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 1.32 2008/08/28 23:15:43 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/signalvar.h> 54 #include <sys/sysctl.h> 55 #include <sys/aio.h> /* for aio_swake proto */ 56 #include <sys/event.h> 57 58 #include <sys/thread2.h> 59 #include <sys/msgport2.h> 60 61 int maxsockets; 62 63 /* 64 * Primitive routines for operating on sockets and socket buffers 65 */ 66 67 u_long sb_max = SB_MAX; 68 u_long sb_max_adj = 69 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */ 70 71 static u_long sb_efficiency = 8; /* parameter for sbreserve() */ 72 73 /************************************************************************ 74 * signalsockbuf procedures * 75 ************************************************************************/ 76 77 /* 78 * Wait for data to arrive at/drain from a socket buffer. 79 */ 80 int 81 ssb_wait(struct signalsockbuf *ssb) 82 { 83 84 ssb->ssb_flags |= SSB_WAIT; 85 return (tsleep((caddr_t)&ssb->ssb_cc, 86 ((ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH), 87 "sbwait", 88 ssb->ssb_timeo)); 89 } 90 91 /* 92 * Lock a sockbuf already known to be locked; 93 * return any error returned from sleep (EINTR). 94 */ 95 int 96 _ssb_lock(struct signalsockbuf *ssb) 97 { 98 int error; 99 100 while (ssb->ssb_flags & SSB_LOCK) { 101 ssb->ssb_flags |= SSB_WANT; 102 error = tsleep((caddr_t)&ssb->ssb_flags, 103 ((ssb->ssb_flags & SSB_NOINTR) ? 0 : PCATCH), 104 "sblock", 0); 105 if (error) 106 return (error); 107 } 108 ssb->ssb_flags |= SSB_LOCK; 109 return (0); 110 } 111 112 /* 113 * This does the same for sockbufs. Note that the xsockbuf structure, 114 * since it is always embedded in a socket, does not include a self 115 * pointer nor a length. We make this entry point public in case 116 * some other mechanism needs it. 117 */ 118 void 119 ssbtoxsockbuf(struct signalsockbuf *ssb, struct xsockbuf *xsb) 120 { 121 xsb->sb_cc = ssb->ssb_cc; 122 xsb->sb_hiwat = ssb->ssb_hiwat; 123 xsb->sb_mbcnt = ssb->ssb_mbcnt; 124 xsb->sb_mbmax = ssb->ssb_mbmax; 125 xsb->sb_lowat = ssb->ssb_lowat; 126 xsb->sb_flags = ssb->ssb_flags; 127 xsb->sb_timeo = ssb->ssb_timeo; 128 } 129 130 131 /************************************************************************ 132 * Procedures which manipulate socket state flags, wakeups, etc. * 133 ************************************************************************ 134 * 135 * Normal sequence from the active (originating) side is that 136 * soisconnecting() is called during processing of connect() call, resulting 137 * in an eventual call to soisconnected() if/when the connection is 138 * established. When the connection is torn down soisdisconnecting() is 139 * called during processing of disconnect() call, and soisdisconnected() is 140 * called when the connection to the peer is totally severed. 141 * 142 * The semantics of these routines are such that connectionless protocols 143 * can call soisconnected() and soisdisconnected() only, bypassing the 144 * in-progress calls when setting up a ``connection'' takes no time. 145 * 146 * From the passive side, a socket is created with two queues of sockets: 147 * so_incomp for connections in progress and so_comp for connections 148 * already made and awaiting user acceptance. As a protocol is preparing 149 * incoming connections, it creates a socket structure queued on so_incomp 150 * by calling sonewconn(). When the connection is established, 151 * soisconnected() is called, and transfers the socket structure to so_comp, 152 * making it available to accept(). 