1 /* 2 * Copyright (c) 2005 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 1982, 1986, 1988, 1990, 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. Neither the name of the University nor the names of its contributors 15 * may be used to endorse or promote products derived from this software 16 * without specific prior written permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 31 * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $ 32 */ 33 34 #include "opt_param.h" 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/domain.h> 38 #include <sys/file.h> /* for maxfiles */ 39 #include <sys/kernel.h> 40 #include <sys/ktr.h> 41 #include <sys/proc.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/protosw.h> 45 #include <sys/resourcevar.h> 46 #include <sys/stat.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/socketops.h> 50 #include <sys/signalvar.h> 51 #include <sys/sysctl.h> 52 #include <sys/event.h> 53 54 #include <sys/msgport2.h> 55 #include <sys/socketvar2.h> 56 57 #include <net/netisr2.h> 58 59 #ifndef KTR_SOWAKEUP 60 #define KTR_SOWAKEUP KTR_ALL 61 #endif 62 KTR_INFO_MASTER(sowakeup); 63 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_start, 0, "newconn sorwakeup start"); 64 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_end, 1, "newconn sorwakeup end"); 65 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_wakeupstart, 2, "newconn wakeup start"); 66 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_wakeupend, 3, "newconn wakeup end"); 67 #define logsowakeup(name) KTR_LOG(sowakeup_ ## name) 68 69 int maxsockets; 70 71 /* 72 * Primitive routines for operating on sockets and socket buffers 73 */ 74 75 u_long sb_max = SB_MAX; 76 u_long sb_max_adj = 77 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 78 79 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 80 81 /************************************************************************ 82 * signalsockbuf procedures * 83 ************************************************************************/ 84 85 /* 86 * Wait for data to arrive at/drain from a socket buffer. 87 * 88 * NOTE: Caller must generally hold the ssb_lock (client side lock) since 89 * WAIT/WAKEUP only works for one client at a time. 90 * 91 * NOTE: Caller always retries whatever operation it was waiting on. 92 */ 93 int 94 ssb_wait(struct signalsockbuf *ssb) 95 { 96 uint32_t flags; 97 int pflags; 98 int error; 99 100 pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH; 101 102 for (;;) { 103 flags = ssb->ssb_flags; 104 cpu_ccfence(); 105 106 /* 107 * WAKEUP and WAIT interlock each other. We can catch the 108 * race by checking to see if WAKEUP has already been set, 109 * and only setting WAIT if WAKEUP is clear. 110 */ 111 if (flags & SSB_WAKEUP) { 112 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 113 flags & ~SSB_WAKEUP)) { 114 error = 0; 115 break; 116 } 117 continue; 118 } 119 120 /* 121 * Only set WAIT if WAKEUP is clear. 122 */ 123 tsleep_interlock(&ssb->ssb_cc, pflags); 124 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 125 flags | SSB_WAIT)) { 126 error = tsleep(&ssb->ssb_cc, pflags | PINTERLOCKED, 127 "sbwait", ssb->ssb_timeo); 128 break; 129 } 130 } 131 return (error); 132 } 133 134 /* 135 * Lock a sockbuf already known to be locked; 136 * return any error returned from sleep (EINTR). 