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. All advertising materials mentioning features or use of this software 15 * must display the following acknowledgement: 16 * This product includes software developed by the University of 17 * California, Berkeley and its contributors. 18 * 4. 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 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93 35 * $FreeBSD: src/sys/kern/uipc_socket2.c,v 1.55.2.17 2002/08/31 19:04:55 dwmalone Exp $ 36 * $DragonFly: src/sys/kern/uipc_socket2.c,v 1.33 2008/09/02 16:17:52 dillon Exp $ 37 */ 38 39 #include "opt_param.h" 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/domain.h> 43 #include <sys/file.h> /* for maxfiles */ 44 #include <sys/kernel.h> 45 #include <sys/proc.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/protosw.h> 49 #include <sys/resourcevar.h> 50 #include <sys/stat.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/socketops.h> 54 #include <sys/signalvar.h> 55 #include <sys/sysctl.h> 56 #include <sys/aio.h> /* for aio_swake proto */ 57 #include <sys/event.h> 58 59 #include <sys/thread2.h> 60 #include <sys/msgport2.h> 61 #include <sys/socketvar2.h> 62 63 int maxsockets; 64 65 /* 66 * Primitive routines for operating on sockets and socket buffers 67 */ 68 69 u_long sb_max = SB_MAX; 70 u_long sb_max_adj = 71 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 72 73 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 74 75 /************************************************************************ 76 * signalsockbuf procedures * 77 ************************************************************************/ 78 79 /* 80 * Wait for data to arrive at/drain from a socket buffer. 81 * 82 * NOTE: Caller must generally hold the ssb_lock (client side lock) since 83 * WAIT/WAKEUP only works for one client at a time. 84 * 85 * NOTE: Caller always retries whatever operation it was waiting on. 86 */ 87 int 88 ssb_wait(struct signalsockbuf *ssb) 89 { 90 uint32_t flags; 91 int pflags; 92 int error; 93 94 pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH; 95 96 for (;;) { 97 flags = ssb->ssb_flags; 98 cpu_ccfence(); 99 100 /* 101 * WAKEUP and WAIT interlock eachother. We can catch the 102 * race by checking to see if WAKEUP has already been set, 103 * and only setting WAIT if WAKEUP is clear. 104 */ 105 if (flags & SSB_WAKEUP) { 106 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 107 flags & ~SSB_WAKEUP)) { 108 error = 0; 109 break; 110 } 111 continue; 112 } 113 114 /* 115 * Only set WAIT if WAKEUP is clear. 116 */ 117 tsleep_interlock(&ssb->ssb_cc, pflags); 118 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 119 flags | SSB_WAIT)) { 120 error = tsleep(&ssb->ssb_cc, pflags | PINTERLOCKED, 121 "sbwait", ssb->ssb_timeo); 122 break; 123 } 124 } 125 return (error); 126 } 127 128 /* 129 * Lock a sockbuf already known to be locked; 130 * return any error returned from sleep (EINTR). 131 */ 132 int 133 _ssb_lock(struct signalsockbuf *ssb) 134 { 135 uint32_t flags; 136 int pflags; 137 int error; 138 139 pflags = (ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH; 140 141 for (;;) { 142 flags = ssb->ssb_flags; 143 cpu_ccfence(); 144 if (flags & SSB_LOCK) { 145 tsleep_interlock(&ssb->ssb_flags, pflags); 146 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 147 flags | SSB_WANT)) { 148 error = tsleep(&ssb->ssb_flags, 149 pflags | PINTERLOCKED, 150 "sblock", 0); 151 if (error) 152 break; 153 } 154 } else { 155 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 156 flags | SSB_LOCK)) { 157 lwkt_gettoken(&ssb->ssb_token); 158 error = 0; 159 break; 160 } 161 } 162 } 163 return (error); 164 } 165 166 /* 167 * This does the same for sockbufs. Note that the xsockbuf structure, 168 * since it is always embedded in a socket, does not include a self 169 * pointer nor a length. We make this entry point public in case 170 * some other mechanism needs it. 171 */ 172 void 173 