153 * 154 * If a socket is closed with sockets on either so_incomp or so_comp, these 155 * sockets are dropped. 156 * 157 * If higher level protocols are implemented in the kernel, the wakeups 158 * done here will sometimes cause software-interrupt process scheduling. 159 */ 160 161 void 162 soisconnecting(struct socket *so) 163 { 164 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING); 165 so->so_state |= SS_ISCONNECTING; 166 } 167 168 void 169 soisconnected(struct socket *so) 170 { 171 struct socket *head = so->so_head; 172 173 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING); 174 so->so_state |= SS_ISCONNECTED; 175 if (head && (so->so_state & SS_INCOMP)) { 176 if ((so->so_options & SO_ACCEPTFILTER) != 0) { 177 so->so_upcall = head->so_accf->so_accept_filter->accf_callback; 178 so->so_upcallarg = head->so_accf->so_accept_filter_arg; 179 so->so_rcv.ssb_flags |= SSB_UPCALL; 180 so->so_options &= ~SO_ACCEPTFILTER; 181 so->so_upcall(so, so->so_upcallarg, 0); 182 return; 183 } 184 TAILQ_REMOVE(&head->so_incomp, so, so_list); 185 head->so_incqlen--; 186 so->so_state &= ~SS_INCOMP; 187 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 188 head->so_qlen++; 189 so->so_state |= SS_COMP; 190 sorwakeup(head); 191 wakeup_one(&head->so_timeo); 192 } else { 193 wakeup(&so->so_timeo); 194 sorwakeup(so); 195 sowwakeup(so); 196 } 197 } 198 199 void 200 soisdisconnecting(struct socket *so) 201 { 202 so->so_state &= ~SS_ISCONNECTING; 203 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE); 204 wakeup((caddr_t)&so->so_timeo); 205 sowwakeup(so); 206 sorwakeup(so); 207 } 208 209 void 210 soisdisconnected(struct socket *so) 211 { 212 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING); 213 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED); 214 wakeup((caddr_t)&so->so_timeo); 215 sbdrop(&so->so_snd.sb, so->so_snd.ssb_cc); 216 sowwakeup(so); 217 sorwakeup(so); 218 } 219 220 /* 221 * When an attempt at a new connection is noted on a socket 222 * which accepts connections, sonewconn is called. If the 223 * connection is possible (subject to space constraints, etc.) 224 * then we allocate a new structure, propoerly linked into the 225 * data structure of the original socket, and return this. 226 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED. 227 */ 228 struct socket * 229 sonewconn(struct socket *head, int connstatus) 230 { 231 struct socket *so; 232 struct socket *sp; 233 struct pru_attach_info ai; 234 235 if (head->so_qlen > 3 * head->so_qlimit / 2) 236 return ((struct socket *)0); 237 so = soalloc(1); 238 if (so == NULL) 239 return (NULL); 240 if ((head->so_options & SO_ACCEPTFILTER) != 0) 241 connstatus = 0; 242 so->so_head = head; 243 so->so_type = head->so_type; 244 so->so_options = head->so_options &~ SO_ACCEPTCONN; 245 so->so_linger = head->so_linger; 246 so->so_state = head->so_state | SS_NOFDREF; 247 so->so_proto = head->so_proto; 248 so->so_timeo = head->so_timeo; 249 so->so_cred = crhold(head->so_cred); 250 ai.sb_rlimit = NULL; 251 ai.p_ucred = NULL; 252 ai.fd_rdir = NULL; /* jail code cruft XXX JH */ 253 if (soreserve(so, head->so_snd.ssb_hiwat, head->so_rcv.ssb_hiwat, NULL) || 254 /* Directly call function since we're already at protocol level. */ 255 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, &ai)) { 256 sodealloc(so); 257 return ((struct socket *)0); 258 } 259 260 if (connstatus) { 261 