137 */ 138 int 139 _ssb_lock(struct signalsockbuf *ssb) 140 { 141 uint32_t flags; 142 int pflags; 143 int error; 144 145 pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH; 146 147 for (;;) { 148 flags = ssb->ssb_flags; 149 cpu_ccfence(); 150 if (flags & SSB_LOCK) { 151 tsleep_interlock(&ssb->ssb_flags, pflags); 152 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 153 flags | SSB_WANT)) { 154 error = tsleep(&ssb->ssb_flags, 155 pflags | PINTERLOCKED, 156 "sblock", 0); 157 if (error) 158 break; 159 } 160 } else { 161 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 162 flags | SSB_LOCK)) { 163 lwkt_gettoken(&ssb->ssb_token); 164 error = 0; 165 break; 166 } 167 } 168 } 169 return (error); 170 } 171 172 /* 173 * This does the same for sockbufs. Note that the xsockbuf structure, 174 * since it is always embedded in a socket, does not include a self 175 * pointer nor a length. We make this entry point public in case 176 * some other mechanism needs it. 177 */ 178 void 179 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb) 180 { 181 xsb->sb_cc = ssb->ssb_cc; 182 xsb->sb_hiwat = ssb->ssb_hiwat; 183 xsb->sb_mbcnt = ssb->ssb_mbcnt; 184 xsb->sb_mbmax = ssb->ssb_mbmax; 185 xsb->sb_lowat = ssb->ssb_lowat; 186 xsb->sb_flags = ssb->ssb_flags; 187 xsb->sb_timeo = ssb->ssb_timeo; 188 } 189 190 191 /************************************************************************ 192 * Procedures which manipulate socket state flags, wakeups, etc. * 193 ************************************************************************ 194 * 195 * Normal sequence from the active (originating) side is that 196 * soisconnecting() is called during processing of connect() call, resulting 197 * in an eventual call to soisconnected() if/when the connection is 198 * established. When the connection is torn down soisdisconnecting() is 199 * called during processing of disconnect() call, and soisdisconnected() is 200 * called when the connection to the peer is totally severed. 201 * 202 * The semantics of these routines are such that connectionless protocols 203 * can call soisconnected() and soisdisconnected() only, bypassing the 204 * in-progress calls when setting up a ``connection'' takes no time. 205 * 206 * From the passive side, a socket is created with two queues of sockets: 207 * so_incomp for connections in progress and so_comp for connections 208 * already made and awaiting user acceptance. As a protocol is preparing 209 * incoming connections, it creates a socket structure queued on so_incomp 210 * by calling sonewconn(). When the connection is established, 211 * soisconnected() is called, and transfers the socket structure to so_comp, 212 * making it available to accept(). 213 * 214 * If a socket is closed with sockets on either so_incomp or so_comp, these 215 * sockets are dropped. 216 * 217 * If higher level protocols are implemented in the kernel, the wakeups 218 * done here will sometimes cause software-interrupt process scheduling. 219 */ 220 221 void 222 soisconnecting(struct socket *so) 223 { 224 soclrstate(so, SS_ISCONNECTED | SS_ISDISCONNECTING); 225 sosetstate(so, SS_ISCONNECTING); 226 } 227 228 void 229 soisconnected(struct socket *so) 230 { 231 struct socket *head; 232 233 while ((head = so->so_head) != NULL) { 234 lwkt_getpooltoken(head); 235 if (so->so_head == head) 236 break; 237 lwkt_relpooltoken(head); 238 } 239 240 soclrstate(so, SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING); 241 sosetstate(so, SS_ISCONNECTED); 242 if (head && (so->so_state & SS_INCOMP)) { 243 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 244 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 245 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 246 atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 247 so->so_options &= ~SO_ACCEPTFILTER; 248 so->so_upcall(so, so->so_upcallarg, 0); 249 lwkt_relpooltoken(head); 250 return; 251 } 252 253 /* 254 * Listen socket are not per-cpu. 255 */ 256 KKASSERT((so->so_state & (SS_COMP | SS_INCOMP)) == SS_INCOMP); 257 TAILQ_REMOVE(&head->so_incomp, so, so_list); 258 head->so_incqlen--; 259 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 260 head->so_qlen++; 261 sosetstate(so, SS_COMP); 262 soclrstate(so, SS_INCOMP); 263 264 /* 265 * XXX head may be on a different protocol thread. 