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb) 174 { 175 xsb->sb_cc = ssb->ssb_cc; 176 xsb->sb_hiwat = ssb->ssb_hiwat; 177 xsb->sb_mbcnt = ssb->ssb_mbcnt; 178 xsb->sb_mbmax = ssb->ssb_mbmax; 179 xsb->sb_lowat = ssb->ssb_lowat; 180 xsb->sb_flags = ssb->ssb_flags; 181 xsb->sb_timeo = ssb->ssb_timeo; 182 } 183 184 185 /************************************************************************ 186 * Procedures which manipulate socket state flags, wakeups, etc. * 187 ************************************************************************ 188 * 189 * Normal sequence from the active (originating) side is that 190 * soisconnecting() is called during processing of connect() call, resulting 191 * in an eventual call to soisconnected() if/when the connection is 192 * established. When the connection is torn down soisdisconnecting() is 193 * called during processing of disconnect() call, and soisdisconnected() is 194 * called when the connection to the peer is totally severed. 195 * 196 * The semantics of these routines are such that connectionless protocols 197 * can call soisconnected() and soisdisconnected() only, bypassing the 198 * in-progress calls when setting up a ``connection'' takes no time. 199 * 200 * From the passive side, a socket is created with two queues of sockets: 201 * so_incomp for connections in progress and so_comp for connections 202 * already made and awaiting user acceptance. As a protocol is preparing 203 * incoming connections, it creates a socket structure queued on so_incomp 204 * by calling sonewconn(). When the connection is established, 205 * soisconnected() is called, and transfers the socket structure to so_comp, 206 * making it available to accept(). 207 * 208 * If a socket is closed with sockets on either so_incomp or so_comp, these 209 * sockets are dropped. 210 * 211 * If higher level protocols are implemented in the kernel, the wakeups 212 * done here will sometimes cause software-interrupt process scheduling. 213 */ 214 215 void 216 soisconnecting(struct socket *so) 217 { 218 soclrstate(so, SS_ISCONNECTED | SS_ISDISCONNECTING); 219 sosetstate(so, SS_ISCONNECTING); 220 } 221 222 void 223 soisconnected(struct socket *so) 224 { 225 struct socket *head; 226 227 while ((head = so->so_head) != NULL) { 228 lwkt_getpooltoken(head); 229 if (so->so_head == head) 230 break; 231 lwkt_relpooltoken(head); 232 } 233 234 soclrstate(so, SS_ISCONNECTING | SS_ISDISCONNECTING | SS_ISCONFIRMING); 235 sosetstate(so, SS_ISCONNECTED); 236 if (head && (so->so_state & SS_INCOMP)) { 237 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 238 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 239 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 240 atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL); 241 so->so_options &= ~SO_ACCEPTFILTER; 242 so->so_upcall(so, so->so_upcallarg, 0); 243 lwkt_relpooltoken(head); 244 return; 245 } 246 247 /* 248 * Listen socket are not per-cpu. 249 */ 250 TAILQ_REMOVE(&head->so_incomp, so, so_list); 251 head->so_incqlen--; 252 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 253 head->so_qlen++; 254 sosetstate(so, SS_COMP); 255 soclrstate(so, SS_INCOMP); 256 257 /* 258 * XXX head may be on a different protocol thread. 259 * sorwakeup()->sowakeup() is hacked atm. 260 */ 261 sorwakeup(head); 262 wakeup_one(&head->so_timeo); 263 } else { 264 wakeup(&so->so_timeo); 265 sorwakeup(so); 266 sowwakeup(so); 267 } 268 if (head) 269 lwkt_relpooltoken(head); 270 } 271 272 void 273 soisdisconnecting(struct socket *so) 274 { 275 soclrstate(so, SS_ISCONNECTING); 276 sosetstate(so, SS_ISDISCONNECTING | SS_CANTRCVMORE | SS_CANTSENDMORE); 277 wakeup((caddr_t)&so->so_timeo); 278 sowwakeup(so); 279 sorwakeup(so); 280 } 281 282 void 283 soisdisconnected(struct socket *so) 284 { 285 soclrstate(so, SS_ISCONNECTING | SS_ISCONNECTED | SS_ISDISCONNECTING); 286 sosetstate(so, SS_CANTRCVMORE | SS_CANTSENDMORE | SS_ISDISCONNECTED); 287 