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list); 262 so->so_state |= SS_COMP; 263 head->so_qlen++; 264 } else { 265 if (head->so_incqlen > head->so_qlimit) { 266 sp = TAILQ_FIRST(&head->so_incomp); 267 TAILQ_REMOVE(&head->so_incomp, sp, so_list); 268 head->so_incqlen--; 269 sp->so_state &= ~SS_INCOMP; 270 sp->so_head = NULL; 271 if ((sp->so_state & SS_ABORTING) == 0) { 272 sp->so_state |= SS_ABORTING; 273 soaborta(sp); 274 } 275 } 276 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list); 277 so->so_state |= SS_INCOMP; 278 head->so_incqlen++; 279 } 280 if (connstatus) { 281 sorwakeup(head); 282 wakeup((caddr_t)&head->so_timeo); 283 so->so_state |= connstatus; 284 } 285 return (so); 286 } 287 288 /* 289 * Socantsendmore indicates that no more data will be sent on the 290 * socket; it would normally be applied to a socket when the user 291 * informs the system that no more data is to be sent, by the protocol 292 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data 293 * will be received, and will normally be applied to the socket by a 294 * protocol when it detects that the peer will send no more data. 295 * Data queued for reading in the socket may yet be read. 296 */ 297 void 298 socantsendmore(struct socket *so) 299 { 300 so->so_state |= SS_CANTSENDMORE; 301 sowwakeup(so); 302 } 303 304 void 305 socantrcvmore(struct socket *so) 306 { 307 so->so_state |= SS_CANTRCVMORE; 308 sorwakeup(so); 309 } 310 311 /* 312 * Wakeup processes waiting on a socket buffer. Do asynchronous notification 313 * via SIGIO if the socket has the SS_ASYNC flag set. 314 */ 315 void 316 sowakeup(struct socket *so, struct signalsockbuf *ssb) 317 { 318 struct selinfo *selinfo = &ssb->ssb_sel; 319 320 selwakeup(selinfo); 321 ssb->ssb_flags &= ~SSB_SEL; 322 if (ssb->ssb_flags & SSB_WAIT) { 323 ssb->ssb_flags &= ~SSB_WAIT; 324 wakeup((caddr_t)&ssb->ssb_cc); 325 } 326 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL) 327 pgsigio(so->so_sigio, SIGIO, 0); 328 if (ssb->ssb_flags & SSB_UPCALL) 329 (*so->so_upcall)(so, so->so_upcallarg, MB_DONTWAIT); 330 if (ssb->ssb_flags & SSB_AIO) 331 aio_swake(so, ssb); 332 KNOTE(&selinfo->si_note, 0); 333 if (ssb->ssb_flags & SSB_MEVENT) { 334 struct netmsg_so_notify *msg, *nmsg; 335 336 TAILQ_FOREACH_MUTABLE(msg, &selinfo->si_mlist, nm_list, nmsg) { 337 if (msg->nm_predicate(&msg->nm_netmsg)) { 338 TAILQ_REMOVE(&selinfo->si_mlist, msg, nm_list); 339 lwkt_replymsg(&msg->nm_netmsg.nm_lmsg, 340 msg->nm_netmsg.nm_lmsg.ms_error); 341 } 342 } 343 if (TAILQ_EMPTY(&ssb->ssb_sel.si_mlist)) 344 ssb->ssb_flags &= ~SSB_MEVENT; 345 } 346 } 347 348 /* 349 * Socket buffer (struct signalsockbuf) utility routines. 350 * 351 * Each socket contains two socket buffers: one for sending data and 352 * one for receiving data. Each buffer contains a queue of mbufs, 353 * information about the number of mbufs and amount of data in the 354 * queue, and other fields allowing select() statements and notification 355 * on data availability to be implemented. 356 * 357 * Data stored in a socket buffer is maintained as a list of records. 358 * Each record is a list of mbufs chained together with the m_next 359 * field. Records are chained together with the m_nextpkt field. The upper 360 * level routine soreceive() expects the following conventions to be 361 * observed when placing information in the receive buffer: 362 * 363 * 1. If the protocol requires each message be preceded by the sender's 364 * name, then a record containing that name must be present before 365 * any associated data (mbuf's must be of type MT_SONAME). 