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/thread2.h> 55 #include <sys/msgport2.h> 56 #include <sys/socketvar2.h> 57 58 #include <net/netisr2.h> 59 60 #ifndef KTR_SOWAKEUP 61 #define KTR_SOWAKEUP KTR_ALL 62 #endif 63 KTR_INFO_MASTER(sowakeup); 64 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_start, 0, "newconn sorwakeup start"); 65 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_end, 1, "newconn sorwakeup end"); 66 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_wakeupstart, 2, "newconn wakeup start"); 67 KTR_INFO(KTR_SOWAKEUP, sowakeup, nconn_wakeupend, 3, "newconn wakeup end"); 68 #define logsowakeup(name) KTR_LOG(sowakeup_ ## name) 69 70 int maxsockets; 71 72 /* 73 * Primitive routines for operating on sockets and socket buffers 74 */ 75 76 u_long sb_max = SB_MAX; 77 u_long sb_max_adj = 78 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 79 80 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 81 82 /************************************************************************ 83 * signalsockbuf procedures * 84 ************************************************************************/ 85 86 /* 87 * Wait for data to arrive at/drain from a socket buffer. 88 * 89 * NOTE: Caller must generally hold the ssb_lock (client side lock) since 90 * WAIT/WAKEUP only works for one client at a time. 91 * 92 * NOTE: Caller always retries whatever operation it was waiting on. 93 */ 94 int 95 ssb_wait(struct signalsockbuf *ssb) 96 { 97 uint32_t flags; 98 int pflags; 99 int error; 100 101 pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH; 102 103 for (;;) { 104 flags = ssb->ssb_flags; 105 cpu_ccfence(); 106 107 /* 108 * WAKEUP and WAIT interlock each other. We can catch the 109 * race by checking to see if WAKEUP has already been set, 110 * and only setting WAIT if WAKEUP is clear. 111 */ 112 if (flags & SSB_WAKEUP) { 113 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 114 flags & ~SSB_WAKEUP)) { 115 error = 0; 116 break; 117 } 118 continue; 119 } 120 121 /* 122 * Only set WAIT if WAKEUP is clear. 123 */ 124 tsleep_interlock(&ssb->ssb_cc, pflags); 125 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 126 flags | SSB_WAIT)) { 127 error = tsleep(&ssb->ssb_cc, pflags | PINTERLOCKED, 128 "sbwait", ssb->ssb_timeo); 129 break; 130 } 131 } 132 return (error); 133 } 134 135 /* 136 * Lock a sockbuf already known to be locked; 137 * return any error returned from sleep (EINTR). 138 */ 139 int 140 _ssb_lock(struct signalsockbuf *ssb) 141 { 142 uint32_t flags; 143 int pflags; 144 int error; 145 146 pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH; 147 148 for (;;) { 149 flags = ssb->ssb_flags; 150 cpu_ccfence(); 151 if (flags & SSB_LOCK) { 152 tsleep_interlock(&ssb->ssb_flags, pflags); 153 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 154 flags | SSB_WANT)) { 155 error = tsleep(&ssb->ssb_flags, 156 pflags | PINTERLOCKED, 157 "sblock", 0); 158 if (error) 159 break; 160 } 161 } else { 162 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 163 flags | SSB_LOCK)) { 164 lwkt_gettoken(&ssb->ssb_token); 165 error = 0; 166 break; 167 } 168 } 169 } 170 return (error); 171 } 172 173 /* 174 * This does the same for sockbufs. Note that the xsockbuf structure, 175 * since it is always embedded in a socket, does not include a self 176 * pointer nor a length. We make this entry point public in case 177 * some other mechanism needs it. 178 */ 179 void 180 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb) 181 { 182 xsb->sb_cc = ssb->ssb_cc; 183 xsb->sb_hiwat = ssb->ssb_hiwat; 184 xsb->sb_mbcnt = ssb->ssb_mbcnt; 185 xsb->sb_mbmax = ssb->ssb_mbmax; 