wakeup((caddr_t)&so->so_timeo); 288 sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc); 289 sowwakeup(so); 290 sorwakeup(so); 291 } 292 293 void 294 soisreconnecting(struct socket *so) 295 { 296 soclrstate(so, SS_ISDISCONNECTING | SS_ISDISCONNECTED | 297 SS_CANTRCVMORE | SS_CANTSENDMORE); 298 sosetstate(so, SS_ISCONNECTING); 299 } 300 301 void 302 soisreconnected(struct socket *so) 303 { 304 soclrstate(so, SS_ISDISCONNECTED | SS_CANTRCVMORE | SS_CANTSENDMORE); 305 soisconnected(so); 306 } 307 308 /* 309 * Set or change the message port a socket receives commands on. 310 * 311 * XXX 312 */ 313 void 314 sosetport(struct socket *so, lwkt_port_t port) 315 { 316 so->so_port = port; 317 } 318 319 /* 320 * When an attempt at a new connection is noted on a socket 321 * which accepts connections, sonewconn is called. If the 322 * connection is possible (subject to space constraints, etc.) 323 * then we allocate a new structure, propoerly linked into the 324 * data structure of the original socket, and return this. 325 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 326 * 327 * The new socket is returned with one ref and so_pcb assigned. 328 * The reference is implied by so_pcb. 329 */ 330 struct socket * 331 sonewconn(struct socket *head, int connstatus) 332 { 333 struct socket *so; 334 struct socket *sp; 335 struct pru_attach_info ai; 336 337 if (head->so_qlen > 3 * head->so_qlimit / 2) 338 return (NULL); 339 so = soalloc(1); 340 if (so == NULL) 341 return (NULL); 342 343 /* 344 * Set the port prior to attaching the inpcb to the current 345 * cpu's protocol thread (which should be the current thread 346 * but might not be in all cases). This serializes any pcb ops 347 * which occur to our cpu allowing us to complete the attachment 348 * without racing anything. 349 */ 350 sosetport(so, cpu_portfn(mycpu->gd_cpuid)); 351 if ((head->so_options & SO_ACCEPTFILTER) != 0) 352 connstatus = 0; 353 so->so_head = head; 354 so->so_type = head->so_type; 355 so->so_options = head->so_options &~ SO_ACCEPTCONN; 356 so->so_linger = head->so_linger; 357 358 /* 359 * NOTE: Clearing NOFDREF implies referencing the so with 360 * soreference(). 361 */ 362 so->so_state = head->so_state | SS_NOFDREF | SS_ASSERTINPROG; 363 so->so_proto = head->so_proto; 364 so->so_cred = crhold(head->so_cred); 365 ai.sb_rlimit = NULL; 366 ai.p_ucred = NULL; 367 ai.fd_rdir = NULL; /* jail code cruft XXX JH */ 368 369 /* 370 * Reserve space and call pru_attach. We can direct-call the 371 * function since we're already in the protocol thread. 372 */ 373 if (soreserve(so, head->so_snd.ssb_hiwat, 374 head->so_rcv.ssb_hiwat, NULL) || 375 so_pru_attach_direct(so, 0, &ai)) { 376 so->so_head = NULL; 377 soclrstate(so, SS_ASSERTINPROG); 378 sofree(so); /* remove implied pcb ref */ 379 return (NULL); 380 } 381 KKASSERT(so->so_refs == 2); /* attach + our base ref */ 382 sofree(so); 383 KKASSERT(so->so_port != NULL); 384 so->so_rcv.ssb_lowat = head->so_rcv.ssb_lowat; 385 so->so_snd.ssb_lowat = head->so_snd.ssb_lowat; 386 so->so_rcv.ssb_timeo = head->so_rcv.ssb_timeo; 387 so->so_snd.ssb_timeo = head->so_snd.ssb_timeo; 388 so->so_rcv.ssb_flags |= head->so_rcv.ssb_flags & 389 (SSB_AUTOSIZE | SSB_AUTOLOWAT); 390 so->so_snd.ssb_flags |= head->so_snd.ssb_flags & 391 (SSB_AUTOSIZE | SSB_AUTOLOWAT); 392 lwkt_getpooltoken(head); 393 if (connstatus) { 394 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 395 sosetstate(so, SS_COMP); 396 head->so_qlen++; 397 } else { 398 if (head->so_incqlen > head->so_qlimit) { 399 sp = TAILQ_FIRST(&head->so_incomp); 400 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 401 head->so_incqlen--; 402 soclrstate(sp, SS_INCOMP); 403 sp->so_head = NULL; 404 soaborta(sp); 405 } 406 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 407 sosetstate(so, SS_INCOMP); 408 head->so_incqlen++; 409 } 410 lwkt_relpooltoken(head); 411 if (connstatus) { 412 /* 413 * XXX head may be on a different protocol thread. 