266 * sorwakeup()->sowakeup() is hacked atm. 267 */ 268 sorwakeup(head); 269 wakeup_one(&head->so_timeo); 270 } else { 271 wakeup(&so->so_timeo); 272 sorwakeup(so); 273 sowwakeup(so); 274 } 275 if (head) 276 lwkt_relpooltoken(head); 277 } 278 279 void 280 soisdisconnecting(struct socket *so) 281 { 282 soclrstate(so, SS_ISCONNECTING); 283 sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE); 284 wakeup((caddr_t)&so->so_timeo); 285 sowwakeup(so); 286 sorwakeup(so); 287 } 288 289 void 290 soisdisconnected(struct socket *so) 291 { 292 soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING); 293 sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED); 294 wakeup((caddr_t)&so->so_timeo); 295 sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc); 296 sowwakeup(so); 297 sorwakeup(so); 298 } 299 300 void 301 soisreconnecting(struct socket *so) 302 { 303 soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED | 304 SS_CANTRCVMORE | SS_CANTSENDMORE); 305 sosetstate(so, SS_ISCONNECTING); 306 } 307 308 void 309 soisreconnected(struct socket *so) 310 { 311 soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE); 312 soisconnected(so); 313 } 314 315 /* 316 * Set or change the message port a socket receives commands on. 317 * 318 * XXX 319 */ 320 void 321 sosetport(struct socket *so, lwkt_port_t port) 322 { 323 so->so_port = port; 324 } 325 326 /* 327 * When an attempt at a new connection is noted on a socket 328 * which accepts connections, sonewconn is called. If the 329 * connection is possible (subject to space constraints, etc.) 330 * then we allocate a new structure, propoerly linked into the 331 * data structure of the original socket, and return this. 332 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 333 * 334 * The new socket is returned with one ref and so_pcb assigned. 335 * The reference is implied by so_pcb. 336 */ 337 struct socket * 338 sonewconn_faddr(struct socket *head, int connstatus, 339 const struct sockaddr *faddr, boolean_t keep_ref) 340 { 341 struct socket *so; 342 struct socket *sp; 343 struct pru_attach_info ai; 344 345 if (head->so_qlen > 3 * head->so_qlimit / 2) 346 return (NULL); 347 so = soalloc(1, head->so_proto); 348 if (so == NULL) 349 return (NULL); 350 351 /* 352 * Set the port prior to attaching the inpcb to the current 353 * cpu's protocol thread (which should be the current thread 354 * but might not be in all cases). This serializes any pcb ops 355 * which occur to our cpu allowing us to complete the attachment 356 * without racing anything. 357 */ 358 if (head->so_proto->pr_flags & PR_SYNC_PORT) 359 sosetport(so, &netisr_sync_port); 360 else 361 sosetport(so, netisr_cpuport(mycpuid)); 362 if ((head->so_options & SO_ACCEPTFILTER) != 0) 363 connstatus = 0; 364 so->so_head = head; 365 so->so_type = head->so_type; 366 so->so_options = head->so_options &~ SO_ACCEPTCONN; 367 so->so_linger = head->so_linger; 368 369 /* 370 * NOTE: Clearing NOFDREF implies referencing the so with 371 * soreference(). 372 */ 373 so->so_state = head->so_state | SS_NOFDREF | SS_ASSERTINPROG; 374 so->so_cred = crhold(head->so_cred); 375 ai.sb_rlimit = NULL; 376 ai.p_ucred = NULL; 377 ai.fd_rdir = NULL; /* jail code cruft XXX JH */ 378 379 /* 380 * Reserve space and call pru_attach. We can direct-call the 381 * function since we're already in the protocol thread. 