186 xsb->sb_lowat = ssb->ssb_lowat; 187 xsb->sb_flags = ssb->ssb_flags; 188 xsb->sb_timeo = ssb->ssb_timeo; 189 } 190 191 192 /************************************************************************ 193 * Procedures which manipulate socket state flags, wakeups, etc. * 194 ************************************************************************ 195 * 196 * Normal sequence from the active (originating) side is that 197 * soisconnecting() is called during processing of connect() call, resulting 198 * in an eventual call to soisconnected() if/when the connection is 199 * established. When the connection is torn down soisdisconnecting() is 200 * called during processing of disconnect() call, and soisdisconnected() is 201 * called when the connection to the peer is totally severed. 202 * 203 * The semantics of these routines are such that connectionless protocols 204 * can call soisconnected() and soisdisconnected() only, bypassing the 205 * in-progress calls when setting up a ``connection'' takes no time. 206 * 207 * From the passive side, a socket is created with two queues of sockets: 208 * so_incomp for connections in progress and so_comp for connections 209 * already made and awaiting user acceptance. As a protocol is preparing 210 * incoming connections, it creates a socket structure queued on so_incomp 211 * by calling sonewconn(). When the connection is established, 212 * soisconnected() is called, and transfers the socket structure to so_comp, 213 * making it available to accept(). 214 * 215 * If a socket is closed with sockets on either so_incomp or so_comp, these 216 * sockets are dropped. 217 * 218 * If higher level protocols are implemented in the kernel, the wakeups 219 * done here will sometimes cause software-interrupt process scheduling. 220 */ 221 222 void 223 soisconnecting(struct socket *so) 224 { 225 soclrstate(so, SS_ISCONNECTED | SS_ISDISCONNECTING); 226 sosetstate(so, SS_ISCONNECTING); 227 } 228 229 void 230 soisconnected(struct socket *so) 231 { 232 struct socket *head; 233 234 while ((head = so->so_head) != NULL) { 235 lwkt_getpooltoken(head); 236 if (so->so_head == head) 237 break; 238 lwkt_relpooltoken(head); 239 } 240 241 soclrstate(so, SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING); 242 sosetstate(so, SS_ISCONNECTED); 243 if (head && (so->so_state & SS_INCOMP)) { 244 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 245 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 246 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 247 atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 248 so->so_options &= ~SO_ACCEPTFILTER; 249 so->so_upcall(so, so->so_upcallarg, 0); 250 lwkt_relpooltoken(head); 251 return; 252 } 253 254 /* 255 * Listen socket are not per-cpu. 256 */ 257 KKASSERT((so->so_state & (SS_COMP | SS_INCOMP)) == SS_INCOMP); 258 TAILQ_REMOVE(&head->so_incomp, so, so_list); 259 head->so_incqlen--; 260 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 261 head->so_qlen++; 262 sosetstate(so, SS_COMP); 263 soclrstate(so, SS_INCOMP); 264 265 /* 266 * XXX head may be on a different protocol thread. 