m->m_len); 4374 if (error != 0) { 4375 m_freem(m0); 4376 return(error); 4377 } 4378 } else 4379 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len); 4380 sopt->sopt_valsize -= m->m_len; 4381 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len; 4382 valsize += m->m_len; 4383 m = m->m_next; 4384 } 4385 if (m != NULL) { 4386 /* enough soopt buffer should be given from user-land */ 4387 m_freem(m0); 4388 return(EINVAL); 4389 } 4390 sopt->sopt_valsize = valsize; 4391 return (0); 4392 } 4393 4394 /* 4395 * sohasoutofband(): protocol notifies socket layer of the arrival of new 4396 * out-of-band data, which will then notify socket consumers. 4397 */ 4398 void 4399 sohasoutofband(struct socket *so) 4400 { 4401 4402 if (so->so_sigio != NULL) 4403 pgsigio(&so->so_sigio, SIGURG, 0); 4404 selwakeuppri(&so->so_rdsel, PSOCK); 4405 } 4406 4407 int 4408 sopoll_generic(struct socket *so, int events, struct thread *td) 4409 { 4410 int revents; 4411 4412 SOCK_LOCK(so); 4413 if (SOLISTENING(so)) { 4414 if (!(events & (POLLIN | POLLRDNORM))) 4415 revents = 0; 4416 else if (!TAILQ_EMPTY(&so->sol_comp)) 4417 revents = events & (POLLIN | POLLRDNORM); 4418 else if ((events & POLLINIGNEOF) == 0 && so->so_error) 4419 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP; 4420 else { 4421 selrecord(td, &so->so_rdsel); 4422 revents = 0; 4423 } 4424 } else { 4425 revents = 0; 4426 SOCK_SENDBUF_LOCK(so); 4427 SOCK_RECVBUF_LOCK(so); 4428 if (events & (POLLIN | POLLRDNORM)) 4429 if (soreadabledata(so) && !isspliced(so)) 4430 revents |= events & (POLLIN | POLLRDNORM); 4431 if (events & (POLLOUT | POLLWRNORM)) 4432 if (sowriteable(so) && !issplicedback(so)) 4433 revents |= events & (POLLOUT | POLLWRNORM); 4434 if (events & (POLLPRI | POLLRDBAND)) 4435 if (so->so_oobmark || 4436 (so->so_rcv.sb_state & SBS_RCVATMARK)) 4437 revents |= events & (POLLPRI | POLLRDBAND); 4438 if ((events & POLLINIGNEOF) == 0) { 4439 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 4440 revents |= events & (POLLIN | POLLRDNORM); 4441 if (so->so_snd.sb_state & SBS_CANTSENDMORE) 4442 revents |= POLLHUP; 4443 } 4444 } 4445 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 4446 revents |= events & POLLRDHUP; 4447 if (revents == 0) { 4448 if (events & 4449 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) { 4450 selrecord(td, &so->so_rdsel); 4451 so->so_rcv.sb_flags |= SB_SEL; 4452 } 4453 if (events & (POLLOUT | POLLWRNORM)) { 4454 selrecord(td, &so->so_wrsel); 4455 so->so_snd.sb_flags |= SB_SEL; 4456 } 4457 } 4458 SOCK_RECVBUF_UNLOCK(so); 4459 SOCK_SENDBUF_UNLOCK(so); 4460 } 4461 SOCK_UNLOCK(so); 4462 return (revents); 4463 } 4464 4465 int 4466 soo_kqfilter(struct file *fp, struct knote *kn) 4467 { 4468 struct socket *so = kn->kn_fp->f_data; 4469 struct sockbuf *sb; 4470 sb_which which; 4471 struct knlist *knl; 4472 4473 switch (kn->kn_filter) { 4474 case EVFILT_READ: 4475 kn->kn_fop = &soread_filtops; 4476 knl = &so->so_rdsel.si_note; 4477 sb = &so->so_rcv; 4478 which = SO_RCV; 4479 break; 4480 case EVFILT_WRITE: 4481 kn->kn_fop = &sowrite_filtops; 4482 knl = &so->so_wrsel.si_note; 4483 sb = &so->so_snd; 4484 which = SO_SND; 4485 break; 4486 case EVFILT_EMPTY: 4487 kn->kn_fop = &soempty_filtops; 4488 knl = &so->so_wrsel.si_note; 4489 sb = &so->so_snd; 4490 which = SO_SND; 4491 break; 4492 default: 4493 return (EINVAL); 4494 } 4495 4496 SOCK_LOCK(so); 4497 if (SOLISTENING(so)) { 4498 knlist_add(knl, kn, 1); 4499 } else { 4500 SOCK_BUF_LOCK(so, which); 4501 knlist_add(knl, kn, 1); 4502 sb->sb_flags |= SB_KNOTE; 4503 SOCK_BUF_UNLOCK(so, which); 4504 } 4505 SOCK_UNLOCK(so); 4506 return (0); 4507 } 4508 4509 static void 4510 filt_sordetach(struct knote *kn) 4511 { 4512 struct socket *so = kn->kn_fp->f_data; 4513 4514 so_rdknl_lock(so); 4515 knlist_remove(&so->so_rdsel.si_note, kn, 1); 4516 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note)) 4517 so->so_rcv.sb_flags &= ~SB_KNOTE; 4518 so_rdknl_unlock(so); 4519 } 4520 4521 /*ARGSUSED*/ 4522 static int 4523 filt_soread(struct knote *kn, long hint) 4524 { 4525 struct socket *so; 4526 4527 so = kn->kn_fp->f_data; 4528 4529 if (SOLISTENING(so)) { 4530 SOCK_LOCK_ASSERT(so); 4531 kn->kn_data = so->sol_qlen; 4532 if (so->so_error) { 4533 kn->kn_flags |= EV_EOF; 4534 kn->kn_fflags = so->so_error; 4535 return (1); 4536 } 4537 return (!TAILQ_EMPTY(&so->sol_comp)); 4538 } 4539 4540 if ((so->so_rcv.sb_flags & SB_SPLICED) != 0) 4541 return (0); 4542 4543 SOCK_RECVBUF_LOCK_ASSERT(so); 4544 4545 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl; 4546 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 4547 kn->kn_flags |= EV_EOF; 4548 kn->kn_fflags = so->so_error; 4549 return (1); 4550 } else if (so->so_error || so->so_rerror) 4551 return (1); 4552 4553 if (kn->kn_sfflags & NOTE_LOWAT) { 4554 if (kn->kn_data >= kn->kn_sdata) 4555 return (1); 4556 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat) 4557 return (1); 4558 4559 #ifdef SOCKET_HHOOK 4560 /* This hook returning non-zero indicates an event, not error */ 4561 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD)); 4562 #else 4563 return (0); 4564 #endif 4565 } 4566 4567 static void 4568 filt_sowdetach(struct knote *kn) 4569 { 4570 struct socket *so = kn->kn_fp->f_data; 4571 4572 so_wrknl_lock(so); 4573 knlist_remove(&so->so_wrsel.si_note, kn, 1); 4574 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note)) 4575 so->so_snd.sb_flags &= ~SB_KNOTE; 4576 so_wrknl_unlock(so); 4577 } 4578 4579 /*ARGSUSED*/ 4580 static int 4581 filt_sowrite(struct knote *kn, long hint) 4582 { 4583 struct socket *so; 4584 4585 so = kn->kn_fp->f_data; 4586 4587 if (SOLISTENING(so)) 4588 return (0); 4589 4590 SOCK_SENDBUF_LOCK_ASSERT(so); 4591 kn->kn_data = sbspace(&so->so_snd); 4592 4593 #ifdef SOCKET_HHOOK 4594 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE); 4595 #endif 4596 4597 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 4598 kn->kn_flags |= EV_EOF; 4599 kn->kn_fflags = so->so_error; 4600 return (1); 4601 } else if (so->so_error) /* temporary udp error */ 4602 return (1); 4603 else if (((so->so_state & SS_ISCONNECTED) == 0) && 4604 (so->so_proto->pr_flags & PR_CONNREQUIRED)) 4605 