414 * sorwakeup()->sowakeup() is hacked atm. 415 */ 416 sorwakeup(head); 417 wakeup((caddr_t)&head->so_timeo); 418 sosetstate(so, connstatus); 419 } 420 soclrstate(so, SS_ASSERTINPROG); 421 return (so); 422 } 423 424 /* 425 * Socantsendmore indicates that no more data will be sent on the 426 * socket; it would normally be applied to a socket when the user 427 * informs the system that no more data is to be sent, by the protocol 428 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 429 * will be received, and will normally be applied to the socket by a 430 * protocol when it detects that the peer will send no more data. 431 * Data queued for reading in the socket may yet be read. 432 */ 433 void 434 socantsendmore(struct socket *so) 435 { 436 sosetstate(so, SS_CANTSENDMORE); 437 sowwakeup(so); 438 } 439 440 void 441 socantrcvmore(struct socket *so) 442 { 443 sosetstate(so, SS_CANTRCVMORE); 444 sorwakeup(so); 445 } 446 447 /* 448 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 449 * via SIGIO if the socket has the SS_ASYNC flag set. 450 * 451 * For users waiting on send/recv try to avoid unnecessary context switch 452 * thrashing. Particularly for senders of large buffers (needs to be 453 * extended to sel and aio? XXX) 454 * 455 * WARNING! Can be called on a foreign socket from the wrong protocol 456 * thread. aka is called on the 'head' listen socket when 457 * a new connection comes in. 458 */ 459 void 460 sowakeup(struct socket *so, struct signalsockbuf *ssb) 461 { 462 struct kqinfo *kqinfo = &ssb->ssb_kq; 463 uint32_t flags; 464 465 /* 466 * Check conditions, set the WAKEUP flag, and clear and signal if 467 * the WAIT flag is found to be set. This interlocks against the 468 * client side. 469 */ 470 for (;;) { 471 flags = ssb->ssb_flags; 472 cpu_ccfence(); 473 474 if ((ssb == &so->so_snd && ssb_space(ssb) >= ssb->ssb_lowat) || 475 (ssb == &so->so_rcv && ssb->ssb_cc >= ssb->ssb_lowat) || 476 (ssb == &so->so_snd && (so->so_state & SS_CANTSENDMORE)) || 477 (ssb == &so->so_rcv && (so->so_state & SS_CANTRCVMORE)) 478 ) { 479 if (atomic_cmpset_int(&ssb->ssb_flags, flags, 480 (flags | SSB_WAKEUP) & ~SSB_WAIT)) { 481 if (flags & SSB_WAIT) 482 wakeup(&ssb->ssb_cc); 483 break; 484 } 485 } else { 486 break; 487 } 488 } 489 490 /* 491 * Misc other events 492 */ 493 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 494 pgsigio(so->so_sigio, SIGIO, 0); 495 if (ssb->ssb_flags & SSB_UPCALL) 496 (*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT); 497 if (ssb->ssb_flags & SSB_AIO) 498 aio_swake(so, ssb); 499 KNOTE(&kqinfo->ki_note, 0); 500 501 /* 502 * This is a bit of a hack. Multiple threads can wind up scanning 503 * ki_mlist concurrently due to the fact that this function can be 504 * called on a foreign socket, so we can't afford to block here. 505 * 506 * We need the pool token for (so) (likely the listne socket if 507 * SSB_MEVENT is set) because the predicate function may have 508 * to access the accept queue. 509 */ 510 if (ssb->ssb_flags & SSB_MEVENT) { 511 struct netmsg_so_notify *msg, *nmsg; 512 513 lwkt_gettoken(&kq_token); 514 lwkt_getpooltoken(so); 515 TAILQ_FOREACH_MUTABLE(msg, &kqinfo->ki_mlist, nm_list, nmsg) { 516 if (msg->nm_predicate(msg)) { 517 TAILQ_REMOVE(&kqinfo->ki_mlist, msg, nm_list); 518 lwkt_replymsg(&msg->base.lmsg, 519 msg->base.lmsg.ms_error); 520 } 521 } 522 if (TAILQ_EMPTY(&ssb->ssb_kq.ki_mlist)) 523 atomic_clear_int(&ssb->ssb_flags, SSB_MEVENT); 524 lwkt_relpooltoken(so); 525 lwkt_reltoken(&kq_token); 526 } 527 } 528 529 /* 530 * Socket buffer (struct signalsockbuf) utility routines. 531 * 532 * Each socket contains two socket buffers: one for sending data and 533 * one for receiving data. Each buffer contains a queue of mbufs, 534 * information about the number of mbufs and amount of data in the 535 * queue, and other fields allowing kevent()/select()/poll() statements 536 * and notification on data availability to be implemented. 