267 * sorwakeup()->sowakeup() is hacked atm. 268 */ 269 sorwakeup(head); 270 wakeup_one(&head->so_timeo); 271 } else { 272 wakeup(&so->so_timeo); 273 sorwakeup(so); 274 sowwakeup(so); 275 } 276 if (head) 277 lwkt_relpooltoken(head); 278 } 279 280 void 281 soisdisconnecting(struct socket *so) 282 { 283 soclrstate(so, SS_ISCONNECTING); 284 sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE); 285 wakeup((caddr_t)&so->so_timeo); 286 sowwakeup(so); 287 sorwakeup(so); 288 } 289 290 void 291 soisdisconnected(struct socket *so) 292 { 293 soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING); 294 sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED); 295 wakeup((caddr_t)&so->so_timeo); 296 sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc); 297 sowwakeup(so); 298 sorwakeup(so); 299 } 300 301 void 302 soisreconnecting(struct socket *so) 303 { 304 soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED | 305 SS_CANTRCVMORE | SS_CANTSENDMORE); 306 sosetstate(so, SS_ISCONNECTING); 307 } 308 309 void 310 soisreconnected(struct socket *so) 311 { 312 soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE); 313 soisconnected(so); 314 } 315 316 /* 317 * Set or change the message port a socket receives commands on. 318 * 319 * XXX 320 */ 321 void 322 sosetport(struct socket *so, lwkt_port_t port) 323 { 324 so->so_port = port; 325 } 326 327 /* 328 * When an attempt at a new connection is noted on a socket 329 * which accepts connections, sonewconn is called. If the 330 * connection is possible (subject to space constraints, etc.) 331 * then we allocate a new structure, propoerly linked into the 332 * data structure of the original socket, and return this. 333 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 334 * 335 * The new socket is returned with one ref and so_pcb assigned. 336 * The reference is implied by so_pcb. 337 */ 338 struct socket * 339 sonewconn_faddr(struct socket *head, int connstatus, 340 const struct sockaddr *faddr, boolean_t keep_ref) 341 { 342 struct socket *so; 343 struct socket *sp; 344 struct pru_attach_info ai; 345 346 if (head->so_qlen > 3 * head->so_qlimit / 2) 347 return (NULL); 348 so = soalloc(1, head->so_proto); 349 if (so == NULL) 350 return (NULL); 351 352 /* 353 * Set the port prior to attaching the inpcb to the current 354 * cpu's protocol thread (which should be the current thread 355 * but might not be in all cases). This serializes any pcb ops 356 * which occur to our cpu allowing us to complete the attachment 357 * without racing anything. 358 */ 359 if (head->so_proto->pr_flags & PR_SYNC_PORT) 360 sosetport(so, &netisr_sync_port); 361 else 362 sosetport(so, netisr_cpuport(mycpuid)); 363 if ((head->so_options & SO_ACCEPTFILTER) != 0) 364 connstatus = 0; 365 so->so_head = head; 366 so->so_type = head->so_type; 367 so->so_options = head->so_options &~ SO_ACCEPTCONN; 368 so->so_linger = head->so_linger; 369 370 /* 371 * NOTE: Clearing NOFDREF implies referencing the so with 372 * soreference(). 373 */ 374 so->so_state = head->so_state | SS_NOFDREF | SS_ASSERTINPROG; 375 so->so_cred = crhold(head->so_cred); 376 ai.sb_rlimit = NULL; 377 ai.p_ucred = NULL; 378 ai.fd_rdir = NULL; /* jail code cruft XXX JH */ 379 380 /* 381 * Reserve space and call pru_attach. We can direct-call the 382 * function since we're already in the protocol thread. 383 */ 384 if (soreserve(so, head->so_snd.ssb_hiwat, 385 head->so_rcv.ssb_hiwat, NULL) || 386 so_pru_attach_direct(so, 0, &ai)) { 387 so->so_head = NULL; 388 soclrstate(so, SS_ASSERTINPROG); 389 sofree(so); /* remove implied pcb ref */ 390 return (NULL); 391 } 392 KKASSERT(((so->so_proto->pr_flags & PR_ASYNC_RCVD) == 0 && 393 so->so_refs == 2) || /* attach + our base ref */ 394 ((so->so_proto->pr_flags & PR_ASYNC_RCVD) && 395 so->so_refs == 3)); /* + async rcvd ref */ 396 if (keep_ref) { 397 /* 398 * Keep the reference; caller will free it. 399 */ 400 } else { 401 sofree(so); 402 } 403 KKASSERT(so->so_port != NULL); 404 so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat; 405 so->so_snd.ssb_lowat = head->so_snd.ssb_lowat; 406 so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo; 407 so->so_snd.ssb_timeo = head->so_snd.ssb_timeo; 408 409 if (head->so_rcv.ssb_flags & SSB_AUTOLOWAT) 410 so->so_rcv.ssb_flags |= SSB_AUTOLOWAT; 411 else 412 so->so_rcv.ssb_flags &= ~SSB_AUTOLOWAT; 413 414 if (head->so_snd.ssb_flags & SSB_AUTOLOWAT) 415 so->so_snd.ssb_flags |= SSB_AUTOLOWAT; 416 else 417 so->so_snd.ssb_flags &= ~SSB_AUTOLOWAT; 418 419 if (head->so_rcv.ssb_flags & SSB_AUTOSIZE) 420 so->so_rcv.ssb_flags |= SSB_AUTOSIZE; 421 else 422 so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE; 423 424 if (head->so_snd.ssb_flags & SSB_AUTOSIZE) 425 so->so_snd.ssb_flags |= SSB_AUTOSIZE; 426 else 427 so->so_snd.ssb_flags &= ~SSB_AUTOSIZE; 428 429 /* 430 * Save the faddr, if the information is provided and 431 * the protocol can perform the saving opertation. 432 */ 433 if (faddr != NULL && so->so_proto->pr_usrreqs->pru_savefaddr != NULL) 434 so->so_proto->pr_usrreqs->pru_savefaddr(so, faddr); 435 436 lwkt_getpooltoken(head); 437 if (connstatus) { 438 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0); 439 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 440 head->so_qlen++; 441 /* 442 * Set connstatus within head token, so that the accepted 443 * socket will have connstatus (SS_ISCONNECTED) set. 444 */ 445 sosetstate(so, SS_COMP | connstatus); 446 } else { 447 if (head->so_incqlen > head->so_qlimit) { 448 sp = TAILQ_FIRST(&head->so_incomp); 449 KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == 450 SS_INCOMP); 451 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 452 head->so_incqlen--; 453 soclrstate(sp, SS_INCOMP); 454 soabort_async(sp, TRUE); 455 } 456 KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0); 457 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 458 head->so_incqlen++; 459 sosetstate(so, SS_INCOMP); 460 } 461 /* 462 * Clear SS_ASSERTINPROG within head token, so that it will not 463 * race against accept-close or abort for "synchronous" sockets, 464 * e.g. unix socket, on other CPUs. 465 */ 466 soclrstate(so, SS_ASSERTINPROG); 467 lwkt_relpooltoken(head); 468 469 if (connstatus) { 470 /* 471 * XXX head may be on a different protocol thread. 472 * sorwakeup()->sowakeup() is hacked atm. 473 */ 474 logsowakeup(nconn_start); 475 sorwakeup(head); 476 logsowakeup(nconn_end); 477 478 logsowakeup(nconn_wakeupstart); 479 wakeup((caddr_t)&head->so_timeo); 480 logsowakeup(nconn_wakeupend); 481 } 482 return (so); 483 } 484 485 struct socket * 486 sonewconn(struct socket *head, int connstatus) 487 { 488 return sonewconn_faddr(head, connstatus, NULL, FALSE /* don't ref */); 489 } 490 491 /* 492 * Socantsendmore indicates that no more data will be sent on the 493 * socket; it would normally be applied to a socket when the user 494 * informs the system that no more data is to be sent, by the protocol 495 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 496 * will be received, and will normally be applied to the socket by a 497 * protocol when it detects that the peer will send no more data. 498 * Data queued for reading in the socket may yet be read. 499 */ 500 void 501 socantsendmore(struct socket *so) 502 { 503 sosetstate(so, SS_CANTSENDMORE); 504 sowwakeup(so); 505 } 506 507 void 508 socantrcvmore(struct socket *so) 509 { 510 sosetstate(so, SS_CANTRCVMORE); 511 sorwakeup(so); 512 } 513 514 /* 515 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 516 * via SIGIO if the socket has the SS_ASYNC flag set. 517 * 518 * For users waiting on send/recv try to avoid unnecessary context switch 519 * thrashing. Particularly for senders of large buffers (needs to be 520 * extended to sel and aio? XXX) 521 * 522 * WARNING! Can be called on a foreign socket from the wrong protocol 523 * thread. aka is called on the 'head' listen socket when 524 * a new connection comes in. 525 */ 526 527 void 528 sowakeup(struct socket *so, struct signalsockbuf *ssb) 529 { 530 uint32_t flags; 531 532 /* 533 * Atomically check the flags. When no special features are being 534 * used, WAIT is clear, and WAKEUP is already set, we can simply 535 * return. The upcoming synchronous waiter will not block. 