return (0); 4606 else if (kn->kn_sfflags & NOTE_LOWAT) 4607 return (kn->kn_data >= kn->kn_sdata); 4608 else 4609 return (kn->kn_data >= so->so_snd.sb_lowat); 4610 } 4611 4612 static int 4613 filt_soempty(struct knote *kn, long hint) 4614 { 4615 struct socket *so; 4616 4617 so = kn->kn_fp->f_data; 4618 4619 if (SOLISTENING(so)) 4620 return (1); 4621 4622 SOCK_SENDBUF_LOCK_ASSERT(so); 4623 kn->kn_data = sbused(&so->so_snd); 4624 4625 if (kn->kn_data == 0) 4626 return (1); 4627 else 4628 return (0); 4629 } 4630 4631 int 4632 socheckuid(struct socket *so, uid_t uid) 4633 { 4634 4635 if (so == NULL) 4636 return (EPERM); 4637 if (so->so_cred->cr_uid != uid) 4638 return (EPERM); 4639 return (0); 4640 } 4641 4642 /* 4643 * These functions are used by protocols to notify the socket layer (and its 4644 * consumers) of state changes in the sockets driven by protocol-side events. 4645 */ 4646 4647 /* 4648 * Procedures to manipulate state flags of socket and do appropriate wakeups. 4649 * 4650 * Normal sequence from the active (originating) side is that 4651 * soisconnecting() is called during processing of connect() call, resulting 4652 * in an eventual call to soisconnected() if/when the connection is 4653 * established. When the connection is torn down soisdisconnecting() is 4654 * called during processing of disconnect() call, and soisdisconnected() is 4655 * called when the connection to the peer is totally severed. The semantics 4656 * of these routines are such that connectionless protocols can call 4657 * soisconnected() and soisdisconnected() only, bypassing the in-progress 4658 * calls when setting up a ``connection'' takes no time. 4659 * 4660 * From the passive side, a socket is created with two queues of sockets: 4661 * so_incomp for connections in progress and so_comp for connections already 4662 * made and awaiting user acceptance. As a protocol is preparing incoming 4663 * connections, it creates a socket structure queued on so_incomp by calling 4664 * sonewconn(). When the connection is established, soisconnected() is 4665 * called, and transfers the socket structure to so_comp, making it available 4666 * to accept(). 4667 * 4668 * If a socket is closed with sockets on either so_incomp or so_comp, these 4669 * sockets are dropped. 4670 * 4671 * If higher-level protocols are implemented in the kernel, the wakeups done 4672 * here will sometimes cause software-interrupt process scheduling. 4673 */ 4674 void 4675 soisconnecting(struct socket *so) 4676 { 4677 4678 SOCK_LOCK(so); 4679 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 4680 so->so_state |= SS_ISCONNECTING; 4681 SOCK_UNLOCK(so); 4682 } 4683 4684 void 4685 soisconnected(struct socket *so) 4686 { 4687 bool last __diagused; 4688 4689 SOCK_LOCK(so); 4690 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING); 4691 so->so_state |= SS_ISCONNECTED; 4692 4693 if (so->so_qstate == SQ_INCOMP) { 4694 struct socket *head = so->so_listen; 4695 int ret; 4696 4697 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so)); 4698 /* 4699 * Promoting a socket from incomplete queue to complete, we 4700 * need to go through reverse order of locking. We first do 4701 * trylock, and if that doesn't succeed, we go the hard way 4702 * leaving a reference and rechecking consistency after proper 4703 * locking. 