382 */ 383 if (soreserve(so, head->so_snd.ssb_hiwat, 384 head->so_rcv.ssb_hiwat, NULL) || 385 so_pru_attach_direct(so, 0, &ai)) { 386 so->so_head = NULL; 387 soclrstate(so, SS_ASSERTINPROG); 388 sofree(so); /* remove implied pcb ref */ 389 return (NULL); 390 } 391 KKASSERT(((so->so_proto->pr_flags & PR_ASYNC_RCVD) == 0 && 392 so->so_refs == 2) || /* attach + our base ref */ 393 ((so->so_proto->pr_flags & PR_ASYNC_RCVD) && 394 so->so_refs == 3)); /* + async rcvd ref */ 395 if (keep_ref) { 396 /* 397 * Keep the reference; caller will free it. 398 */ 399 } else { 400 sofree(so); 401 } 402 KKASSERT(so->so_port != NULL); 403 so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat; 404 so->so_snd.ssb_lowat = head->so_snd.ssb_lowat; 405 so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo; 406 so->so_snd.ssb_timeo = head->so_snd.ssb_timeo; 407 408 if (head->so_rcv.ssb_flags & SSB_AUTOLOWAT) 409 so->so_rcv.ssb_flags |= SSB_AUTOLOWAT; 410 else 411 so->so_rcv.ssb_flags &= ~SSB_AUTOLOWAT; 412 413 if (head->so_snd.ssb_flags & SSB_AUTOLOWAT) 414 so->so_snd.ssb_flags |= SSB_AUTOLOWAT; 415 else 416 so->so_snd.ssb_flags &= ~SSB_AUTOLOWAT; 417 418 if (head->so_rcv.ssb_flags & SSB_AUTOSIZE) 419 so->so_rcv.ssb_flags |= SSB_AUTOSIZE; 420 else 421 so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE; 422 423 if (head->so_snd.ssb_flags & SSB_AUTOSIZE) 424 so->so_snd.ssb_flags |= SSB_AUTOSIZE; 425 else 426 so->so_snd.ssb_flags &= ~SSB_AUTOSIZE; 427 428 /* 429 * Save the faddr, if the information is provided and 430 * the protocol can perform the saving opertation. 431 */ 432 if (faddr != NULL && so->so_proto->pr_usrreqs->pru_savefaddr != NULL) 433 so->so_proto->pr_usrreqs->pru_savefaddr(so, faddr); 434 435 lwkt_getpooltoken(head); 436 if (connstatus) { 437 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0); 438 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 439 head->so_qlen++; 440 /* 441 * Set connstatus within head token, so that the accepted 442 * socket will have connstatus (SS_ISCONNECTED) set. 443 */ 444 sosetstate(so, SS_COMP | connstatus); 445 } else { 446 if (head->so_incqlen > head->so_qlimit) { 447 sp = TAILQ_FIRST(&head->so_incomp); 448 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == 449 SS_INCOMP); 450 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 451 head->so_incqlen--; 452 soclrstate(sp, SS_INCOMP); 453 soabort_async(sp, TRUE); 454 } 455 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0); 456 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 457 head->so_incqlen++; 458 sosetstate(so, SS_INCOMP); 459 } 460 /* 461 * Clear SS_ASSERTINPROG within head token, so that it will not 462 * race against accept-close or abort for "synchronous" sockets, 463 * e.g. unix socket, on other CPUs. 464 */ 465 soclrstate(so, SS_ASSERTINPROG); 466 lwkt_relpooltoken(head); 467 468 if (connstatus) { 469 /* 470 * XXX head may be on a different protocol thread. 471 * sorwakeup()->sowakeup() is hacked atm. 472 */ 473 logsowakeup(nconn_start); 474 sorwakeup(head); 475 logsowakeup(nconn_end); 476 477 logsowakeup(nconn_wakeupstart); 478 wakeup((caddr_t)&head->so_timeo); 479 logsowakeup(nconn_wakeupend); 480 } 481 return (so); 482 } 483 484 struct socket * 485 sonewconn(struct socket *head, int connstatus) 486 { 487 return sonewconn_faddr(head, connstatus, NULL, FALSE /* don't ref */); 488 } 489 490 /* 491 * Socantsendmore indicates that no more data will be sent on the 492 * socket; it would normally be applied to a socket when the user 493 * informs the system that no more data is to be sent, by the protocol 494 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 495 * will be received, and will normally be applied to the socket by a 496 * protocol when it detects that the peer will send no more data. 497 * Data queued for reading in the socket may yet be read. 498 */ 499 void 500 socantsendmore(struct socket *so) 501 { 502 sosetstate(so, SS_CANTSENDMORE); 503 sowwakeup(so); 504 } 505 506 void 507 socantrcvmore(struct socket *so) 508 { 509 sosetstate(so, SS_CANTRCVMORE); 510 sorwakeup(so); 511 } 