537 * 538 * Data stored in a socket buffer is maintained as a list of records. 539 * Each record is a list of mbufs chained together with the m_next 540 * field. Records are chained together with the m_nextpkt field. The upper 541 * level routine soreceive() expects the following conventions to be 542 * observed when placing information in the receive buffer: 543 * 544 * 1. If the protocol requires each message be preceded by the sender's 545 * name, then a record containing that name must be present before 546 * any associated data (mbuf's must be of type MT_SONAME). 547 * 2. If the protocol supports the exchange of ``access rights'' (really 548 * just additional data associated with the message), and there are 549 * ``rights'' to be received, then a record containing this data 550 * should be present (mbuf's must be of type MT_RIGHTS). 551 * 3. If a name or rights record exists, then it must be followed by 552 * a data record, perhaps of zero length. 553 * 554 * Before using a new socket structure it is first necessary to reserve 555 * buffer space to the socket, by calling sbreserve(). This should commit 556 * some of the available buffer space in the system buffer pool for the 557 * socket (currently, it does nothing but enforce limits). The space 558 * should be released by calling ssb_release() when the socket is destroyed. 559 */ 560 int 561 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) 562 { 563 if (so->so_snd.ssb_lowat == 0) 564 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOLOWAT); 565 if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0) 566 goto bad; 567 if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0) 568 goto bad2; 569 if (so->so_rcv.ssb_lowat == 0) 570 so->so_rcv.ssb_lowat = 1; 571 if (so->so_snd.ssb_lowat == 0) 572 so->so_snd.ssb_lowat = MCLBYTES; 573 if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat) 574 so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat; 575 return (0); 576 bad2: 577 ssb_release(&so->so_snd, so); 578 bad: 579 return (ENOBUFS); 580 } 581 582 static int 583 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 584 { 585 int error = 0; 586 u_long old_sb_max = sb_max; 587 588 error = SYSCTL_OUT(req, arg1, sizeof(int)); 589 if (error || !req->newptr) 590 return (error); 591 error = SYSCTL_IN(req, arg1, sizeof(int)); 592 if (error) 593 return (error); 594 if (sb_max < MSIZE + MCLBYTES) { 595 sb_max = old_sb_max; 596 return (EINVAL); 597 } 598 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 599 return (0); 600 } 601 602 /* 603 * Allot mbufs to a signalsockbuf. 604 * 605 * Attempt to scale mbmax so that mbcnt doesn't become limiting 606 * if buffering efficiency is near the normal case. 607 * 608 * sb_max only applies to user-sockets (where rl != NULL). It does 609 * not apply to kernel sockets or kernel-controlled sockets. Note 610 * that NFS overrides the sockbuf limits created when nfsd creates 611 * a socket. 612 */ 613 int 614 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so, 615 struct rlimit *rl) 616 { 617 /* 618 * rl will only be NULL when we're in an interrupt (eg, in tcp_input) 619 * or when called from netgraph (ie, ngd_attach) 620 */ 621 if (rl && cc > sb_max_adj) 622 cc = sb_max_adj; 623 if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc, 624 rl ? rl->rlim_cur : RLIM_INFINITY)) { 625 return (0); 626 } 627 if (rl) 628 ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max); 629 else 630 ssb->ssb_mbmax = cc * sb_efficiency; 631 632 /* 633 * AUTOLOWAT is set on send buffers and prevents large writes 634 * from generating a huge number of context switches. 635 */ 636 if (ssb->ssb_flags & SSB_AUTOLOWAT) { 637 ssb->ssb_lowat = ssb->ssb_hiwat / 2; 638 if (ssb->ssb_lowat < MCLBYTES) 639 ssb->ssb_lowat = MCLBYTES; 640 } 641 if (ssb->ssb_lowat > ssb->ssb_hiwat) 642 ssb->ssb_lowat = ssb->ssb_hiwat; 643 return (1); 644 } 645 646 /* 647 * Free mbufs held by a socket, and reserved mbuf space. 