4704 */ 4705 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) { 4706 soref(head); 4707 SOCK_UNLOCK(so); 4708 SOLISTEN_LOCK(head); 4709 SOCK_LOCK(so); 4710 if (__predict_false(head != so->so_listen)) { 4711 /* 4712 * The socket went off the listen queue, 4713 * should be lost race to close(2) of sol. 4714 * The socket is about to soabort(). 4715 */ 4716 SOCK_UNLOCK(so); 4717 sorele_locked(head); 4718 return; 4719 } 4720 last = refcount_release(&head->so_count); 4721 KASSERT(!last, ("%s: released last reference for %p", 4722 __func__, head)); 4723 } 4724 again: 4725 if ((so->so_options & SO_ACCEPTFILTER) == 0) { 4726 TAILQ_REMOVE(&head->sol_incomp, so, so_list); 4727 head->sol_incqlen--; 4728 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list); 4729 head->sol_qlen++; 4730 so->so_qstate = SQ_COMP; 4731 SOCK_UNLOCK(so); 4732 solisten_wakeup(head); /* unlocks */ 4733 } else { 4734 SOCK_RECVBUF_LOCK(so); 4735 soupcall_set(so, SO_RCV, 4736 head->sol_accept_filter->accf_callback, 4737 head->sol_accept_filter_arg); 4738 so->so_options &= ~SO_ACCEPTFILTER; 4739 ret = head->sol_accept_filter->accf_callback(so, 4740 head->sol_accept_filter_arg, M_NOWAIT); 4741 if (ret == SU_ISCONNECTED) { 4742 soupcall_clear(so, SO_RCV); 4743 SOCK_RECVBUF_UNLOCK(so); 4744 goto again; 4745 } 4746 SOCK_RECVBUF_UNLOCK(so); 4747 SOCK_UNLOCK(so); 4748 SOLISTEN_UNLOCK(head); 4749 } 4750 return; 4751 } 4752 SOCK_UNLOCK(so); 4753 wakeup(&so->so_timeo); 4754 sorwakeup(so); 4755 sowwakeup(so); 4756 } 4757 4758 void 4759 soisdisconnecting(struct socket *so) 4760 { 4761 4762 SOCK_LOCK(so); 4763 so->so_state &= ~SS_ISCONNECTING; 4764 so->so_state |= SS_ISDISCONNECTING; 4765 4766 if (!SOLISTENING(so)) { 4767 SOCK_RECVBUF_LOCK(so); 4768 socantrcvmore_locked(so); 4769 SOCK_SENDBUF_LOCK(so); 4770 socantsendmore_locked(so); 4771 } 4772 SOCK_UNLOCK(so); 4773 wakeup(&so->so_timeo); 4774 } 4775 4776 void 4777 soisdisconnected(struct socket *so) 4778 { 4779 4780 SOCK_LOCK(so); 4781 4782 /* 4783 * There is at least one reader of so_state that does not 4784 * acquire socket lock, namely soreceive_generic(). Ensure 4785 * that it never sees all flags that track connection status 4786 * cleared, by ordering the update with a barrier semantic of 4787 * our release thread fence. 4788 */ 4789 so->so_state |= SS_ISDISCONNECTED; 4790 atomic_thread_fence_rel(); 4791 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 4792 4793 if (!SOLISTENING(so)) { 4794 SOCK_UNLOCK(so); 4795 SOCK_RECVBUF_LOCK(so); 4796 socantrcvmore_locked(so); 4797 SOCK_SENDBUF_LOCK(so); 4798 sbdrop_locked(&so->so_snd, sbused(&so->so_snd)); 4799 socantsendmore_locked(so); 4800 } else 4801 SOCK_UNLOCK(so); 4802 wakeup(&so->so_timeo); 4803 } 4804 4805 int 4806 soiolock(struct socket *so, struct sx *sx, int flags) 4807 { 4808 int error; 4809 4810 KASSERT((flags & SBL_VALID) == flags, 