512 513 /* 514 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 515 * via SIGIO if the socket has the SS_ASYNC flag set. 516 * 517 * For users waiting on send/recv try to avoid unnecessary context switch 518 * thrashing. Particularly for senders of large buffers (needs to be 519 * extended to sel and aio? XXX) 520 * 521 * WARNING! Can be called on a foreign socket from the wrong protocol 522 * thread. aka is called on the 'head' listen socket when 523 * a new connection comes in. 524 */ 525 526 void 527 sowakeup(struct socket *so, struct signalsockbuf *ssb) 528 { 529 uint32_t flags; 530 531 /* 532 * Atomically check the flags. When no special features are being 533 * used, WAIT is clear, and WAKEUP is already set, we can simply 534 * return. The upcoming synchronous waiter will not block. 535 */ 536 flags = atomic_fetchadd_int(&ssb->ssb_flags, 0); 537 if ((flags & SSB_NOTIFY_MASK) == 0) { 538 if (flags & SSB_WAKEUP) 539 return; 540 } 541 542 /* 543 * Check conditions, set the WAKEUP flag, and clear and signal if 544 * the WAIT flag is found to be set. This interlocks against the 545 * client side. 546 */ 547 for (;;) { 548 long space; 549 550 flags = ssb->ssb_flags; 551 cpu_ccfence(); 552 if (ssb->ssb_flags & SSB_PREALLOC) 553 space = ssb_space_prealloc(ssb); 554 else 555 space = ssb_space(ssb); 556 557 if ((ssb == &so->so_snd && space >= ssb->ssb_lowat) || 558 (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) || 559 (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) || 560 (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE)) 561 ) { 562 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 563 (flags | SSB_WAKEUP) & ~SSB_WAIT)) { 564 if (flags & SSB_WAIT) 565 wakeup(&ssb->ssb_cc); 566 break; 567 } 568 } else { 569 break; 570 } 571 } 572 573 /* 574 * Misc other events 575 */ 576 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 577 pgsigio(so->so_sigio, SIGIO, 0); 578 if (ssb->ssb_flags & SSB_UPCALL) 579 (*so->so_upcall)(so, so->so_upcallarg, M_NOWAIT); 580 KNOTE(&ssb->ssb_kq.ki_note, 0); 581 582 /* 583 * This is a bit of a hack. Multiple threads can wind up scanning 584 * ssb_mlist concurrently due to the fact that this function can be 585 * called on a foreign socket, so we can't afford to block here. 586 * 587 * We need the pool token for (so) (likely the listne socket if 588 * SSB_MEVENT is set) because the predicate function may have 589 * to access the accept queue. 590 */ 591 if (ssb->ssb_flags & SSB_MEVENT) { 592 struct netmsg_so_notify *msg, *nmsg; 593 594 lwkt_getpooltoken(so); 595 TAILQ_FOREACH_MUTABLE(msg, &ssb->ssb_mlist, nm_list, nmsg) { 596 if (msg->nm_predicate(msg)) { 597 TAILQ_REMOVE(&ssb->ssb_mlist, msg, nm_list); 598 lwkt_replymsg(&msg->base.lmsg, 599 msg->base.lmsg.ms_error); 600 } 601 } 602 if (TAILQ_EMPTY(&ssb->ssb_mlist)) 603 atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT); 604 lwkt_relpooltoken(so); 605 } 606 } 607 608 /* 609 * Socket buffer (struct signalsockbuf) utility routines. 610 * 611 * Each socket contains two socket buffers: one for sending data and 612 * one for receiving data. Each buffer contains a queue of mbufs, 613 * information about the number of mbufs and amount of data in the 614 * queue, and other fields allowing kevent()/select()/poll() statements 615 * and notification on data availability to be implemented. 616 * 617 * Data stored in a socket buffer is maintained as a list of records. 618 * Each record is a list of mbufs chained together with the m_next 619 * field. Records are chained together with the m_nextpkt field. The upper 620 * level routine soreceive() expects the following conventions to be 621 * observed when placing information in the receive buffer: 622 * 623 * 1. If the protocol requires each message be preceded by the sender's 624 * name, then a record containing that name must be present before 625 * any associated data (mbuf's must be of type MT_SONAME). 