648 */ 649 void 650 ssb_release(struct signalsockbuf *ssb, struct socket *so) 651 { 652 sbflush(&ssb->sb); 653 (void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0, 654 RLIM_INFINITY); 655 ssb->ssb_mbmax = 0; 656 } 657 658 /* 659 * Some routines that return EOPNOTSUPP for entry points that are not 660 * supported by a protocol. Fill in as needed. 661 */ 662 void 663 pr_generic_notsupp(netmsg_t msg) 664 { 665 lwkt_replymsg(&msg->lmsg, EOPNOTSUPP); 666 } 667 668 int 669 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio, 670 struct mbuf *top, struct mbuf *control, int flags, 671 struct thread *td) 672 { 673 if (top) 674 m_freem(top); 675 if (control) 676 m_freem(control); 677 return (EOPNOTSUPP); 678 } 679 680 int 681 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr, 682 struct uio *uio, struct sockbuf *sio, 683 struct mbuf **controlp, int *flagsp) 684 { 685 return (EOPNOTSUPP); 686 } 687 688 /* 689 * This isn't really a ``null'' operation, but it's the default one 690 * and doesn't do anything destructive. 691 */ 692 void 693 pru_sense_null(netmsg_t msg) 694 { 695 msg->sense.nm_stat->st_blksize = msg->base.nm_so->so_snd.ssb_hiwat; 696 lwkt_replymsg(&msg->lmsg, 0); 697 } 698 699 /* 700 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers 701 * of this routine assume that it always succeeds, so we have to use a 702 * blockable allocation even though we might be called from a critical thread. 703 */ 704 struct sockaddr * 705 dup_sockaddr(const struct sockaddr *sa) 706 { 707 struct sockaddr *sa2; 708 709 sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT); 710 bcopy(sa, sa2, sa->sa_len); 711 return (sa2); 712 } 713 714 /* 715 * Create an external-format (``xsocket'') structure using the information 716 * in the kernel-format socket structure pointed to by so. This is done 717 * to reduce the spew of irrelevant information over this interface, 718 * to isolate user code from changes in the kernel structure, and 719 * potentially to provide information-hiding if we decide that 720 * some of this information should be hidden from users. 721 */ 722 void 723 sotoxsocket(struct socket *so, struct xsocket *xso) 724 { 725 xso->xso_len = sizeof *xso; 726 xso->xso_so = so; 727 xso->so_type = so->so_type; 728 xso->so_options = so->so_options; 729 xso->so_linger = so->so_linger; 730 xso->so_state = so->so_state; 731 xso->so_pcb = so->so_pcb; 732 xso->xso_protocol = so->so_proto->pr_protocol; 733 xso->xso_family = so->so_proto->pr_domain->dom_family; 734 xso->so_qlen = so->so_qlen; 735 xso->so_incqlen = so->so_incqlen; 736 xso->so_qlimit = so->so_qlimit; 737 xso->so_timeo = so->so_timeo; 738 xso->so_error = so->so_error; 739 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 740 xso->so_oobmark = so->so_oobmark; 741 ssbtoxsockbuf(&so->so_snd, &xso->so_snd); 742 ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 743 xso->so_uid = so->so_cred->cr_uid; 744 } 745 746 /* 747 * Here is the definition of some of the basic objects in the kern.ipc 748 * branch of the MIB. 749 */ 750 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 751 752 /* 753 * This takes the place of kern.maxsockbuf, which moved to kern.ipc. 754 * 755 * NOTE! sb_max only applies to user-created socket buffers. 756 */ 757 static int dummy; 758 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 759 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 760 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 761 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 762 &maxsockets, 0, "Maximum number of sockets available"); 763 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 764 &sb_efficiency, 0, ""); 765 766 /* 767 * Initialize maxsockets 768 */ 769 static void 770 init_maxsockets(void *ignored) 771 { 772 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 773 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 774 } 775 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY, 776 init_maxsockets, NULL); 777 778