4811 ("soiolock: invalid flags %#x", flags)); 4812 4813 if ((flags & SBL_WAIT) != 0) { 4814 if ((flags & SBL_NOINTR) != 0) { 4815 sx_xlock(sx); 4816 } else { 4817 error = sx_xlock_sig(sx); 4818 if (error != 0) 4819 return (error); 4820 } 4821 } else if (!sx_try_xlock(sx)) { 4822 return (EWOULDBLOCK); 4823 } 4824 4825 if (__predict_false(SOLISTENING(so))) { 4826 sx_xunlock(sx); 4827 return (ENOTCONN); 4828 } 4829 return (0); 4830 } 4831 4832 void 4833 soiounlock(struct sx *sx) 4834 { 4835 sx_xunlock(sx); 4836 } 4837 4838 /* 4839 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. 4840 */ 4841 struct sockaddr * 4842 sodupsockaddr(const struct sockaddr *sa, int mflags) 4843 { 4844 struct sockaddr *sa2; 4845 4846 sa2 = malloc(sa->sa_len, M_SONAME, mflags); 4847 if (sa2) 4848 bcopy(sa, sa2, sa->sa_len); 4849 return sa2; 4850 } 4851 4852 /* 4853 * Register per-socket destructor. 4854 */ 4855 void 4856 sodtor_set(struct socket *so, so_dtor_t *func) 4857 { 4858 4859 SOCK_LOCK_ASSERT(so); 4860 so->so_dtor = func; 4861 } 4862 4863 /* 4864 * Register per-socket buffer upcalls. 4865 */ 4866 void 4867 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg) 4868 { 4869 struct sockbuf *sb; 4870 4871 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4872 4873 switch (which) { 4874 case SO_RCV: 4875 sb = &so->so_rcv; 4876 break; 4877 case SO_SND: 4878 sb = &so->so_snd; 4879 break; 4880 } 4881 SOCK_BUF_LOCK_ASSERT(so, which); 4882 sb->sb_upcall = func; 4883 sb->sb_upcallarg = arg; 4884 sb->sb_flags |= SB_UPCALL; 4885 } 4886 4887 void 4888 soupcall_clear(struct socket *so, sb_which which) 4889 { 4890 struct sockbuf *sb; 4891 4892 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so)); 4893 4894 switch (which) { 4895 case SO_RCV: 4896 sb = &so->so_rcv; 4897 break; 4898 case SO_SND: 4899 sb = &so->so_snd; 4900 break; 4901 } 4902 SOCK_BUF_LOCK_ASSERT(so, which); 4903 KASSERT(sb->sb_upcall != NULL, 4904 ("%s: so %p no upcall to clear", __func__, so)); 4905 sb->sb_upcall = NULL; 4906 sb->sb_upcallarg = NULL; 4907 sb->sb_flags &= ~SB_UPCALL; 4908 } 4909 4910 void 4911 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg) 4912 { 4913 4914 SOLISTEN_LOCK_ASSERT(so); 4915 so->sol_upcall = func; 4916 so->sol_upcallarg = arg; 4917 } 4918 4919 static void 4920 so_rdknl_lock(void *arg) 4921 { 4922 struct socket *so = arg; 4923 4924 retry: 4925 if (SOLISTENING(so)) { 4926 SOLISTEN_LOCK(so); 4927 } else { 4928 SOCK_RECVBUF_LOCK(so); 4929 if (__predict_false(SOLISTENING(so))) { 4930 SOCK_RECVBUF_UNLOCK(so); 4931 goto retry; 4932 } 4933 } 4934 } 4935 4936 static void 4937 so_rdknl_unlock(void *arg) 4938 { 4939 struct socket *so = arg; 4940 4941 if (SOLISTENING(so)) 4942 SOLISTEN_UNLOCK(so); 4943 else 4944 SOCK_RECVBUF_UNLOCK(so); 4945 } 4946 4947 static void 4948 so_rdknl_assert_lock(void *arg, int what) 4949 { 4950 struct socket *so = arg; 4951 4952 if (what == LA_LOCKED) { 4953 if (SOLISTENING(so)) 4954 SOLISTEN_LOCK_ASSERT(so); 4955 else 4956 SOCK_RECVBUF_LOCK_ASSERT(so); 4957 } else { 4958 if (SOLISTENING(so)) 4959 SOLISTEN_UNLOCK_ASSERT(so); 4960 