366 * 2. If the protocol supports the exchange of ``access rights'' (really 367 * just additional data associated with the message), and there are 368 * ``rights'' to be received, then a record containing this data 369 * should be present (mbuf's must be of type MT_RIGHTS). 370 * 3. If a name or rights record exists, then it must be followed by 371 * a data record, perhaps of zero length. 372 * 373 * Before using a new socket structure it is first necessary to reserve 374 * buffer space to the socket, by calling sbreserve(). This should commit 375 * some of the available buffer space in the system buffer pool for the 376 * socket (currently, it does nothing but enforce limits). The space 377 * should be released by calling ssb_release() when the socket is destroyed. 378 */ 379 int 380 soreserve(struct socket *so, u_long sndcc, u_long rcvcc, struct rlimit *rl) 381 { 382 if (ssb_reserve(&so->so_snd, sndcc, so, rl) == 0) 383 goto bad; 384 if (ssb_reserve(&so->so_rcv, rcvcc, so, rl) == 0) 385 goto bad2; 386 if (so->so_rcv.ssb_lowat == 0) 387 so->so_rcv.ssb_lowat = 1; 388 if (so->so_snd.ssb_lowat == 0) 389 so->so_snd.ssb_lowat = MCLBYTES; 390 if (so->so_snd.ssb_lowat > so->so_snd.ssb_hiwat) 391 so->so_snd.ssb_lowat = so->so_snd.ssb_hiwat; 392 return (0); 393 bad2: 394 ssb_release(&so->so_snd, so); 395 bad: 396 return (ENOBUFS); 397 } 398 399 static int 400 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS) 401 { 402 int error = 0; 403 u_long old_sb_max = sb_max; 404 405 error = SYSCTL_OUT(req, arg1, sizeof(int)); 406 if (error || !req->newptr) 407 return (error); 408 error = SYSCTL_IN(req, arg1, sizeof(int)); 409 if (error) 410 return (error); 411 if (sb_max < MSIZE + MCLBYTES) { 412 sb_max = old_sb_max; 413 return (EINVAL); 414 } 415 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES); 416 return (0); 417 } 418 419 /* 420 * Allot mbufs to a signalsockbuf. 421 * Attempt to scale mbmax so that mbcnt doesn't become limiting 422 * if buffering efficiency is near the normal case. 423 */ 424 int 425 ssb_reserve(struct signalsockbuf *ssb, u_long cc, struct socket *so, 426 struct rlimit *rl) 427 { 428 /* 429 * rl will only be NULL when we're in an interrupt (eg, in tcp_input) 430 * or when called from netgraph (ie, ngd_attach) 431 */ 432 if (cc > sb_max_adj) 433 cc = sb_max_adj; 434 if (!chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, cc, 435 rl ? rl->rlim_cur : RLIM_INFINITY)) { 436 return (0); 437 } 438 ssb->ssb_mbmax = min(cc * sb_efficiency, sb_max); 439 if (ssb->ssb_lowat > ssb->ssb_hiwat) 440 ssb->ssb_lowat = ssb->ssb_hiwat; 441 return (1); 442 } 443 444 /* 445 * Free mbufs held by a socket, and reserved mbuf space. 446 */ 447 void 448 ssb_release(struct signalsockbuf *ssb, struct socket *so) 449 { 450 sbflush(&ssb->sb); 451 (void)chgsbsize(so->so_cred->cr_uidinfo, &ssb->ssb_hiwat, 0, 452 RLIM_INFINITY); 453 ssb->ssb_mbmax = 0; 454 } 455 456 /* 457 * Some routines that return EOPNOTSUPP for entry points that are not 458 * supported by a protocol. Fill in as needed. 