536 */ 537 flags = atomic_fetchadd_int(&ssb->ssb_flags, 0); 538 if ((flags & SSB_NOTIFY_MASK) == 0) { 539 if (flags & SSB_WAKEUP) 540 return; 541 } 542 543 /* 544 * Check conditions, set the WAKEUP flag, and clear and signal if 545 * the WAIT flag is found to be set. This interlocks against the 546 * client side. 547 */ 548 for (;;) { 549 long space; 550 551 flags = ssb->ssb_flags; 552 cpu_ccfence(); 553 if (ssb->ssb_flags & SSB_PREALLOC) 554 space = ssb_space_prealloc(ssb); 555 else 556 space = ssb_space(ssb); 557 558 if ((ssb == &so->so_snd && space >= ssb->ssb_lowat) || 559 (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) || 560 (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) || 561 (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE)) 562 ) { 563 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 564 (flags | SSB_WAKEUP) & ~SSB_WAIT)) { 565 if (flags & SSB_WAIT) 566 wakeup(&ssb->ssb_cc); 567 break; 568 } 569 } else { 570 break; 571 } 572 } 573 574 /* 575 * Misc other events 576 */ 577 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 578 pgsigio(so->so_sigio, SIGIO, 0); 579 if (ssb->ssb_flags & SSB_UPCALL) 580 (*so->so_upcall)(so, so->so_upcallarg, M_NOWAIT); 581 KNOTE(&ssb->ssb_kq.ki_note, 0); 582 583 /* 584 * This is a bit of a hack. Multiple threads can wind up scanning 585 * ssb_mlist concurrently due to the fact that this function can be 586 * called on a foreign socket, so we can't afford to block here. 587 * 588 * We need the pool token for (so) (likely the listne socket if 589 * SSB_MEVENT is set) because the predicate function may have 590 * to access the accept queue. 591 */ 592 if (ssb->ssb_flags & SSB_MEVENT) { 593 struct netmsg_so_notify *msg, *nmsg; 594 595 lwkt_getpooltoken(so); 596 TAILQ_FOREACH_MUTABLE(msg, &ssb->ssb_mlist, nm_list, nmsg) { 597 if (msg->nm_predicate(msg)) { 598 TAILQ_REMOVE(&ssb->ssb_mlist, msg, nm_list); 599 lwkt_replymsg(&msg->base.lmsg, 600 msg->base.lmsg.ms_error); 601 } 602 } 603 if (TAILQ_EMPTY(&ssb->ssb_mlist)) 604 atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT); 605 lwkt_relpooltoken(so); 606 } 607 } 608 609 /* 610 * Socket buffer (struct signalsockbuf) utility routines. 611 * 612 * Each socket contains two socket buffers: one for sending data and 613 * one for receiving data. Each buffer contains a queue of mbufs, 614 * information about the number of mbufs and amount of data in the 615 * queue, and other fields allowing kevent()/select()/poll() statements 616 * and notification on data availability to be implemented. 617 * 618 * Data stored in a socket buffer is maintained as a list of records. 619 * Each record is a list of mbufs chained together with the m_next 620 * field. Records are chained together with the m_nextpkt field. The upper 621 * level routine soreceive() expects the following conventions to be 622 * observed when placing information in the receive buffer: 623 * 624 * 1. If the protocol requires each message be preceded by the sender's 625 * name, then a record containing that name must be present before 626 * any associated data (mbuf's must be of type MT_SONAME). 