else 4961 SOCK_RECVBUF_UNLOCK_ASSERT(so); 4962 } 4963 } 4964 4965 static void 4966 so_wrknl_lock(void *arg) 4967 { 4968 struct socket *so = arg; 4969 4970 retry: 4971 if (SOLISTENING(so)) { 4972 SOLISTEN_LOCK(so); 4973 } else { 4974 SOCK_SENDBUF_LOCK(so); 4975 if (__predict_false(SOLISTENING(so))) { 4976 SOCK_SENDBUF_UNLOCK(so); 4977 goto retry; 4978 } 4979 } 4980 } 4981 4982 static void 4983 so_wrknl_unlock(void *arg) 4984 { 4985 struct socket *so = arg; 4986 4987 if (SOLISTENING(so)) 4988 SOLISTEN_UNLOCK(so); 4989 else 4990 SOCK_SENDBUF_UNLOCK(so); 4991 } 4992 4993 static void 4994 so_wrknl_assert_lock(void *arg, int what) 4995 { 4996 struct socket *so = arg; 4997 4998 if (what == LA_LOCKED) { 4999 if (SOLISTENING(so)) 5000 SOLISTEN_LOCK_ASSERT(so); 5001 else 5002 SOCK_SENDBUF_LOCK_ASSERT(so); 5003 } else { 5004 if (SOLISTENING(so)) 5005 SOLISTEN_UNLOCK_ASSERT(so); 5006 else 5007 SOCK_SENDBUF_UNLOCK_ASSERT(so); 5008 } 5009 } 5010 5011 /* 5012 * Create an external-format (``xsocket'') structure using the information in 5013 * the kernel-format socket structure pointed to by so. This is done to 5014 * reduce the spew of irrelevant information over this interface, to isolate 5015 * user code from changes in the kernel structure, and potentially to provide 5016 * information-hiding if we decide that some of this information should be 5017 * hidden from users. 5018 */ 5019 void 5020 sotoxsocket(struct socket *so, struct xsocket *xso) 5021 { 5022 5023 bzero(xso, sizeof(*xso)); 5024 xso->xso_len = sizeof *xso; 5025 xso->xso_so = (uintptr_t)so; 5026 xso->so_type = so->so_type; 5027 xso->so_options = so->so_options; 5028 xso->so_linger = so->so_linger; 5029 xso->so_state = so->so_state; 5030 xso->so_pcb = (uintptr_t)so->so_pcb; 5031 xso->xso_protocol = so->so_proto->pr_protocol; 5032 xso->xso_family = so->so_proto->pr_domain->dom_family; 5033 xso->so_timeo = so->so_timeo; 5034 xso->so_error = so->so_error; 5035 xso->so_uid = so->so_cred->cr_uid; 5036 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 5037 SOCK_LOCK(so); 5038 xso->so_fibnum = so->so_fibnum; 5039 if (SOLISTENING(so)) { 5040 xso->so_qlen = so->sol_qlen; 5041 xso->so_incqlen = so->sol_incqlen; 5042 xso->so_qlimit = so->sol_qlimit; 5043 xso->so_oobmark = 0; 5044 } else { 5045 xso->so_state |= so->so_qstate; 5046 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0; 5047 xso->so_oobmark = so->so_oobmark; 5048 sbtoxsockbuf(&so->so_snd, &xso->so_snd); 5049 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 5050 if ((so->so_rcv.sb_flags & SB_SPLICED) != 0) 5051 xso->so_splice_so = (uintptr_t)so->so_splice->dst; 5052 } 5053 SOCK_UNLOCK(so); 5054 } 5055 5056 int 5057 so_options_get(const struct socket *so) 5058 { 5059 5060 return (so->so_options); 5061 } 5062 5063 void 5064 so_options_set(struct socket *so, int val) 5065 { 5066 5067 so->so_options = val; 5068 } 5069 5070 int 5071 so_error_get(const struct socket *so) 5072 { 5073 5074 return (so->so_error); 5075 } 5076 5077 void 5078 so_error_set(struct socket *so, int val) 5079 { 5080 5081 so->so_error = val; 5082 } 5083