459 */ 460 int 461 pru_accept_notsupp(struct socket *so, struct sockaddr **nam) 462 { 463 return EOPNOTSUPP; 464 } 465 466 int 467 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td) 468 { 469 return EOPNOTSUPP; 470 } 471 472 int 473 pru_connect2_notsupp(struct socket *so1, struct socket *so2) 474 { 475 return EOPNOTSUPP; 476 } 477 478 int 479 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data, 480 struct ifnet *ifp, struct thread *td) 481 { 482 return EOPNOTSUPP; 483 } 484 485 int 486 pru_listen_notsupp(struct socket *so, struct thread *td) 487 { 488 return EOPNOTSUPP; 489 } 490 491 int 492 pru_rcvd_notsupp(struct socket *so, int flags) 493 { 494 return EOPNOTSUPP; 495 } 496 497 int 498 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags) 499 { 500 return EOPNOTSUPP; 501 } 502 503 /* 504 * This isn't really a ``null'' operation, but it's the default one 505 * and doesn't do anything destructive. 506 */ 507 int 508 pru_sense_null(struct socket *so, struct stat *sb) 509 { 510 sb->st_blksize = so->so_snd.ssb_hiwat; 511 return 0; 512 } 513 514 /* 515 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME. Callers 516 * of this routine assume that it always succeeds, so we have to use a 517 * blockable allocation even though we might be called from a critical thread. 518 */ 519 struct sockaddr * 520 dup_sockaddr(const struct sockaddr *sa) 521 { 522 struct sockaddr *sa2; 523 524 sa2 = kmalloc(sa->sa_len, M_SONAME, M_INTWAIT); 525 bcopy(sa, sa2, sa->sa_len); 526 return (sa2); 527 } 528 529 /* 530 * Create an external-format (``xsocket'') structure using the information 531 * in the kernel-format socket structure pointed to by so. This is done 532 * to reduce the spew of irrelevant information over this interface, 533 * to isolate user code from changes in the kernel structure, and 534 * potentially to provide information-hiding if we decide that 535 * some of this information should be hidden from users. 536 */ 537 void 538 sotoxsocket(struct socket *so, struct xsocket *xso) 539 { 540 xso->xso_len = sizeof *xso; 541 xso->xso_so = so; 542 xso->so_type = so->so_type; 543 xso->so_options = so->so_options; 544 xso->so_linger = so->so_linger; 545 xso->so_state = so->so_state; 546 xso->so_pcb = so->so_pcb; 547 xso->xso_protocol = so->so_proto->pr_protocol; 548 xso->xso_family = so->so_proto->pr_domain->dom_family; 549 xso->so_qlen = so->so_qlen; 550 xso->so_incqlen = so->so_incqlen; 551 xso->so_qlimit = so->so_qlimit; 552 xso->so_timeo = so->so_timeo; 553 xso->so_error = so->so_error; 554 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0; 555 xso->so_oobmark = so->so_oobmark; 556 ssbtoxsockbuf(&so->so_snd, &xso->so_snd); 557 ssbtoxsockbuf(&so->so_rcv, &xso->so_rcv); 558 xso->so_uid = so->so_cred->cr_uid; 559 } 560 561 /* 562 * Here is the definition of some of the basic objects in the kern.ipc 563 * branch of the MIB. 564 */ 565 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC"); 566 567 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */ 568 static int dummy; 569 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, ""); 570 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_INT|CTLFLAG_RW, 571 &sb_max, 0, sysctl_handle_sb_max, "I", "Maximum socket buffer size"); 572 SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RD, 573 &maxsockets, 0, "Maximum number of sockets available"); 574 SYSCTL_INT(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW, 575 &sb_efficiency, 0, ""); 576 577 /* 578 * Initialize maxsockets 579 */ 580 static void 581 init_maxsockets(void *ignored) 582 { 583 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets); 584 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters)); 585 } 586 SYSINIT(param, SI_BOOT1_TUNABLES, SI_ORDER_ANY, 587 init_maxsockets, NULL); 588 589