627 * 2. If the protocol supports the exchange of ``access rights'' (really 628 * just additional data associated with the message), and there are 629 * ``rights'' to be received, then a record containing this data 630 * should be present (mbuf's must be of type MT_RIGHTS). 631 * 3. If a name or rights record exists, then it must be followed by 632 * a data record, perhaps of zero length. 633 * 634 * Before using a new socket structure it is first necessary to reserve 635 * buffer space to the socket, by calling sbreserve(). This should commit 636 * some of the available buffer space in the system buffer pool for the 637 * socket (currently, it does nothing but enforce limits). The space 638 * should be released by calling ssb_release() when the socket is destroyed. 639 */ 640 int 641 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) 642 { 643 if (so->so_snd.ssb_lowat == 0) 644 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT); 645 if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0) 646 goto bad; 647 if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0) 648 goto bad2; 649 if (so->so_rcv.ssb_lowat == 0) 650 so->so_rcv.ssb_lowat = 1; 651 if (so->so_snd.ssb_lowat == 0) 652 so->so_snd.ssb_lowat = MCLBYTES; 653 if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat) 654 so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat; 655 return (0); 656 bad2: 657 ssb_release(&so->so_snd, so); 658 bad: 659 return (ENOBUFS); 660 } 661 662 static int 663 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 664 { 665 int error = 0; 666 u_long old_sb_max = sb_max; 667 668 error = SYSCTL_OUT(req, arg1, sizeof(int)); 669 if (error || !req->newptr) 670 return (error); 671 error = SYSCTL_IN(req, arg1, sizeof(int)); 672 if (error) 673 return (error); 674 if (sb_max < MSIZE + MCLBYTES) { 675 sb_max = old_sb_max; 676 return (EINVAL); 677 } 678 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 679 return (0); 680 } 681 682 /* 683 * Allot mbufs to a signalsockbuf. 684 * 685 * Attempt to scale mbmax so that mbcnt doesn't become limiting 686 * if buffering efficiency is near the normal case. 687 * 688 * sb_max only applies to user-sockets (where rl != NULL). It does 689 * not apply to kernel sockets or kernel-controlled sockets. Note 690 * that NFS overrides the sockbuf limits created when nfsd creates 691 * a socket. 692 */ 693 int 694 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so, 695 struct rlimit *rl) 696 { 697 /* 698 * rl will only be NULL when we're in an interrupt (eg, in tcp_input) 699 * or when called from netgraph (ie, ngd_attach) 700 */ 701 if (rl && cc > sb_max_adj) 702 cc = sb_max_adj; 703 if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc, 704 rl ? rl->rlim_cur : RLIM_INFINITY)) { 705 return (0); 706 } 707 if (rl) 708 ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max); 709 else 710 ssb->ssb_mbmax = cc * sb_efficiency; 711 712 /* 713 * AUTOLOWAT is set on send buffers and prevents large writes 714 * from generating a huge number of context switches. 715 */ 716 if (ssb->ssb_flags & SSB_AUTOLOWAT) { 717 ssb->ssb_lowat = ssb->ssb_hiwat / 4; 718 if (ssb->ssb_lowat < MCLBYTES) 719 ssb->ssb_lowat = MCLBYTES; 720 } 721 if (ssb->ssb_lowat > ssb->ssb_hiwat) 722 ssb->ssb_lowat = ssb->ssb_hiwat; 723 return (1); 724 } 725 726 /* 727 * Free mbufs held by a socket, and reserved mbuf space. 728 */ 729 void 730 ssb_release(struct signalsockbuf *ssb, struct socket *so) 731 { 732 sbflush(&ssb->sb); 733 (void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0, 734 RLIM_INFINITY); 735 ssb->ssb_mbmax = 0; 736 } 737 738 /* 739 * Some routines that return EOPNOTSUPP for entry points that are not 740 * supported by a protocol. Fill in as needed. 