626 * 2. If the protocol supports the exchange of ``access rights'' (really 627 * just additional data associated with the message), and there are 628 * ``rights'' to be received, then a record containing this data 629 * should be present (mbuf's must be of type MT_RIGHTS). 630 * 3. If a name or rights record exists, then it must be followed by 631 * a data record, perhaps of zero length. 632 * 633 * Before using a new socket structure it is first necessary to reserve 634 * buffer space to the socket, by calling sbreserve(). This should commit 635 * some of the available buffer space in the system buffer pool for the 636 * socket (currently, it does nothing but enforce limits). The space 637 * should be released by calling ssb_release() when the socket is destroyed. 638 */ 639 int 640 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) 641 { 642 if (so->so_snd.ssb_lowat == 0) 643 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT); 644 if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0) 645 goto bad; 646 if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0) 647 goto bad2; 648 if (so->so_rcv.ssb_lowat == 0) 649 so->so_rcv.ssb_lowat = 1; 650 if (so->so_snd.ssb_lowat == 0) 651 so->so_snd.ssb_lowat = MCLBYTES; 652 if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat) 653 so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat; 654 return (0); 655 bad2: 656 ssb_release(&so->so_snd, so); 657 bad: 658 return (ENOBUFS); 659 } 660 661 static int 662 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 663 { 664 int error = 0; 665 u_long old_sb_max = sb_max; 666 667 error = SYSCTL_OUT(req, arg1, sizeof(int)); 668 if (error || !req->newptr) 669 return (error); 670 error = SYSCTL_IN(req, arg1, sizeof(int)); 671 if (error) 672 return (error); 673 if (sb_max < MSIZE + MCLBYTES) { 674 sb_max = old_sb_max; 675 return (EINVAL); 676 } 677 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 678 return (0); 679 } 680 681 /* 682 * Allot mbufs to a signalsockbuf. 683 * 684 * Attempt to scale mbmax so that mbcnt doesn't become limiting 685 * if buffering efficiency is near the normal case. 686 * 687 * sb_max only applies to user-sockets (where rl != NULL). It does 688 * not apply to kernel sockets or kernel-controlled sockets. Note 689 * that NFS overrides the sockbuf limits created when nfsd creates 690 * a socket. 691 */ 692 int 693 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so, 694 struct rlimit *rl) 695 { 696 /* 697 * rl will only be NULL when we're in an interrupt (eg, in tcp_input) 698 * or when called from netgraph (ie, ngd_attach) 699 */ 700 if (rl && cc > sb_max_adj) 701 cc = sb_max_adj; 702 if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc, 703 rl ? rl->rlim_cur : RLIM_INFINITY)) { 704 return (0); 705 } 706 if (rl) 707 ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max); 708 else 709 ssb->ssb_mbmax = cc * sb_efficiency; 710 711 /* 712 * AUTOLOWAT is set on send buffers and prevents large writes 713 * from generating a huge number of context switches. 714 */ 715 if (ssb->ssb_flags & SSB_AUTOLOWAT) { 716 ssb->ssb_lowat = ssb->ssb_hiwat / 4; 717 if (ssb->ssb_lowat < MCLBYTES) 718 ssb->ssb_lowat = MCLBYTES; 719 } 720 if (ssb->ssb_lowat > ssb->ssb_hiwat) 721 ssb->ssb_lowat = ssb->ssb_hiwat; 722 return (1); 723 } 724 725 /* 726 * Free mbufs held by a socket, and reserved mbuf space. 727 */ 728 void 729 ssb_release(struct signalsockbuf *ssb, struct socket *so) 730 { 731 sbflush(&ssb->sb); 732 (void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0, 733 RLIM_INFINITY); 734 ssb->ssb_mbmax = 0; 735 } 736 737 /* 738 * Some routines that return EOPNOTSUPP for entry points that are not 739 * supported by a protocol. Fill in as needed. 