741 */ 742 void 743 pr_generic_notsupp(netmsg_t msg) 744 { 745 lwkt_replymsg(&msg->lmsg, EOPNOTSUPP); 746 } 747 748 int 749 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 750 struct mbuf *top, struct mbuf *control, int flags, 751 struct thread *td) 752 { 753 if (top) 754 m_freem(top); 755 if (control) 756 m_freem(control); 757 return (EOPNOTSUPP); 758 } 759 760 int 761 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 762 struct uio *uio, struct sockbuf *sio, 763 struct mbuf **controlp, int *flagsp) 764 { 765 return (EOPNOTSUPP); 766 } 767 768 /* 769 * This isn't really a ``null'' operation, but it's the default one 770 * and doesn't do anything destructive. 771 */ 772 void 773 pru_sense_null(netmsg_t msg) 774 { 775 msg->sense.nm_stat->st_blksize = msg->base.nm_so->so_snd.ssb_hiwat; 776 lwkt_replymsg(&msg->lmsg, 0); 777 } 778 779 /* 780 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers 781 * of this routine assume that it always succeeds, so we have to use a 782 * blockable allocation even though we might be called from a critical thread. 783 */ 784 struct sockaddr * 785 dup_sockaddr(const struct sockaddr *sa) 786 { 787 struct sockaddr *sa2; 788 789 sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT); 790 bcopy(sa, sa2, sa->sa_len); 791 return (sa2); 792 } 793 794 /* 795 * Create an external-format (``xsocket'') structure using the information 796 * in the kernel-format socket structure pointed to by so. This is done 797 * to reduce the spew of irrelevant information over this interface, 798 * to isolate user code from changes in the kernel structure, and 799 * potentially to provide information-hiding if we decide that 800 * some of this information should be hidden from users. 801 */ 802 void 803 sotoxsocket(struct socket *so, struct xsocket *xso) 804 { 805 xso->xso_len = sizeof *xso; 806 xso->xso_so = so; 807 xso->so_type = so->so_type; 808 xso->so_options = so->so_options; 809 xso->so_linger = so->so_linger; 810 xso->so_state = so->so_state; 811 xso->so_pcb = so->so_pcb; 812 xso->xso_protocol = so->so_proto->pr_protocol; 813 xso->xso_family = so->so_proto->pr_domain->dom_family; 814 xso->so_qlen = so->so_qlen; 815 xso->so_incqlen = so->so_incqlen; 816 xso->so_qlimit = so->so_qlimit; 817 xso->so_timeo = so->so_timeo; 818 xso->so_error = so->so_error; 819 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 820 xso->so_oobmark = so->so_oobmark; 821 ssbtoxsockbuf(&so->so_snd, &xso->so_snd); 822 ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 823 xso->so_uid = so->so_cred->cr_uid; 824 } 825 826 /* 827 * Here is the definition of some of the basic objects in the kern.ipc 828 * branch of the MIB. 829 */ 830 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 831 832 /* 833 * This takes the place of kern.maxsockbuf, which moved to kern.ipc. 834 * 835 * NOTE! sb_max only applies to user-created socket buffers. 836 */ 837 static int dummy; 838 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 839 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 840 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 841 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 842 &maxsockets, 0, "Maximum number of sockets available"); 843 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 844 &sb_efficiency, 0, 845 "Socket buffer limit scaler"); 846 847 /* 848 * Initialize maxsockets 849 */ 850 static void 851 init_maxsockets(void *ignored) 852 { 853 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 854 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 855 } 856 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY, 857 init_maxsockets, NULL); 858 859