740 */ 741 void 742 pr_generic_notsupp(netmsg_t msg) 743 { 744 lwkt_replymsg(&msg->lmsg, EOPNOTSUPP); 745 } 746 747 int 748 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 749 struct mbuf *top, struct mbuf *control, int flags, 750 struct thread *td) 751 { 752 if (top) 753 m_freem(top); 754 if (control) 755 m_freem(control); 756 return (EOPNOTSUPP); 757 } 758 759 int 760 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 761 struct uio *uio, struct sockbuf *sio, 762 struct mbuf **controlp, int *flagsp) 763 { 764 return (EOPNOTSUPP); 765 } 766 767 /* 768 * This isn't really a ``null'' operation, but it's the default one 769 * and doesn't do anything destructive. 770 */ 771 void 772 pru_sense_null(netmsg_t msg) 773 { 774 msg->sense.nm_stat->st_blksize = msg->base.nm_so->so_snd.ssb_hiwat; 775 lwkt_replymsg(&msg->lmsg, 0); 776 } 777 778 /* 779 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers 780 * of this routine assume that it always succeeds, so we have to use a 781 * blockable allocation even though we might be called from a critical thread. 782 */ 783 struct sockaddr * 784 dup_sockaddr(const struct sockaddr *sa) 785 { 786 struct sockaddr *sa2; 787 788 sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT); 789 bcopy(sa, sa2, sa->sa_len); 790 return (sa2); 791 } 792 793 /* 794 * Create an external-format (``xsocket'') structure using the information 795 * in the kernel-format socket structure pointed to by so. This is done 796 * to reduce the spew of irrelevant information over this interface, 797 * to isolate user code from changes in the kernel structure, and 798 * potentially to provide information-hiding if we decide that 799 * some of this information should be hidden from users. 800 */ 801 void 802 sotoxsocket(struct socket *so, struct xsocket *xso) 803 { 804 xso->xso_len = sizeof *xso; 805 xso->xso_so = so; 806 xso->so_type = so->so_type; 807 xso->so_options = so->so_options; 808 xso->so_linger = so->so_linger; 809 xso->so_state = so->so_state; 810 xso->so_pcb = so->so_pcb; 811 xso->xso_protocol = so->so_proto->pr_protocol; 812 xso->xso_family = so->so_proto->pr_domain->dom_family; 813 xso->so_qlen = so->so_qlen; 814 xso->so_incqlen = so->so_incqlen; 815 xso->so_qlimit = so->so_qlimit; 816 xso->so_timeo = so->so_timeo; 817 xso->so_error = so->so_error; 818 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 819 xso->so_oobmark = so->so_oobmark; 820 ssbtoxsockbuf(&so->so_snd, &xso->so_snd); 821 ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 822 xso->so_uid = so->so_cred->cr_uid; 823 } 824 825 /* 826 * Here is the definition of some of the basic objects in the kern.ipc 827 * branch of the MIB. 828 */ 829 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 830 831 /* 832 * This takes the place of kern.maxsockbuf, which moved to kern.ipc. 833 * 834 * NOTE! sb_max only applies to user-created socket buffers. 835 */ 836 static int dummy; 837 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 838 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 839 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 840 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 841 &maxsockets, 0, "Maximum number of sockets available"); 842 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 843 &sb_efficiency, 0, 844 "Socket buffer limit scaler"); 845 846 /* 847 * Initialize maxsockets 848 */ 849 static void 850 init_maxsockets(void *ignored) 851 { 852 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 853 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 854 } 855 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY, 856 init_maxsockets, NULL); 857 858